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/futex.h> 41 #include <linux/compat.h> 42 #include <linux/kthread.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/proc_fs.h> 51 #include <linux/profile.h> 52 #include <linux/rmap.h> 53 #include <linux/ksm.h> 54 #include <linux/acct.h> 55 #include <linux/tsacct_kern.h> 56 #include <linux/cn_proc.h> 57 #include <linux/freezer.h> 58 #include <linux/delayacct.h> 59 #include <linux/taskstats_kern.h> 60 #include <linux/random.h> 61 #include <linux/tty.h> 62 #include <linux/blkdev.h> 63 #include <linux/fs_struct.h> 64 #include <linux/magic.h> 65 #include <linux/perf_event.h> 66 #include <linux/posix-timers.h> 67 #include <linux/user-return-notifier.h> 68 #include <linux/oom.h> 69 #include <linux/khugepaged.h> 70 #include <linux/signalfd.h> 71 72 #include <asm/pgtable.h> 73 #include <asm/pgalloc.h> 74 #include <asm/uaccess.h> 75 #include <asm/mmu_context.h> 76 #include <asm/cacheflush.h> 77 #include <asm/tlbflush.h> 78 79 #include <trace/events/sched.h> 80 81 #define CREATE_TRACE_POINTS 82 #include <trace/events/task.h> 83 84 /* 85 * Protected counters by write_lock_irq(&tasklist_lock) 86 */ 87 unsigned long total_forks; /* Handle normal Linux uptimes. */ 88 int nr_threads; /* The idle threads do not count.. */ 89 90 int max_threads; /* tunable limit on nr_threads */ 91 92 DEFINE_PER_CPU(unsigned long, process_counts) = 0; 93 94 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */ 95 96 #ifdef CONFIG_PROVE_RCU 97 int lockdep_tasklist_lock_is_held(void) 98 { 99 return lockdep_is_held(&tasklist_lock); 100 } 101 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held); 102 #endif /* #ifdef CONFIG_PROVE_RCU */ 103 104 int nr_processes(void) 105 { 106 int cpu; 107 int total = 0; 108 109 for_each_possible_cpu(cpu) 110 total += per_cpu(process_counts, cpu); 111 112 return total; 113 } 114 115 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR 116 # define alloc_task_struct_node(node) \ 117 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node) 118 # define free_task_struct(tsk) \ 119 kmem_cache_free(task_struct_cachep, (tsk)) 120 static struct kmem_cache *task_struct_cachep; 121 #endif 122 123 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR 124 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk, 125 int node) 126 { 127 #ifdef CONFIG_DEBUG_STACK_USAGE 128 gfp_t mask = GFP_KERNEL | __GFP_ZERO; 129 #else 130 gfp_t mask = GFP_KERNEL; 131 #endif 132 struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER); 133 134 return page ? page_address(page) : NULL; 135 } 136 137 static inline void free_thread_info(struct thread_info *ti) 138 { 139 free_pages((unsigned long)ti, THREAD_SIZE_ORDER); 140 } 141 #endif 142 143 /* SLAB cache for signal_struct structures (tsk->signal) */ 144 static struct kmem_cache *signal_cachep; 145 146 /* SLAB cache for sighand_struct structures (tsk->sighand) */ 147 struct kmem_cache *sighand_cachep; 148 149 /* SLAB cache for files_struct structures (tsk->files) */ 150 struct kmem_cache *files_cachep; 151 152 /* SLAB cache for fs_struct structures (tsk->fs) */ 153 struct kmem_cache *fs_cachep; 154 155 /* SLAB cache for vm_area_struct structures */ 156 struct kmem_cache *vm_area_cachep; 157 158 /* SLAB cache for mm_struct structures (tsk->mm) */ 159 static struct kmem_cache *mm_cachep; 160 161 static void account_kernel_stack(struct thread_info *ti, int account) 162 { 163 struct zone *zone = page_zone(virt_to_page(ti)); 164 165 mod_zone_page_state(zone, NR_KERNEL_STACK, account); 166 } 167 168 void free_task(struct task_struct *tsk) 169 { 170 account_kernel_stack(tsk->stack, -1); 171 free_thread_info(tsk->stack); 172 rt_mutex_debug_task_free(tsk); 173 ftrace_graph_exit_task(tsk); 174 free_task_struct(tsk); 175 } 176 EXPORT_SYMBOL(free_task); 177 178 static inline void free_signal_struct(struct signal_struct *sig) 179 { 180 taskstats_tgid_free(sig); 181 sched_autogroup_exit(sig); 182 kmem_cache_free(signal_cachep, sig); 183 } 184 185 static inline void put_signal_struct(struct signal_struct *sig) 186 { 187 if (atomic_dec_and_test(&sig->sigcnt)) 188 free_signal_struct(sig); 189 } 190 191 void __put_task_struct(struct task_struct *tsk) 192 { 193 WARN_ON(!tsk->exit_state); 194 WARN_ON(atomic_read(&tsk->usage)); 195 WARN_ON(tsk == current); 196 197 security_task_free(tsk); 198 exit_creds(tsk); 199 delayacct_tsk_free(tsk); 200 put_signal_struct(tsk->signal); 201 202 if (!profile_handoff_task(tsk)) 203 free_task(tsk); 204 } 205 EXPORT_SYMBOL_GPL(__put_task_struct); 206 207 /* 208 * macro override instead of weak attribute alias, to workaround 209 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions. 210 */ 211 #ifndef arch_task_cache_init 212 #define arch_task_cache_init() 213 #endif 214 215 void __init fork_init(unsigned long mempages) 216 { 217 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR 218 #ifndef ARCH_MIN_TASKALIGN 219 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES 220 #endif 221 /* create a slab on which task_structs can be allocated */ 222 task_struct_cachep = 223 kmem_cache_create("task_struct", sizeof(struct task_struct), 224 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL); 225 #endif 226 227 /* do the arch specific task caches init */ 228 arch_task_cache_init(); 229 230 /* 231 * The default maximum number of threads is set to a safe 232 * value: the thread structures can take up at most half 233 * of memory. 234 */ 235 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE); 236 237 /* 238 * we need to allow at least 20 threads to boot a system 239 */ 240 if (max_threads < 20) 241 max_threads = 20; 242 243 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 244 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 245 init_task.signal->rlim[RLIMIT_SIGPENDING] = 246 init_task.signal->rlim[RLIMIT_NPROC]; 247 } 248 249 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst, 250 struct task_struct *src) 251 { 252 *dst = *src; 253 return 0; 254 } 255 256 static struct task_struct *dup_task_struct(struct task_struct *orig) 257 { 258 struct task_struct *tsk; 259 struct thread_info *ti; 260 unsigned long *stackend; 261 int node = tsk_fork_get_node(orig); 262 int err; 263 264 prepare_to_copy(orig); 265 266 tsk = alloc_task_struct_node(node); 267 if (!tsk) 268 return NULL; 269 270 ti = alloc_thread_info_node(tsk, node); 271 if (!ti) { 272 free_task_struct(tsk); 273 return NULL; 274 } 275 276 err = arch_dup_task_struct(tsk, orig); 277 if (err) 278 goto out; 279 280 tsk->stack = ti; 281 282 setup_thread_stack(tsk, orig); 283 clear_user_return_notifier(tsk); 284 clear_tsk_need_resched(tsk); 285 stackend = end_of_stack(tsk); 286 *stackend = STACK_END_MAGIC; /* for overflow detection */ 287 288 #ifdef CONFIG_CC_STACKPROTECTOR 289 tsk->stack_canary = get_random_int(); 290 #endif 291 292 /* 293 * One for us, one for whoever does the "release_task()" (usually 294 * parent) 295 */ 296 atomic_set(&tsk->usage, 2); 297 #ifdef CONFIG_BLK_DEV_IO_TRACE 298 tsk->btrace_seq = 0; 299 #endif 300 tsk->splice_pipe = NULL; 301 302 account_kernel_stack(ti, 1); 303 304 return tsk; 305 306 out: 307 free_thread_info(ti); 308 free_task_struct(tsk); 309 return NULL; 310 } 311 312 #ifdef CONFIG_MMU 313 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) 314 { 315 struct vm_area_struct *mpnt, *tmp, *prev, **pprev; 316 struct rb_node **rb_link, *rb_parent; 317 int retval; 318 unsigned long charge; 319 struct mempolicy *pol; 320 321 down_write(&oldmm->mmap_sem); 322 flush_cache_dup_mm(oldmm); 323 /* 324 * Not linked in yet - no deadlock potential: 325 */ 326 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING); 327 328 mm->locked_vm = 0; 329 mm->mmap = NULL; 330 mm->mmap_cache = NULL; 331 mm->free_area_cache = oldmm->mmap_base; 332 mm->cached_hole_size = ~0UL; 333 mm->map_count = 0; 334 cpumask_clear(mm_cpumask(mm)); 335 mm->mm_rb = RB_ROOT; 336 rb_link = &mm->mm_rb.rb_node; 337 rb_parent = NULL; 338 pprev = &mm->mmap; 339 retval = ksm_fork(mm, oldmm); 340 if (retval) 341 goto out; 342 retval = khugepaged_fork(mm, oldmm); 343 if (retval) 344 goto out; 345 346 prev = NULL; 347 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { 348 struct file *file; 349 350 if (mpnt->vm_flags & VM_DONTCOPY) { 351 long pages = vma_pages(mpnt); 352 mm->total_vm -= pages; 353 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file, 354 -pages); 355 continue; 356 } 357 charge = 0; 358 if (mpnt->vm_flags & VM_ACCOUNT) { 359 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT; 360 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */ 361 goto fail_nomem; 362 charge = len; 363 } 364 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 365 if (!tmp) 366 goto fail_nomem; 367 *tmp = *mpnt; 368 INIT_LIST_HEAD(&tmp->anon_vma_chain); 369 pol = mpol_dup(vma_policy(mpnt)); 370 retval = PTR_ERR(pol); 371 if (IS_ERR(pol)) 372 goto fail_nomem_policy; 373 vma_set_policy(tmp, pol); 374 tmp->vm_mm = mm; 375 if (anon_vma_fork(tmp, mpnt)) 376 goto fail_nomem_anon_vma_fork; 377 tmp->vm_flags &= ~VM_LOCKED; 378 tmp->vm_next = tmp->vm_prev = NULL; 379 file = tmp->vm_file; 380 if (file) { 381 struct inode *inode = file->f_path.dentry->d_inode; 382 struct address_space *mapping = file->f_mapping; 383 384 get_file(file); 385 if (tmp->vm_flags & VM_DENYWRITE) 386 atomic_dec(&inode->i_writecount); 387 mutex_lock(&mapping->i_mmap_mutex); 388 if (tmp->vm_flags & VM_SHARED) 389 mapping->i_mmap_writable++; 390 flush_dcache_mmap_lock(mapping); 391 /* insert tmp into the share list, just after mpnt */ 392 vma_prio_tree_add(tmp, mpnt); 393 flush_dcache_mmap_unlock(mapping); 394 mutex_unlock(&mapping->i_mmap_mutex); 395 } 396 397 /* 398 * Clear hugetlb-related page reserves for children. This only 399 * affects MAP_PRIVATE mappings. Faults generated by the child 400 * are not guaranteed to succeed, even if read-only 401 */ 402 if (is_vm_hugetlb_page(tmp)) 403 reset_vma_resv_huge_pages(tmp); 404 405 /* 406 * Link in the new vma and copy the page table entries. 407 */ 408 *pprev = tmp; 409 pprev = &tmp->vm_next; 410 tmp->vm_prev = prev; 411 prev = tmp; 412 413 __vma_link_rb(mm, tmp, rb_link, rb_parent); 414 rb_link = &tmp->vm_rb.rb_right; 415 rb_parent = &tmp->vm_rb; 416 417 mm->map_count++; 418 retval = copy_page_range(mm, oldmm, mpnt); 419 420 if (tmp->vm_ops && tmp->vm_ops->open) 421 tmp->vm_ops->open(tmp); 422 423 if (retval) 424 goto out; 425 } 426 /* a new mm has just been created */ 427 arch_dup_mmap(oldmm, mm); 428 retval = 0; 429 out: 430 up_write(&mm->mmap_sem); 431 flush_tlb_mm(oldmm); 432 up_write(&oldmm->mmap_sem); 433 return retval; 434 fail_nomem_anon_vma_fork: 435 mpol_put(pol); 436 fail_nomem_policy: 437 kmem_cache_free(vm_area_cachep, tmp); 438 fail_nomem: 439 retval = -ENOMEM; 440 vm_unacct_memory(charge); 441 goto out; 442 } 443 444 static inline int mm_alloc_pgd(struct mm_struct *mm) 445 { 446 mm->pgd = pgd_alloc(mm); 447 if (unlikely(!mm->pgd)) 448 return -ENOMEM; 449 return 0; 450 } 451 452 static inline void mm_free_pgd(struct mm_struct *mm) 453 { 454 pgd_free(mm, mm->pgd); 455 } 456 #else 457 #define dup_mmap(mm, oldmm) (0) 458 #define mm_alloc_pgd(mm) (0) 459 #define mm_free_pgd(mm) 460 #endif /* CONFIG_MMU */ 461 462 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); 463 464 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL)) 465 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) 466 467 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT; 468 469 static int __init coredump_filter_setup(char *s) 470 { 471 default_dump_filter = 472 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) & 473 MMF_DUMP_FILTER_MASK; 474 return 1; 475 } 476 477 __setup("coredump_filter=", coredump_filter_setup); 478 479 #include <linux/init_task.h> 480 481 static void mm_init_aio(struct mm_struct *mm) 482 { 483 #ifdef CONFIG_AIO 484 spin_lock_init(&mm->ioctx_lock); 485 INIT_HLIST_HEAD(&mm->ioctx_list); 486 #endif 487 } 488 489 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p) 490 { 491 atomic_set(&mm->mm_users, 1); 492 atomic_set(&mm->mm_count, 1); 493 init_rwsem(&mm->mmap_sem); 494 INIT_LIST_HEAD(&mm->mmlist); 495 mm->flags = (current->mm) ? 496 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter; 497 mm->core_state = NULL; 498 mm->nr_ptes = 0; 499 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat)); 500 spin_lock_init(&mm->page_table_lock); 501 mm->free_area_cache = TASK_UNMAPPED_BASE; 502 mm->cached_hole_size = ~0UL; 503 mm_init_aio(mm); 504 mm_init_owner(mm, p); 505 506 if (likely(!mm_alloc_pgd(mm))) { 507 mm->def_flags = 0; 508 mmu_notifier_mm_init(mm); 509 return mm; 510 } 511 512 free_mm(mm); 513 return NULL; 514 } 515 516 static void check_mm(struct mm_struct *mm) 517 { 518 int i; 519 520 for (i = 0; i < NR_MM_COUNTERS; i++) { 521 long x = atomic_long_read(&mm->rss_stat.count[i]); 522 523 if (unlikely(x)) 524 printk(KERN_ALERT "BUG: Bad rss-counter state " 525 "mm:%p idx:%d val:%ld\n", mm, i, x); 526 } 527 528 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 529 VM_BUG_ON(mm->pmd_huge_pte); 530 #endif 531 } 532 533 /* 534 * Allocate and initialize an mm_struct. 535 */ 536 struct mm_struct *mm_alloc(void) 537 { 538 struct mm_struct *mm; 539 540 mm = allocate_mm(); 541 if (!mm) 542 return NULL; 543 544 memset(mm, 0, sizeof(*mm)); 545 mm_init_cpumask(mm); 546 return mm_init(mm, current); 547 } 548 549 /* 550 * Called when the last reference to the mm 551 * is dropped: either by a lazy thread or by 552 * mmput. Free the page directory and the mm. 553 */ 554 void __mmdrop(struct mm_struct *mm) 555 { 556 BUG_ON(mm == &init_mm); 557 mm_free_pgd(mm); 558 destroy_context(mm); 559 mmu_notifier_mm_destroy(mm); 560 check_mm(mm); 561 free_mm(mm); 562 } 563 EXPORT_SYMBOL_GPL(__mmdrop); 564 565 /* 566 * Decrement the use count and release all resources for an mm. 567 */ 568 void mmput(struct mm_struct *mm) 569 { 570 might_sleep(); 571 572 if (atomic_dec_and_test(&mm->mm_users)) { 573 exit_aio(mm); 574 ksm_exit(mm); 575 khugepaged_exit(mm); /* must run before exit_mmap */ 576 exit_mmap(mm); 577 set_mm_exe_file(mm, NULL); 578 if (!list_empty(&mm->mmlist)) { 579 spin_lock(&mmlist_lock); 580 list_del(&mm->mmlist); 581 spin_unlock(&mmlist_lock); 582 } 583 put_swap_token(mm); 584 if (mm->binfmt) 585 module_put(mm->binfmt->module); 586 mmdrop(mm); 587 } 588 } 589 EXPORT_SYMBOL_GPL(mmput); 590 591 /* 592 * We added or removed a vma mapping the executable. The vmas are only mapped 593 * during exec and are not mapped with the mmap system call. 594 * Callers must hold down_write() on the mm's mmap_sem for these 595 */ 596 void added_exe_file_vma(struct mm_struct *mm) 597 { 598 mm->num_exe_file_vmas++; 599 } 600 601 void removed_exe_file_vma(struct mm_struct *mm) 602 { 603 mm->num_exe_file_vmas--; 604 if ((mm->num_exe_file_vmas == 0) && mm->exe_file) { 605 fput(mm->exe_file); 606 mm->exe_file = NULL; 607 } 608 609 } 610 611 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 612 { 613 if (new_exe_file) 614 get_file(new_exe_file); 615 if (mm->exe_file) 616 fput(mm->exe_file); 617 mm->exe_file = new_exe_file; 618 mm->num_exe_file_vmas = 0; 619 } 620 621 struct file *get_mm_exe_file(struct mm_struct *mm) 622 { 623 struct file *exe_file; 624 625 /* We need mmap_sem to protect against races with removal of 626 * VM_EXECUTABLE vmas */ 627 down_read(&mm->mmap_sem); 628 exe_file = mm->exe_file; 629 if (exe_file) 630 get_file(exe_file); 631 up_read(&mm->mmap_sem); 632 return exe_file; 633 } 634 635 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm) 636 { 637 /* It's safe to write the exe_file pointer without exe_file_lock because 638 * this is called during fork when the task is not yet in /proc */ 639 newmm->exe_file = get_mm_exe_file(oldmm); 640 } 641 642 /** 643 * get_task_mm - acquire a reference to the task's mm 644 * 645 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning 646 * this kernel workthread has transiently adopted a user mm with use_mm, 647 * to do its AIO) is not set and if so returns a reference to it, after 648 * bumping up the use count. User must release the mm via mmput() 649 * after use. Typically used by /proc and ptrace. 650 */ 651 struct mm_struct *get_task_mm(struct task_struct *task) 652 { 653 struct mm_struct *mm; 654 655 task_lock(task); 656 mm = task->mm; 657 if (mm) { 658 if (task->flags & PF_KTHREAD) 659 mm = NULL; 660 else 661 atomic_inc(&mm->mm_users); 662 } 663 task_unlock(task); 664 return mm; 665 } 666 EXPORT_SYMBOL_GPL(get_task_mm); 667 668 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode) 669 { 670 struct mm_struct *mm; 671 int err; 672 673 err = mutex_lock_killable(&task->signal->cred_guard_mutex); 674 if (err) 675 return ERR_PTR(err); 676 677 mm = get_task_mm(task); 678 if (mm && mm != current->mm && 679 !ptrace_may_access(task, mode)) { 680 mmput(mm); 681 mm = ERR_PTR(-EACCES); 682 } 683 mutex_unlock(&task->signal->cred_guard_mutex); 684 685 return mm; 686 } 687 688 static void complete_vfork_done(struct task_struct *tsk) 689 { 690 struct completion *vfork; 691 692 task_lock(tsk); 693 vfork = tsk->vfork_done; 694 if (likely(vfork)) { 695 tsk->vfork_done = NULL; 696 complete(vfork); 697 } 698 task_unlock(tsk); 699 } 700 701 static int wait_for_vfork_done(struct task_struct *child, 702 struct completion *vfork) 703 { 704 int killed; 705 706 freezer_do_not_count(); 707 killed = wait_for_completion_killable(vfork); 708 freezer_count(); 709 710 if (killed) { 711 task_lock(child); 712 child->vfork_done = NULL; 713 task_unlock(child); 714 } 715 716 put_task_struct(child); 717 return killed; 718 } 719 720 /* Please note the differences between mmput and mm_release. 721 * mmput is called whenever we stop holding onto a mm_struct, 722 * error success whatever. 723 * 724 * mm_release is called after a mm_struct has been removed 725 * from the current process. 726 * 727 * This difference is important for error handling, when we 728 * only half set up a mm_struct for a new process and need to restore 729 * the old one. Because we mmput the new mm_struct before 730 * restoring the old one. . . 731 * Eric Biederman 10 January 1998 732 */ 733 void mm_release(struct task_struct *tsk, struct mm_struct *mm) 734 { 735 /* Get rid of any futexes when releasing the mm */ 736 #ifdef CONFIG_FUTEX 737 if (unlikely(tsk->robust_list)) { 738 exit_robust_list(tsk); 739 tsk->robust_list = NULL; 740 } 741 #ifdef CONFIG_COMPAT 742 if (unlikely(tsk->compat_robust_list)) { 743 compat_exit_robust_list(tsk); 744 tsk->compat_robust_list = NULL; 745 } 746 #endif 747 if (unlikely(!list_empty(&tsk->pi_state_list))) 748 exit_pi_state_list(tsk); 749 #endif 750 751 /* Get rid of any cached register state */ 752 deactivate_mm(tsk, mm); 753 754 if (tsk->vfork_done) 755 complete_vfork_done(tsk); 756 757 /* 758 * If we're exiting normally, clear a user-space tid field if 759 * requested. We leave this alone when dying by signal, to leave 760 * the value intact in a core dump, and to save the unnecessary 761 * trouble, say, a killed vfork parent shouldn't touch this mm. 762 * Userland only wants this done for a sys_exit. 763 */ 764 if (tsk->clear_child_tid) { 765 if (!(tsk->flags & PF_SIGNALED) && 766 atomic_read(&mm->mm_users) > 1) { 767 /* 768 * We don't check the error code - if userspace has 769 * not set up a proper pointer then tough luck. 770 */ 771 put_user(0, tsk->clear_child_tid); 772 sys_futex(tsk->clear_child_tid, FUTEX_WAKE, 773 1, NULL, NULL, 0); 774 } 775 tsk->clear_child_tid = NULL; 776 } 777 } 778 779 /* 780 * Allocate a new mm structure and copy contents from the 781 * mm structure of the passed in task structure. 782 */ 783 struct mm_struct *dup_mm(struct task_struct *tsk) 784 { 785 struct mm_struct *mm, *oldmm = current->mm; 786 int err; 787 788 if (!oldmm) 789 return NULL; 790 791 mm = allocate_mm(); 792 if (!mm) 793 goto fail_nomem; 794 795 memcpy(mm, oldmm, sizeof(*mm)); 796 mm_init_cpumask(mm); 797 798 /* Initializing for Swap token stuff */ 799 mm->token_priority = 0; 800 mm->last_interval = 0; 801 802 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 803 mm->pmd_huge_pte = NULL; 804 #endif 805 806 if (!mm_init(mm, tsk)) 807 goto fail_nomem; 808 809 if (init_new_context(tsk, mm)) 810 goto fail_nocontext; 811 812 dup_mm_exe_file(oldmm, mm); 813 814 err = dup_mmap(mm, oldmm); 815 if (err) 816 goto free_pt; 817 818 mm->hiwater_rss = get_mm_rss(mm); 819 mm->hiwater_vm = mm->total_vm; 820 821 if (mm->binfmt && !try_module_get(mm->binfmt->module)) 822 goto free_pt; 823 824 return mm; 825 826 free_pt: 827 /* don't put binfmt in mmput, we haven't got module yet */ 828 mm->binfmt = NULL; 829 mmput(mm); 830 831 fail_nomem: 832 return NULL; 833 834 fail_nocontext: 835 /* 836 * If init_new_context() failed, we cannot use mmput() to free the mm 837 * because it calls destroy_context() 838 */ 839 mm_free_pgd(mm); 840 free_mm(mm); 841 return NULL; 842 } 843 844 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk) 845 { 846 struct mm_struct *mm, *oldmm; 847 int retval; 848 849 tsk->min_flt = tsk->maj_flt = 0; 850 tsk->nvcsw = tsk->nivcsw = 0; 851 #ifdef CONFIG_DETECT_HUNG_TASK 852 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; 853 #endif 854 855 tsk->mm = NULL; 856 tsk->active_mm = NULL; 857 858 /* 859 * Are we cloning a kernel thread? 860 * 861 * We need to steal a active VM for that.. 862 */ 863 oldmm = current->mm; 864 if (!oldmm) 865 return 0; 866 867 if (clone_flags & CLONE_VM) { 868 atomic_inc(&oldmm->mm_users); 869 mm = oldmm; 870 goto good_mm; 871 } 872 873 retval = -ENOMEM; 874 mm = dup_mm(tsk); 875 if (!mm) 876 goto fail_nomem; 877 878 good_mm: 879 /* Initializing for Swap token stuff */ 880 mm->token_priority = 0; 881 mm->last_interval = 0; 882 883 tsk->mm = mm; 884 tsk->active_mm = mm; 885 return 0; 886 887 fail_nomem: 888 return retval; 889 } 890 891 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 892 { 893 struct fs_struct *fs = current->fs; 894 if (clone_flags & CLONE_FS) { 895 /* tsk->fs is already what we want */ 896 spin_lock(&fs->lock); 897 if (fs->in_exec) { 898 spin_unlock(&fs->lock); 899 return -EAGAIN; 900 } 901 fs->users++; 902 spin_unlock(&fs->lock); 903 return 0; 904 } 905 tsk->fs = copy_fs_struct(fs); 906 if (!tsk->fs) 907 return -ENOMEM; 908 return 0; 909 } 910 911 static int copy_files(unsigned long clone_flags, struct task_struct *tsk) 912 { 913 struct files_struct *oldf, *newf; 914 int error = 0; 915 916 /* 917 * A background process may not have any files ... 918 */ 919 oldf = current->files; 920 if (!oldf) 921 goto out; 922 923 if (clone_flags & CLONE_FILES) { 924 atomic_inc(&oldf->count); 925 goto out; 926 } 927 928 newf = dup_fd(oldf, &error); 929 if (!newf) 930 goto out; 931 932 tsk->files = newf; 933 error = 0; 934 out: 935 return error; 936 } 937 938 static int copy_io(unsigned long clone_flags, struct task_struct *tsk) 939 { 940 #ifdef CONFIG_BLOCK 941 struct io_context *ioc = current->io_context; 942 struct io_context *new_ioc; 943 944 if (!ioc) 945 return 0; 946 /* 947 * Share io context with parent, if CLONE_IO is set 948 */ 949 if (clone_flags & CLONE_IO) { 950 tsk->io_context = ioc_task_link(ioc); 951 if (unlikely(!tsk->io_context)) 952 return -ENOMEM; 953 } else if (ioprio_valid(ioc->ioprio)) { 954 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE); 955 if (unlikely(!new_ioc)) 956 return -ENOMEM; 957 958 new_ioc->ioprio = ioc->ioprio; 959 put_io_context(new_ioc); 960 } 961 #endif 962 return 0; 963 } 964 965 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk) 966 { 967 struct sighand_struct *sig; 968 969 if (clone_flags & CLONE_SIGHAND) { 970 atomic_inc(¤t->sighand->count); 971 return 0; 972 } 973 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 974 rcu_assign_pointer(tsk->sighand, sig); 975 if (!sig) 976 return -ENOMEM; 977 atomic_set(&sig->count, 1); 978 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 979 return 0; 980 } 981 982 void __cleanup_sighand(struct sighand_struct *sighand) 983 { 984 if (atomic_dec_and_test(&sighand->count)) { 985 signalfd_cleanup(sighand); 986 kmem_cache_free(sighand_cachep, sighand); 987 } 988 } 989 990 991 /* 992 * Initialize POSIX timer handling for a thread group. 993 */ 994 static void posix_cpu_timers_init_group(struct signal_struct *sig) 995 { 996 unsigned long cpu_limit; 997 998 /* Thread group counters. */ 999 thread_group_cputime_init(sig); 1000 1001 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); 1002 if (cpu_limit != RLIM_INFINITY) { 1003 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit); 1004 sig->cputimer.running = 1; 1005 } 1006 1007 /* The timer lists. */ 1008 INIT_LIST_HEAD(&sig->cpu_timers[0]); 1009 INIT_LIST_HEAD(&sig->cpu_timers[1]); 1010 INIT_LIST_HEAD(&sig->cpu_timers[2]); 1011 } 1012 1013 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk) 1014 { 1015 struct signal_struct *sig; 1016 1017 if (clone_flags & CLONE_THREAD) 1018 return 0; 1019 1020 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL); 1021 tsk->signal = sig; 1022 if (!sig) 1023 return -ENOMEM; 1024 1025 sig->nr_threads = 1; 1026 atomic_set(&sig->live, 1); 1027 atomic_set(&sig->sigcnt, 1); 1028 init_waitqueue_head(&sig->wait_chldexit); 1029 if (clone_flags & CLONE_NEWPID) 1030 sig->flags |= SIGNAL_UNKILLABLE; 1031 sig->curr_target = tsk; 1032 init_sigpending(&sig->shared_pending); 1033 INIT_LIST_HEAD(&sig->posix_timers); 1034 1035 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1036 sig->real_timer.function = it_real_fn; 1037 1038 task_lock(current->group_leader); 1039 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 1040 task_unlock(current->group_leader); 1041 1042 posix_cpu_timers_init_group(sig); 1043 1044 tty_audit_fork(sig); 1045 sched_autogroup_fork(sig); 1046 1047 #ifdef CONFIG_CGROUPS 1048 init_rwsem(&sig->group_rwsem); 1049 #endif 1050 1051 sig->oom_adj = current->signal->oom_adj; 1052 sig->oom_score_adj = current->signal->oom_score_adj; 1053 sig->oom_score_adj_min = current->signal->oom_score_adj_min; 1054 1055 sig->has_child_subreaper = current->signal->has_child_subreaper || 1056 current->signal->is_child_subreaper; 1057 1058 mutex_init(&sig->cred_guard_mutex); 1059 1060 return 0; 1061 } 1062 1063 static void copy_flags(unsigned long clone_flags, struct task_struct *p) 1064 { 1065 unsigned long new_flags = p->flags; 1066 1067 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER); 1068 new_flags |= PF_FORKNOEXEC; 1069 p->flags = new_flags; 1070 } 1071 1072 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr) 1073 { 1074 current->clear_child_tid = tidptr; 1075 1076 return task_pid_vnr(current); 1077 } 1078 1079 static void rt_mutex_init_task(struct task_struct *p) 1080 { 1081 raw_spin_lock_init(&p->pi_lock); 1082 #ifdef CONFIG_RT_MUTEXES 1083 plist_head_init(&p->pi_waiters); 1084 p->pi_blocked_on = NULL; 1085 #endif 1086 } 1087 1088 #ifdef CONFIG_MM_OWNER 1089 void mm_init_owner(struct mm_struct *mm, struct task_struct *p) 1090 { 1091 mm->owner = p; 1092 } 1093 #endif /* CONFIG_MM_OWNER */ 1094 1095 /* 1096 * Initialize POSIX timer handling for a single task. 1097 */ 1098 static void posix_cpu_timers_init(struct task_struct *tsk) 1099 { 1100 tsk->cputime_expires.prof_exp = 0; 1101 tsk->cputime_expires.virt_exp = 0; 1102 tsk->cputime_expires.sched_exp = 0; 1103 INIT_LIST_HEAD(&tsk->cpu_timers[0]); 1104 INIT_LIST_HEAD(&tsk->cpu_timers[1]); 1105 INIT_LIST_HEAD(&tsk->cpu_timers[2]); 1106 } 1107 1108 /* 1109 * This creates a new process as a copy of the old one, 1110 * but does not actually start it yet. 1111 * 1112 * It copies the registers, and all the appropriate 1113 * parts of the process environment (as per the clone 1114 * flags). The actual kick-off is left to the caller. 1115 */ 1116 static struct task_struct *copy_process(unsigned long clone_flags, 1117 unsigned long stack_start, 1118 struct pt_regs *regs, 1119 unsigned long stack_size, 1120 int __user *child_tidptr, 1121 struct pid *pid, 1122 int trace) 1123 { 1124 int retval; 1125 struct task_struct *p; 1126 int cgroup_callbacks_done = 0; 1127 1128 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 1129 return ERR_PTR(-EINVAL); 1130 1131 /* 1132 * Thread groups must share signals as well, and detached threads 1133 * can only be started up within the thread group. 1134 */ 1135 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 1136 return ERR_PTR(-EINVAL); 1137 1138 /* 1139 * Shared signal handlers imply shared VM. By way of the above, 1140 * thread groups also imply shared VM. Blocking this case allows 1141 * for various simplifications in other code. 1142 */ 1143 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 1144 return ERR_PTR(-EINVAL); 1145 1146 /* 1147 * Siblings of global init remain as zombies on exit since they are 1148 * not reaped by their parent (swapper). To solve this and to avoid 1149 * multi-rooted process trees, prevent global and container-inits 1150 * from creating siblings. 1151 */ 1152 if ((clone_flags & CLONE_PARENT) && 1153 current->signal->flags & SIGNAL_UNKILLABLE) 1154 return ERR_PTR(-EINVAL); 1155 1156 retval = security_task_create(clone_flags); 1157 if (retval) 1158 goto fork_out; 1159 1160 retval = -ENOMEM; 1161 p = dup_task_struct(current); 1162 if (!p) 1163 goto fork_out; 1164 1165 ftrace_graph_init_task(p); 1166 1167 rt_mutex_init_task(p); 1168 1169 #ifdef CONFIG_PROVE_LOCKING 1170 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); 1171 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 1172 #endif 1173 retval = -EAGAIN; 1174 if (atomic_read(&p->real_cred->user->processes) >= 1175 task_rlimit(p, RLIMIT_NPROC)) { 1176 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) && 1177 p->real_cred->user != INIT_USER) 1178 goto bad_fork_free; 1179 } 1180 current->flags &= ~PF_NPROC_EXCEEDED; 1181 1182 retval = copy_creds(p, clone_flags); 1183 if (retval < 0) 1184 goto bad_fork_free; 1185 1186 /* 1187 * If multiple threads are within copy_process(), then this check 1188 * triggers too late. This doesn't hurt, the check is only there 1189 * to stop root fork bombs. 1190 */ 1191 retval = -EAGAIN; 1192 if (nr_threads >= max_threads) 1193 goto bad_fork_cleanup_count; 1194 1195 if (!try_module_get(task_thread_info(p)->exec_domain->module)) 1196 goto bad_fork_cleanup_count; 1197 1198 p->did_exec = 0; 1199 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */ 1200 copy_flags(clone_flags, p); 1201 INIT_LIST_HEAD(&p->children); 1202 INIT_LIST_HEAD(&p->sibling); 1203 rcu_copy_process(p); 1204 p->vfork_done = NULL; 1205 spin_lock_init(&p->alloc_lock); 1206 1207 init_sigpending(&p->pending); 1208 1209 p->utime = p->stime = p->gtime = 0; 1210 p->utimescaled = p->stimescaled = 0; 1211 #ifndef CONFIG_VIRT_CPU_ACCOUNTING 1212 p->prev_utime = p->prev_stime = 0; 1213 #endif 1214 #if defined(SPLIT_RSS_COUNTING) 1215 memset(&p->rss_stat, 0, sizeof(p->rss_stat)); 1216 #endif 1217 1218 p->default_timer_slack_ns = current->timer_slack_ns; 1219 1220 task_io_accounting_init(&p->ioac); 1221 acct_clear_integrals(p); 1222 1223 posix_cpu_timers_init(p); 1224 1225 do_posix_clock_monotonic_gettime(&p->start_time); 1226 p->real_start_time = p->start_time; 1227 monotonic_to_bootbased(&p->real_start_time); 1228 p->io_context = NULL; 1229 p->audit_context = NULL; 1230 if (clone_flags & CLONE_THREAD) 1231 threadgroup_change_begin(current); 1232 cgroup_fork(p); 1233 #ifdef CONFIG_NUMA 1234 p->mempolicy = mpol_dup(p->mempolicy); 1235 if (IS_ERR(p->mempolicy)) { 1236 retval = PTR_ERR(p->mempolicy); 1237 p->mempolicy = NULL; 1238 goto bad_fork_cleanup_cgroup; 1239 } 1240 mpol_fix_fork_child_flag(p); 1241 #endif 1242 #ifdef CONFIG_CPUSETS 1243 p->cpuset_mem_spread_rotor = NUMA_NO_NODE; 1244 p->cpuset_slab_spread_rotor = NUMA_NO_NODE; 1245 seqcount_init(&p->mems_allowed_seq); 1246 #endif 1247 #ifdef CONFIG_TRACE_IRQFLAGS 1248 p->irq_events = 0; 1249 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW 1250 p->hardirqs_enabled = 1; 1251 #else 1252 p->hardirqs_enabled = 0; 1253 #endif 1254 p->hardirq_enable_ip = 0; 1255 p->hardirq_enable_event = 0; 1256 p->hardirq_disable_ip = _THIS_IP_; 1257 p->hardirq_disable_event = 0; 1258 p->softirqs_enabled = 1; 1259 p->softirq_enable_ip = _THIS_IP_; 1260 p->softirq_enable_event = 0; 1261 p->softirq_disable_ip = 0; 1262 p->softirq_disable_event = 0; 1263 p->hardirq_context = 0; 1264 p->softirq_context = 0; 1265 #endif 1266 #ifdef CONFIG_LOCKDEP 1267 p->lockdep_depth = 0; /* no locks held yet */ 1268 p->curr_chain_key = 0; 1269 p->lockdep_recursion = 0; 1270 #endif 1271 1272 #ifdef CONFIG_DEBUG_MUTEXES 1273 p->blocked_on = NULL; /* not blocked yet */ 1274 #endif 1275 #ifdef CONFIG_CGROUP_MEM_RES_CTLR 1276 p->memcg_batch.do_batch = 0; 1277 p->memcg_batch.memcg = NULL; 1278 #endif 1279 1280 /* Perform scheduler related setup. Assign this task to a CPU. */ 1281 sched_fork(p); 1282 1283 retval = perf_event_init_task(p); 1284 if (retval) 1285 goto bad_fork_cleanup_policy; 1286 retval = audit_alloc(p); 1287 if (retval) 1288 goto bad_fork_cleanup_policy; 1289 /* copy all the process information */ 1290 retval = copy_semundo(clone_flags, p); 1291 if (retval) 1292 goto bad_fork_cleanup_audit; 1293 retval = copy_files(clone_flags, p); 1294 if (retval) 1295 goto bad_fork_cleanup_semundo; 1296 retval = copy_fs(clone_flags, p); 1297 if (retval) 1298 goto bad_fork_cleanup_files; 1299 retval = copy_sighand(clone_flags, p); 1300 if (retval) 1301 goto bad_fork_cleanup_fs; 1302 retval = copy_signal(clone_flags, p); 1303 if (retval) 1304 goto bad_fork_cleanup_sighand; 1305 retval = copy_mm(clone_flags, p); 1306 if (retval) 1307 goto bad_fork_cleanup_signal; 1308 retval = copy_namespaces(clone_flags, p); 1309 if (retval) 1310 goto bad_fork_cleanup_mm; 1311 retval = copy_io(clone_flags, p); 1312 if (retval) 1313 goto bad_fork_cleanup_namespaces; 1314 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs); 1315 if (retval) 1316 goto bad_fork_cleanup_io; 1317 1318 if (pid != &init_struct_pid) { 1319 retval = -ENOMEM; 1320 pid = alloc_pid(p->nsproxy->pid_ns); 1321 if (!pid) 1322 goto bad_fork_cleanup_io; 1323 } 1324 1325 p->pid = pid_nr(pid); 1326 p->tgid = p->pid; 1327 if (clone_flags & CLONE_THREAD) 1328 p->tgid = current->tgid; 1329 1330 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; 1331 /* 1332 * Clear TID on mm_release()? 1333 */ 1334 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL; 1335 #ifdef CONFIG_BLOCK 1336 p->plug = NULL; 1337 #endif 1338 #ifdef CONFIG_FUTEX 1339 p->robust_list = NULL; 1340 #ifdef CONFIG_COMPAT 1341 p->compat_robust_list = NULL; 1342 #endif 1343 INIT_LIST_HEAD(&p->pi_state_list); 1344 p->pi_state_cache = NULL; 1345 #endif 1346 /* 1347 * sigaltstack should be cleared when sharing the same VM 1348 */ 1349 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM) 1350 p->sas_ss_sp = p->sas_ss_size = 0; 1351 1352 /* 1353 * Syscall tracing and stepping should be turned off in the 1354 * child regardless of CLONE_PTRACE. 1355 */ 1356 user_disable_single_step(p); 1357 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); 1358 #ifdef TIF_SYSCALL_EMU 1359 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); 1360 #endif 1361 clear_all_latency_tracing(p); 1362 1363 /* ok, now we should be set up.. */ 1364 if (clone_flags & CLONE_THREAD) 1365 p->exit_signal = -1; 1366 else if (clone_flags & CLONE_PARENT) 1367 p->exit_signal = current->group_leader->exit_signal; 1368 else 1369 p->exit_signal = (clone_flags & CSIGNAL); 1370 1371 p->pdeath_signal = 0; 1372 p->exit_state = 0; 1373 1374 p->nr_dirtied = 0; 1375 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10); 1376 p->dirty_paused_when = 0; 1377 1378 /* 1379 * Ok, make it visible to the rest of the system. 1380 * We dont wake it up yet. 1381 */ 1382 p->group_leader = p; 1383 INIT_LIST_HEAD(&p->thread_group); 1384 1385 /* Now that the task is set up, run cgroup callbacks if 1386 * necessary. We need to run them before the task is visible 1387 * on the tasklist. */ 1388 cgroup_fork_callbacks(p); 1389 cgroup_callbacks_done = 1; 1390 1391 /* Need tasklist lock for parent etc handling! */ 1392 write_lock_irq(&tasklist_lock); 1393 1394 /* CLONE_PARENT re-uses the old parent */ 1395 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { 1396 p->real_parent = current->real_parent; 1397 p->parent_exec_id = current->parent_exec_id; 1398 } else { 1399 p->real_parent = current; 1400 p->parent_exec_id = current->self_exec_id; 1401 } 1402 1403 spin_lock(¤t->sighand->siglock); 1404 1405 /* 1406 * Process group and session signals need to be delivered to just the 1407 * parent before the fork or both the parent and the child after the 1408 * fork. Restart if a signal comes in before we add the new process to 1409 * it's process group. 1410 * A fatal signal pending means that current will exit, so the new 1411 * thread can't slip out of an OOM kill (or normal SIGKILL). 1412 */ 1413 recalc_sigpending(); 1414 if (signal_pending(current)) { 1415 spin_unlock(¤t->sighand->siglock); 1416 write_unlock_irq(&tasklist_lock); 1417 retval = -ERESTARTNOINTR; 1418 goto bad_fork_free_pid; 1419 } 1420 1421 if (clone_flags & CLONE_THREAD) { 1422 current->signal->nr_threads++; 1423 atomic_inc(¤t->signal->live); 1424 atomic_inc(¤t->signal->sigcnt); 1425 p->group_leader = current->group_leader; 1426 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group); 1427 } 1428 1429 if (likely(p->pid)) { 1430 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace); 1431 1432 if (thread_group_leader(p)) { 1433 if (is_child_reaper(pid)) 1434 p->nsproxy->pid_ns->child_reaper = p; 1435 1436 p->signal->leader_pid = pid; 1437 p->signal->tty = tty_kref_get(current->signal->tty); 1438 attach_pid(p, PIDTYPE_PGID, task_pgrp(current)); 1439 attach_pid(p, PIDTYPE_SID, task_session(current)); 1440 list_add_tail(&p->sibling, &p->real_parent->children); 1441 list_add_tail_rcu(&p->tasks, &init_task.tasks); 1442 __this_cpu_inc(process_counts); 1443 } 1444 attach_pid(p, PIDTYPE_PID, pid); 1445 nr_threads++; 1446 } 1447 1448 total_forks++; 1449 spin_unlock(¤t->sighand->siglock); 1450 write_unlock_irq(&tasklist_lock); 1451 proc_fork_connector(p); 1452 cgroup_post_fork(p); 1453 if (clone_flags & CLONE_THREAD) 1454 threadgroup_change_end(current); 1455 perf_event_fork(p); 1456 1457 trace_task_newtask(p, clone_flags); 1458 1459 return p; 1460 1461 bad_fork_free_pid: 1462 if (pid != &init_struct_pid) 1463 free_pid(pid); 1464 bad_fork_cleanup_io: 1465 if (p->io_context) 1466 exit_io_context(p); 1467 bad_fork_cleanup_namespaces: 1468 if (unlikely(clone_flags & CLONE_NEWPID)) 1469 pid_ns_release_proc(p->nsproxy->pid_ns); 1470 exit_task_namespaces(p); 1471 bad_fork_cleanup_mm: 1472 if (p->mm) 1473 mmput(p->mm); 1474 bad_fork_cleanup_signal: 1475 if (!(clone_flags & CLONE_THREAD)) 1476 free_signal_struct(p->signal); 1477 bad_fork_cleanup_sighand: 1478 __cleanup_sighand(p->sighand); 1479 bad_fork_cleanup_fs: 1480 exit_fs(p); /* blocking */ 1481 bad_fork_cleanup_files: 1482 exit_files(p); /* blocking */ 1483 bad_fork_cleanup_semundo: 1484 exit_sem(p); 1485 bad_fork_cleanup_audit: 1486 audit_free(p); 1487 bad_fork_cleanup_policy: 1488 perf_event_free_task(p); 1489 #ifdef CONFIG_NUMA 1490 mpol_put(p->mempolicy); 1491 bad_fork_cleanup_cgroup: 1492 #endif 1493 if (clone_flags & CLONE_THREAD) 1494 threadgroup_change_end(current); 1495 cgroup_exit(p, cgroup_callbacks_done); 1496 delayacct_tsk_free(p); 1497 module_put(task_thread_info(p)->exec_domain->module); 1498 bad_fork_cleanup_count: 1499 atomic_dec(&p->cred->user->processes); 1500 exit_creds(p); 1501 bad_fork_free: 1502 free_task(p); 1503 fork_out: 1504 return ERR_PTR(retval); 1505 } 1506 1507 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs) 1508 { 1509 memset(regs, 0, sizeof(struct pt_regs)); 1510 return regs; 1511 } 1512 1513 static inline void init_idle_pids(struct pid_link *links) 1514 { 1515 enum pid_type type; 1516 1517 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 1518 INIT_HLIST_NODE(&links[type].node); /* not really needed */ 1519 links[type].pid = &init_struct_pid; 1520 } 1521 } 1522 1523 struct task_struct * __cpuinit fork_idle(int cpu) 1524 { 1525 struct task_struct *task; 1526 struct pt_regs regs; 1527 1528 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, 1529 &init_struct_pid, 0); 1530 if (!IS_ERR(task)) { 1531 init_idle_pids(task->pids); 1532 init_idle(task, cpu); 1533 } 1534 1535 return task; 1536 } 1537 1538 /* 1539 * Ok, this is the main fork-routine. 1540 * 1541 * It copies the process, and if successful kick-starts 1542 * it and waits for it to finish using the VM if required. 1543 */ 1544 long do_fork(unsigned long clone_flags, 1545 unsigned long stack_start, 1546 struct pt_regs *regs, 1547 unsigned long stack_size, 1548 int __user *parent_tidptr, 1549 int __user *child_tidptr) 1550 { 1551 struct task_struct *p; 1552 int trace = 0; 1553 long nr; 1554 1555 /* 1556 * Do some preliminary argument and permissions checking before we 1557 * actually start allocating stuff 1558 */ 1559 if (clone_flags & CLONE_NEWUSER) { 1560 if (clone_flags & CLONE_THREAD) 1561 return -EINVAL; 1562 /* hopefully this check will go away when userns support is 1563 * complete 1564 */ 1565 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) || 1566 !capable(CAP_SETGID)) 1567 return -EPERM; 1568 } 1569 1570 /* 1571 * Determine whether and which event to report to ptracer. When 1572 * called from kernel_thread or CLONE_UNTRACED is explicitly 1573 * requested, no event is reported; otherwise, report if the event 1574 * for the type of forking is enabled. 1575 */ 1576 if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) { 1577 if (clone_flags & CLONE_VFORK) 1578 trace = PTRACE_EVENT_VFORK; 1579 else if ((clone_flags & CSIGNAL) != SIGCHLD) 1580 trace = PTRACE_EVENT_CLONE; 1581 else 1582 trace = PTRACE_EVENT_FORK; 1583 1584 if (likely(!ptrace_event_enabled(current, trace))) 1585 trace = 0; 1586 } 1587 1588 p = copy_process(clone_flags, stack_start, regs, stack_size, 1589 child_tidptr, NULL, trace); 1590 /* 1591 * Do this prior waking up the new thread - the thread pointer 1592 * might get invalid after that point, if the thread exits quickly. 1593 */ 1594 if (!IS_ERR(p)) { 1595 struct completion vfork; 1596 1597 trace_sched_process_fork(current, p); 1598 1599 nr = task_pid_vnr(p); 1600 1601 if (clone_flags & CLONE_PARENT_SETTID) 1602 put_user(nr, parent_tidptr); 1603 1604 if (clone_flags & CLONE_VFORK) { 1605 p->vfork_done = &vfork; 1606 init_completion(&vfork); 1607 get_task_struct(p); 1608 } 1609 1610 wake_up_new_task(p); 1611 1612 /* forking complete and child started to run, tell ptracer */ 1613 if (unlikely(trace)) 1614 ptrace_event(trace, nr); 1615 1616 if (clone_flags & CLONE_VFORK) { 1617 if (!wait_for_vfork_done(p, &vfork)) 1618 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr); 1619 } 1620 } else { 1621 nr = PTR_ERR(p); 1622 } 1623 return nr; 1624 } 1625 1626 #ifndef ARCH_MIN_MMSTRUCT_ALIGN 1627 #define ARCH_MIN_MMSTRUCT_ALIGN 0 1628 #endif 1629 1630 static void sighand_ctor(void *data) 1631 { 1632 struct sighand_struct *sighand = data; 1633 1634 spin_lock_init(&sighand->siglock); 1635 init_waitqueue_head(&sighand->signalfd_wqh); 1636 } 1637 1638 void __init proc_caches_init(void) 1639 { 1640 sighand_cachep = kmem_cache_create("sighand_cache", 1641 sizeof(struct sighand_struct), 0, 1642 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU| 1643 SLAB_NOTRACK, sighand_ctor); 1644 signal_cachep = kmem_cache_create("signal_cache", 1645 sizeof(struct signal_struct), 0, 1646 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1647 files_cachep = kmem_cache_create("files_cache", 1648 sizeof(struct files_struct), 0, 1649 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1650 fs_cachep = kmem_cache_create("fs_cache", 1651 sizeof(struct fs_struct), 0, 1652 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1653 /* 1654 * FIXME! The "sizeof(struct mm_struct)" currently includes the 1655 * whole struct cpumask for the OFFSTACK case. We could change 1656 * this to *only* allocate as much of it as required by the 1657 * maximum number of CPU's we can ever have. The cpumask_allocation 1658 * is at the end of the structure, exactly for that reason. 1659 */ 1660 mm_cachep = kmem_cache_create("mm_struct", 1661 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN, 1662 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1663 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC); 1664 mmap_init(); 1665 nsproxy_cache_init(); 1666 } 1667 1668 /* 1669 * Check constraints on flags passed to the unshare system call. 1670 */ 1671 static int check_unshare_flags(unsigned long unshare_flags) 1672 { 1673 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND| 1674 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM| 1675 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET)) 1676 return -EINVAL; 1677 /* 1678 * Not implemented, but pretend it works if there is nothing to 1679 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND 1680 * needs to unshare vm. 1681 */ 1682 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) { 1683 /* FIXME: get_task_mm() increments ->mm_users */ 1684 if (atomic_read(¤t->mm->mm_users) > 1) 1685 return -EINVAL; 1686 } 1687 1688 return 0; 1689 } 1690 1691 /* 1692 * Unshare the filesystem structure if it is being shared 1693 */ 1694 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) 1695 { 1696 struct fs_struct *fs = current->fs; 1697 1698 if (!(unshare_flags & CLONE_FS) || !fs) 1699 return 0; 1700 1701 /* don't need lock here; in the worst case we'll do useless copy */ 1702 if (fs->users == 1) 1703 return 0; 1704 1705 *new_fsp = copy_fs_struct(fs); 1706 if (!*new_fsp) 1707 return -ENOMEM; 1708 1709 return 0; 1710 } 1711 1712 /* 1713 * Unshare file descriptor table if it is being shared 1714 */ 1715 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp) 1716 { 1717 struct files_struct *fd = current->files; 1718 int error = 0; 1719 1720 if ((unshare_flags & CLONE_FILES) && 1721 (fd && atomic_read(&fd->count) > 1)) { 1722 *new_fdp = dup_fd(fd, &error); 1723 if (!*new_fdp) 1724 return error; 1725 } 1726 1727 return 0; 1728 } 1729 1730 /* 1731 * unshare allows a process to 'unshare' part of the process 1732 * context which was originally shared using clone. copy_* 1733 * functions used by do_fork() cannot be used here directly 1734 * because they modify an inactive task_struct that is being 1735 * constructed. Here we are modifying the current, active, 1736 * task_struct. 1737 */ 1738 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags) 1739 { 1740 struct fs_struct *fs, *new_fs = NULL; 1741 struct files_struct *fd, *new_fd = NULL; 1742 struct nsproxy *new_nsproxy = NULL; 1743 int do_sysvsem = 0; 1744 int err; 1745 1746 err = check_unshare_flags(unshare_flags); 1747 if (err) 1748 goto bad_unshare_out; 1749 1750 /* 1751 * If unsharing namespace, must also unshare filesystem information. 1752 */ 1753 if (unshare_flags & CLONE_NEWNS) 1754 unshare_flags |= CLONE_FS; 1755 /* 1756 * CLONE_NEWIPC must also detach from the undolist: after switching 1757 * to a new ipc namespace, the semaphore arrays from the old 1758 * namespace are unreachable. 1759 */ 1760 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) 1761 do_sysvsem = 1; 1762 err = unshare_fs(unshare_flags, &new_fs); 1763 if (err) 1764 goto bad_unshare_out; 1765 err = unshare_fd(unshare_flags, &new_fd); 1766 if (err) 1767 goto bad_unshare_cleanup_fs; 1768 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs); 1769 if (err) 1770 goto bad_unshare_cleanup_fd; 1771 1772 if (new_fs || new_fd || do_sysvsem || new_nsproxy) { 1773 if (do_sysvsem) { 1774 /* 1775 * CLONE_SYSVSEM is equivalent to sys_exit(). 1776 */ 1777 exit_sem(current); 1778 } 1779 1780 if (new_nsproxy) { 1781 switch_task_namespaces(current, new_nsproxy); 1782 new_nsproxy = NULL; 1783 } 1784 1785 task_lock(current); 1786 1787 if (new_fs) { 1788 fs = current->fs; 1789 spin_lock(&fs->lock); 1790 current->fs = new_fs; 1791 if (--fs->users) 1792 new_fs = NULL; 1793 else 1794 new_fs = fs; 1795 spin_unlock(&fs->lock); 1796 } 1797 1798 if (new_fd) { 1799 fd = current->files; 1800 current->files = new_fd; 1801 new_fd = fd; 1802 } 1803 1804 task_unlock(current); 1805 } 1806 1807 if (new_nsproxy) 1808 put_nsproxy(new_nsproxy); 1809 1810 bad_unshare_cleanup_fd: 1811 if (new_fd) 1812 put_files_struct(new_fd); 1813 1814 bad_unshare_cleanup_fs: 1815 if (new_fs) 1816 free_fs_struct(new_fs); 1817 1818 bad_unshare_out: 1819 return err; 1820 } 1821 1822 /* 1823 * Helper to unshare the files of the current task. 1824 * We don't want to expose copy_files internals to 1825 * the exec layer of the kernel. 1826 */ 1827 1828 int unshare_files(struct files_struct **displaced) 1829 { 1830 struct task_struct *task = current; 1831 struct files_struct *copy = NULL; 1832 int error; 1833 1834 error = unshare_fd(CLONE_FILES, ©); 1835 if (error || !copy) { 1836 *displaced = NULL; 1837 return error; 1838 } 1839 *displaced = task->files; 1840 task_lock(task); 1841 task->files = copy; 1842 task_unlock(task); 1843 return 0; 1844 } 1845