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/slab.h> 15 #include <linux/init.h> 16 #include <linux/unistd.h> 17 #include <linux/module.h> 18 #include <linux/vmalloc.h> 19 #include <linux/completion.h> 20 #include <linux/personality.h> 21 #include <linux/mempolicy.h> 22 #include <linux/sem.h> 23 #include <linux/file.h> 24 #include <linux/fdtable.h> 25 #include <linux/iocontext.h> 26 #include <linux/key.h> 27 #include <linux/binfmts.h> 28 #include <linux/mman.h> 29 #include <linux/mmu_notifier.h> 30 #include <linux/fs.h> 31 #include <linux/nsproxy.h> 32 #include <linux/capability.h> 33 #include <linux/cpu.h> 34 #include <linux/cgroup.h> 35 #include <linux/security.h> 36 #include <linux/hugetlb.h> 37 #include <linux/swap.h> 38 #include <linux/syscalls.h> 39 #include <linux/jiffies.h> 40 #include <linux/tracehook.h> 41 #include <linux/futex.h> 42 #include <linux/compat.h> 43 #include <linux/task_io_accounting_ops.h> 44 #include <linux/rcupdate.h> 45 #include <linux/ptrace.h> 46 #include <linux/mount.h> 47 #include <linux/audit.h> 48 #include <linux/memcontrol.h> 49 #include <linux/ftrace.h> 50 #include <linux/profile.h> 51 #include <linux/rmap.h> 52 #include <linux/acct.h> 53 #include <linux/tsacct_kern.h> 54 #include <linux/cn_proc.h> 55 #include <linux/freezer.h> 56 #include <linux/delayacct.h> 57 #include <linux/taskstats_kern.h> 58 #include <linux/random.h> 59 #include <linux/tty.h> 60 #include <linux/proc_fs.h> 61 #include <linux/blkdev.h> 62 #include <linux/fs_struct.h> 63 #include <linux/magic.h> 64 #include <linux/perf_counter.h> 65 66 #include <asm/pgtable.h> 67 #include <asm/pgalloc.h> 68 #include <asm/uaccess.h> 69 #include <asm/mmu_context.h> 70 #include <asm/cacheflush.h> 71 #include <asm/tlbflush.h> 72 73 #include <trace/events/sched.h> 74 75 /* 76 * Protected counters by write_lock_irq(&tasklist_lock) 77 */ 78 unsigned long total_forks; /* Handle normal Linux uptimes. */ 79 int nr_threads; /* The idle threads do not count.. */ 80 81 int max_threads; /* tunable limit on nr_threads */ 82 83 DEFINE_PER_CPU(unsigned long, process_counts) = 0; 84 85 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */ 86 87 int nr_processes(void) 88 { 89 int cpu; 90 int total = 0; 91 92 for_each_online_cpu(cpu) 93 total += per_cpu(process_counts, cpu); 94 95 return total; 96 } 97 98 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR 99 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL) 100 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk)) 101 static struct kmem_cache *task_struct_cachep; 102 #endif 103 104 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR 105 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk) 106 { 107 #ifdef CONFIG_DEBUG_STACK_USAGE 108 gfp_t mask = GFP_KERNEL | __GFP_ZERO; 109 #else 110 gfp_t mask = GFP_KERNEL; 111 #endif 112 return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER); 113 } 114 115 static inline void free_thread_info(struct thread_info *ti) 116 { 117 free_pages((unsigned long)ti, THREAD_SIZE_ORDER); 118 } 119 #endif 120 121 /* SLAB cache for signal_struct structures (tsk->signal) */ 122 static struct kmem_cache *signal_cachep; 123 124 /* SLAB cache for sighand_struct structures (tsk->sighand) */ 125 struct kmem_cache *sighand_cachep; 126 127 /* SLAB cache for files_struct structures (tsk->files) */ 128 struct kmem_cache *files_cachep; 129 130 /* SLAB cache for fs_struct structures (tsk->fs) */ 131 struct kmem_cache *fs_cachep; 132 133 /* SLAB cache for vm_area_struct structures */ 134 struct kmem_cache *vm_area_cachep; 135 136 /* SLAB cache for mm_struct structures (tsk->mm) */ 137 static struct kmem_cache *mm_cachep; 138 139 void free_task(struct task_struct *tsk) 140 { 141 prop_local_destroy_single(&tsk->dirties); 142 free_thread_info(tsk->stack); 143 rt_mutex_debug_task_free(tsk); 144 ftrace_graph_exit_task(tsk); 145 free_task_struct(tsk); 146 } 147 EXPORT_SYMBOL(free_task); 148 149 void __put_task_struct(struct task_struct *tsk) 150 { 151 WARN_ON(!tsk->exit_state); 152 WARN_ON(atomic_read(&tsk->usage)); 153 WARN_ON(tsk == current); 154 155 put_cred(tsk->real_cred); 156 put_cred(tsk->cred); 157 delayacct_tsk_free(tsk); 158 159 if (!profile_handoff_task(tsk)) 160 free_task(tsk); 161 } 162 163 /* 164 * macro override instead of weak attribute alias, to workaround 165 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions. 166 */ 167 #ifndef arch_task_cache_init 168 #define arch_task_cache_init() 169 #endif 170 171 void __init fork_init(unsigned long mempages) 172 { 173 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR 174 #ifndef ARCH_MIN_TASKALIGN 175 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES 176 #endif 177 /* create a slab on which task_structs can be allocated */ 178 task_struct_cachep = 179 kmem_cache_create("task_struct", sizeof(struct task_struct), 180 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL); 181 #endif 182 183 /* do the arch specific task caches init */ 184 arch_task_cache_init(); 185 186 /* 187 * The default maximum number of threads is set to a safe 188 * value: the thread structures can take up at most half 189 * of memory. 190 */ 191 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE); 192 193 /* 194 * we need to allow at least 20 threads to boot a system 195 */ 196 if(max_threads < 20) 197 max_threads = 20; 198 199 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 200 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 201 init_task.signal->rlim[RLIMIT_SIGPENDING] = 202 init_task.signal->rlim[RLIMIT_NPROC]; 203 } 204 205 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst, 206 struct task_struct *src) 207 { 208 *dst = *src; 209 return 0; 210 } 211 212 static struct task_struct *dup_task_struct(struct task_struct *orig) 213 { 214 struct task_struct *tsk; 215 struct thread_info *ti; 216 unsigned long *stackend; 217 218 int err; 219 220 prepare_to_copy(orig); 221 222 tsk = alloc_task_struct(); 223 if (!tsk) 224 return NULL; 225 226 ti = alloc_thread_info(tsk); 227 if (!ti) { 228 free_task_struct(tsk); 229 return NULL; 230 } 231 232 err = arch_dup_task_struct(tsk, orig); 233 if (err) 234 goto out; 235 236 tsk->stack = ti; 237 238 err = prop_local_init_single(&tsk->dirties); 239 if (err) 240 goto out; 241 242 setup_thread_stack(tsk, orig); 243 stackend = end_of_stack(tsk); 244 *stackend = STACK_END_MAGIC; /* for overflow detection */ 245 246 #ifdef CONFIG_CC_STACKPROTECTOR 247 tsk->stack_canary = get_random_int(); 248 #endif 249 250 /* One for us, one for whoever does the "release_task()" (usually parent) */ 251 atomic_set(&tsk->usage,2); 252 atomic_set(&tsk->fs_excl, 0); 253 #ifdef CONFIG_BLK_DEV_IO_TRACE 254 tsk->btrace_seq = 0; 255 #endif 256 tsk->splice_pipe = NULL; 257 return tsk; 258 259 out: 260 free_thread_info(ti); 261 free_task_struct(tsk); 262 return NULL; 263 } 264 265 #ifdef CONFIG_MMU 266 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) 267 { 268 struct vm_area_struct *mpnt, *tmp, **pprev; 269 struct rb_node **rb_link, *rb_parent; 270 int retval; 271 unsigned long charge; 272 struct mempolicy *pol; 273 274 down_write(&oldmm->mmap_sem); 275 flush_cache_dup_mm(oldmm); 276 /* 277 * Not linked in yet - no deadlock potential: 278 */ 279 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING); 280 281 mm->locked_vm = 0; 282 mm->mmap = NULL; 283 mm->mmap_cache = NULL; 284 mm->free_area_cache = oldmm->mmap_base; 285 mm->cached_hole_size = ~0UL; 286 mm->map_count = 0; 287 cpumask_clear(mm_cpumask(mm)); 288 mm->mm_rb = RB_ROOT; 289 rb_link = &mm->mm_rb.rb_node; 290 rb_parent = NULL; 291 pprev = &mm->mmap; 292 293 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { 294 struct file *file; 295 296 if (mpnt->vm_flags & VM_DONTCOPY) { 297 long pages = vma_pages(mpnt); 298 mm->total_vm -= pages; 299 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file, 300 -pages); 301 continue; 302 } 303 charge = 0; 304 if (mpnt->vm_flags & VM_ACCOUNT) { 305 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT; 306 if (security_vm_enough_memory(len)) 307 goto fail_nomem; 308 charge = len; 309 } 310 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 311 if (!tmp) 312 goto fail_nomem; 313 *tmp = *mpnt; 314 pol = mpol_dup(vma_policy(mpnt)); 315 retval = PTR_ERR(pol); 316 if (IS_ERR(pol)) 317 goto fail_nomem_policy; 318 vma_set_policy(tmp, pol); 319 tmp->vm_flags &= ~VM_LOCKED; 320 tmp->vm_mm = mm; 321 tmp->vm_next = NULL; 322 anon_vma_link(tmp); 323 file = tmp->vm_file; 324 if (file) { 325 struct inode *inode = file->f_path.dentry->d_inode; 326 struct address_space *mapping = file->f_mapping; 327 328 get_file(file); 329 if (tmp->vm_flags & VM_DENYWRITE) 330 atomic_dec(&inode->i_writecount); 331 spin_lock(&mapping->i_mmap_lock); 332 if (tmp->vm_flags & VM_SHARED) 333 mapping->i_mmap_writable++; 334 tmp->vm_truncate_count = mpnt->vm_truncate_count; 335 flush_dcache_mmap_lock(mapping); 336 /* insert tmp into the share list, just after mpnt */ 337 vma_prio_tree_add(tmp, mpnt); 338 flush_dcache_mmap_unlock(mapping); 339 spin_unlock(&mapping->i_mmap_lock); 340 } 341 342 /* 343 * Clear hugetlb-related page reserves for children. This only 344 * affects MAP_PRIVATE mappings. Faults generated by the child 345 * are not guaranteed to succeed, even if read-only 346 */ 347 if (is_vm_hugetlb_page(tmp)) 348 reset_vma_resv_huge_pages(tmp); 349 350 /* 351 * Link in the new vma and copy the page table entries. 352 */ 353 *pprev = tmp; 354 pprev = &tmp->vm_next; 355 356 __vma_link_rb(mm, tmp, rb_link, rb_parent); 357 rb_link = &tmp->vm_rb.rb_right; 358 rb_parent = &tmp->vm_rb; 359 360 mm->map_count++; 361 retval = copy_page_range(mm, oldmm, mpnt); 362 363 if (tmp->vm_ops && tmp->vm_ops->open) 364 tmp->vm_ops->open(tmp); 365 366 if (retval) 367 goto out; 368 } 369 /* a new mm has just been created */ 370 arch_dup_mmap(oldmm, mm); 371 retval = 0; 372 out: 373 up_write(&mm->mmap_sem); 374 flush_tlb_mm(oldmm); 375 up_write(&oldmm->mmap_sem); 376 return retval; 377 fail_nomem_policy: 378 kmem_cache_free(vm_area_cachep, tmp); 379 fail_nomem: 380 retval = -ENOMEM; 381 vm_unacct_memory(charge); 382 goto out; 383 } 384 385 static inline int mm_alloc_pgd(struct mm_struct * mm) 386 { 387 mm->pgd = pgd_alloc(mm); 388 if (unlikely(!mm->pgd)) 389 return -ENOMEM; 390 return 0; 391 } 392 393 static inline void mm_free_pgd(struct mm_struct * mm) 394 { 395 pgd_free(mm, mm->pgd); 396 } 397 #else 398 #define dup_mmap(mm, oldmm) (0) 399 #define mm_alloc_pgd(mm) (0) 400 #define mm_free_pgd(mm) 401 #endif /* CONFIG_MMU */ 402 403 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); 404 405 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL)) 406 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) 407 408 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT; 409 410 static int __init coredump_filter_setup(char *s) 411 { 412 default_dump_filter = 413 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) & 414 MMF_DUMP_FILTER_MASK; 415 return 1; 416 } 417 418 __setup("coredump_filter=", coredump_filter_setup); 419 420 #include <linux/init_task.h> 421 422 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p) 423 { 424 atomic_set(&mm->mm_users, 1); 425 atomic_set(&mm->mm_count, 1); 426 init_rwsem(&mm->mmap_sem); 427 INIT_LIST_HEAD(&mm->mmlist); 428 mm->flags = (current->mm) ? current->mm->flags : default_dump_filter; 429 mm->core_state = NULL; 430 mm->nr_ptes = 0; 431 set_mm_counter(mm, file_rss, 0); 432 set_mm_counter(mm, anon_rss, 0); 433 spin_lock_init(&mm->page_table_lock); 434 spin_lock_init(&mm->ioctx_lock); 435 INIT_HLIST_HEAD(&mm->ioctx_list); 436 mm->free_area_cache = TASK_UNMAPPED_BASE; 437 mm->cached_hole_size = ~0UL; 438 mm_init_owner(mm, p); 439 440 if (likely(!mm_alloc_pgd(mm))) { 441 mm->def_flags = 0; 442 mmu_notifier_mm_init(mm); 443 return mm; 444 } 445 446 free_mm(mm); 447 return NULL; 448 } 449 450 /* 451 * Allocate and initialize an mm_struct. 452 */ 453 struct mm_struct * mm_alloc(void) 454 { 455 struct mm_struct * mm; 456 457 mm = allocate_mm(); 458 if (mm) { 459 memset(mm, 0, sizeof(*mm)); 460 mm = mm_init(mm, current); 461 } 462 return mm; 463 } 464 465 /* 466 * Called when the last reference to the mm 467 * is dropped: either by a lazy thread or by 468 * mmput. Free the page directory and the mm. 469 */ 470 void __mmdrop(struct mm_struct *mm) 471 { 472 BUG_ON(mm == &init_mm); 473 mm_free_pgd(mm); 474 destroy_context(mm); 475 mmu_notifier_mm_destroy(mm); 476 free_mm(mm); 477 } 478 EXPORT_SYMBOL_GPL(__mmdrop); 479 480 /* 481 * Decrement the use count and release all resources for an mm. 482 */ 483 void mmput(struct mm_struct *mm) 484 { 485 might_sleep(); 486 487 if (atomic_dec_and_test(&mm->mm_users)) { 488 exit_aio(mm); 489 exit_mmap(mm); 490 set_mm_exe_file(mm, NULL); 491 if (!list_empty(&mm->mmlist)) { 492 spin_lock(&mmlist_lock); 493 list_del(&mm->mmlist); 494 spin_unlock(&mmlist_lock); 495 } 496 put_swap_token(mm); 497 mmdrop(mm); 498 } 499 } 500 EXPORT_SYMBOL_GPL(mmput); 501 502 /** 503 * get_task_mm - acquire a reference to the task's mm 504 * 505 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning 506 * this kernel workthread has transiently adopted a user mm with use_mm, 507 * to do its AIO) is not set and if so returns a reference to it, after 508 * bumping up the use count. User must release the mm via mmput() 509 * after use. Typically used by /proc and ptrace. 510 */ 511 struct mm_struct *get_task_mm(struct task_struct *task) 512 { 513 struct mm_struct *mm; 514 515 task_lock(task); 516 mm = task->mm; 517 if (mm) { 518 if (task->flags & PF_KTHREAD) 519 mm = NULL; 520 else 521 atomic_inc(&mm->mm_users); 522 } 523 task_unlock(task); 524 return mm; 525 } 526 EXPORT_SYMBOL_GPL(get_task_mm); 527 528 /* Please note the differences between mmput and mm_release. 529 * mmput is called whenever we stop holding onto a mm_struct, 530 * error success whatever. 531 * 532 * mm_release is called after a mm_struct has been removed 533 * from the current process. 534 * 535 * This difference is important for error handling, when we 536 * only half set up a mm_struct for a new process and need to restore 537 * the old one. Because we mmput the new mm_struct before 538 * restoring the old one. . . 539 * Eric Biederman 10 January 1998 540 */ 541 void mm_release(struct task_struct *tsk, struct mm_struct *mm) 542 { 543 struct completion *vfork_done = tsk->vfork_done; 544 545 /* Get rid of any futexes when releasing the mm */ 546 #ifdef CONFIG_FUTEX 547 if (unlikely(tsk->robust_list)) 548 exit_robust_list(tsk); 549 #ifdef CONFIG_COMPAT 550 if (unlikely(tsk->compat_robust_list)) 551 compat_exit_robust_list(tsk); 552 #endif 553 #endif 554 555 /* Get rid of any cached register state */ 556 deactivate_mm(tsk, mm); 557 558 /* notify parent sleeping on vfork() */ 559 if (vfork_done) { 560 tsk->vfork_done = NULL; 561 complete(vfork_done); 562 } 563 564 /* 565 * If we're exiting normally, clear a user-space tid field if 566 * requested. We leave this alone when dying by signal, to leave 567 * the value intact in a core dump, and to save the unnecessary 568 * trouble otherwise. Userland only wants this done for a sys_exit. 569 */ 570 if (tsk->clear_child_tid 571 && !(tsk->flags & PF_SIGNALED) 572 && atomic_read(&mm->mm_users) > 1) { 573 u32 __user * tidptr = tsk->clear_child_tid; 574 tsk->clear_child_tid = NULL; 575 576 /* 577 * We don't check the error code - if userspace has 578 * not set up a proper pointer then tough luck. 579 */ 580 put_user(0, tidptr); 581 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0); 582 } 583 } 584 585 /* 586 * Allocate a new mm structure and copy contents from the 587 * mm structure of the passed in task structure. 588 */ 589 struct mm_struct *dup_mm(struct task_struct *tsk) 590 { 591 struct mm_struct *mm, *oldmm = current->mm; 592 int err; 593 594 if (!oldmm) 595 return NULL; 596 597 mm = allocate_mm(); 598 if (!mm) 599 goto fail_nomem; 600 601 memcpy(mm, oldmm, sizeof(*mm)); 602 603 /* Initializing for Swap token stuff */ 604 mm->token_priority = 0; 605 mm->last_interval = 0; 606 607 if (!mm_init(mm, tsk)) 608 goto fail_nomem; 609 610 if (init_new_context(tsk, mm)) 611 goto fail_nocontext; 612 613 dup_mm_exe_file(oldmm, mm); 614 615 err = dup_mmap(mm, oldmm); 616 if (err) 617 goto free_pt; 618 619 mm->hiwater_rss = get_mm_rss(mm); 620 mm->hiwater_vm = mm->total_vm; 621 622 return mm; 623 624 free_pt: 625 mmput(mm); 626 627 fail_nomem: 628 return NULL; 629 630 fail_nocontext: 631 /* 632 * If init_new_context() failed, we cannot use mmput() to free the mm 633 * because it calls destroy_context() 634 */ 635 mm_free_pgd(mm); 636 free_mm(mm); 637 return NULL; 638 } 639 640 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk) 641 { 642 struct mm_struct * mm, *oldmm; 643 int retval; 644 645 tsk->min_flt = tsk->maj_flt = 0; 646 tsk->nvcsw = tsk->nivcsw = 0; 647 #ifdef CONFIG_DETECT_HUNG_TASK 648 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; 649 #endif 650 651 tsk->mm = NULL; 652 tsk->active_mm = NULL; 653 654 /* 655 * Are we cloning a kernel thread? 656 * 657 * We need to steal a active VM for that.. 658 */ 659 oldmm = current->mm; 660 if (!oldmm) 661 return 0; 662 663 if (clone_flags & CLONE_VM) { 664 atomic_inc(&oldmm->mm_users); 665 mm = oldmm; 666 goto good_mm; 667 } 668 669 retval = -ENOMEM; 670 mm = dup_mm(tsk); 671 if (!mm) 672 goto fail_nomem; 673 674 good_mm: 675 /* Initializing for Swap token stuff */ 676 mm->token_priority = 0; 677 mm->last_interval = 0; 678 679 tsk->mm = mm; 680 tsk->active_mm = mm; 681 return 0; 682 683 fail_nomem: 684 return retval; 685 } 686 687 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 688 { 689 struct fs_struct *fs = current->fs; 690 if (clone_flags & CLONE_FS) { 691 /* tsk->fs is already what we want */ 692 write_lock(&fs->lock); 693 if (fs->in_exec) { 694 write_unlock(&fs->lock); 695 return -EAGAIN; 696 } 697 fs->users++; 698 write_unlock(&fs->lock); 699 return 0; 700 } 701 tsk->fs = copy_fs_struct(fs); 702 if (!tsk->fs) 703 return -ENOMEM; 704 return 0; 705 } 706 707 static int copy_files(unsigned long clone_flags, struct task_struct * tsk) 708 { 709 struct files_struct *oldf, *newf; 710 int error = 0; 711 712 /* 713 * A background process may not have any files ... 714 */ 715 oldf = current->files; 716 if (!oldf) 717 goto out; 718 719 if (clone_flags & CLONE_FILES) { 720 atomic_inc(&oldf->count); 721 goto out; 722 } 723 724 newf = dup_fd(oldf, &error); 725 if (!newf) 726 goto out; 727 728 tsk->files = newf; 729 error = 0; 730 out: 731 return error; 732 } 733 734 static int copy_io(unsigned long clone_flags, struct task_struct *tsk) 735 { 736 #ifdef CONFIG_BLOCK 737 struct io_context *ioc = current->io_context; 738 739 if (!ioc) 740 return 0; 741 /* 742 * Share io context with parent, if CLONE_IO is set 743 */ 744 if (clone_flags & CLONE_IO) { 745 tsk->io_context = ioc_task_link(ioc); 746 if (unlikely(!tsk->io_context)) 747 return -ENOMEM; 748 } else if (ioprio_valid(ioc->ioprio)) { 749 tsk->io_context = alloc_io_context(GFP_KERNEL, -1); 750 if (unlikely(!tsk->io_context)) 751 return -ENOMEM; 752 753 tsk->io_context->ioprio = ioc->ioprio; 754 } 755 #endif 756 return 0; 757 } 758 759 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk) 760 { 761 struct sighand_struct *sig; 762 763 if (clone_flags & CLONE_SIGHAND) { 764 atomic_inc(¤t->sighand->count); 765 return 0; 766 } 767 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 768 rcu_assign_pointer(tsk->sighand, sig); 769 if (!sig) 770 return -ENOMEM; 771 atomic_set(&sig->count, 1); 772 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 773 return 0; 774 } 775 776 void __cleanup_sighand(struct sighand_struct *sighand) 777 { 778 if (atomic_dec_and_test(&sighand->count)) 779 kmem_cache_free(sighand_cachep, sighand); 780 } 781 782 783 /* 784 * Initialize POSIX timer handling for a thread group. 785 */ 786 static void posix_cpu_timers_init_group(struct signal_struct *sig) 787 { 788 /* Thread group counters. */ 789 thread_group_cputime_init(sig); 790 791 /* Expiration times and increments. */ 792 sig->it_virt_expires = cputime_zero; 793 sig->it_virt_incr = cputime_zero; 794 sig->it_prof_expires = cputime_zero; 795 sig->it_prof_incr = cputime_zero; 796 797 /* Cached expiration times. */ 798 sig->cputime_expires.prof_exp = cputime_zero; 799 sig->cputime_expires.virt_exp = cputime_zero; 800 sig->cputime_expires.sched_exp = 0; 801 802 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) { 803 sig->cputime_expires.prof_exp = 804 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur); 805 sig->cputimer.running = 1; 806 } 807 808 /* The timer lists. */ 809 INIT_LIST_HEAD(&sig->cpu_timers[0]); 810 INIT_LIST_HEAD(&sig->cpu_timers[1]); 811 INIT_LIST_HEAD(&sig->cpu_timers[2]); 812 } 813 814 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk) 815 { 816 struct signal_struct *sig; 817 818 if (clone_flags & CLONE_THREAD) { 819 atomic_inc(¤t->signal->count); 820 atomic_inc(¤t->signal->live); 821 return 0; 822 } 823 824 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL); 825 tsk->signal = sig; 826 if (!sig) 827 return -ENOMEM; 828 829 atomic_set(&sig->count, 1); 830 atomic_set(&sig->live, 1); 831 init_waitqueue_head(&sig->wait_chldexit); 832 sig->flags = 0; 833 if (clone_flags & CLONE_NEWPID) 834 sig->flags |= SIGNAL_UNKILLABLE; 835 sig->group_exit_code = 0; 836 sig->group_exit_task = NULL; 837 sig->group_stop_count = 0; 838 sig->curr_target = tsk; 839 init_sigpending(&sig->shared_pending); 840 INIT_LIST_HEAD(&sig->posix_timers); 841 842 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 843 sig->it_real_incr.tv64 = 0; 844 sig->real_timer.function = it_real_fn; 845 846 sig->leader = 0; /* session leadership doesn't inherit */ 847 sig->tty_old_pgrp = NULL; 848 sig->tty = NULL; 849 850 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero; 851 sig->gtime = cputime_zero; 852 sig->cgtime = cputime_zero; 853 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0; 854 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0; 855 sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0; 856 task_io_accounting_init(&sig->ioac); 857 sig->sum_sched_runtime = 0; 858 taskstats_tgid_init(sig); 859 860 task_lock(current->group_leader); 861 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 862 task_unlock(current->group_leader); 863 864 posix_cpu_timers_init_group(sig); 865 866 acct_init_pacct(&sig->pacct); 867 868 tty_audit_fork(sig); 869 870 return 0; 871 } 872 873 void __cleanup_signal(struct signal_struct *sig) 874 { 875 thread_group_cputime_free(sig); 876 tty_kref_put(sig->tty); 877 kmem_cache_free(signal_cachep, sig); 878 } 879 880 static void cleanup_signal(struct task_struct *tsk) 881 { 882 struct signal_struct *sig = tsk->signal; 883 884 atomic_dec(&sig->live); 885 886 if (atomic_dec_and_test(&sig->count)) 887 __cleanup_signal(sig); 888 } 889 890 static void copy_flags(unsigned long clone_flags, struct task_struct *p) 891 { 892 unsigned long new_flags = p->flags; 893 894 new_flags &= ~PF_SUPERPRIV; 895 new_flags |= PF_FORKNOEXEC; 896 new_flags |= PF_STARTING; 897 p->flags = new_flags; 898 clear_freeze_flag(p); 899 } 900 901 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr) 902 { 903 current->clear_child_tid = tidptr; 904 905 return task_pid_vnr(current); 906 } 907 908 static void rt_mutex_init_task(struct task_struct *p) 909 { 910 spin_lock_init(&p->pi_lock); 911 #ifdef CONFIG_RT_MUTEXES 912 plist_head_init(&p->pi_waiters, &p->pi_lock); 913 p->pi_blocked_on = NULL; 914 #endif 915 } 916 917 #ifdef CONFIG_MM_OWNER 918 void mm_init_owner(struct mm_struct *mm, struct task_struct *p) 919 { 920 mm->owner = p; 921 } 922 #endif /* CONFIG_MM_OWNER */ 923 924 /* 925 * Initialize POSIX timer handling for a single task. 926 */ 927 static void posix_cpu_timers_init(struct task_struct *tsk) 928 { 929 tsk->cputime_expires.prof_exp = cputime_zero; 930 tsk->cputime_expires.virt_exp = cputime_zero; 931 tsk->cputime_expires.sched_exp = 0; 932 INIT_LIST_HEAD(&tsk->cpu_timers[0]); 933 INIT_LIST_HEAD(&tsk->cpu_timers[1]); 934 INIT_LIST_HEAD(&tsk->cpu_timers[2]); 935 } 936 937 /* 938 * This creates a new process as a copy of the old one, 939 * but does not actually start it yet. 940 * 941 * It copies the registers, and all the appropriate 942 * parts of the process environment (as per the clone 943 * flags). The actual kick-off is left to the caller. 944 */ 945 static struct task_struct *copy_process(unsigned long clone_flags, 946 unsigned long stack_start, 947 struct pt_regs *regs, 948 unsigned long stack_size, 949 int __user *child_tidptr, 950 struct pid *pid, 951 int trace) 952 { 953 int retval; 954 struct task_struct *p; 955 int cgroup_callbacks_done = 0; 956 957 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 958 return ERR_PTR(-EINVAL); 959 960 /* 961 * Thread groups must share signals as well, and detached threads 962 * can only be started up within the thread group. 963 */ 964 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 965 return ERR_PTR(-EINVAL); 966 967 /* 968 * Shared signal handlers imply shared VM. By way of the above, 969 * thread groups also imply shared VM. Blocking this case allows 970 * for various simplifications in other code. 971 */ 972 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 973 return ERR_PTR(-EINVAL); 974 975 retval = security_task_create(clone_flags); 976 if (retval) 977 goto fork_out; 978 979 retval = -ENOMEM; 980 p = dup_task_struct(current); 981 if (!p) 982 goto fork_out; 983 984 ftrace_graph_init_task(p); 985 986 rt_mutex_init_task(p); 987 988 #ifdef CONFIG_PROVE_LOCKING 989 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); 990 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 991 #endif 992 retval = -EAGAIN; 993 if (atomic_read(&p->real_cred->user->processes) >= 994 p->signal->rlim[RLIMIT_NPROC].rlim_cur) { 995 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) && 996 p->real_cred->user != INIT_USER) 997 goto bad_fork_free; 998 } 999 1000 retval = copy_creds(p, clone_flags); 1001 if (retval < 0) 1002 goto bad_fork_free; 1003 1004 /* 1005 * If multiple threads are within copy_process(), then this check 1006 * triggers too late. This doesn't hurt, the check is only there 1007 * to stop root fork bombs. 1008 */ 1009 retval = -EAGAIN; 1010 if (nr_threads >= max_threads) 1011 goto bad_fork_cleanup_count; 1012 1013 if (!try_module_get(task_thread_info(p)->exec_domain->module)) 1014 goto bad_fork_cleanup_count; 1015 1016 if (p->binfmt && !try_module_get(p->binfmt->module)) 1017 goto bad_fork_cleanup_put_domain; 1018 1019 p->did_exec = 0; 1020 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */ 1021 copy_flags(clone_flags, p); 1022 INIT_LIST_HEAD(&p->children); 1023 INIT_LIST_HEAD(&p->sibling); 1024 #ifdef CONFIG_PREEMPT_RCU 1025 p->rcu_read_lock_nesting = 0; 1026 p->rcu_flipctr_idx = 0; 1027 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 1028 p->vfork_done = NULL; 1029 spin_lock_init(&p->alloc_lock); 1030 1031 init_sigpending(&p->pending); 1032 1033 p->utime = cputime_zero; 1034 p->stime = cputime_zero; 1035 p->gtime = cputime_zero; 1036 p->utimescaled = cputime_zero; 1037 p->stimescaled = cputime_zero; 1038 p->prev_utime = cputime_zero; 1039 p->prev_stime = cputime_zero; 1040 1041 p->default_timer_slack_ns = current->timer_slack_ns; 1042 1043 task_io_accounting_init(&p->ioac); 1044 acct_clear_integrals(p); 1045 1046 posix_cpu_timers_init(p); 1047 1048 p->lock_depth = -1; /* -1 = no lock */ 1049 do_posix_clock_monotonic_gettime(&p->start_time); 1050 p->real_start_time = p->start_time; 1051 monotonic_to_bootbased(&p->real_start_time); 1052 p->io_context = NULL; 1053 p->audit_context = NULL; 1054 cgroup_fork(p); 1055 #ifdef CONFIG_NUMA 1056 p->mempolicy = mpol_dup(p->mempolicy); 1057 if (IS_ERR(p->mempolicy)) { 1058 retval = PTR_ERR(p->mempolicy); 1059 p->mempolicy = NULL; 1060 goto bad_fork_cleanup_cgroup; 1061 } 1062 mpol_fix_fork_child_flag(p); 1063 #endif 1064 #ifdef CONFIG_TRACE_IRQFLAGS 1065 p->irq_events = 0; 1066 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW 1067 p->hardirqs_enabled = 1; 1068 #else 1069 p->hardirqs_enabled = 0; 1070 #endif 1071 p->hardirq_enable_ip = 0; 1072 p->hardirq_enable_event = 0; 1073 p->hardirq_disable_ip = _THIS_IP_; 1074 p->hardirq_disable_event = 0; 1075 p->softirqs_enabled = 1; 1076 p->softirq_enable_ip = _THIS_IP_; 1077 p->softirq_enable_event = 0; 1078 p->softirq_disable_ip = 0; 1079 p->softirq_disable_event = 0; 1080 p->hardirq_context = 0; 1081 p->softirq_context = 0; 1082 #endif 1083 #ifdef CONFIG_LOCKDEP 1084 p->lockdep_depth = 0; /* no locks held yet */ 1085 p->curr_chain_key = 0; 1086 p->lockdep_recursion = 0; 1087 #endif 1088 1089 #ifdef CONFIG_DEBUG_MUTEXES 1090 p->blocked_on = NULL; /* not blocked yet */ 1091 #endif 1092 1093 p->bts = NULL; 1094 1095 /* Perform scheduler related setup. Assign this task to a CPU. */ 1096 sched_fork(p, clone_flags); 1097 1098 retval = perf_counter_init_task(p); 1099 if (retval) 1100 goto bad_fork_cleanup_policy; 1101 1102 if ((retval = audit_alloc(p))) 1103 goto bad_fork_cleanup_policy; 1104 /* copy all the process information */ 1105 if ((retval = copy_semundo(clone_flags, p))) 1106 goto bad_fork_cleanup_audit; 1107 if ((retval = copy_files(clone_flags, p))) 1108 goto bad_fork_cleanup_semundo; 1109 if ((retval = copy_fs(clone_flags, p))) 1110 goto bad_fork_cleanup_files; 1111 if ((retval = copy_sighand(clone_flags, p))) 1112 goto bad_fork_cleanup_fs; 1113 if ((retval = copy_signal(clone_flags, p))) 1114 goto bad_fork_cleanup_sighand; 1115 if ((retval = copy_mm(clone_flags, p))) 1116 goto bad_fork_cleanup_signal; 1117 if ((retval = copy_namespaces(clone_flags, p))) 1118 goto bad_fork_cleanup_mm; 1119 if ((retval = copy_io(clone_flags, p))) 1120 goto bad_fork_cleanup_namespaces; 1121 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs); 1122 if (retval) 1123 goto bad_fork_cleanup_io; 1124 1125 if (pid != &init_struct_pid) { 1126 retval = -ENOMEM; 1127 pid = alloc_pid(p->nsproxy->pid_ns); 1128 if (!pid) 1129 goto bad_fork_cleanup_io; 1130 1131 if (clone_flags & CLONE_NEWPID) { 1132 retval = pid_ns_prepare_proc(p->nsproxy->pid_ns); 1133 if (retval < 0) 1134 goto bad_fork_free_pid; 1135 } 1136 } 1137 1138 p->pid = pid_nr(pid); 1139 p->tgid = p->pid; 1140 if (clone_flags & CLONE_THREAD) 1141 p->tgid = current->tgid; 1142 1143 if (current->nsproxy != p->nsproxy) { 1144 retval = ns_cgroup_clone(p, pid); 1145 if (retval) 1146 goto bad_fork_free_pid; 1147 } 1148 1149 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; 1150 /* 1151 * Clear TID on mm_release()? 1152 */ 1153 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL; 1154 #ifdef CONFIG_FUTEX 1155 p->robust_list = NULL; 1156 #ifdef CONFIG_COMPAT 1157 p->compat_robust_list = NULL; 1158 #endif 1159 INIT_LIST_HEAD(&p->pi_state_list); 1160 p->pi_state_cache = NULL; 1161 #endif 1162 /* 1163 * sigaltstack should be cleared when sharing the same VM 1164 */ 1165 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM) 1166 p->sas_ss_sp = p->sas_ss_size = 0; 1167 1168 /* 1169 * Syscall tracing should be turned off in the child regardless 1170 * of CLONE_PTRACE. 1171 */ 1172 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); 1173 #ifdef TIF_SYSCALL_EMU 1174 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); 1175 #endif 1176 clear_all_latency_tracing(p); 1177 1178 /* ok, now we should be set up.. */ 1179 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL); 1180 p->pdeath_signal = 0; 1181 p->exit_state = 0; 1182 1183 /* 1184 * Ok, make it visible to the rest of the system. 1185 * We dont wake it up yet. 1186 */ 1187 p->group_leader = p; 1188 INIT_LIST_HEAD(&p->thread_group); 1189 1190 /* Now that the task is set up, run cgroup callbacks if 1191 * necessary. We need to run them before the task is visible 1192 * on the tasklist. */ 1193 cgroup_fork_callbacks(p); 1194 cgroup_callbacks_done = 1; 1195 1196 /* Need tasklist lock for parent etc handling! */ 1197 write_lock_irq(&tasklist_lock); 1198 1199 /* 1200 * The task hasn't been attached yet, so its cpus_allowed mask will 1201 * not be changed, nor will its assigned CPU. 1202 * 1203 * The cpus_allowed mask of the parent may have changed after it was 1204 * copied first time - so re-copy it here, then check the child's CPU 1205 * to ensure it is on a valid CPU (and if not, just force it back to 1206 * parent's CPU). This avoids alot of nasty races. 1207 */ 1208 p->cpus_allowed = current->cpus_allowed; 1209 p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed; 1210 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) || 1211 !cpu_online(task_cpu(p)))) 1212 set_task_cpu(p, smp_processor_id()); 1213 1214 /* CLONE_PARENT re-uses the old parent */ 1215 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { 1216 p->real_parent = current->real_parent; 1217 p->parent_exec_id = current->parent_exec_id; 1218 } else { 1219 p->real_parent = current; 1220 p->parent_exec_id = current->self_exec_id; 1221 } 1222 1223 spin_lock(¤t->sighand->siglock); 1224 1225 /* 1226 * Process group and session signals need to be delivered to just the 1227 * parent before the fork or both the parent and the child after the 1228 * fork. Restart if a signal comes in before we add the new process to 1229 * it's process group. 1230 * A fatal signal pending means that current will exit, so the new 1231 * thread can't slip out of an OOM kill (or normal SIGKILL). 1232 */ 1233 recalc_sigpending(); 1234 if (signal_pending(current)) { 1235 spin_unlock(¤t->sighand->siglock); 1236 write_unlock_irq(&tasklist_lock); 1237 retval = -ERESTARTNOINTR; 1238 goto bad_fork_free_pid; 1239 } 1240 1241 if (clone_flags & CLONE_THREAD) { 1242 p->group_leader = current->group_leader; 1243 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group); 1244 } 1245 1246 if (likely(p->pid)) { 1247 list_add_tail(&p->sibling, &p->real_parent->children); 1248 tracehook_finish_clone(p, clone_flags, trace); 1249 1250 if (thread_group_leader(p)) { 1251 if (clone_flags & CLONE_NEWPID) 1252 p->nsproxy->pid_ns->child_reaper = p; 1253 1254 p->signal->leader_pid = pid; 1255 tty_kref_put(p->signal->tty); 1256 p->signal->tty = tty_kref_get(current->signal->tty); 1257 attach_pid(p, PIDTYPE_PGID, task_pgrp(current)); 1258 attach_pid(p, PIDTYPE_SID, task_session(current)); 1259 list_add_tail_rcu(&p->tasks, &init_task.tasks); 1260 __get_cpu_var(process_counts)++; 1261 } 1262 attach_pid(p, PIDTYPE_PID, pid); 1263 nr_threads++; 1264 } 1265 1266 total_forks++; 1267 spin_unlock(¤t->sighand->siglock); 1268 write_unlock_irq(&tasklist_lock); 1269 proc_fork_connector(p); 1270 cgroup_post_fork(p); 1271 return p; 1272 1273 bad_fork_free_pid: 1274 if (pid != &init_struct_pid) 1275 free_pid(pid); 1276 bad_fork_cleanup_io: 1277 put_io_context(p->io_context); 1278 bad_fork_cleanup_namespaces: 1279 exit_task_namespaces(p); 1280 bad_fork_cleanup_mm: 1281 if (p->mm) 1282 mmput(p->mm); 1283 bad_fork_cleanup_signal: 1284 cleanup_signal(p); 1285 bad_fork_cleanup_sighand: 1286 __cleanup_sighand(p->sighand); 1287 bad_fork_cleanup_fs: 1288 exit_fs(p); /* blocking */ 1289 bad_fork_cleanup_files: 1290 exit_files(p); /* blocking */ 1291 bad_fork_cleanup_semundo: 1292 exit_sem(p); 1293 bad_fork_cleanup_audit: 1294 audit_free(p); 1295 bad_fork_cleanup_policy: 1296 perf_counter_free_task(p); 1297 #ifdef CONFIG_NUMA 1298 mpol_put(p->mempolicy); 1299 bad_fork_cleanup_cgroup: 1300 #endif 1301 cgroup_exit(p, cgroup_callbacks_done); 1302 delayacct_tsk_free(p); 1303 if (p->binfmt) 1304 module_put(p->binfmt->module); 1305 bad_fork_cleanup_put_domain: 1306 module_put(task_thread_info(p)->exec_domain->module); 1307 bad_fork_cleanup_count: 1308 atomic_dec(&p->cred->user->processes); 1309 put_cred(p->real_cred); 1310 put_cred(p->cred); 1311 bad_fork_free: 1312 free_task(p); 1313 fork_out: 1314 return ERR_PTR(retval); 1315 } 1316 1317 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs) 1318 { 1319 memset(regs, 0, sizeof(struct pt_regs)); 1320 return regs; 1321 } 1322 1323 struct task_struct * __cpuinit fork_idle(int cpu) 1324 { 1325 struct task_struct *task; 1326 struct pt_regs regs; 1327 1328 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, 1329 &init_struct_pid, 0); 1330 if (!IS_ERR(task)) 1331 init_idle(task, cpu); 1332 1333 return task; 1334 } 1335 1336 /* 1337 * Ok, this is the main fork-routine. 1338 * 1339 * It copies the process, and if successful kick-starts 1340 * it and waits for it to finish using the VM if required. 1341 */ 1342 long do_fork(unsigned long clone_flags, 1343 unsigned long stack_start, 1344 struct pt_regs *regs, 1345 unsigned long stack_size, 1346 int __user *parent_tidptr, 1347 int __user *child_tidptr) 1348 { 1349 struct task_struct *p; 1350 int trace = 0; 1351 long nr; 1352 1353 /* 1354 * Do some preliminary argument and permissions checking before we 1355 * actually start allocating stuff 1356 */ 1357 if (clone_flags & CLONE_NEWUSER) { 1358 if (clone_flags & CLONE_THREAD) 1359 return -EINVAL; 1360 /* hopefully this check will go away when userns support is 1361 * complete 1362 */ 1363 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) || 1364 !capable(CAP_SETGID)) 1365 return -EPERM; 1366 } 1367 1368 /* 1369 * We hope to recycle these flags after 2.6.26 1370 */ 1371 if (unlikely(clone_flags & CLONE_STOPPED)) { 1372 static int __read_mostly count = 100; 1373 1374 if (count > 0 && printk_ratelimit()) { 1375 char comm[TASK_COMM_LEN]; 1376 1377 count--; 1378 printk(KERN_INFO "fork(): process `%s' used deprecated " 1379 "clone flags 0x%lx\n", 1380 get_task_comm(comm, current), 1381 clone_flags & CLONE_STOPPED); 1382 } 1383 } 1384 1385 /* 1386 * When called from kernel_thread, don't do user tracing stuff. 1387 */ 1388 if (likely(user_mode(regs))) 1389 trace = tracehook_prepare_clone(clone_flags); 1390 1391 p = copy_process(clone_flags, stack_start, regs, stack_size, 1392 child_tidptr, NULL, trace); 1393 /* 1394 * Do this prior waking up the new thread - the thread pointer 1395 * might get invalid after that point, if the thread exits quickly. 1396 */ 1397 if (!IS_ERR(p)) { 1398 struct completion vfork; 1399 1400 trace_sched_process_fork(current, p); 1401 1402 nr = task_pid_vnr(p); 1403 1404 if (clone_flags & CLONE_PARENT_SETTID) 1405 put_user(nr, parent_tidptr); 1406 1407 if (clone_flags & CLONE_VFORK) { 1408 p->vfork_done = &vfork; 1409 init_completion(&vfork); 1410 } 1411 1412 if (!(clone_flags & CLONE_THREAD)) 1413 perf_counter_fork(p); 1414 1415 audit_finish_fork(p); 1416 tracehook_report_clone(regs, clone_flags, nr, p); 1417 1418 /* 1419 * We set PF_STARTING at creation in case tracing wants to 1420 * use this to distinguish a fully live task from one that 1421 * hasn't gotten to tracehook_report_clone() yet. Now we 1422 * clear it and set the child going. 1423 */ 1424 p->flags &= ~PF_STARTING; 1425 1426 if (unlikely(clone_flags & CLONE_STOPPED)) { 1427 /* 1428 * We'll start up with an immediate SIGSTOP. 1429 */ 1430 sigaddset(&p->pending.signal, SIGSTOP); 1431 set_tsk_thread_flag(p, TIF_SIGPENDING); 1432 __set_task_state(p, TASK_STOPPED); 1433 } else { 1434 wake_up_new_task(p, clone_flags); 1435 } 1436 1437 tracehook_report_clone_complete(trace, regs, 1438 clone_flags, nr, p); 1439 1440 if (clone_flags & CLONE_VFORK) { 1441 freezer_do_not_count(); 1442 wait_for_completion(&vfork); 1443 freezer_count(); 1444 tracehook_report_vfork_done(p, nr); 1445 } 1446 } else { 1447 nr = PTR_ERR(p); 1448 } 1449 return nr; 1450 } 1451 1452 #ifndef ARCH_MIN_MMSTRUCT_ALIGN 1453 #define ARCH_MIN_MMSTRUCT_ALIGN 0 1454 #endif 1455 1456 static void sighand_ctor(void *data) 1457 { 1458 struct sighand_struct *sighand = data; 1459 1460 spin_lock_init(&sighand->siglock); 1461 init_waitqueue_head(&sighand->signalfd_wqh); 1462 } 1463 1464 void __init proc_caches_init(void) 1465 { 1466 sighand_cachep = kmem_cache_create("sighand_cache", 1467 sizeof(struct sighand_struct), 0, 1468 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU| 1469 SLAB_NOTRACK, sighand_ctor); 1470 signal_cachep = kmem_cache_create("signal_cache", 1471 sizeof(struct signal_struct), 0, 1472 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1473 files_cachep = kmem_cache_create("files_cache", 1474 sizeof(struct files_struct), 0, 1475 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1476 fs_cachep = kmem_cache_create("fs_cache", 1477 sizeof(struct fs_struct), 0, 1478 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1479 mm_cachep = kmem_cache_create("mm_struct", 1480 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN, 1481 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1482 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC); 1483 mmap_init(); 1484 } 1485 1486 /* 1487 * Check constraints on flags passed to the unshare system call and 1488 * force unsharing of additional process context as appropriate. 1489 */ 1490 static void check_unshare_flags(unsigned long *flags_ptr) 1491 { 1492 /* 1493 * If unsharing a thread from a thread group, must also 1494 * unshare vm. 1495 */ 1496 if (*flags_ptr & CLONE_THREAD) 1497 *flags_ptr |= CLONE_VM; 1498 1499 /* 1500 * If unsharing vm, must also unshare signal handlers. 1501 */ 1502 if (*flags_ptr & CLONE_VM) 1503 *flags_ptr |= CLONE_SIGHAND; 1504 1505 /* 1506 * If unsharing signal handlers and the task was created 1507 * using CLONE_THREAD, then must unshare the thread 1508 */ 1509 if ((*flags_ptr & CLONE_SIGHAND) && 1510 (atomic_read(¤t->signal->count) > 1)) 1511 *flags_ptr |= CLONE_THREAD; 1512 1513 /* 1514 * If unsharing namespace, must also unshare filesystem information. 1515 */ 1516 if (*flags_ptr & CLONE_NEWNS) 1517 *flags_ptr |= CLONE_FS; 1518 } 1519 1520 /* 1521 * Unsharing of tasks created with CLONE_THREAD is not supported yet 1522 */ 1523 static int unshare_thread(unsigned long unshare_flags) 1524 { 1525 if (unshare_flags & CLONE_THREAD) 1526 return -EINVAL; 1527 1528 return 0; 1529 } 1530 1531 /* 1532 * Unshare the filesystem structure if it is being shared 1533 */ 1534 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) 1535 { 1536 struct fs_struct *fs = current->fs; 1537 1538 if (!(unshare_flags & CLONE_FS) || !fs) 1539 return 0; 1540 1541 /* don't need lock here; in the worst case we'll do useless copy */ 1542 if (fs->users == 1) 1543 return 0; 1544 1545 *new_fsp = copy_fs_struct(fs); 1546 if (!*new_fsp) 1547 return -ENOMEM; 1548 1549 return 0; 1550 } 1551 1552 /* 1553 * Unsharing of sighand is not supported yet 1554 */ 1555 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp) 1556 { 1557 struct sighand_struct *sigh = current->sighand; 1558 1559 if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1) 1560 return -EINVAL; 1561 else 1562 return 0; 1563 } 1564 1565 /* 1566 * Unshare vm if it is being shared 1567 */ 1568 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp) 1569 { 1570 struct mm_struct *mm = current->mm; 1571 1572 if ((unshare_flags & CLONE_VM) && 1573 (mm && atomic_read(&mm->mm_users) > 1)) { 1574 return -EINVAL; 1575 } 1576 1577 return 0; 1578 } 1579 1580 /* 1581 * Unshare file descriptor table if it is being shared 1582 */ 1583 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp) 1584 { 1585 struct files_struct *fd = current->files; 1586 int error = 0; 1587 1588 if ((unshare_flags & CLONE_FILES) && 1589 (fd && atomic_read(&fd->count) > 1)) { 1590 *new_fdp = dup_fd(fd, &error); 1591 if (!*new_fdp) 1592 return error; 1593 } 1594 1595 return 0; 1596 } 1597 1598 /* 1599 * unshare allows a process to 'unshare' part of the process 1600 * context which was originally shared using clone. copy_* 1601 * functions used by do_fork() cannot be used here directly 1602 * because they modify an inactive task_struct that is being 1603 * constructed. Here we are modifying the current, active, 1604 * task_struct. 1605 */ 1606 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags) 1607 { 1608 int err = 0; 1609 struct fs_struct *fs, *new_fs = NULL; 1610 struct sighand_struct *new_sigh = NULL; 1611 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL; 1612 struct files_struct *fd, *new_fd = NULL; 1613 struct nsproxy *new_nsproxy = NULL; 1614 int do_sysvsem = 0; 1615 1616 check_unshare_flags(&unshare_flags); 1617 1618 /* Return -EINVAL for all unsupported flags */ 1619 err = -EINVAL; 1620 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND| 1621 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM| 1622 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET)) 1623 goto bad_unshare_out; 1624 1625 /* 1626 * CLONE_NEWIPC must also detach from the undolist: after switching 1627 * to a new ipc namespace, the semaphore arrays from the old 1628 * namespace are unreachable. 1629 */ 1630 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) 1631 do_sysvsem = 1; 1632 if ((err = unshare_thread(unshare_flags))) 1633 goto bad_unshare_out; 1634 if ((err = unshare_fs(unshare_flags, &new_fs))) 1635 goto bad_unshare_cleanup_thread; 1636 if ((err = unshare_sighand(unshare_flags, &new_sigh))) 1637 goto bad_unshare_cleanup_fs; 1638 if ((err = unshare_vm(unshare_flags, &new_mm))) 1639 goto bad_unshare_cleanup_sigh; 1640 if ((err = unshare_fd(unshare_flags, &new_fd))) 1641 goto bad_unshare_cleanup_vm; 1642 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, 1643 new_fs))) 1644 goto bad_unshare_cleanup_fd; 1645 1646 if (new_fs || new_mm || new_fd || do_sysvsem || new_nsproxy) { 1647 if (do_sysvsem) { 1648 /* 1649 * CLONE_SYSVSEM is equivalent to sys_exit(). 1650 */ 1651 exit_sem(current); 1652 } 1653 1654 if (new_nsproxy) { 1655 switch_task_namespaces(current, new_nsproxy); 1656 new_nsproxy = NULL; 1657 } 1658 1659 task_lock(current); 1660 1661 if (new_fs) { 1662 fs = current->fs; 1663 write_lock(&fs->lock); 1664 current->fs = new_fs; 1665 if (--fs->users) 1666 new_fs = NULL; 1667 else 1668 new_fs = fs; 1669 write_unlock(&fs->lock); 1670 } 1671 1672 if (new_mm) { 1673 mm = current->mm; 1674 active_mm = current->active_mm; 1675 current->mm = new_mm; 1676 current->active_mm = new_mm; 1677 activate_mm(active_mm, new_mm); 1678 new_mm = mm; 1679 } 1680 1681 if (new_fd) { 1682 fd = current->files; 1683 current->files = new_fd; 1684 new_fd = fd; 1685 } 1686 1687 task_unlock(current); 1688 } 1689 1690 if (new_nsproxy) 1691 put_nsproxy(new_nsproxy); 1692 1693 bad_unshare_cleanup_fd: 1694 if (new_fd) 1695 put_files_struct(new_fd); 1696 1697 bad_unshare_cleanup_vm: 1698 if (new_mm) 1699 mmput(new_mm); 1700 1701 bad_unshare_cleanup_sigh: 1702 if (new_sigh) 1703 if (atomic_dec_and_test(&new_sigh->count)) 1704 kmem_cache_free(sighand_cachep, new_sigh); 1705 1706 bad_unshare_cleanup_fs: 1707 if (new_fs) 1708 free_fs_struct(new_fs); 1709 1710 bad_unshare_cleanup_thread: 1711 bad_unshare_out: 1712 return err; 1713 } 1714 1715 /* 1716 * Helper to unshare the files of the current task. 1717 * We don't want to expose copy_files internals to 1718 * the exec layer of the kernel. 1719 */ 1720 1721 int unshare_files(struct files_struct **displaced) 1722 { 1723 struct task_struct *task = current; 1724 struct files_struct *copy = NULL; 1725 int error; 1726 1727 error = unshare_fd(CLONE_FILES, ©); 1728 if (error || !copy) { 1729 *displaced = NULL; 1730 return error; 1731 } 1732 *displaced = task->files; 1733 task_lock(task); 1734 task->files = copy; 1735 task_unlock(task); 1736 return 0; 1737 } 1738