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/seccomp.h> 38 #include <linux/swap.h> 39 #include <linux/syscalls.h> 40 #include <linux/jiffies.h> 41 #include <linux/futex.h> 42 #include <linux/compat.h> 43 #include <linux/kthread.h> 44 #include <linux/task_io_accounting_ops.h> 45 #include <linux/rcupdate.h> 46 #include <linux/ptrace.h> 47 #include <linux/mount.h> 48 #include <linux/audit.h> 49 #include <linux/memcontrol.h> 50 #include <linux/ftrace.h> 51 #include <linux/proc_fs.h> 52 #include <linux/profile.h> 53 #include <linux/rmap.h> 54 #include <linux/ksm.h> 55 #include <linux/acct.h> 56 #include <linux/tsacct_kern.h> 57 #include <linux/cn_proc.h> 58 #include <linux/freezer.h> 59 #include <linux/delayacct.h> 60 #include <linux/taskstats_kern.h> 61 #include <linux/random.h> 62 #include <linux/tty.h> 63 #include <linux/blkdev.h> 64 #include <linux/fs_struct.h> 65 #include <linux/magic.h> 66 #include <linux/perf_event.h> 67 #include <linux/posix-timers.h> 68 #include <linux/user-return-notifier.h> 69 #include <linux/oom.h> 70 #include <linux/khugepaged.h> 71 #include <linux/signalfd.h> 72 #include <linux/uprobes.h> 73 #include <linux/aio.h> 74 75 #include <asm/pgtable.h> 76 #include <asm/pgalloc.h> 77 #include <asm/uaccess.h> 78 #include <asm/mmu_context.h> 79 #include <asm/cacheflush.h> 80 #include <asm/tlbflush.h> 81 82 #include <trace/events/sched.h> 83 84 #define CREATE_TRACE_POINTS 85 #include <trace/events/task.h> 86 87 /* 88 * Protected counters by write_lock_irq(&tasklist_lock) 89 */ 90 unsigned long total_forks; /* Handle normal Linux uptimes. */ 91 int nr_threads; /* The idle threads do not count.. */ 92 93 int max_threads; /* tunable limit on nr_threads */ 94 95 DEFINE_PER_CPU(unsigned long, process_counts) = 0; 96 97 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */ 98 99 #ifdef CONFIG_PROVE_RCU 100 int lockdep_tasklist_lock_is_held(void) 101 { 102 return lockdep_is_held(&tasklist_lock); 103 } 104 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held); 105 #endif /* #ifdef CONFIG_PROVE_RCU */ 106 107 int nr_processes(void) 108 { 109 int cpu; 110 int total = 0; 111 112 for_each_possible_cpu(cpu) 113 total += per_cpu(process_counts, cpu); 114 115 return total; 116 } 117 118 void __weak arch_release_task_struct(struct task_struct *tsk) 119 { 120 } 121 122 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 123 static struct kmem_cache *task_struct_cachep; 124 125 static inline struct task_struct *alloc_task_struct_node(int node) 126 { 127 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node); 128 } 129 130 static inline void free_task_struct(struct task_struct *tsk) 131 { 132 kmem_cache_free(task_struct_cachep, tsk); 133 } 134 #endif 135 136 void __weak arch_release_thread_info(struct thread_info *ti) 137 { 138 } 139 140 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR 141 142 /* 143 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a 144 * kmemcache based allocator. 145 */ 146 # if THREAD_SIZE >= PAGE_SIZE 147 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk, 148 int node) 149 { 150 struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED, 151 THREAD_SIZE_ORDER); 152 153 return page ? page_address(page) : NULL; 154 } 155 156 static inline void free_thread_info(struct thread_info *ti) 157 { 158 free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER); 159 } 160 # else 161 static struct kmem_cache *thread_info_cache; 162 163 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk, 164 int node) 165 { 166 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node); 167 } 168 169 static void free_thread_info(struct thread_info *ti) 170 { 171 kmem_cache_free(thread_info_cache, ti); 172 } 173 174 void thread_info_cache_init(void) 175 { 176 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE, 177 THREAD_SIZE, 0, NULL); 178 BUG_ON(thread_info_cache == NULL); 179 } 180 # endif 181 #endif 182 183 /* SLAB cache for signal_struct structures (tsk->signal) */ 184 static struct kmem_cache *signal_cachep; 185 186 /* SLAB cache for sighand_struct structures (tsk->sighand) */ 187 struct kmem_cache *sighand_cachep; 188 189 /* SLAB cache for files_struct structures (tsk->files) */ 190 struct kmem_cache *files_cachep; 191 192 /* SLAB cache for fs_struct structures (tsk->fs) */ 193 struct kmem_cache *fs_cachep; 194 195 /* SLAB cache for vm_area_struct structures */ 196 struct kmem_cache *vm_area_cachep; 197 198 /* SLAB cache for mm_struct structures (tsk->mm) */ 199 static struct kmem_cache *mm_cachep; 200 201 static void account_kernel_stack(struct thread_info *ti, int account) 202 { 203 struct zone *zone = page_zone(virt_to_page(ti)); 204 205 mod_zone_page_state(zone, NR_KERNEL_STACK, account); 206 } 207 208 void free_task(struct task_struct *tsk) 209 { 210 account_kernel_stack(tsk->stack, -1); 211 arch_release_thread_info(tsk->stack); 212 free_thread_info(tsk->stack); 213 rt_mutex_debug_task_free(tsk); 214 ftrace_graph_exit_task(tsk); 215 put_seccomp_filter(tsk); 216 arch_release_task_struct(tsk); 217 free_task_struct(tsk); 218 } 219 EXPORT_SYMBOL(free_task); 220 221 static inline void free_signal_struct(struct signal_struct *sig) 222 { 223 taskstats_tgid_free(sig); 224 sched_autogroup_exit(sig); 225 kmem_cache_free(signal_cachep, sig); 226 } 227 228 static inline void put_signal_struct(struct signal_struct *sig) 229 { 230 if (atomic_dec_and_test(&sig->sigcnt)) 231 free_signal_struct(sig); 232 } 233 234 void __put_task_struct(struct task_struct *tsk) 235 { 236 WARN_ON(!tsk->exit_state); 237 WARN_ON(atomic_read(&tsk->usage)); 238 WARN_ON(tsk == current); 239 240 security_task_free(tsk); 241 exit_creds(tsk); 242 delayacct_tsk_free(tsk); 243 put_signal_struct(tsk->signal); 244 245 if (!profile_handoff_task(tsk)) 246 free_task(tsk); 247 } 248 EXPORT_SYMBOL_GPL(__put_task_struct); 249 250 void __init __weak arch_task_cache_init(void) { } 251 252 void __init fork_init(unsigned long mempages) 253 { 254 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 255 #ifndef ARCH_MIN_TASKALIGN 256 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES 257 #endif 258 /* create a slab on which task_structs can be allocated */ 259 task_struct_cachep = 260 kmem_cache_create("task_struct", sizeof(struct task_struct), 261 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL); 262 #endif 263 264 /* do the arch specific task caches init */ 265 arch_task_cache_init(); 266 267 /* 268 * The default maximum number of threads is set to a safe 269 * value: the thread structures can take up at most half 270 * of memory. 271 */ 272 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE); 273 274 /* 275 * we need to allow at least 20 threads to boot a system 276 */ 277 if (max_threads < 20) 278 max_threads = 20; 279 280 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 281 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 282 init_task.signal->rlim[RLIMIT_SIGPENDING] = 283 init_task.signal->rlim[RLIMIT_NPROC]; 284 } 285 286 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst, 287 struct task_struct *src) 288 { 289 *dst = *src; 290 return 0; 291 } 292 293 static struct task_struct *dup_task_struct(struct task_struct *orig) 294 { 295 struct task_struct *tsk; 296 struct thread_info *ti; 297 unsigned long *stackend; 298 int node = tsk_fork_get_node(orig); 299 int err; 300 301 tsk = alloc_task_struct_node(node); 302 if (!tsk) 303 return NULL; 304 305 ti = alloc_thread_info_node(tsk, node); 306 if (!ti) 307 goto free_tsk; 308 309 err = arch_dup_task_struct(tsk, orig); 310 if (err) 311 goto free_ti; 312 313 tsk->stack = ti; 314 315 setup_thread_stack(tsk, orig); 316 clear_user_return_notifier(tsk); 317 clear_tsk_need_resched(tsk); 318 stackend = end_of_stack(tsk); 319 *stackend = STACK_END_MAGIC; /* for overflow detection */ 320 321 #ifdef CONFIG_CC_STACKPROTECTOR 322 tsk->stack_canary = get_random_int(); 323 #endif 324 325 /* 326 * One for us, one for whoever does the "release_task()" (usually 327 * parent) 328 */ 329 atomic_set(&tsk->usage, 2); 330 #ifdef CONFIG_BLK_DEV_IO_TRACE 331 tsk->btrace_seq = 0; 332 #endif 333 tsk->splice_pipe = NULL; 334 tsk->task_frag.page = NULL; 335 336 account_kernel_stack(ti, 1); 337 338 return tsk; 339 340 free_ti: 341 free_thread_info(ti); 342 free_tsk: 343 free_task_struct(tsk); 344 return NULL; 345 } 346 347 #ifdef CONFIG_MMU 348 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) 349 { 350 struct vm_area_struct *mpnt, *tmp, *prev, **pprev; 351 struct rb_node **rb_link, *rb_parent; 352 int retval; 353 unsigned long charge; 354 struct mempolicy *pol; 355 356 uprobe_start_dup_mmap(); 357 down_write(&oldmm->mmap_sem); 358 flush_cache_dup_mm(oldmm); 359 uprobe_dup_mmap(oldmm, mm); 360 /* 361 * Not linked in yet - no deadlock potential: 362 */ 363 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING); 364 365 mm->locked_vm = 0; 366 mm->mmap = NULL; 367 mm->mmap_cache = NULL; 368 mm->map_count = 0; 369 cpumask_clear(mm_cpumask(mm)); 370 mm->mm_rb = RB_ROOT; 371 rb_link = &mm->mm_rb.rb_node; 372 rb_parent = NULL; 373 pprev = &mm->mmap; 374 retval = ksm_fork(mm, oldmm); 375 if (retval) 376 goto out; 377 retval = khugepaged_fork(mm, oldmm); 378 if (retval) 379 goto out; 380 381 prev = NULL; 382 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { 383 struct file *file; 384 385 if (mpnt->vm_flags & VM_DONTCOPY) { 386 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file, 387 -vma_pages(mpnt)); 388 continue; 389 } 390 charge = 0; 391 if (mpnt->vm_flags & VM_ACCOUNT) { 392 unsigned long len = vma_pages(mpnt); 393 394 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */ 395 goto fail_nomem; 396 charge = len; 397 } 398 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 399 if (!tmp) 400 goto fail_nomem; 401 *tmp = *mpnt; 402 INIT_LIST_HEAD(&tmp->anon_vma_chain); 403 pol = mpol_dup(vma_policy(mpnt)); 404 retval = PTR_ERR(pol); 405 if (IS_ERR(pol)) 406 goto fail_nomem_policy; 407 vma_set_policy(tmp, pol); 408 tmp->vm_mm = mm; 409 if (anon_vma_fork(tmp, mpnt)) 410 goto fail_nomem_anon_vma_fork; 411 tmp->vm_flags &= ~VM_LOCKED; 412 tmp->vm_next = tmp->vm_prev = NULL; 413 file = tmp->vm_file; 414 if (file) { 415 struct inode *inode = file_inode(file); 416 struct address_space *mapping = file->f_mapping; 417 418 get_file(file); 419 if (tmp->vm_flags & VM_DENYWRITE) 420 atomic_dec(&inode->i_writecount); 421 mutex_lock(&mapping->i_mmap_mutex); 422 if (tmp->vm_flags & VM_SHARED) 423 mapping->i_mmap_writable++; 424 flush_dcache_mmap_lock(mapping); 425 /* insert tmp into the share list, just after mpnt */ 426 if (unlikely(tmp->vm_flags & VM_NONLINEAR)) 427 vma_nonlinear_insert(tmp, 428 &mapping->i_mmap_nonlinear); 429 else 430 vma_interval_tree_insert_after(tmp, mpnt, 431 &mapping->i_mmap); 432 flush_dcache_mmap_unlock(mapping); 433 mutex_unlock(&mapping->i_mmap_mutex); 434 } 435 436 /* 437 * Clear hugetlb-related page reserves for children. This only 438 * affects MAP_PRIVATE mappings. Faults generated by the child 439 * are not guaranteed to succeed, even if read-only 440 */ 441 if (is_vm_hugetlb_page(tmp)) 442 reset_vma_resv_huge_pages(tmp); 443 444 /* 445 * Link in the new vma and copy the page table entries. 446 */ 447 *pprev = tmp; 448 pprev = &tmp->vm_next; 449 tmp->vm_prev = prev; 450 prev = tmp; 451 452 __vma_link_rb(mm, tmp, rb_link, rb_parent); 453 rb_link = &tmp->vm_rb.rb_right; 454 rb_parent = &tmp->vm_rb; 455 456 mm->map_count++; 457 retval = copy_page_range(mm, oldmm, mpnt); 458 459 if (tmp->vm_ops && tmp->vm_ops->open) 460 tmp->vm_ops->open(tmp); 461 462 if (retval) 463 goto out; 464 } 465 /* a new mm has just been created */ 466 arch_dup_mmap(oldmm, mm); 467 retval = 0; 468 out: 469 up_write(&mm->mmap_sem); 470 flush_tlb_mm(oldmm); 471 up_write(&oldmm->mmap_sem); 472 uprobe_end_dup_mmap(); 473 return retval; 474 fail_nomem_anon_vma_fork: 475 mpol_put(pol); 476 fail_nomem_policy: 477 kmem_cache_free(vm_area_cachep, tmp); 478 fail_nomem: 479 retval = -ENOMEM; 480 vm_unacct_memory(charge); 481 goto out; 482 } 483 484 static inline int mm_alloc_pgd(struct mm_struct *mm) 485 { 486 mm->pgd = pgd_alloc(mm); 487 if (unlikely(!mm->pgd)) 488 return -ENOMEM; 489 return 0; 490 } 491 492 static inline void mm_free_pgd(struct mm_struct *mm) 493 { 494 pgd_free(mm, mm->pgd); 495 } 496 #else 497 #define dup_mmap(mm, oldmm) (0) 498 #define mm_alloc_pgd(mm) (0) 499 #define mm_free_pgd(mm) 500 #endif /* CONFIG_MMU */ 501 502 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); 503 504 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL)) 505 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) 506 507 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT; 508 509 static int __init coredump_filter_setup(char *s) 510 { 511 default_dump_filter = 512 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) & 513 MMF_DUMP_FILTER_MASK; 514 return 1; 515 } 516 517 __setup("coredump_filter=", coredump_filter_setup); 518 519 #include <linux/init_task.h> 520 521 static void mm_init_aio(struct mm_struct *mm) 522 { 523 #ifdef CONFIG_AIO 524 spin_lock_init(&mm->ioctx_lock); 525 INIT_HLIST_HEAD(&mm->ioctx_list); 526 #endif 527 } 528 529 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p) 530 { 531 atomic_set(&mm->mm_users, 1); 532 atomic_set(&mm->mm_count, 1); 533 init_rwsem(&mm->mmap_sem); 534 INIT_LIST_HEAD(&mm->mmlist); 535 mm->flags = (current->mm) ? 536 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter; 537 mm->core_state = NULL; 538 mm->nr_ptes = 0; 539 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat)); 540 spin_lock_init(&mm->page_table_lock); 541 mm_init_aio(mm); 542 mm_init_owner(mm, p); 543 544 if (likely(!mm_alloc_pgd(mm))) { 545 mm->def_flags = 0; 546 mmu_notifier_mm_init(mm); 547 return mm; 548 } 549 550 free_mm(mm); 551 return NULL; 552 } 553 554 static void check_mm(struct mm_struct *mm) 555 { 556 int i; 557 558 for (i = 0; i < NR_MM_COUNTERS; i++) { 559 long x = atomic_long_read(&mm->rss_stat.count[i]); 560 561 if (unlikely(x)) 562 printk(KERN_ALERT "BUG: Bad rss-counter state " 563 "mm:%p idx:%d val:%ld\n", mm, i, x); 564 } 565 566 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 567 VM_BUG_ON(mm->pmd_huge_pte); 568 #endif 569 } 570 571 /* 572 * Allocate and initialize an mm_struct. 573 */ 574 struct mm_struct *mm_alloc(void) 575 { 576 struct mm_struct *mm; 577 578 mm = allocate_mm(); 579 if (!mm) 580 return NULL; 581 582 memset(mm, 0, sizeof(*mm)); 583 mm_init_cpumask(mm); 584 return mm_init(mm, current); 585 } 586 587 /* 588 * Called when the last reference to the mm 589 * is dropped: either by a lazy thread or by 590 * mmput. Free the page directory and the mm. 591 */ 592 void __mmdrop(struct mm_struct *mm) 593 { 594 BUG_ON(mm == &init_mm); 595 mm_free_pgd(mm); 596 destroy_context(mm); 597 mmu_notifier_mm_destroy(mm); 598 check_mm(mm); 599 free_mm(mm); 600 } 601 EXPORT_SYMBOL_GPL(__mmdrop); 602 603 /* 604 * Decrement the use count and release all resources for an mm. 605 */ 606 void mmput(struct mm_struct *mm) 607 { 608 might_sleep(); 609 610 if (atomic_dec_and_test(&mm->mm_users)) { 611 uprobe_clear_state(mm); 612 exit_aio(mm); 613 ksm_exit(mm); 614 khugepaged_exit(mm); /* must run before exit_mmap */ 615 exit_mmap(mm); 616 set_mm_exe_file(mm, NULL); 617 if (!list_empty(&mm->mmlist)) { 618 spin_lock(&mmlist_lock); 619 list_del(&mm->mmlist); 620 spin_unlock(&mmlist_lock); 621 } 622 if (mm->binfmt) 623 module_put(mm->binfmt->module); 624 mmdrop(mm); 625 } 626 } 627 EXPORT_SYMBOL_GPL(mmput); 628 629 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 630 { 631 if (new_exe_file) 632 get_file(new_exe_file); 633 if (mm->exe_file) 634 fput(mm->exe_file); 635 mm->exe_file = new_exe_file; 636 } 637 638 struct file *get_mm_exe_file(struct mm_struct *mm) 639 { 640 struct file *exe_file; 641 642 /* We need mmap_sem to protect against races with removal of exe_file */ 643 down_read(&mm->mmap_sem); 644 exe_file = mm->exe_file; 645 if (exe_file) 646 get_file(exe_file); 647 up_read(&mm->mmap_sem); 648 return exe_file; 649 } 650 651 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm) 652 { 653 /* It's safe to write the exe_file pointer without exe_file_lock because 654 * this is called during fork when the task is not yet in /proc */ 655 newmm->exe_file = get_mm_exe_file(oldmm); 656 } 657 658 /** 659 * get_task_mm - acquire a reference to the task's mm 660 * 661 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning 662 * this kernel workthread has transiently adopted a user mm with use_mm, 663 * to do its AIO) is not set and if so returns a reference to it, after 664 * bumping up the use count. User must release the mm via mmput() 665 * after use. Typically used by /proc and ptrace. 666 */ 667 struct mm_struct *get_task_mm(struct task_struct *task) 668 { 669 struct mm_struct *mm; 670 671 task_lock(task); 672 mm = task->mm; 673 if (mm) { 674 if (task->flags & PF_KTHREAD) 675 mm = NULL; 676 else 677 atomic_inc(&mm->mm_users); 678 } 679 task_unlock(task); 680 return mm; 681 } 682 EXPORT_SYMBOL_GPL(get_task_mm); 683 684 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode) 685 { 686 struct mm_struct *mm; 687 int err; 688 689 err = mutex_lock_killable(&task->signal->cred_guard_mutex); 690 if (err) 691 return ERR_PTR(err); 692 693 mm = get_task_mm(task); 694 if (mm && mm != current->mm && 695 !ptrace_may_access(task, mode)) { 696 mmput(mm); 697 mm = ERR_PTR(-EACCES); 698 } 699 mutex_unlock(&task->signal->cred_guard_mutex); 700 701 return mm; 702 } 703 704 static void complete_vfork_done(struct task_struct *tsk) 705 { 706 struct completion *vfork; 707 708 task_lock(tsk); 709 vfork = tsk->vfork_done; 710 if (likely(vfork)) { 711 tsk->vfork_done = NULL; 712 complete(vfork); 713 } 714 task_unlock(tsk); 715 } 716 717 static int wait_for_vfork_done(struct task_struct *child, 718 struct completion *vfork) 719 { 720 int killed; 721 722 freezer_do_not_count(); 723 killed = wait_for_completion_killable(vfork); 724 freezer_count(); 725 726 if (killed) { 727 task_lock(child); 728 child->vfork_done = NULL; 729 task_unlock(child); 730 } 731 732 put_task_struct(child); 733 return killed; 734 } 735 736 /* Please note the differences between mmput and mm_release. 737 * mmput is called whenever we stop holding onto a mm_struct, 738 * error success whatever. 739 * 740 * mm_release is called after a mm_struct has been removed 741 * from the current process. 742 * 743 * This difference is important for error handling, when we 744 * only half set up a mm_struct for a new process and need to restore 745 * the old one. Because we mmput the new mm_struct before 746 * restoring the old one. . . 747 * Eric Biederman 10 January 1998 748 */ 749 void mm_release(struct task_struct *tsk, struct mm_struct *mm) 750 { 751 /* Get rid of any futexes when releasing the mm */ 752 #ifdef CONFIG_FUTEX 753 if (unlikely(tsk->robust_list)) { 754 exit_robust_list(tsk); 755 tsk->robust_list = NULL; 756 } 757 #ifdef CONFIG_COMPAT 758 if (unlikely(tsk->compat_robust_list)) { 759 compat_exit_robust_list(tsk); 760 tsk->compat_robust_list = NULL; 761 } 762 #endif 763 if (unlikely(!list_empty(&tsk->pi_state_list))) 764 exit_pi_state_list(tsk); 765 #endif 766 767 uprobe_free_utask(tsk); 768 769 /* Get rid of any cached register state */ 770 deactivate_mm(tsk, mm); 771 772 /* 773 * If we're exiting normally, clear a user-space tid field if 774 * requested. We leave this alone when dying by signal, to leave 775 * the value intact in a core dump, and to save the unnecessary 776 * trouble, say, a killed vfork parent shouldn't touch this mm. 777 * Userland only wants this done for a sys_exit. 778 */ 779 if (tsk->clear_child_tid) { 780 if (!(tsk->flags & PF_SIGNALED) && 781 atomic_read(&mm->mm_users) > 1) { 782 /* 783 * We don't check the error code - if userspace has 784 * not set up a proper pointer then tough luck. 785 */ 786 put_user(0, tsk->clear_child_tid); 787 sys_futex(tsk->clear_child_tid, FUTEX_WAKE, 788 1, NULL, NULL, 0); 789 } 790 tsk->clear_child_tid = NULL; 791 } 792 793 /* 794 * All done, finally we can wake up parent and return this mm to him. 795 * Also kthread_stop() uses this completion for synchronization. 796 */ 797 if (tsk->vfork_done) 798 complete_vfork_done(tsk); 799 } 800 801 /* 802 * Allocate a new mm structure and copy contents from the 803 * mm structure of the passed in task structure. 804 */ 805 struct mm_struct *dup_mm(struct task_struct *tsk) 806 { 807 struct mm_struct *mm, *oldmm = current->mm; 808 int err; 809 810 if (!oldmm) 811 return NULL; 812 813 mm = allocate_mm(); 814 if (!mm) 815 goto fail_nomem; 816 817 memcpy(mm, oldmm, sizeof(*mm)); 818 mm_init_cpumask(mm); 819 820 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 821 mm->pmd_huge_pte = NULL; 822 #endif 823 #ifdef CONFIG_NUMA_BALANCING 824 mm->first_nid = NUMA_PTE_SCAN_INIT; 825 #endif 826 if (!mm_init(mm, tsk)) 827 goto fail_nomem; 828 829 if (init_new_context(tsk, mm)) 830 goto fail_nocontext; 831 832 dup_mm_exe_file(oldmm, mm); 833 834 err = dup_mmap(mm, oldmm); 835 if (err) 836 goto free_pt; 837 838 mm->hiwater_rss = get_mm_rss(mm); 839 mm->hiwater_vm = mm->total_vm; 840 841 if (mm->binfmt && !try_module_get(mm->binfmt->module)) 842 goto free_pt; 843 844 return mm; 845 846 free_pt: 847 /* don't put binfmt in mmput, we haven't got module yet */ 848 mm->binfmt = NULL; 849 mmput(mm); 850 851 fail_nomem: 852 return NULL; 853 854 fail_nocontext: 855 /* 856 * If init_new_context() failed, we cannot use mmput() to free the mm 857 * because it calls destroy_context() 858 */ 859 mm_free_pgd(mm); 860 free_mm(mm); 861 return NULL; 862 } 863 864 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk) 865 { 866 struct mm_struct *mm, *oldmm; 867 int retval; 868 869 tsk->min_flt = tsk->maj_flt = 0; 870 tsk->nvcsw = tsk->nivcsw = 0; 871 #ifdef CONFIG_DETECT_HUNG_TASK 872 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; 873 #endif 874 875 tsk->mm = NULL; 876 tsk->active_mm = NULL; 877 878 /* 879 * Are we cloning a kernel thread? 880 * 881 * We need to steal a active VM for that.. 882 */ 883 oldmm = current->mm; 884 if (!oldmm) 885 return 0; 886 887 if (clone_flags & CLONE_VM) { 888 atomic_inc(&oldmm->mm_users); 889 mm = oldmm; 890 goto good_mm; 891 } 892 893 retval = -ENOMEM; 894 mm = dup_mm(tsk); 895 if (!mm) 896 goto fail_nomem; 897 898 good_mm: 899 tsk->mm = mm; 900 tsk->active_mm = mm; 901 return 0; 902 903 fail_nomem: 904 return retval; 905 } 906 907 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 908 { 909 struct fs_struct *fs = current->fs; 910 if (clone_flags & CLONE_FS) { 911 /* tsk->fs is already what we want */ 912 spin_lock(&fs->lock); 913 if (fs->in_exec) { 914 spin_unlock(&fs->lock); 915 return -EAGAIN; 916 } 917 fs->users++; 918 spin_unlock(&fs->lock); 919 return 0; 920 } 921 tsk->fs = copy_fs_struct(fs); 922 if (!tsk->fs) 923 return -ENOMEM; 924 return 0; 925 } 926 927 static int copy_files(unsigned long clone_flags, struct task_struct *tsk) 928 { 929 struct files_struct *oldf, *newf; 930 int error = 0; 931 932 /* 933 * A background process may not have any files ... 934 */ 935 oldf = current->files; 936 if (!oldf) 937 goto out; 938 939 if (clone_flags & CLONE_FILES) { 940 atomic_inc(&oldf->count); 941 goto out; 942 } 943 944 newf = dup_fd(oldf, &error); 945 if (!newf) 946 goto out; 947 948 tsk->files = newf; 949 error = 0; 950 out: 951 return error; 952 } 953 954 static int copy_io(unsigned long clone_flags, struct task_struct *tsk) 955 { 956 #ifdef CONFIG_BLOCK 957 struct io_context *ioc = current->io_context; 958 struct io_context *new_ioc; 959 960 if (!ioc) 961 return 0; 962 /* 963 * Share io context with parent, if CLONE_IO is set 964 */ 965 if (clone_flags & CLONE_IO) { 966 ioc_task_link(ioc); 967 tsk->io_context = ioc; 968 } else if (ioprio_valid(ioc->ioprio)) { 969 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE); 970 if (unlikely(!new_ioc)) 971 return -ENOMEM; 972 973 new_ioc->ioprio = ioc->ioprio; 974 put_io_context(new_ioc); 975 } 976 #endif 977 return 0; 978 } 979 980 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk) 981 { 982 struct sighand_struct *sig; 983 984 if (clone_flags & CLONE_SIGHAND) { 985 atomic_inc(¤t->sighand->count); 986 return 0; 987 } 988 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 989 rcu_assign_pointer(tsk->sighand, sig); 990 if (!sig) 991 return -ENOMEM; 992 atomic_set(&sig->count, 1); 993 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 994 return 0; 995 } 996 997 void __cleanup_sighand(struct sighand_struct *sighand) 998 { 999 if (atomic_dec_and_test(&sighand->count)) { 1000 signalfd_cleanup(sighand); 1001 kmem_cache_free(sighand_cachep, sighand); 1002 } 1003 } 1004 1005 1006 /* 1007 * Initialize POSIX timer handling for a thread group. 1008 */ 1009 static void posix_cpu_timers_init_group(struct signal_struct *sig) 1010 { 1011 unsigned long cpu_limit; 1012 1013 /* Thread group counters. */ 1014 thread_group_cputime_init(sig); 1015 1016 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); 1017 if (cpu_limit != RLIM_INFINITY) { 1018 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit); 1019 sig->cputimer.running = 1; 1020 } 1021 1022 /* The timer lists. */ 1023 INIT_LIST_HEAD(&sig->cpu_timers[0]); 1024 INIT_LIST_HEAD(&sig->cpu_timers[1]); 1025 INIT_LIST_HEAD(&sig->cpu_timers[2]); 1026 } 1027 1028 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk) 1029 { 1030 struct signal_struct *sig; 1031 1032 if (clone_flags & CLONE_THREAD) 1033 return 0; 1034 1035 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL); 1036 tsk->signal = sig; 1037 if (!sig) 1038 return -ENOMEM; 1039 1040 sig->nr_threads = 1; 1041 atomic_set(&sig->live, 1); 1042 atomic_set(&sig->sigcnt, 1); 1043 init_waitqueue_head(&sig->wait_chldexit); 1044 sig->curr_target = tsk; 1045 init_sigpending(&sig->shared_pending); 1046 INIT_LIST_HEAD(&sig->posix_timers); 1047 1048 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1049 sig->real_timer.function = it_real_fn; 1050 1051 task_lock(current->group_leader); 1052 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 1053 task_unlock(current->group_leader); 1054 1055 posix_cpu_timers_init_group(sig); 1056 1057 tty_audit_fork(sig); 1058 sched_autogroup_fork(sig); 1059 1060 #ifdef CONFIG_CGROUPS 1061 init_rwsem(&sig->group_rwsem); 1062 #endif 1063 1064 sig->oom_score_adj = current->signal->oom_score_adj; 1065 sig->oom_score_adj_min = current->signal->oom_score_adj_min; 1066 1067 sig->has_child_subreaper = current->signal->has_child_subreaper || 1068 current->signal->is_child_subreaper; 1069 1070 mutex_init(&sig->cred_guard_mutex); 1071 1072 return 0; 1073 } 1074 1075 static void copy_flags(unsigned long clone_flags, struct task_struct *p) 1076 { 1077 unsigned long new_flags = p->flags; 1078 1079 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER); 1080 new_flags |= PF_FORKNOEXEC; 1081 p->flags = new_flags; 1082 } 1083 1084 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr) 1085 { 1086 current->clear_child_tid = tidptr; 1087 1088 return task_pid_vnr(current); 1089 } 1090 1091 static void rt_mutex_init_task(struct task_struct *p) 1092 { 1093 raw_spin_lock_init(&p->pi_lock); 1094 #ifdef CONFIG_RT_MUTEXES 1095 plist_head_init(&p->pi_waiters); 1096 p->pi_blocked_on = NULL; 1097 #endif 1098 } 1099 1100 #ifdef CONFIG_MM_OWNER 1101 void mm_init_owner(struct mm_struct *mm, struct task_struct *p) 1102 { 1103 mm->owner = p; 1104 } 1105 #endif /* CONFIG_MM_OWNER */ 1106 1107 /* 1108 * Initialize POSIX timer handling for a single task. 1109 */ 1110 static void posix_cpu_timers_init(struct task_struct *tsk) 1111 { 1112 tsk->cputime_expires.prof_exp = 0; 1113 tsk->cputime_expires.virt_exp = 0; 1114 tsk->cputime_expires.sched_exp = 0; 1115 INIT_LIST_HEAD(&tsk->cpu_timers[0]); 1116 INIT_LIST_HEAD(&tsk->cpu_timers[1]); 1117 INIT_LIST_HEAD(&tsk->cpu_timers[2]); 1118 } 1119 1120 static inline void 1121 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid) 1122 { 1123 task->pids[type].pid = pid; 1124 } 1125 1126 /* 1127 * This creates a new process as a copy of the old one, 1128 * but does not actually start it yet. 1129 * 1130 * It copies the registers, and all the appropriate 1131 * parts of the process environment (as per the clone 1132 * flags). The actual kick-off is left to the caller. 1133 */ 1134 static struct task_struct *copy_process(unsigned long clone_flags, 1135 unsigned long stack_start, 1136 unsigned long stack_size, 1137 int __user *child_tidptr, 1138 struct pid *pid, 1139 int trace) 1140 { 1141 int retval; 1142 struct task_struct *p; 1143 1144 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 1145 return ERR_PTR(-EINVAL); 1146 1147 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS)) 1148 return ERR_PTR(-EINVAL); 1149 1150 /* 1151 * Thread groups must share signals as well, and detached threads 1152 * can only be started up within the thread group. 1153 */ 1154 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 1155 return ERR_PTR(-EINVAL); 1156 1157 /* 1158 * Shared signal handlers imply shared VM. By way of the above, 1159 * thread groups also imply shared VM. Blocking this case allows 1160 * for various simplifications in other code. 1161 */ 1162 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 1163 return ERR_PTR(-EINVAL); 1164 1165 /* 1166 * Siblings of global init remain as zombies on exit since they are 1167 * not reaped by their parent (swapper). To solve this and to avoid 1168 * multi-rooted process trees, prevent global and container-inits 1169 * from creating siblings. 1170 */ 1171 if ((clone_flags & CLONE_PARENT) && 1172 current->signal->flags & SIGNAL_UNKILLABLE) 1173 return ERR_PTR(-EINVAL); 1174 1175 /* 1176 * If the new process will be in a different pid namespace 1177 * don't allow the creation of threads. 1178 */ 1179 if ((clone_flags & (CLONE_VM|CLONE_NEWPID)) && 1180 (task_active_pid_ns(current) != current->nsproxy->pid_ns)) 1181 return ERR_PTR(-EINVAL); 1182 1183 retval = security_task_create(clone_flags); 1184 if (retval) 1185 goto fork_out; 1186 1187 retval = -ENOMEM; 1188 p = dup_task_struct(current); 1189 if (!p) 1190 goto fork_out; 1191 1192 ftrace_graph_init_task(p); 1193 get_seccomp_filter(p); 1194 1195 rt_mutex_init_task(p); 1196 1197 #ifdef CONFIG_PROVE_LOCKING 1198 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); 1199 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 1200 #endif 1201 retval = -EAGAIN; 1202 if (atomic_read(&p->real_cred->user->processes) >= 1203 task_rlimit(p, RLIMIT_NPROC)) { 1204 if (p->real_cred->user != INIT_USER && 1205 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) 1206 goto bad_fork_free; 1207 } 1208 current->flags &= ~PF_NPROC_EXCEEDED; 1209 1210 retval = copy_creds(p, clone_flags); 1211 if (retval < 0) 1212 goto bad_fork_free; 1213 1214 /* 1215 * If multiple threads are within copy_process(), then this check 1216 * triggers too late. This doesn't hurt, the check is only there 1217 * to stop root fork bombs. 1218 */ 1219 retval = -EAGAIN; 1220 if (nr_threads >= max_threads) 1221 goto bad_fork_cleanup_count; 1222 1223 if (!try_module_get(task_thread_info(p)->exec_domain->module)) 1224 goto bad_fork_cleanup_count; 1225 1226 p->did_exec = 0; 1227 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */ 1228 copy_flags(clone_flags, p); 1229 INIT_LIST_HEAD(&p->children); 1230 INIT_LIST_HEAD(&p->sibling); 1231 rcu_copy_process(p); 1232 p->vfork_done = NULL; 1233 spin_lock_init(&p->alloc_lock); 1234 1235 init_sigpending(&p->pending); 1236 1237 p->utime = p->stime = p->gtime = 0; 1238 p->utimescaled = p->stimescaled = 0; 1239 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE 1240 p->prev_cputime.utime = p->prev_cputime.stime = 0; 1241 #endif 1242 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN 1243 seqlock_init(&p->vtime_seqlock); 1244 p->vtime_snap = 0; 1245 p->vtime_snap_whence = VTIME_SLEEPING; 1246 #endif 1247 1248 #if defined(SPLIT_RSS_COUNTING) 1249 memset(&p->rss_stat, 0, sizeof(p->rss_stat)); 1250 #endif 1251 1252 p->default_timer_slack_ns = current->timer_slack_ns; 1253 1254 task_io_accounting_init(&p->ioac); 1255 acct_clear_integrals(p); 1256 1257 posix_cpu_timers_init(p); 1258 1259 do_posix_clock_monotonic_gettime(&p->start_time); 1260 p->real_start_time = p->start_time; 1261 monotonic_to_bootbased(&p->real_start_time); 1262 p->io_context = NULL; 1263 p->audit_context = NULL; 1264 if (clone_flags & CLONE_THREAD) 1265 threadgroup_change_begin(current); 1266 cgroup_fork(p); 1267 #ifdef CONFIG_NUMA 1268 p->mempolicy = mpol_dup(p->mempolicy); 1269 if (IS_ERR(p->mempolicy)) { 1270 retval = PTR_ERR(p->mempolicy); 1271 p->mempolicy = NULL; 1272 goto bad_fork_cleanup_cgroup; 1273 } 1274 mpol_fix_fork_child_flag(p); 1275 #endif 1276 #ifdef CONFIG_CPUSETS 1277 p->cpuset_mem_spread_rotor = NUMA_NO_NODE; 1278 p->cpuset_slab_spread_rotor = NUMA_NO_NODE; 1279 seqcount_init(&p->mems_allowed_seq); 1280 #endif 1281 #ifdef CONFIG_TRACE_IRQFLAGS 1282 p->irq_events = 0; 1283 p->hardirqs_enabled = 0; 1284 p->hardirq_enable_ip = 0; 1285 p->hardirq_enable_event = 0; 1286 p->hardirq_disable_ip = _THIS_IP_; 1287 p->hardirq_disable_event = 0; 1288 p->softirqs_enabled = 1; 1289 p->softirq_enable_ip = _THIS_IP_; 1290 p->softirq_enable_event = 0; 1291 p->softirq_disable_ip = 0; 1292 p->softirq_disable_event = 0; 1293 p->hardirq_context = 0; 1294 p->softirq_context = 0; 1295 #endif 1296 #ifdef CONFIG_LOCKDEP 1297 p->lockdep_depth = 0; /* no locks held yet */ 1298 p->curr_chain_key = 0; 1299 p->lockdep_recursion = 0; 1300 #endif 1301 1302 #ifdef CONFIG_DEBUG_MUTEXES 1303 p->blocked_on = NULL; /* not blocked yet */ 1304 #endif 1305 #ifdef CONFIG_MEMCG 1306 p->memcg_batch.do_batch = 0; 1307 p->memcg_batch.memcg = NULL; 1308 #endif 1309 #ifdef CONFIG_BCACHE 1310 p->sequential_io = 0; 1311 p->sequential_io_avg = 0; 1312 #endif 1313 1314 /* Perform scheduler related setup. Assign this task to a CPU. */ 1315 sched_fork(p); 1316 1317 retval = perf_event_init_task(p); 1318 if (retval) 1319 goto bad_fork_cleanup_policy; 1320 retval = audit_alloc(p); 1321 if (retval) 1322 goto bad_fork_cleanup_policy; 1323 /* copy all the process information */ 1324 retval = copy_semundo(clone_flags, p); 1325 if (retval) 1326 goto bad_fork_cleanup_audit; 1327 retval = copy_files(clone_flags, p); 1328 if (retval) 1329 goto bad_fork_cleanup_semundo; 1330 retval = copy_fs(clone_flags, p); 1331 if (retval) 1332 goto bad_fork_cleanup_files; 1333 retval = copy_sighand(clone_flags, p); 1334 if (retval) 1335 goto bad_fork_cleanup_fs; 1336 retval = copy_signal(clone_flags, p); 1337 if (retval) 1338 goto bad_fork_cleanup_sighand; 1339 retval = copy_mm(clone_flags, p); 1340 if (retval) 1341 goto bad_fork_cleanup_signal; 1342 retval = copy_namespaces(clone_flags, p); 1343 if (retval) 1344 goto bad_fork_cleanup_mm; 1345 retval = copy_io(clone_flags, p); 1346 if (retval) 1347 goto bad_fork_cleanup_namespaces; 1348 retval = copy_thread(clone_flags, stack_start, stack_size, p); 1349 if (retval) 1350 goto bad_fork_cleanup_io; 1351 1352 if (pid != &init_struct_pid) { 1353 retval = -ENOMEM; 1354 pid = alloc_pid(p->nsproxy->pid_ns); 1355 if (!pid) 1356 goto bad_fork_cleanup_io; 1357 } 1358 1359 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; 1360 /* 1361 * Clear TID on mm_release()? 1362 */ 1363 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL; 1364 #ifdef CONFIG_BLOCK 1365 p->plug = NULL; 1366 #endif 1367 #ifdef CONFIG_FUTEX 1368 p->robust_list = NULL; 1369 #ifdef CONFIG_COMPAT 1370 p->compat_robust_list = NULL; 1371 #endif 1372 INIT_LIST_HEAD(&p->pi_state_list); 1373 p->pi_state_cache = NULL; 1374 #endif 1375 uprobe_copy_process(p); 1376 /* 1377 * sigaltstack should be cleared when sharing the same VM 1378 */ 1379 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM) 1380 p->sas_ss_sp = p->sas_ss_size = 0; 1381 1382 /* 1383 * Syscall tracing and stepping should be turned off in the 1384 * child regardless of CLONE_PTRACE. 1385 */ 1386 user_disable_single_step(p); 1387 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); 1388 #ifdef TIF_SYSCALL_EMU 1389 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); 1390 #endif 1391 clear_all_latency_tracing(p); 1392 1393 /* ok, now we should be set up.. */ 1394 p->pid = pid_nr(pid); 1395 if (clone_flags & CLONE_THREAD) { 1396 p->exit_signal = -1; 1397 p->group_leader = current->group_leader; 1398 p->tgid = current->tgid; 1399 } else { 1400 if (clone_flags & CLONE_PARENT) 1401 p->exit_signal = current->group_leader->exit_signal; 1402 else 1403 p->exit_signal = (clone_flags & CSIGNAL); 1404 p->group_leader = p; 1405 p->tgid = p->pid; 1406 } 1407 1408 p->pdeath_signal = 0; 1409 p->exit_state = 0; 1410 1411 p->nr_dirtied = 0; 1412 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10); 1413 p->dirty_paused_when = 0; 1414 1415 INIT_LIST_HEAD(&p->thread_group); 1416 p->task_works = NULL; 1417 1418 /* 1419 * Make it visible to the rest of the system, but dont wake it up yet. 1420 * Need tasklist lock for parent etc handling! 1421 */ 1422 write_lock_irq(&tasklist_lock); 1423 1424 /* CLONE_PARENT re-uses the old parent */ 1425 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { 1426 p->real_parent = current->real_parent; 1427 p->parent_exec_id = current->parent_exec_id; 1428 } else { 1429 p->real_parent = current; 1430 p->parent_exec_id = current->self_exec_id; 1431 } 1432 1433 spin_lock(¤t->sighand->siglock); 1434 1435 /* 1436 * Process group and session signals need to be delivered to just the 1437 * parent before the fork or both the parent and the child after the 1438 * fork. Restart if a signal comes in before we add the new process to 1439 * it's process group. 1440 * A fatal signal pending means that current will exit, so the new 1441 * thread can't slip out of an OOM kill (or normal SIGKILL). 1442 */ 1443 recalc_sigpending(); 1444 if (signal_pending(current)) { 1445 spin_unlock(¤t->sighand->siglock); 1446 write_unlock_irq(&tasklist_lock); 1447 retval = -ERESTARTNOINTR; 1448 goto bad_fork_free_pid; 1449 } 1450 1451 if (likely(p->pid)) { 1452 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace); 1453 1454 init_task_pid(p, PIDTYPE_PID, pid); 1455 if (thread_group_leader(p)) { 1456 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current)); 1457 init_task_pid(p, PIDTYPE_SID, task_session(current)); 1458 1459 if (is_child_reaper(pid)) { 1460 ns_of_pid(pid)->child_reaper = p; 1461 p->signal->flags |= SIGNAL_UNKILLABLE; 1462 } 1463 1464 p->signal->leader_pid = pid; 1465 p->signal->tty = tty_kref_get(current->signal->tty); 1466 list_add_tail(&p->sibling, &p->real_parent->children); 1467 list_add_tail_rcu(&p->tasks, &init_task.tasks); 1468 attach_pid(p, PIDTYPE_PGID); 1469 attach_pid(p, PIDTYPE_SID); 1470 __this_cpu_inc(process_counts); 1471 } else { 1472 current->signal->nr_threads++; 1473 atomic_inc(¤t->signal->live); 1474 atomic_inc(¤t->signal->sigcnt); 1475 list_add_tail_rcu(&p->thread_group, 1476 &p->group_leader->thread_group); 1477 } 1478 attach_pid(p, PIDTYPE_PID); 1479 nr_threads++; 1480 } 1481 1482 total_forks++; 1483 spin_unlock(¤t->sighand->siglock); 1484 write_unlock_irq(&tasklist_lock); 1485 proc_fork_connector(p); 1486 cgroup_post_fork(p); 1487 if (clone_flags & CLONE_THREAD) 1488 threadgroup_change_end(current); 1489 perf_event_fork(p); 1490 1491 trace_task_newtask(p, clone_flags); 1492 1493 return p; 1494 1495 bad_fork_free_pid: 1496 if (pid != &init_struct_pid) 1497 free_pid(pid); 1498 bad_fork_cleanup_io: 1499 if (p->io_context) 1500 exit_io_context(p); 1501 bad_fork_cleanup_namespaces: 1502 exit_task_namespaces(p); 1503 bad_fork_cleanup_mm: 1504 if (p->mm) 1505 mmput(p->mm); 1506 bad_fork_cleanup_signal: 1507 if (!(clone_flags & CLONE_THREAD)) 1508 free_signal_struct(p->signal); 1509 bad_fork_cleanup_sighand: 1510 __cleanup_sighand(p->sighand); 1511 bad_fork_cleanup_fs: 1512 exit_fs(p); /* blocking */ 1513 bad_fork_cleanup_files: 1514 exit_files(p); /* blocking */ 1515 bad_fork_cleanup_semundo: 1516 exit_sem(p); 1517 bad_fork_cleanup_audit: 1518 audit_free(p); 1519 bad_fork_cleanup_policy: 1520 perf_event_free_task(p); 1521 #ifdef CONFIG_NUMA 1522 mpol_put(p->mempolicy); 1523 bad_fork_cleanup_cgroup: 1524 #endif 1525 if (clone_flags & CLONE_THREAD) 1526 threadgroup_change_end(current); 1527 cgroup_exit(p, 0); 1528 delayacct_tsk_free(p); 1529 module_put(task_thread_info(p)->exec_domain->module); 1530 bad_fork_cleanup_count: 1531 atomic_dec(&p->cred->user->processes); 1532 exit_creds(p); 1533 bad_fork_free: 1534 free_task(p); 1535 fork_out: 1536 return ERR_PTR(retval); 1537 } 1538 1539 static inline void init_idle_pids(struct pid_link *links) 1540 { 1541 enum pid_type type; 1542 1543 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 1544 INIT_HLIST_NODE(&links[type].node); /* not really needed */ 1545 links[type].pid = &init_struct_pid; 1546 } 1547 } 1548 1549 struct task_struct * __cpuinit fork_idle(int cpu) 1550 { 1551 struct task_struct *task; 1552 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0); 1553 if (!IS_ERR(task)) { 1554 init_idle_pids(task->pids); 1555 init_idle(task, cpu); 1556 } 1557 1558 return task; 1559 } 1560 1561 /* 1562 * Ok, this is the main fork-routine. 1563 * 1564 * It copies the process, and if successful kick-starts 1565 * it and waits for it to finish using the VM if required. 1566 */ 1567 long do_fork(unsigned long clone_flags, 1568 unsigned long stack_start, 1569 unsigned long stack_size, 1570 int __user *parent_tidptr, 1571 int __user *child_tidptr) 1572 { 1573 struct task_struct *p; 1574 int trace = 0; 1575 long nr; 1576 1577 /* 1578 * Do some preliminary argument and permissions checking before we 1579 * actually start allocating stuff 1580 */ 1581 if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) { 1582 if (clone_flags & (CLONE_THREAD|CLONE_PARENT)) 1583 return -EINVAL; 1584 } 1585 1586 /* 1587 * Determine whether and which event to report to ptracer. When 1588 * called from kernel_thread or CLONE_UNTRACED is explicitly 1589 * requested, no event is reported; otherwise, report if the event 1590 * for the type of forking is enabled. 1591 */ 1592 if (!(clone_flags & CLONE_UNTRACED)) { 1593 if (clone_flags & CLONE_VFORK) 1594 trace = PTRACE_EVENT_VFORK; 1595 else if ((clone_flags & CSIGNAL) != SIGCHLD) 1596 trace = PTRACE_EVENT_CLONE; 1597 else 1598 trace = PTRACE_EVENT_FORK; 1599 1600 if (likely(!ptrace_event_enabled(current, trace))) 1601 trace = 0; 1602 } 1603 1604 p = copy_process(clone_flags, stack_start, stack_size, 1605 child_tidptr, NULL, trace); 1606 /* 1607 * Do this prior waking up the new thread - the thread pointer 1608 * might get invalid after that point, if the thread exits quickly. 1609 */ 1610 if (!IS_ERR(p)) { 1611 struct completion vfork; 1612 1613 trace_sched_process_fork(current, p); 1614 1615 nr = task_pid_vnr(p); 1616 1617 if (clone_flags & CLONE_PARENT_SETTID) 1618 put_user(nr, parent_tidptr); 1619 1620 if (clone_flags & CLONE_VFORK) { 1621 p->vfork_done = &vfork; 1622 init_completion(&vfork); 1623 get_task_struct(p); 1624 } 1625 1626 wake_up_new_task(p); 1627 1628 /* forking complete and child started to run, tell ptracer */ 1629 if (unlikely(trace)) 1630 ptrace_event(trace, nr); 1631 1632 if (clone_flags & CLONE_VFORK) { 1633 if (!wait_for_vfork_done(p, &vfork)) 1634 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr); 1635 } 1636 } else { 1637 nr = PTR_ERR(p); 1638 } 1639 return nr; 1640 } 1641 1642 /* 1643 * Create a kernel thread. 1644 */ 1645 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags) 1646 { 1647 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn, 1648 (unsigned long)arg, NULL, NULL); 1649 } 1650 1651 #ifdef __ARCH_WANT_SYS_FORK 1652 SYSCALL_DEFINE0(fork) 1653 { 1654 #ifdef CONFIG_MMU 1655 return do_fork(SIGCHLD, 0, 0, NULL, NULL); 1656 #else 1657 /* can not support in nommu mode */ 1658 return(-EINVAL); 1659 #endif 1660 } 1661 #endif 1662 1663 #ifdef __ARCH_WANT_SYS_VFORK 1664 SYSCALL_DEFINE0(vfork) 1665 { 1666 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0, 1667 0, NULL, NULL); 1668 } 1669 #endif 1670 1671 #ifdef __ARCH_WANT_SYS_CLONE 1672 #ifdef CONFIG_CLONE_BACKWARDS 1673 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 1674 int __user *, parent_tidptr, 1675 int, tls_val, 1676 int __user *, child_tidptr) 1677 #elif defined(CONFIG_CLONE_BACKWARDS2) 1678 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags, 1679 int __user *, parent_tidptr, 1680 int __user *, child_tidptr, 1681 int, tls_val) 1682 #else 1683 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 1684 int __user *, parent_tidptr, 1685 int __user *, child_tidptr, 1686 int, tls_val) 1687 #endif 1688 { 1689 return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr); 1690 } 1691 #endif 1692 1693 #ifndef ARCH_MIN_MMSTRUCT_ALIGN 1694 #define ARCH_MIN_MMSTRUCT_ALIGN 0 1695 #endif 1696 1697 static void sighand_ctor(void *data) 1698 { 1699 struct sighand_struct *sighand = data; 1700 1701 spin_lock_init(&sighand->siglock); 1702 init_waitqueue_head(&sighand->signalfd_wqh); 1703 } 1704 1705 void __init proc_caches_init(void) 1706 { 1707 sighand_cachep = kmem_cache_create("sighand_cache", 1708 sizeof(struct sighand_struct), 0, 1709 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU| 1710 SLAB_NOTRACK, sighand_ctor); 1711 signal_cachep = kmem_cache_create("signal_cache", 1712 sizeof(struct signal_struct), 0, 1713 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1714 files_cachep = kmem_cache_create("files_cache", 1715 sizeof(struct files_struct), 0, 1716 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1717 fs_cachep = kmem_cache_create("fs_cache", 1718 sizeof(struct fs_struct), 0, 1719 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1720 /* 1721 * FIXME! The "sizeof(struct mm_struct)" currently includes the 1722 * whole struct cpumask for the OFFSTACK case. We could change 1723 * this to *only* allocate as much of it as required by the 1724 * maximum number of CPU's we can ever have. The cpumask_allocation 1725 * is at the end of the structure, exactly for that reason. 1726 */ 1727 mm_cachep = kmem_cache_create("mm_struct", 1728 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN, 1729 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1730 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC); 1731 mmap_init(); 1732 nsproxy_cache_init(); 1733 } 1734 1735 /* 1736 * Check constraints on flags passed to the unshare system call. 1737 */ 1738 static int check_unshare_flags(unsigned long unshare_flags) 1739 { 1740 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND| 1741 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM| 1742 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET| 1743 CLONE_NEWUSER|CLONE_NEWPID)) 1744 return -EINVAL; 1745 /* 1746 * Not implemented, but pretend it works if there is nothing to 1747 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND 1748 * needs to unshare vm. 1749 */ 1750 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) { 1751 /* FIXME: get_task_mm() increments ->mm_users */ 1752 if (atomic_read(¤t->mm->mm_users) > 1) 1753 return -EINVAL; 1754 } 1755 1756 return 0; 1757 } 1758 1759 /* 1760 * Unshare the filesystem structure if it is being shared 1761 */ 1762 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) 1763 { 1764 struct fs_struct *fs = current->fs; 1765 1766 if (!(unshare_flags & CLONE_FS) || !fs) 1767 return 0; 1768 1769 /* don't need lock here; in the worst case we'll do useless copy */ 1770 if (fs->users == 1) 1771 return 0; 1772 1773 *new_fsp = copy_fs_struct(fs); 1774 if (!*new_fsp) 1775 return -ENOMEM; 1776 1777 return 0; 1778 } 1779 1780 /* 1781 * Unshare file descriptor table if it is being shared 1782 */ 1783 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp) 1784 { 1785 struct files_struct *fd = current->files; 1786 int error = 0; 1787 1788 if ((unshare_flags & CLONE_FILES) && 1789 (fd && atomic_read(&fd->count) > 1)) { 1790 *new_fdp = dup_fd(fd, &error); 1791 if (!*new_fdp) 1792 return error; 1793 } 1794 1795 return 0; 1796 } 1797 1798 /* 1799 * unshare allows a process to 'unshare' part of the process 1800 * context which was originally shared using clone. copy_* 1801 * functions used by do_fork() cannot be used here directly 1802 * because they modify an inactive task_struct that is being 1803 * constructed. Here we are modifying the current, active, 1804 * task_struct. 1805 */ 1806 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags) 1807 { 1808 struct fs_struct *fs, *new_fs = NULL; 1809 struct files_struct *fd, *new_fd = NULL; 1810 struct cred *new_cred = NULL; 1811 struct nsproxy *new_nsproxy = NULL; 1812 int do_sysvsem = 0; 1813 int err; 1814 1815 /* 1816 * If unsharing a user namespace must also unshare the thread. 1817 */ 1818 if (unshare_flags & CLONE_NEWUSER) 1819 unshare_flags |= CLONE_THREAD | CLONE_FS; 1820 /* 1821 * If unsharing a pid namespace must also unshare the thread. 1822 */ 1823 if (unshare_flags & CLONE_NEWPID) 1824 unshare_flags |= CLONE_THREAD; 1825 /* 1826 * If unsharing a thread from a thread group, must also unshare vm. 1827 */ 1828 if (unshare_flags & CLONE_THREAD) 1829 unshare_flags |= CLONE_VM; 1830 /* 1831 * If unsharing vm, must also unshare signal handlers. 1832 */ 1833 if (unshare_flags & CLONE_VM) 1834 unshare_flags |= CLONE_SIGHAND; 1835 /* 1836 * If unsharing namespace, must also unshare filesystem information. 1837 */ 1838 if (unshare_flags & CLONE_NEWNS) 1839 unshare_flags |= CLONE_FS; 1840 1841 err = check_unshare_flags(unshare_flags); 1842 if (err) 1843 goto bad_unshare_out; 1844 /* 1845 * CLONE_NEWIPC must also detach from the undolist: after switching 1846 * to a new ipc namespace, the semaphore arrays from the old 1847 * namespace are unreachable. 1848 */ 1849 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) 1850 do_sysvsem = 1; 1851 err = unshare_fs(unshare_flags, &new_fs); 1852 if (err) 1853 goto bad_unshare_out; 1854 err = unshare_fd(unshare_flags, &new_fd); 1855 if (err) 1856 goto bad_unshare_cleanup_fs; 1857 err = unshare_userns(unshare_flags, &new_cred); 1858 if (err) 1859 goto bad_unshare_cleanup_fd; 1860 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, 1861 new_cred, new_fs); 1862 if (err) 1863 goto bad_unshare_cleanup_cred; 1864 1865 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) { 1866 if (do_sysvsem) { 1867 /* 1868 * CLONE_SYSVSEM is equivalent to sys_exit(). 1869 */ 1870 exit_sem(current); 1871 } 1872 1873 if (new_nsproxy) 1874 switch_task_namespaces(current, new_nsproxy); 1875 1876 task_lock(current); 1877 1878 if (new_fs) { 1879 fs = current->fs; 1880 spin_lock(&fs->lock); 1881 current->fs = new_fs; 1882 if (--fs->users) 1883 new_fs = NULL; 1884 else 1885 new_fs = fs; 1886 spin_unlock(&fs->lock); 1887 } 1888 1889 if (new_fd) { 1890 fd = current->files; 1891 current->files = new_fd; 1892 new_fd = fd; 1893 } 1894 1895 task_unlock(current); 1896 1897 if (new_cred) { 1898 /* Install the new user namespace */ 1899 commit_creds(new_cred); 1900 new_cred = NULL; 1901 } 1902 } 1903 1904 bad_unshare_cleanup_cred: 1905 if (new_cred) 1906 put_cred(new_cred); 1907 bad_unshare_cleanup_fd: 1908 if (new_fd) 1909 put_files_struct(new_fd); 1910 1911 bad_unshare_cleanup_fs: 1912 if (new_fs) 1913 free_fs_struct(new_fs); 1914 1915 bad_unshare_out: 1916 return err; 1917 } 1918 1919 /* 1920 * Helper to unshare the files of the current task. 1921 * We don't want to expose copy_files internals to 1922 * the exec layer of the kernel. 1923 */ 1924 1925 int unshare_files(struct files_struct **displaced) 1926 { 1927 struct task_struct *task = current; 1928 struct files_struct *copy = NULL; 1929 int error; 1930 1931 error = unshare_fd(CLONE_FILES, ©); 1932 if (error || !copy) { 1933 *displaced = NULL; 1934 return error; 1935 } 1936 *displaced = task->files; 1937 task_lock(task); 1938 task->files = copy; 1939 task_unlock(task); 1940 return 0; 1941 } 1942