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/mm.h> 32 #include <linux/vmacache.h> 33 #include <linux/nsproxy.h> 34 #include <linux/capability.h> 35 #include <linux/cpu.h> 36 #include <linux/cgroup.h> 37 #include <linux/security.h> 38 #include <linux/hugetlb.h> 39 #include <linux/seccomp.h> 40 #include <linux/swap.h> 41 #include <linux/syscalls.h> 42 #include <linux/jiffies.h> 43 #include <linux/futex.h> 44 #include <linux/compat.h> 45 #include <linux/kthread.h> 46 #include <linux/task_io_accounting_ops.h> 47 #include <linux/rcupdate.h> 48 #include <linux/ptrace.h> 49 #include <linux/mount.h> 50 #include <linux/audit.h> 51 #include <linux/memcontrol.h> 52 #include <linux/ftrace.h> 53 #include <linux/proc_fs.h> 54 #include <linux/profile.h> 55 #include <linux/rmap.h> 56 #include <linux/ksm.h> 57 #include <linux/acct.h> 58 #include <linux/tsacct_kern.h> 59 #include <linux/cn_proc.h> 60 #include <linux/freezer.h> 61 #include <linux/delayacct.h> 62 #include <linux/taskstats_kern.h> 63 #include <linux/random.h> 64 #include <linux/tty.h> 65 #include <linux/blkdev.h> 66 #include <linux/fs_struct.h> 67 #include <linux/magic.h> 68 #include <linux/perf_event.h> 69 #include <linux/posix-timers.h> 70 #include <linux/user-return-notifier.h> 71 #include <linux/oom.h> 72 #include <linux/khugepaged.h> 73 #include <linux/signalfd.h> 74 #include <linux/uprobes.h> 75 #include <linux/aio.h> 76 #include <linux/compiler.h> 77 78 #include <asm/pgtable.h> 79 #include <asm/pgalloc.h> 80 #include <asm/uaccess.h> 81 #include <asm/mmu_context.h> 82 #include <asm/cacheflush.h> 83 #include <asm/tlbflush.h> 84 85 #include <trace/events/sched.h> 86 87 #define CREATE_TRACE_POINTS 88 #include <trace/events/task.h> 89 90 /* 91 * Protected counters by write_lock_irq(&tasklist_lock) 92 */ 93 unsigned long total_forks; /* Handle normal Linux uptimes. */ 94 int nr_threads; /* The idle threads do not count.. */ 95 96 int max_threads; /* tunable limit on nr_threads */ 97 98 DEFINE_PER_CPU(unsigned long, process_counts) = 0; 99 100 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */ 101 102 #ifdef CONFIG_PROVE_RCU 103 int lockdep_tasklist_lock_is_held(void) 104 { 105 return lockdep_is_held(&tasklist_lock); 106 } 107 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held); 108 #endif /* #ifdef CONFIG_PROVE_RCU */ 109 110 int nr_processes(void) 111 { 112 int cpu; 113 int total = 0; 114 115 for_each_possible_cpu(cpu) 116 total += per_cpu(process_counts, cpu); 117 118 return total; 119 } 120 121 void __weak arch_release_task_struct(struct task_struct *tsk) 122 { 123 } 124 125 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 126 static struct kmem_cache *task_struct_cachep; 127 128 static inline struct task_struct *alloc_task_struct_node(int node) 129 { 130 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node); 131 } 132 133 static inline void free_task_struct(struct task_struct *tsk) 134 { 135 kmem_cache_free(task_struct_cachep, tsk); 136 } 137 #endif 138 139 void __weak arch_release_thread_info(struct thread_info *ti) 140 { 141 } 142 143 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR 144 145 /* 146 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a 147 * kmemcache based allocator. 148 */ 149 # if THREAD_SIZE >= PAGE_SIZE 150 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk, 151 int node) 152 { 153 struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP, 154 THREAD_SIZE_ORDER); 155 156 return page ? page_address(page) : NULL; 157 } 158 159 static inline void free_thread_info(struct thread_info *ti) 160 { 161 free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER); 162 } 163 # else 164 static struct kmem_cache *thread_info_cache; 165 166 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk, 167 int node) 168 { 169 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node); 170 } 171 172 static void free_thread_info(struct thread_info *ti) 173 { 174 kmem_cache_free(thread_info_cache, ti); 175 } 176 177 void thread_info_cache_init(void) 178 { 179 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE, 180 THREAD_SIZE, 0, NULL); 181 BUG_ON(thread_info_cache == NULL); 182 } 183 # endif 184 #endif 185 186 /* SLAB cache for signal_struct structures (tsk->signal) */ 187 static struct kmem_cache *signal_cachep; 188 189 /* SLAB cache for sighand_struct structures (tsk->sighand) */ 190 struct kmem_cache *sighand_cachep; 191 192 /* SLAB cache for files_struct structures (tsk->files) */ 193 struct kmem_cache *files_cachep; 194 195 /* SLAB cache for fs_struct structures (tsk->fs) */ 196 struct kmem_cache *fs_cachep; 197 198 /* SLAB cache for vm_area_struct structures */ 199 struct kmem_cache *vm_area_cachep; 200 201 /* SLAB cache for mm_struct structures (tsk->mm) */ 202 static struct kmem_cache *mm_cachep; 203 204 static void account_kernel_stack(struct thread_info *ti, int account) 205 { 206 struct zone *zone = page_zone(virt_to_page(ti)); 207 208 mod_zone_page_state(zone, NR_KERNEL_STACK, account); 209 } 210 211 void free_task(struct task_struct *tsk) 212 { 213 account_kernel_stack(tsk->stack, -1); 214 arch_release_thread_info(tsk->stack); 215 free_thread_info(tsk->stack); 216 rt_mutex_debug_task_free(tsk); 217 ftrace_graph_exit_task(tsk); 218 put_seccomp_filter(tsk); 219 arch_release_task_struct(tsk); 220 free_task_struct(tsk); 221 } 222 EXPORT_SYMBOL(free_task); 223 224 static inline void free_signal_struct(struct signal_struct *sig) 225 { 226 taskstats_tgid_free(sig); 227 sched_autogroup_exit(sig); 228 kmem_cache_free(signal_cachep, sig); 229 } 230 231 static inline void put_signal_struct(struct signal_struct *sig) 232 { 233 if (atomic_dec_and_test(&sig->sigcnt)) 234 free_signal_struct(sig); 235 } 236 237 void __put_task_struct(struct task_struct *tsk) 238 { 239 WARN_ON(!tsk->exit_state); 240 WARN_ON(atomic_read(&tsk->usage)); 241 WARN_ON(tsk == current); 242 243 task_numa_free(tsk); 244 security_task_free(tsk); 245 exit_creds(tsk); 246 delayacct_tsk_free(tsk); 247 put_signal_struct(tsk->signal); 248 249 if (!profile_handoff_task(tsk)) 250 free_task(tsk); 251 } 252 EXPORT_SYMBOL_GPL(__put_task_struct); 253 254 void __init __weak arch_task_cache_init(void) { } 255 256 void __init fork_init(unsigned long mempages) 257 { 258 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 259 #ifndef ARCH_MIN_TASKALIGN 260 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES 261 #endif 262 /* create a slab on which task_structs can be allocated */ 263 task_struct_cachep = 264 kmem_cache_create("task_struct", sizeof(struct task_struct), 265 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL); 266 #endif 267 268 /* do the arch specific task caches init */ 269 arch_task_cache_init(); 270 271 /* 272 * The default maximum number of threads is set to a safe 273 * value: the thread structures can take up at most half 274 * of memory. 275 */ 276 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE); 277 278 /* 279 * we need to allow at least 20 threads to boot a system 280 */ 281 if (max_threads < 20) 282 max_threads = 20; 283 284 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 285 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 286 init_task.signal->rlim[RLIMIT_SIGPENDING] = 287 init_task.signal->rlim[RLIMIT_NPROC]; 288 } 289 290 int __weak arch_dup_task_struct(struct task_struct *dst, 291 struct task_struct *src) 292 { 293 *dst = *src; 294 return 0; 295 } 296 297 void set_task_stack_end_magic(struct task_struct *tsk) 298 { 299 unsigned long *stackend; 300 301 stackend = end_of_stack(tsk); 302 *stackend = STACK_END_MAGIC; /* for overflow detection */ 303 } 304 305 static struct task_struct *dup_task_struct(struct task_struct *orig) 306 { 307 struct task_struct *tsk; 308 struct thread_info *ti; 309 int node = tsk_fork_get_node(orig); 310 int err; 311 312 tsk = alloc_task_struct_node(node); 313 if (!tsk) 314 return NULL; 315 316 ti = alloc_thread_info_node(tsk, node); 317 if (!ti) 318 goto free_tsk; 319 320 err = arch_dup_task_struct(tsk, orig); 321 if (err) 322 goto free_ti; 323 324 tsk->stack = ti; 325 #ifdef CONFIG_SECCOMP 326 /* 327 * We must handle setting up seccomp filters once we're under 328 * the sighand lock in case orig has changed between now and 329 * then. Until then, filter must be NULL to avoid messing up 330 * the usage counts on the error path calling free_task. 331 */ 332 tsk->seccomp.filter = NULL; 333 #endif 334 335 setup_thread_stack(tsk, orig); 336 clear_user_return_notifier(tsk); 337 clear_tsk_need_resched(tsk); 338 set_task_stack_end_magic(tsk); 339 340 #ifdef CONFIG_CC_STACKPROTECTOR 341 tsk->stack_canary = get_random_int(); 342 #endif 343 344 /* 345 * One for us, one for whoever does the "release_task()" (usually 346 * parent) 347 */ 348 atomic_set(&tsk->usage, 2); 349 #ifdef CONFIG_BLK_DEV_IO_TRACE 350 tsk->btrace_seq = 0; 351 #endif 352 tsk->splice_pipe = NULL; 353 tsk->task_frag.page = NULL; 354 355 account_kernel_stack(ti, 1); 356 357 return tsk; 358 359 free_ti: 360 free_thread_info(ti); 361 free_tsk: 362 free_task_struct(tsk); 363 return NULL; 364 } 365 366 #ifdef CONFIG_MMU 367 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) 368 { 369 struct vm_area_struct *mpnt, *tmp, *prev, **pprev; 370 struct rb_node **rb_link, *rb_parent; 371 int retval; 372 unsigned long charge; 373 374 uprobe_start_dup_mmap(); 375 down_write(&oldmm->mmap_sem); 376 flush_cache_dup_mm(oldmm); 377 uprobe_dup_mmap(oldmm, mm); 378 /* 379 * Not linked in yet - no deadlock potential: 380 */ 381 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING); 382 383 mm->total_vm = oldmm->total_vm; 384 mm->shared_vm = oldmm->shared_vm; 385 mm->exec_vm = oldmm->exec_vm; 386 mm->stack_vm = oldmm->stack_vm; 387 388 rb_link = &mm->mm_rb.rb_node; 389 rb_parent = NULL; 390 pprev = &mm->mmap; 391 retval = ksm_fork(mm, oldmm); 392 if (retval) 393 goto out; 394 retval = khugepaged_fork(mm, oldmm); 395 if (retval) 396 goto out; 397 398 prev = NULL; 399 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { 400 struct file *file; 401 402 if (mpnt->vm_flags & VM_DONTCOPY) { 403 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file, 404 -vma_pages(mpnt)); 405 continue; 406 } 407 charge = 0; 408 if (mpnt->vm_flags & VM_ACCOUNT) { 409 unsigned long len = vma_pages(mpnt); 410 411 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */ 412 goto fail_nomem; 413 charge = len; 414 } 415 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 416 if (!tmp) 417 goto fail_nomem; 418 *tmp = *mpnt; 419 INIT_LIST_HEAD(&tmp->anon_vma_chain); 420 retval = vma_dup_policy(mpnt, tmp); 421 if (retval) 422 goto fail_nomem_policy; 423 tmp->vm_mm = mm; 424 if (anon_vma_fork(tmp, mpnt)) 425 goto fail_nomem_anon_vma_fork; 426 tmp->vm_flags &= ~VM_LOCKED; 427 tmp->vm_next = tmp->vm_prev = NULL; 428 file = tmp->vm_file; 429 if (file) { 430 struct inode *inode = file_inode(file); 431 struct address_space *mapping = file->f_mapping; 432 433 get_file(file); 434 if (tmp->vm_flags & VM_DENYWRITE) 435 atomic_dec(&inode->i_writecount); 436 i_mmap_lock_write(mapping); 437 if (tmp->vm_flags & VM_SHARED) 438 atomic_inc(&mapping->i_mmap_writable); 439 flush_dcache_mmap_lock(mapping); 440 /* insert tmp into the share list, just after mpnt */ 441 vma_interval_tree_insert_after(tmp, mpnt, 442 &mapping->i_mmap); 443 flush_dcache_mmap_unlock(mapping); 444 i_mmap_unlock_write(mapping); 445 } 446 447 /* 448 * Clear hugetlb-related page reserves for children. This only 449 * affects MAP_PRIVATE mappings. Faults generated by the child 450 * are not guaranteed to succeed, even if read-only 451 */ 452 if (is_vm_hugetlb_page(tmp)) 453 reset_vma_resv_huge_pages(tmp); 454 455 /* 456 * Link in the new vma and copy the page table entries. 457 */ 458 *pprev = tmp; 459 pprev = &tmp->vm_next; 460 tmp->vm_prev = prev; 461 prev = tmp; 462 463 __vma_link_rb(mm, tmp, rb_link, rb_parent); 464 rb_link = &tmp->vm_rb.rb_right; 465 rb_parent = &tmp->vm_rb; 466 467 mm->map_count++; 468 retval = copy_page_range(mm, oldmm, mpnt); 469 470 if (tmp->vm_ops && tmp->vm_ops->open) 471 tmp->vm_ops->open(tmp); 472 473 if (retval) 474 goto out; 475 } 476 /* a new mm has just been created */ 477 arch_dup_mmap(oldmm, mm); 478 retval = 0; 479 out: 480 up_write(&mm->mmap_sem); 481 flush_tlb_mm(oldmm); 482 up_write(&oldmm->mmap_sem); 483 uprobe_end_dup_mmap(); 484 return retval; 485 fail_nomem_anon_vma_fork: 486 mpol_put(vma_policy(tmp)); 487 fail_nomem_policy: 488 kmem_cache_free(vm_area_cachep, tmp); 489 fail_nomem: 490 retval = -ENOMEM; 491 vm_unacct_memory(charge); 492 goto out; 493 } 494 495 static inline int mm_alloc_pgd(struct mm_struct *mm) 496 { 497 mm->pgd = pgd_alloc(mm); 498 if (unlikely(!mm->pgd)) 499 return -ENOMEM; 500 return 0; 501 } 502 503 static inline void mm_free_pgd(struct mm_struct *mm) 504 { 505 pgd_free(mm, mm->pgd); 506 } 507 #else 508 #define dup_mmap(mm, oldmm) (0) 509 #define mm_alloc_pgd(mm) (0) 510 #define mm_free_pgd(mm) 511 #endif /* CONFIG_MMU */ 512 513 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); 514 515 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL)) 516 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) 517 518 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT; 519 520 static int __init coredump_filter_setup(char *s) 521 { 522 default_dump_filter = 523 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) & 524 MMF_DUMP_FILTER_MASK; 525 return 1; 526 } 527 528 __setup("coredump_filter=", coredump_filter_setup); 529 530 #include <linux/init_task.h> 531 532 static void mm_init_aio(struct mm_struct *mm) 533 { 534 #ifdef CONFIG_AIO 535 spin_lock_init(&mm->ioctx_lock); 536 mm->ioctx_table = NULL; 537 #endif 538 } 539 540 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p) 541 { 542 #ifdef CONFIG_MEMCG 543 mm->owner = p; 544 #endif 545 } 546 547 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p) 548 { 549 mm->mmap = NULL; 550 mm->mm_rb = RB_ROOT; 551 mm->vmacache_seqnum = 0; 552 atomic_set(&mm->mm_users, 1); 553 atomic_set(&mm->mm_count, 1); 554 init_rwsem(&mm->mmap_sem); 555 INIT_LIST_HEAD(&mm->mmlist); 556 mm->core_state = NULL; 557 atomic_long_set(&mm->nr_ptes, 0); 558 mm->map_count = 0; 559 mm->locked_vm = 0; 560 mm->pinned_vm = 0; 561 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat)); 562 spin_lock_init(&mm->page_table_lock); 563 mm_init_cpumask(mm); 564 mm_init_aio(mm); 565 mm_init_owner(mm, p); 566 mmu_notifier_mm_init(mm); 567 clear_tlb_flush_pending(mm); 568 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 569 mm->pmd_huge_pte = NULL; 570 #endif 571 572 if (current->mm) { 573 mm->flags = current->mm->flags & MMF_INIT_MASK; 574 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK; 575 } else { 576 mm->flags = default_dump_filter; 577 mm->def_flags = 0; 578 } 579 580 if (mm_alloc_pgd(mm)) 581 goto fail_nopgd; 582 583 if (init_new_context(p, mm)) 584 goto fail_nocontext; 585 586 return mm; 587 588 fail_nocontext: 589 mm_free_pgd(mm); 590 fail_nopgd: 591 free_mm(mm); 592 return NULL; 593 } 594 595 static void check_mm(struct mm_struct *mm) 596 { 597 int i; 598 599 for (i = 0; i < NR_MM_COUNTERS; i++) { 600 long x = atomic_long_read(&mm->rss_stat.count[i]); 601 602 if (unlikely(x)) 603 printk(KERN_ALERT "BUG: Bad rss-counter state " 604 "mm:%p idx:%d val:%ld\n", mm, i, x); 605 } 606 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 607 VM_BUG_ON_MM(mm->pmd_huge_pte, mm); 608 #endif 609 } 610 611 /* 612 * Allocate and initialize an mm_struct. 613 */ 614 struct mm_struct *mm_alloc(void) 615 { 616 struct mm_struct *mm; 617 618 mm = allocate_mm(); 619 if (!mm) 620 return NULL; 621 622 memset(mm, 0, sizeof(*mm)); 623 return mm_init(mm, current); 624 } 625 626 /* 627 * Called when the last reference to the mm 628 * is dropped: either by a lazy thread or by 629 * mmput. Free the page directory and the mm. 630 */ 631 void __mmdrop(struct mm_struct *mm) 632 { 633 BUG_ON(mm == &init_mm); 634 mm_free_pgd(mm); 635 destroy_context(mm); 636 mmu_notifier_mm_destroy(mm); 637 check_mm(mm); 638 free_mm(mm); 639 } 640 EXPORT_SYMBOL_GPL(__mmdrop); 641 642 /* 643 * Decrement the use count and release all resources for an mm. 644 */ 645 void mmput(struct mm_struct *mm) 646 { 647 might_sleep(); 648 649 if (atomic_dec_and_test(&mm->mm_users)) { 650 uprobe_clear_state(mm); 651 exit_aio(mm); 652 ksm_exit(mm); 653 khugepaged_exit(mm); /* must run before exit_mmap */ 654 exit_mmap(mm); 655 set_mm_exe_file(mm, NULL); 656 if (!list_empty(&mm->mmlist)) { 657 spin_lock(&mmlist_lock); 658 list_del(&mm->mmlist); 659 spin_unlock(&mmlist_lock); 660 } 661 if (mm->binfmt) 662 module_put(mm->binfmt->module); 663 mmdrop(mm); 664 } 665 } 666 EXPORT_SYMBOL_GPL(mmput); 667 668 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 669 { 670 if (new_exe_file) 671 get_file(new_exe_file); 672 if (mm->exe_file) 673 fput(mm->exe_file); 674 mm->exe_file = new_exe_file; 675 } 676 677 struct file *get_mm_exe_file(struct mm_struct *mm) 678 { 679 struct file *exe_file; 680 681 /* We need mmap_sem to protect against races with removal of exe_file */ 682 down_read(&mm->mmap_sem); 683 exe_file = mm->exe_file; 684 if (exe_file) 685 get_file(exe_file); 686 up_read(&mm->mmap_sem); 687 return exe_file; 688 } 689 690 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm) 691 { 692 /* It's safe to write the exe_file pointer without exe_file_lock because 693 * this is called during fork when the task is not yet in /proc */ 694 newmm->exe_file = get_mm_exe_file(oldmm); 695 } 696 697 /** 698 * get_task_mm - acquire a reference to the task's mm 699 * 700 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning 701 * this kernel workthread has transiently adopted a user mm with use_mm, 702 * to do its AIO) is not set and if so returns a reference to it, after 703 * bumping up the use count. User must release the mm via mmput() 704 * after use. Typically used by /proc and ptrace. 705 */ 706 struct mm_struct *get_task_mm(struct task_struct *task) 707 { 708 struct mm_struct *mm; 709 710 task_lock(task); 711 mm = task->mm; 712 if (mm) { 713 if (task->flags & PF_KTHREAD) 714 mm = NULL; 715 else 716 atomic_inc(&mm->mm_users); 717 } 718 task_unlock(task); 719 return mm; 720 } 721 EXPORT_SYMBOL_GPL(get_task_mm); 722 723 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode) 724 { 725 struct mm_struct *mm; 726 int err; 727 728 err = mutex_lock_killable(&task->signal->cred_guard_mutex); 729 if (err) 730 return ERR_PTR(err); 731 732 mm = get_task_mm(task); 733 if (mm && mm != current->mm && 734 !ptrace_may_access(task, mode)) { 735 mmput(mm); 736 mm = ERR_PTR(-EACCES); 737 } 738 mutex_unlock(&task->signal->cred_guard_mutex); 739 740 return mm; 741 } 742 743 static void complete_vfork_done(struct task_struct *tsk) 744 { 745 struct completion *vfork; 746 747 task_lock(tsk); 748 vfork = tsk->vfork_done; 749 if (likely(vfork)) { 750 tsk->vfork_done = NULL; 751 complete(vfork); 752 } 753 task_unlock(tsk); 754 } 755 756 static int wait_for_vfork_done(struct task_struct *child, 757 struct completion *vfork) 758 { 759 int killed; 760 761 freezer_do_not_count(); 762 killed = wait_for_completion_killable(vfork); 763 freezer_count(); 764 765 if (killed) { 766 task_lock(child); 767 child->vfork_done = NULL; 768 task_unlock(child); 769 } 770 771 put_task_struct(child); 772 return killed; 773 } 774 775 /* Please note the differences between mmput and mm_release. 776 * mmput is called whenever we stop holding onto a mm_struct, 777 * error success whatever. 778 * 779 * mm_release is called after a mm_struct has been removed 780 * from the current process. 781 * 782 * This difference is important for error handling, when we 783 * only half set up a mm_struct for a new process and need to restore 784 * the old one. Because we mmput the new mm_struct before 785 * restoring the old one. . . 786 * Eric Biederman 10 January 1998 787 */ 788 void mm_release(struct task_struct *tsk, struct mm_struct *mm) 789 { 790 /* Get rid of any futexes when releasing the mm */ 791 #ifdef CONFIG_FUTEX 792 if (unlikely(tsk->robust_list)) { 793 exit_robust_list(tsk); 794 tsk->robust_list = NULL; 795 } 796 #ifdef CONFIG_COMPAT 797 if (unlikely(tsk->compat_robust_list)) { 798 compat_exit_robust_list(tsk); 799 tsk->compat_robust_list = NULL; 800 } 801 #endif 802 if (unlikely(!list_empty(&tsk->pi_state_list))) 803 exit_pi_state_list(tsk); 804 #endif 805 806 uprobe_free_utask(tsk); 807 808 /* Get rid of any cached register state */ 809 deactivate_mm(tsk, mm); 810 811 /* 812 * If we're exiting normally, clear a user-space tid field if 813 * requested. We leave this alone when dying by signal, to leave 814 * the value intact in a core dump, and to save the unnecessary 815 * trouble, say, a killed vfork parent shouldn't touch this mm. 816 * Userland only wants this done for a sys_exit. 817 */ 818 if (tsk->clear_child_tid) { 819 if (!(tsk->flags & PF_SIGNALED) && 820 atomic_read(&mm->mm_users) > 1) { 821 /* 822 * We don't check the error code - if userspace has 823 * not set up a proper pointer then tough luck. 824 */ 825 put_user(0, tsk->clear_child_tid); 826 sys_futex(tsk->clear_child_tid, FUTEX_WAKE, 827 1, NULL, NULL, 0); 828 } 829 tsk->clear_child_tid = NULL; 830 } 831 832 /* 833 * All done, finally we can wake up parent and return this mm to him. 834 * Also kthread_stop() uses this completion for synchronization. 835 */ 836 if (tsk->vfork_done) 837 complete_vfork_done(tsk); 838 } 839 840 /* 841 * Allocate a new mm structure and copy contents from the 842 * mm structure of the passed in task structure. 843 */ 844 static struct mm_struct *dup_mm(struct task_struct *tsk) 845 { 846 struct mm_struct *mm, *oldmm = current->mm; 847 int err; 848 849 mm = allocate_mm(); 850 if (!mm) 851 goto fail_nomem; 852 853 memcpy(mm, oldmm, sizeof(*mm)); 854 855 if (!mm_init(mm, tsk)) 856 goto fail_nomem; 857 858 dup_mm_exe_file(oldmm, mm); 859 860 err = dup_mmap(mm, oldmm); 861 if (err) 862 goto free_pt; 863 864 mm->hiwater_rss = get_mm_rss(mm); 865 mm->hiwater_vm = mm->total_vm; 866 867 if (mm->binfmt && !try_module_get(mm->binfmt->module)) 868 goto free_pt; 869 870 return mm; 871 872 free_pt: 873 /* don't put binfmt in mmput, we haven't got module yet */ 874 mm->binfmt = NULL; 875 mmput(mm); 876 877 fail_nomem: 878 return NULL; 879 } 880 881 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk) 882 { 883 struct mm_struct *mm, *oldmm; 884 int retval; 885 886 tsk->min_flt = tsk->maj_flt = 0; 887 tsk->nvcsw = tsk->nivcsw = 0; 888 #ifdef CONFIG_DETECT_HUNG_TASK 889 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; 890 #endif 891 892 tsk->mm = NULL; 893 tsk->active_mm = NULL; 894 895 /* 896 * Are we cloning a kernel thread? 897 * 898 * We need to steal a active VM for that.. 899 */ 900 oldmm = current->mm; 901 if (!oldmm) 902 return 0; 903 904 /* initialize the new vmacache entries */ 905 vmacache_flush(tsk); 906 907 if (clone_flags & CLONE_VM) { 908 atomic_inc(&oldmm->mm_users); 909 mm = oldmm; 910 goto good_mm; 911 } 912 913 retval = -ENOMEM; 914 mm = dup_mm(tsk); 915 if (!mm) 916 goto fail_nomem; 917 918 good_mm: 919 tsk->mm = mm; 920 tsk->active_mm = mm; 921 return 0; 922 923 fail_nomem: 924 return retval; 925 } 926 927 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 928 { 929 struct fs_struct *fs = current->fs; 930 if (clone_flags & CLONE_FS) { 931 /* tsk->fs is already what we want */ 932 spin_lock(&fs->lock); 933 if (fs->in_exec) { 934 spin_unlock(&fs->lock); 935 return -EAGAIN; 936 } 937 fs->users++; 938 spin_unlock(&fs->lock); 939 return 0; 940 } 941 tsk->fs = copy_fs_struct(fs); 942 if (!tsk->fs) 943 return -ENOMEM; 944 return 0; 945 } 946 947 static int copy_files(unsigned long clone_flags, struct task_struct *tsk) 948 { 949 struct files_struct *oldf, *newf; 950 int error = 0; 951 952 /* 953 * A background process may not have any files ... 954 */ 955 oldf = current->files; 956 if (!oldf) 957 goto out; 958 959 if (clone_flags & CLONE_FILES) { 960 atomic_inc(&oldf->count); 961 goto out; 962 } 963 964 newf = dup_fd(oldf, &error); 965 if (!newf) 966 goto out; 967 968 tsk->files = newf; 969 error = 0; 970 out: 971 return error; 972 } 973 974 static int copy_io(unsigned long clone_flags, struct task_struct *tsk) 975 { 976 #ifdef CONFIG_BLOCK 977 struct io_context *ioc = current->io_context; 978 struct io_context *new_ioc; 979 980 if (!ioc) 981 return 0; 982 /* 983 * Share io context with parent, if CLONE_IO is set 984 */ 985 if (clone_flags & CLONE_IO) { 986 ioc_task_link(ioc); 987 tsk->io_context = ioc; 988 } else if (ioprio_valid(ioc->ioprio)) { 989 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE); 990 if (unlikely(!new_ioc)) 991 return -ENOMEM; 992 993 new_ioc->ioprio = ioc->ioprio; 994 put_io_context(new_ioc); 995 } 996 #endif 997 return 0; 998 } 999 1000 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk) 1001 { 1002 struct sighand_struct *sig; 1003 1004 if (clone_flags & CLONE_SIGHAND) { 1005 atomic_inc(¤t->sighand->count); 1006 return 0; 1007 } 1008 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 1009 rcu_assign_pointer(tsk->sighand, sig); 1010 if (!sig) 1011 return -ENOMEM; 1012 atomic_set(&sig->count, 1); 1013 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 1014 return 0; 1015 } 1016 1017 void __cleanup_sighand(struct sighand_struct *sighand) 1018 { 1019 if (atomic_dec_and_test(&sighand->count)) { 1020 signalfd_cleanup(sighand); 1021 /* 1022 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it 1023 * without an RCU grace period, see __lock_task_sighand(). 1024 */ 1025 kmem_cache_free(sighand_cachep, sighand); 1026 } 1027 } 1028 1029 /* 1030 * Initialize POSIX timer handling for a thread group. 1031 */ 1032 static void posix_cpu_timers_init_group(struct signal_struct *sig) 1033 { 1034 unsigned long cpu_limit; 1035 1036 /* Thread group counters. */ 1037 thread_group_cputime_init(sig); 1038 1039 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); 1040 if (cpu_limit != RLIM_INFINITY) { 1041 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit); 1042 sig->cputimer.running = 1; 1043 } 1044 1045 /* The timer lists. */ 1046 INIT_LIST_HEAD(&sig->cpu_timers[0]); 1047 INIT_LIST_HEAD(&sig->cpu_timers[1]); 1048 INIT_LIST_HEAD(&sig->cpu_timers[2]); 1049 } 1050 1051 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk) 1052 { 1053 struct signal_struct *sig; 1054 1055 if (clone_flags & CLONE_THREAD) 1056 return 0; 1057 1058 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL); 1059 tsk->signal = sig; 1060 if (!sig) 1061 return -ENOMEM; 1062 1063 sig->nr_threads = 1; 1064 atomic_set(&sig->live, 1); 1065 atomic_set(&sig->sigcnt, 1); 1066 1067 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */ 1068 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node); 1069 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head); 1070 1071 init_waitqueue_head(&sig->wait_chldexit); 1072 sig->curr_target = tsk; 1073 init_sigpending(&sig->shared_pending); 1074 INIT_LIST_HEAD(&sig->posix_timers); 1075 seqlock_init(&sig->stats_lock); 1076 1077 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1078 sig->real_timer.function = it_real_fn; 1079 1080 task_lock(current->group_leader); 1081 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 1082 task_unlock(current->group_leader); 1083 1084 posix_cpu_timers_init_group(sig); 1085 1086 tty_audit_fork(sig); 1087 sched_autogroup_fork(sig); 1088 1089 #ifdef CONFIG_CGROUPS 1090 init_rwsem(&sig->group_rwsem); 1091 #endif 1092 1093 sig->oom_score_adj = current->signal->oom_score_adj; 1094 sig->oom_score_adj_min = current->signal->oom_score_adj_min; 1095 1096 sig->has_child_subreaper = current->signal->has_child_subreaper || 1097 current->signal->is_child_subreaper; 1098 1099 mutex_init(&sig->cred_guard_mutex); 1100 1101 return 0; 1102 } 1103 1104 static void copy_seccomp(struct task_struct *p) 1105 { 1106 #ifdef CONFIG_SECCOMP 1107 /* 1108 * Must be called with sighand->lock held, which is common to 1109 * all threads in the group. Holding cred_guard_mutex is not 1110 * needed because this new task is not yet running and cannot 1111 * be racing exec. 1112 */ 1113 assert_spin_locked(¤t->sighand->siglock); 1114 1115 /* Ref-count the new filter user, and assign it. */ 1116 get_seccomp_filter(current); 1117 p->seccomp = current->seccomp; 1118 1119 /* 1120 * Explicitly enable no_new_privs here in case it got set 1121 * between the task_struct being duplicated and holding the 1122 * sighand lock. The seccomp state and nnp must be in sync. 1123 */ 1124 if (task_no_new_privs(current)) 1125 task_set_no_new_privs(p); 1126 1127 /* 1128 * If the parent gained a seccomp mode after copying thread 1129 * flags and between before we held the sighand lock, we have 1130 * to manually enable the seccomp thread flag here. 1131 */ 1132 if (p->seccomp.mode != SECCOMP_MODE_DISABLED) 1133 set_tsk_thread_flag(p, TIF_SECCOMP); 1134 #endif 1135 } 1136 1137 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr) 1138 { 1139 current->clear_child_tid = tidptr; 1140 1141 return task_pid_vnr(current); 1142 } 1143 1144 static void rt_mutex_init_task(struct task_struct *p) 1145 { 1146 raw_spin_lock_init(&p->pi_lock); 1147 #ifdef CONFIG_RT_MUTEXES 1148 p->pi_waiters = RB_ROOT; 1149 p->pi_waiters_leftmost = NULL; 1150 p->pi_blocked_on = NULL; 1151 #endif 1152 } 1153 1154 /* 1155 * Initialize POSIX timer handling for a single task. 1156 */ 1157 static void posix_cpu_timers_init(struct task_struct *tsk) 1158 { 1159 tsk->cputime_expires.prof_exp = 0; 1160 tsk->cputime_expires.virt_exp = 0; 1161 tsk->cputime_expires.sched_exp = 0; 1162 INIT_LIST_HEAD(&tsk->cpu_timers[0]); 1163 INIT_LIST_HEAD(&tsk->cpu_timers[1]); 1164 INIT_LIST_HEAD(&tsk->cpu_timers[2]); 1165 } 1166 1167 static inline void 1168 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid) 1169 { 1170 task->pids[type].pid = pid; 1171 } 1172 1173 /* 1174 * This creates a new process as a copy of the old one, 1175 * but does not actually start it yet. 1176 * 1177 * It copies the registers, and all the appropriate 1178 * parts of the process environment (as per the clone 1179 * flags). The actual kick-off is left to the caller. 1180 */ 1181 static struct task_struct *copy_process(unsigned long clone_flags, 1182 unsigned long stack_start, 1183 unsigned long stack_size, 1184 int __user *child_tidptr, 1185 struct pid *pid, 1186 int trace) 1187 { 1188 int retval; 1189 struct task_struct *p; 1190 1191 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 1192 return ERR_PTR(-EINVAL); 1193 1194 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS)) 1195 return ERR_PTR(-EINVAL); 1196 1197 /* 1198 * Thread groups must share signals as well, and detached threads 1199 * can only be started up within the thread group. 1200 */ 1201 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 1202 return ERR_PTR(-EINVAL); 1203 1204 /* 1205 * Shared signal handlers imply shared VM. By way of the above, 1206 * thread groups also imply shared VM. Blocking this case allows 1207 * for various simplifications in other code. 1208 */ 1209 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 1210 return ERR_PTR(-EINVAL); 1211 1212 /* 1213 * Siblings of global init remain as zombies on exit since they are 1214 * not reaped by their parent (swapper). To solve this and to avoid 1215 * multi-rooted process trees, prevent global and container-inits 1216 * from creating siblings. 1217 */ 1218 if ((clone_flags & CLONE_PARENT) && 1219 current->signal->flags & SIGNAL_UNKILLABLE) 1220 return ERR_PTR(-EINVAL); 1221 1222 /* 1223 * If the new process will be in a different pid or user namespace 1224 * do not allow it to share a thread group or signal handlers or 1225 * parent with the forking task. 1226 */ 1227 if (clone_flags & CLONE_SIGHAND) { 1228 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) || 1229 (task_active_pid_ns(current) != 1230 current->nsproxy->pid_ns_for_children)) 1231 return ERR_PTR(-EINVAL); 1232 } 1233 1234 retval = security_task_create(clone_flags); 1235 if (retval) 1236 goto fork_out; 1237 1238 retval = -ENOMEM; 1239 p = dup_task_struct(current); 1240 if (!p) 1241 goto fork_out; 1242 1243 ftrace_graph_init_task(p); 1244 1245 rt_mutex_init_task(p); 1246 1247 #ifdef CONFIG_PROVE_LOCKING 1248 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); 1249 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 1250 #endif 1251 retval = -EAGAIN; 1252 if (atomic_read(&p->real_cred->user->processes) >= 1253 task_rlimit(p, RLIMIT_NPROC)) { 1254 if (p->real_cred->user != INIT_USER && 1255 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) 1256 goto bad_fork_free; 1257 } 1258 current->flags &= ~PF_NPROC_EXCEEDED; 1259 1260 retval = copy_creds(p, clone_flags); 1261 if (retval < 0) 1262 goto bad_fork_free; 1263 1264 /* 1265 * If multiple threads are within copy_process(), then this check 1266 * triggers too late. This doesn't hurt, the check is only there 1267 * to stop root fork bombs. 1268 */ 1269 retval = -EAGAIN; 1270 if (nr_threads >= max_threads) 1271 goto bad_fork_cleanup_count; 1272 1273 if (!try_module_get(task_thread_info(p)->exec_domain->module)) 1274 goto bad_fork_cleanup_count; 1275 1276 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */ 1277 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER); 1278 p->flags |= PF_FORKNOEXEC; 1279 INIT_LIST_HEAD(&p->children); 1280 INIT_LIST_HEAD(&p->sibling); 1281 rcu_copy_process(p); 1282 p->vfork_done = NULL; 1283 spin_lock_init(&p->alloc_lock); 1284 1285 init_sigpending(&p->pending); 1286 1287 p->utime = p->stime = p->gtime = 0; 1288 p->utimescaled = p->stimescaled = 0; 1289 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE 1290 p->prev_cputime.utime = p->prev_cputime.stime = 0; 1291 #endif 1292 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN 1293 seqlock_init(&p->vtime_seqlock); 1294 p->vtime_snap = 0; 1295 p->vtime_snap_whence = VTIME_SLEEPING; 1296 #endif 1297 1298 #if defined(SPLIT_RSS_COUNTING) 1299 memset(&p->rss_stat, 0, sizeof(p->rss_stat)); 1300 #endif 1301 1302 p->default_timer_slack_ns = current->timer_slack_ns; 1303 1304 task_io_accounting_init(&p->ioac); 1305 acct_clear_integrals(p); 1306 1307 posix_cpu_timers_init(p); 1308 1309 p->start_time = ktime_get_ns(); 1310 p->real_start_time = ktime_get_boot_ns(); 1311 p->io_context = NULL; 1312 p->audit_context = NULL; 1313 if (clone_flags & CLONE_THREAD) 1314 threadgroup_change_begin(current); 1315 cgroup_fork(p); 1316 #ifdef CONFIG_NUMA 1317 p->mempolicy = mpol_dup(p->mempolicy); 1318 if (IS_ERR(p->mempolicy)) { 1319 retval = PTR_ERR(p->mempolicy); 1320 p->mempolicy = NULL; 1321 goto bad_fork_cleanup_threadgroup_lock; 1322 } 1323 #endif 1324 #ifdef CONFIG_CPUSETS 1325 p->cpuset_mem_spread_rotor = NUMA_NO_NODE; 1326 p->cpuset_slab_spread_rotor = NUMA_NO_NODE; 1327 seqcount_init(&p->mems_allowed_seq); 1328 #endif 1329 #ifdef CONFIG_TRACE_IRQFLAGS 1330 p->irq_events = 0; 1331 p->hardirqs_enabled = 0; 1332 p->hardirq_enable_ip = 0; 1333 p->hardirq_enable_event = 0; 1334 p->hardirq_disable_ip = _THIS_IP_; 1335 p->hardirq_disable_event = 0; 1336 p->softirqs_enabled = 1; 1337 p->softirq_enable_ip = _THIS_IP_; 1338 p->softirq_enable_event = 0; 1339 p->softirq_disable_ip = 0; 1340 p->softirq_disable_event = 0; 1341 p->hardirq_context = 0; 1342 p->softirq_context = 0; 1343 #endif 1344 #ifdef CONFIG_LOCKDEP 1345 p->lockdep_depth = 0; /* no locks held yet */ 1346 p->curr_chain_key = 0; 1347 p->lockdep_recursion = 0; 1348 #endif 1349 1350 #ifdef CONFIG_DEBUG_MUTEXES 1351 p->blocked_on = NULL; /* not blocked yet */ 1352 #endif 1353 #ifdef CONFIG_BCACHE 1354 p->sequential_io = 0; 1355 p->sequential_io_avg = 0; 1356 #endif 1357 1358 /* Perform scheduler related setup. Assign this task to a CPU. */ 1359 retval = sched_fork(clone_flags, p); 1360 if (retval) 1361 goto bad_fork_cleanup_policy; 1362 1363 retval = perf_event_init_task(p); 1364 if (retval) 1365 goto bad_fork_cleanup_policy; 1366 retval = audit_alloc(p); 1367 if (retval) 1368 goto bad_fork_cleanup_perf; 1369 /* copy all the process information */ 1370 shm_init_task(p); 1371 retval = copy_semundo(clone_flags, p); 1372 if (retval) 1373 goto bad_fork_cleanup_audit; 1374 retval = copy_files(clone_flags, p); 1375 if (retval) 1376 goto bad_fork_cleanup_semundo; 1377 retval = copy_fs(clone_flags, p); 1378 if (retval) 1379 goto bad_fork_cleanup_files; 1380 retval = copy_sighand(clone_flags, p); 1381 if (retval) 1382 goto bad_fork_cleanup_fs; 1383 retval = copy_signal(clone_flags, p); 1384 if (retval) 1385 goto bad_fork_cleanup_sighand; 1386 retval = copy_mm(clone_flags, p); 1387 if (retval) 1388 goto bad_fork_cleanup_signal; 1389 retval = copy_namespaces(clone_flags, p); 1390 if (retval) 1391 goto bad_fork_cleanup_mm; 1392 retval = copy_io(clone_flags, p); 1393 if (retval) 1394 goto bad_fork_cleanup_namespaces; 1395 retval = copy_thread(clone_flags, stack_start, stack_size, p); 1396 if (retval) 1397 goto bad_fork_cleanup_io; 1398 1399 if (pid != &init_struct_pid) { 1400 retval = -ENOMEM; 1401 pid = alloc_pid(p->nsproxy->pid_ns_for_children); 1402 if (!pid) 1403 goto bad_fork_cleanup_io; 1404 } 1405 1406 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; 1407 /* 1408 * Clear TID on mm_release()? 1409 */ 1410 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL; 1411 #ifdef CONFIG_BLOCK 1412 p->plug = NULL; 1413 #endif 1414 #ifdef CONFIG_FUTEX 1415 p->robust_list = NULL; 1416 #ifdef CONFIG_COMPAT 1417 p->compat_robust_list = NULL; 1418 #endif 1419 INIT_LIST_HEAD(&p->pi_state_list); 1420 p->pi_state_cache = NULL; 1421 #endif 1422 /* 1423 * sigaltstack should be cleared when sharing the same VM 1424 */ 1425 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM) 1426 p->sas_ss_sp = p->sas_ss_size = 0; 1427 1428 /* 1429 * Syscall tracing and stepping should be turned off in the 1430 * child regardless of CLONE_PTRACE. 1431 */ 1432 user_disable_single_step(p); 1433 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); 1434 #ifdef TIF_SYSCALL_EMU 1435 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); 1436 #endif 1437 clear_all_latency_tracing(p); 1438 1439 /* ok, now we should be set up.. */ 1440 p->pid = pid_nr(pid); 1441 if (clone_flags & CLONE_THREAD) { 1442 p->exit_signal = -1; 1443 p->group_leader = current->group_leader; 1444 p->tgid = current->tgid; 1445 } else { 1446 if (clone_flags & CLONE_PARENT) 1447 p->exit_signal = current->group_leader->exit_signal; 1448 else 1449 p->exit_signal = (clone_flags & CSIGNAL); 1450 p->group_leader = p; 1451 p->tgid = p->pid; 1452 } 1453 1454 p->nr_dirtied = 0; 1455 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10); 1456 p->dirty_paused_when = 0; 1457 1458 p->pdeath_signal = 0; 1459 INIT_LIST_HEAD(&p->thread_group); 1460 p->task_works = NULL; 1461 1462 /* 1463 * Make it visible to the rest of the system, but dont wake it up yet. 1464 * Need tasklist lock for parent etc handling! 1465 */ 1466 write_lock_irq(&tasklist_lock); 1467 1468 /* CLONE_PARENT re-uses the old parent */ 1469 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { 1470 p->real_parent = current->real_parent; 1471 p->parent_exec_id = current->parent_exec_id; 1472 } else { 1473 p->real_parent = current; 1474 p->parent_exec_id = current->self_exec_id; 1475 } 1476 1477 spin_lock(¤t->sighand->siglock); 1478 1479 /* 1480 * Copy seccomp details explicitly here, in case they were changed 1481 * before holding sighand lock. 1482 */ 1483 copy_seccomp(p); 1484 1485 /* 1486 * Process group and session signals need to be delivered to just the 1487 * parent before the fork or both the parent and the child after the 1488 * fork. Restart if a signal comes in before we add the new process to 1489 * it's process group. 1490 * A fatal signal pending means that current will exit, so the new 1491 * thread can't slip out of an OOM kill (or normal SIGKILL). 1492 */ 1493 recalc_sigpending(); 1494 if (signal_pending(current)) { 1495 spin_unlock(¤t->sighand->siglock); 1496 write_unlock_irq(&tasklist_lock); 1497 retval = -ERESTARTNOINTR; 1498 goto bad_fork_free_pid; 1499 } 1500 1501 if (likely(p->pid)) { 1502 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace); 1503 1504 init_task_pid(p, PIDTYPE_PID, pid); 1505 if (thread_group_leader(p)) { 1506 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current)); 1507 init_task_pid(p, PIDTYPE_SID, task_session(current)); 1508 1509 if (is_child_reaper(pid)) { 1510 ns_of_pid(pid)->child_reaper = p; 1511 p->signal->flags |= SIGNAL_UNKILLABLE; 1512 } 1513 1514 p->signal->leader_pid = pid; 1515 p->signal->tty = tty_kref_get(current->signal->tty); 1516 list_add_tail(&p->sibling, &p->real_parent->children); 1517 list_add_tail_rcu(&p->tasks, &init_task.tasks); 1518 attach_pid(p, PIDTYPE_PGID); 1519 attach_pid(p, PIDTYPE_SID); 1520 __this_cpu_inc(process_counts); 1521 } else { 1522 current->signal->nr_threads++; 1523 atomic_inc(¤t->signal->live); 1524 atomic_inc(¤t->signal->sigcnt); 1525 list_add_tail_rcu(&p->thread_group, 1526 &p->group_leader->thread_group); 1527 list_add_tail_rcu(&p->thread_node, 1528 &p->signal->thread_head); 1529 } 1530 attach_pid(p, PIDTYPE_PID); 1531 nr_threads++; 1532 } 1533 1534 total_forks++; 1535 spin_unlock(¤t->sighand->siglock); 1536 syscall_tracepoint_update(p); 1537 write_unlock_irq(&tasklist_lock); 1538 1539 proc_fork_connector(p); 1540 cgroup_post_fork(p); 1541 if (clone_flags & CLONE_THREAD) 1542 threadgroup_change_end(current); 1543 perf_event_fork(p); 1544 1545 trace_task_newtask(p, clone_flags); 1546 uprobe_copy_process(p, clone_flags); 1547 1548 return p; 1549 1550 bad_fork_free_pid: 1551 if (pid != &init_struct_pid) 1552 free_pid(pid); 1553 bad_fork_cleanup_io: 1554 if (p->io_context) 1555 exit_io_context(p); 1556 bad_fork_cleanup_namespaces: 1557 exit_task_namespaces(p); 1558 bad_fork_cleanup_mm: 1559 if (p->mm) 1560 mmput(p->mm); 1561 bad_fork_cleanup_signal: 1562 if (!(clone_flags & CLONE_THREAD)) 1563 free_signal_struct(p->signal); 1564 bad_fork_cleanup_sighand: 1565 __cleanup_sighand(p->sighand); 1566 bad_fork_cleanup_fs: 1567 exit_fs(p); /* blocking */ 1568 bad_fork_cleanup_files: 1569 exit_files(p); /* blocking */ 1570 bad_fork_cleanup_semundo: 1571 exit_sem(p); 1572 bad_fork_cleanup_audit: 1573 audit_free(p); 1574 bad_fork_cleanup_perf: 1575 perf_event_free_task(p); 1576 bad_fork_cleanup_policy: 1577 #ifdef CONFIG_NUMA 1578 mpol_put(p->mempolicy); 1579 bad_fork_cleanup_threadgroup_lock: 1580 #endif 1581 if (clone_flags & CLONE_THREAD) 1582 threadgroup_change_end(current); 1583 delayacct_tsk_free(p); 1584 module_put(task_thread_info(p)->exec_domain->module); 1585 bad_fork_cleanup_count: 1586 atomic_dec(&p->cred->user->processes); 1587 exit_creds(p); 1588 bad_fork_free: 1589 free_task(p); 1590 fork_out: 1591 return ERR_PTR(retval); 1592 } 1593 1594 static inline void init_idle_pids(struct pid_link *links) 1595 { 1596 enum pid_type type; 1597 1598 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 1599 INIT_HLIST_NODE(&links[type].node); /* not really needed */ 1600 links[type].pid = &init_struct_pid; 1601 } 1602 } 1603 1604 struct task_struct *fork_idle(int cpu) 1605 { 1606 struct task_struct *task; 1607 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0); 1608 if (!IS_ERR(task)) { 1609 init_idle_pids(task->pids); 1610 init_idle(task, cpu); 1611 } 1612 1613 return task; 1614 } 1615 1616 /* 1617 * Ok, this is the main fork-routine. 1618 * 1619 * It copies the process, and if successful kick-starts 1620 * it and waits for it to finish using the VM if required. 1621 */ 1622 long do_fork(unsigned long clone_flags, 1623 unsigned long stack_start, 1624 unsigned long stack_size, 1625 int __user *parent_tidptr, 1626 int __user *child_tidptr) 1627 { 1628 struct task_struct *p; 1629 int trace = 0; 1630 long nr; 1631 1632 /* 1633 * Determine whether and which event to report to ptracer. When 1634 * called from kernel_thread or CLONE_UNTRACED is explicitly 1635 * requested, no event is reported; otherwise, report if the event 1636 * for the type of forking is enabled. 1637 */ 1638 if (!(clone_flags & CLONE_UNTRACED)) { 1639 if (clone_flags & CLONE_VFORK) 1640 trace = PTRACE_EVENT_VFORK; 1641 else if ((clone_flags & CSIGNAL) != SIGCHLD) 1642 trace = PTRACE_EVENT_CLONE; 1643 else 1644 trace = PTRACE_EVENT_FORK; 1645 1646 if (likely(!ptrace_event_enabled(current, trace))) 1647 trace = 0; 1648 } 1649 1650 p = copy_process(clone_flags, stack_start, stack_size, 1651 child_tidptr, NULL, trace); 1652 /* 1653 * Do this prior waking up the new thread - the thread pointer 1654 * might get invalid after that point, if the thread exits quickly. 1655 */ 1656 if (!IS_ERR(p)) { 1657 struct completion vfork; 1658 struct pid *pid; 1659 1660 trace_sched_process_fork(current, p); 1661 1662 pid = get_task_pid(p, PIDTYPE_PID); 1663 nr = pid_vnr(pid); 1664 1665 if (clone_flags & CLONE_PARENT_SETTID) 1666 put_user(nr, parent_tidptr); 1667 1668 if (clone_flags & CLONE_VFORK) { 1669 p->vfork_done = &vfork; 1670 init_completion(&vfork); 1671 get_task_struct(p); 1672 } 1673 1674 wake_up_new_task(p); 1675 1676 /* forking complete and child started to run, tell ptracer */ 1677 if (unlikely(trace)) 1678 ptrace_event_pid(trace, pid); 1679 1680 if (clone_flags & CLONE_VFORK) { 1681 if (!wait_for_vfork_done(p, &vfork)) 1682 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid); 1683 } 1684 1685 put_pid(pid); 1686 } else { 1687 nr = PTR_ERR(p); 1688 } 1689 return nr; 1690 } 1691 1692 /* 1693 * Create a kernel thread. 1694 */ 1695 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags) 1696 { 1697 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn, 1698 (unsigned long)arg, NULL, NULL); 1699 } 1700 1701 #ifdef __ARCH_WANT_SYS_FORK 1702 SYSCALL_DEFINE0(fork) 1703 { 1704 #ifdef CONFIG_MMU 1705 return do_fork(SIGCHLD, 0, 0, NULL, NULL); 1706 #else 1707 /* can not support in nommu mode */ 1708 return -EINVAL; 1709 #endif 1710 } 1711 #endif 1712 1713 #ifdef __ARCH_WANT_SYS_VFORK 1714 SYSCALL_DEFINE0(vfork) 1715 { 1716 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0, 1717 0, NULL, NULL); 1718 } 1719 #endif 1720 1721 #ifdef __ARCH_WANT_SYS_CLONE 1722 #ifdef CONFIG_CLONE_BACKWARDS 1723 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 1724 int __user *, parent_tidptr, 1725 int, tls_val, 1726 int __user *, child_tidptr) 1727 #elif defined(CONFIG_CLONE_BACKWARDS2) 1728 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags, 1729 int __user *, parent_tidptr, 1730 int __user *, child_tidptr, 1731 int, tls_val) 1732 #elif defined(CONFIG_CLONE_BACKWARDS3) 1733 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp, 1734 int, stack_size, 1735 int __user *, parent_tidptr, 1736 int __user *, child_tidptr, 1737 int, tls_val) 1738 #else 1739 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 1740 int __user *, parent_tidptr, 1741 int __user *, child_tidptr, 1742 int, tls_val) 1743 #endif 1744 { 1745 return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr); 1746 } 1747 #endif 1748 1749 #ifndef ARCH_MIN_MMSTRUCT_ALIGN 1750 #define ARCH_MIN_MMSTRUCT_ALIGN 0 1751 #endif 1752 1753 static void sighand_ctor(void *data) 1754 { 1755 struct sighand_struct *sighand = data; 1756 1757 spin_lock_init(&sighand->siglock); 1758 init_waitqueue_head(&sighand->signalfd_wqh); 1759 } 1760 1761 void __init proc_caches_init(void) 1762 { 1763 sighand_cachep = kmem_cache_create("sighand_cache", 1764 sizeof(struct sighand_struct), 0, 1765 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU| 1766 SLAB_NOTRACK, sighand_ctor); 1767 signal_cachep = kmem_cache_create("signal_cache", 1768 sizeof(struct signal_struct), 0, 1769 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1770 files_cachep = kmem_cache_create("files_cache", 1771 sizeof(struct files_struct), 0, 1772 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1773 fs_cachep = kmem_cache_create("fs_cache", 1774 sizeof(struct fs_struct), 0, 1775 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1776 /* 1777 * FIXME! The "sizeof(struct mm_struct)" currently includes the 1778 * whole struct cpumask for the OFFSTACK case. We could change 1779 * this to *only* allocate as much of it as required by the 1780 * maximum number of CPU's we can ever have. The cpumask_allocation 1781 * is at the end of the structure, exactly for that reason. 1782 */ 1783 mm_cachep = kmem_cache_create("mm_struct", 1784 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN, 1785 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); 1786 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC); 1787 mmap_init(); 1788 nsproxy_cache_init(); 1789 } 1790 1791 /* 1792 * Check constraints on flags passed to the unshare system call. 1793 */ 1794 static int check_unshare_flags(unsigned long unshare_flags) 1795 { 1796 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND| 1797 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM| 1798 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET| 1799 CLONE_NEWUSER|CLONE_NEWPID)) 1800 return -EINVAL; 1801 /* 1802 * Not implemented, but pretend it works if there is nothing to 1803 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND 1804 * needs to unshare vm. 1805 */ 1806 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) { 1807 /* FIXME: get_task_mm() increments ->mm_users */ 1808 if (atomic_read(¤t->mm->mm_users) > 1) 1809 return -EINVAL; 1810 } 1811 1812 return 0; 1813 } 1814 1815 /* 1816 * Unshare the filesystem structure if it is being shared 1817 */ 1818 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) 1819 { 1820 struct fs_struct *fs = current->fs; 1821 1822 if (!(unshare_flags & CLONE_FS) || !fs) 1823 return 0; 1824 1825 /* don't need lock here; in the worst case we'll do useless copy */ 1826 if (fs->users == 1) 1827 return 0; 1828 1829 *new_fsp = copy_fs_struct(fs); 1830 if (!*new_fsp) 1831 return -ENOMEM; 1832 1833 return 0; 1834 } 1835 1836 /* 1837 * Unshare file descriptor table if it is being shared 1838 */ 1839 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp) 1840 { 1841 struct files_struct *fd = current->files; 1842 int error = 0; 1843 1844 if ((unshare_flags & CLONE_FILES) && 1845 (fd && atomic_read(&fd->count) > 1)) { 1846 *new_fdp = dup_fd(fd, &error); 1847 if (!*new_fdp) 1848 return error; 1849 } 1850 1851 return 0; 1852 } 1853 1854 /* 1855 * unshare allows a process to 'unshare' part of the process 1856 * context which was originally shared using clone. copy_* 1857 * functions used by do_fork() cannot be used here directly 1858 * because they modify an inactive task_struct that is being 1859 * constructed. Here we are modifying the current, active, 1860 * task_struct. 1861 */ 1862 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags) 1863 { 1864 struct fs_struct *fs, *new_fs = NULL; 1865 struct files_struct *fd, *new_fd = NULL; 1866 struct cred *new_cred = NULL; 1867 struct nsproxy *new_nsproxy = NULL; 1868 int do_sysvsem = 0; 1869 int err; 1870 1871 /* 1872 * If unsharing a user namespace must also unshare the thread. 1873 */ 1874 if (unshare_flags & CLONE_NEWUSER) 1875 unshare_flags |= CLONE_THREAD | CLONE_FS; 1876 /* 1877 * If unsharing a thread from a thread group, must also unshare vm. 1878 */ 1879 if (unshare_flags & CLONE_THREAD) 1880 unshare_flags |= CLONE_VM; 1881 /* 1882 * If unsharing vm, must also unshare signal handlers. 1883 */ 1884 if (unshare_flags & CLONE_VM) 1885 unshare_flags |= CLONE_SIGHAND; 1886 /* 1887 * If unsharing namespace, must also unshare filesystem information. 1888 */ 1889 if (unshare_flags & CLONE_NEWNS) 1890 unshare_flags |= CLONE_FS; 1891 1892 err = check_unshare_flags(unshare_flags); 1893 if (err) 1894 goto bad_unshare_out; 1895 /* 1896 * CLONE_NEWIPC must also detach from the undolist: after switching 1897 * to a new ipc namespace, the semaphore arrays from the old 1898 * namespace are unreachable. 1899 */ 1900 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) 1901 do_sysvsem = 1; 1902 err = unshare_fs(unshare_flags, &new_fs); 1903 if (err) 1904 goto bad_unshare_out; 1905 err = unshare_fd(unshare_flags, &new_fd); 1906 if (err) 1907 goto bad_unshare_cleanup_fs; 1908 err = unshare_userns(unshare_flags, &new_cred); 1909 if (err) 1910 goto bad_unshare_cleanup_fd; 1911 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, 1912 new_cred, new_fs); 1913 if (err) 1914 goto bad_unshare_cleanup_cred; 1915 1916 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) { 1917 if (do_sysvsem) { 1918 /* 1919 * CLONE_SYSVSEM is equivalent to sys_exit(). 1920 */ 1921 exit_sem(current); 1922 } 1923 if (unshare_flags & CLONE_NEWIPC) { 1924 /* Orphan segments in old ns (see sem above). */ 1925 exit_shm(current); 1926 shm_init_task(current); 1927 } 1928 1929 if (new_nsproxy) 1930 switch_task_namespaces(current, new_nsproxy); 1931 1932 task_lock(current); 1933 1934 if (new_fs) { 1935 fs = current->fs; 1936 spin_lock(&fs->lock); 1937 current->fs = new_fs; 1938 if (--fs->users) 1939 new_fs = NULL; 1940 else 1941 new_fs = fs; 1942 spin_unlock(&fs->lock); 1943 } 1944 1945 if (new_fd) { 1946 fd = current->files; 1947 current->files = new_fd; 1948 new_fd = fd; 1949 } 1950 1951 task_unlock(current); 1952 1953 if (new_cred) { 1954 /* Install the new user namespace */ 1955 commit_creds(new_cred); 1956 new_cred = NULL; 1957 } 1958 } 1959 1960 bad_unshare_cleanup_cred: 1961 if (new_cred) 1962 put_cred(new_cred); 1963 bad_unshare_cleanup_fd: 1964 if (new_fd) 1965 put_files_struct(new_fd); 1966 1967 bad_unshare_cleanup_fs: 1968 if (new_fs) 1969 free_fs_struct(new_fs); 1970 1971 bad_unshare_out: 1972 return err; 1973 } 1974 1975 /* 1976 * Helper to unshare the files of the current task. 1977 * We don't want to expose copy_files internals to 1978 * the exec layer of the kernel. 1979 */ 1980 1981 int unshare_files(struct files_struct **displaced) 1982 { 1983 struct task_struct *task = current; 1984 struct files_struct *copy = NULL; 1985 int error; 1986 1987 error = unshare_fd(CLONE_FILES, ©); 1988 if (error || !copy) { 1989 *displaced = NULL; 1990 return error; 1991 } 1992 *displaced = task->files; 1993 task_lock(task); 1994 task->files = copy; 1995 task_unlock(task); 1996 return 0; 1997 } 1998