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/sched/autogroup.h> 16 #include <linux/sched/mm.h> 17 #include <linux/sched/coredump.h> 18 #include <linux/sched/user.h> 19 #include <linux/sched/numa_balancing.h> 20 #include <linux/sched/stat.h> 21 #include <linux/sched/task.h> 22 #include <linux/sched/task_stack.h> 23 #include <linux/sched/cputime.h> 24 #include <linux/rtmutex.h> 25 #include <linux/init.h> 26 #include <linux/unistd.h> 27 #include <linux/module.h> 28 #include <linux/vmalloc.h> 29 #include <linux/completion.h> 30 #include <linux/personality.h> 31 #include <linux/mempolicy.h> 32 #include <linux/sem.h> 33 #include <linux/file.h> 34 #include <linux/fdtable.h> 35 #include <linux/iocontext.h> 36 #include <linux/key.h> 37 #include <linux/binfmts.h> 38 #include <linux/mman.h> 39 #include <linux/mmu_notifier.h> 40 #include <linux/hmm.h> 41 #include <linux/fs.h> 42 #include <linux/mm.h> 43 #include <linux/vmacache.h> 44 #include <linux/nsproxy.h> 45 #include <linux/capability.h> 46 #include <linux/cpu.h> 47 #include <linux/cgroup.h> 48 #include <linux/security.h> 49 #include <linux/hugetlb.h> 50 #include <linux/seccomp.h> 51 #include <linux/swap.h> 52 #include <linux/syscalls.h> 53 #include <linux/jiffies.h> 54 #include <linux/futex.h> 55 #include <linux/compat.h> 56 #include <linux/kthread.h> 57 #include <linux/task_io_accounting_ops.h> 58 #include <linux/rcupdate.h> 59 #include <linux/ptrace.h> 60 #include <linux/mount.h> 61 #include <linux/audit.h> 62 #include <linux/memcontrol.h> 63 #include <linux/ftrace.h> 64 #include <linux/proc_fs.h> 65 #include <linux/profile.h> 66 #include <linux/rmap.h> 67 #include <linux/ksm.h> 68 #include <linux/acct.h> 69 #include <linux/userfaultfd_k.h> 70 #include <linux/tsacct_kern.h> 71 #include <linux/cn_proc.h> 72 #include <linux/freezer.h> 73 #include <linux/delayacct.h> 74 #include <linux/taskstats_kern.h> 75 #include <linux/random.h> 76 #include <linux/tty.h> 77 #include <linux/blkdev.h> 78 #include <linux/fs_struct.h> 79 #include <linux/magic.h> 80 #include <linux/sched/mm.h> 81 #include <linux/perf_event.h> 82 #include <linux/posix-timers.h> 83 #include <linux/user-return-notifier.h> 84 #include <linux/oom.h> 85 #include <linux/khugepaged.h> 86 #include <linux/signalfd.h> 87 #include <linux/uprobes.h> 88 #include <linux/aio.h> 89 #include <linux/compiler.h> 90 #include <linux/sysctl.h> 91 #include <linux/kcov.h> 92 #include <linux/livepatch.h> 93 #include <linux/thread_info.h> 94 #include <linux/stackleak.h> 95 96 #include <asm/pgtable.h> 97 #include <asm/pgalloc.h> 98 #include <linux/uaccess.h> 99 #include <asm/mmu_context.h> 100 #include <asm/cacheflush.h> 101 #include <asm/tlbflush.h> 102 103 #include <trace/events/sched.h> 104 105 #define CREATE_TRACE_POINTS 106 #include <trace/events/task.h> 107 108 /* 109 * Minimum number of threads to boot the kernel 110 */ 111 #define MIN_THREADS 20 112 113 /* 114 * Maximum number of threads 115 */ 116 #define MAX_THREADS FUTEX_TID_MASK 117 118 /* 119 * Protected counters by write_lock_irq(&tasklist_lock) 120 */ 121 unsigned long total_forks; /* Handle normal Linux uptimes. */ 122 int nr_threads; /* The idle threads do not count.. */ 123 124 int max_threads; /* tunable limit on nr_threads */ 125 126 DEFINE_PER_CPU(unsigned long, process_counts) = 0; 127 128 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */ 129 130 #ifdef CONFIG_PROVE_RCU 131 int lockdep_tasklist_lock_is_held(void) 132 { 133 return lockdep_is_held(&tasklist_lock); 134 } 135 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held); 136 #endif /* #ifdef CONFIG_PROVE_RCU */ 137 138 int nr_processes(void) 139 { 140 int cpu; 141 int total = 0; 142 143 for_each_possible_cpu(cpu) 144 total += per_cpu(process_counts, cpu); 145 146 return total; 147 } 148 149 void __weak arch_release_task_struct(struct task_struct *tsk) 150 { 151 } 152 153 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 154 static struct kmem_cache *task_struct_cachep; 155 156 static inline struct task_struct *alloc_task_struct_node(int node) 157 { 158 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node); 159 } 160 161 static inline void free_task_struct(struct task_struct *tsk) 162 { 163 kmem_cache_free(task_struct_cachep, tsk); 164 } 165 #endif 166 167 void __weak arch_release_thread_stack(unsigned long *stack) 168 { 169 } 170 171 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR 172 173 /* 174 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a 175 * kmemcache based allocator. 176 */ 177 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) 178 179 #ifdef CONFIG_VMAP_STACK 180 /* 181 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB 182 * flush. Try to minimize the number of calls by caching stacks. 183 */ 184 #define NR_CACHED_STACKS 2 185 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]); 186 187 static int free_vm_stack_cache(unsigned int cpu) 188 { 189 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu); 190 int i; 191 192 for (i = 0; i < NR_CACHED_STACKS; i++) { 193 struct vm_struct *vm_stack = cached_vm_stacks[i]; 194 195 if (!vm_stack) 196 continue; 197 198 vfree(vm_stack->addr); 199 cached_vm_stacks[i] = NULL; 200 } 201 202 return 0; 203 } 204 #endif 205 206 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node) 207 { 208 #ifdef CONFIG_VMAP_STACK 209 void *stack; 210 int i; 211 212 for (i = 0; i < NR_CACHED_STACKS; i++) { 213 struct vm_struct *s; 214 215 s = this_cpu_xchg(cached_stacks[i], NULL); 216 217 if (!s) 218 continue; 219 220 /* Clear stale pointers from reused stack. */ 221 memset(s->addr, 0, THREAD_SIZE); 222 223 tsk->stack_vm_area = s; 224 return s->addr; 225 } 226 227 /* 228 * Allocated stacks are cached and later reused by new threads, 229 * so memcg accounting is performed manually on assigning/releasing 230 * stacks to tasks. Drop __GFP_ACCOUNT. 231 */ 232 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN, 233 VMALLOC_START, VMALLOC_END, 234 THREADINFO_GFP & ~__GFP_ACCOUNT, 235 PAGE_KERNEL, 236 0, node, __builtin_return_address(0)); 237 238 /* 239 * We can't call find_vm_area() in interrupt context, and 240 * free_thread_stack() can be called in interrupt context, 241 * so cache the vm_struct. 242 */ 243 if (stack) { 244 tsk->stack_vm_area = find_vm_area(stack); 245 tsk->stack = stack; 246 } 247 return stack; 248 #else 249 struct page *page = alloc_pages_node(node, THREADINFO_GFP, 250 THREAD_SIZE_ORDER); 251 252 return page ? page_address(page) : NULL; 253 #endif 254 } 255 256 static inline void free_thread_stack(struct task_struct *tsk) 257 { 258 #ifdef CONFIG_VMAP_STACK 259 struct vm_struct *vm = task_stack_vm_area(tsk); 260 261 if (vm) { 262 int i; 263 264 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 265 mod_memcg_page_state(vm->pages[i], 266 MEMCG_KERNEL_STACK_KB, 267 -(int)(PAGE_SIZE / 1024)); 268 269 memcg_kmem_uncharge(vm->pages[i], 0); 270 } 271 272 for (i = 0; i < NR_CACHED_STACKS; i++) { 273 if (this_cpu_cmpxchg(cached_stacks[i], 274 NULL, tsk->stack_vm_area) != NULL) 275 continue; 276 277 return; 278 } 279 280 vfree_atomic(tsk->stack); 281 return; 282 } 283 #endif 284 285 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER); 286 } 287 # else 288 static struct kmem_cache *thread_stack_cache; 289 290 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, 291 int node) 292 { 293 unsigned long *stack; 294 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node); 295 tsk->stack = stack; 296 return stack; 297 } 298 299 static void free_thread_stack(struct task_struct *tsk) 300 { 301 kmem_cache_free(thread_stack_cache, tsk->stack); 302 } 303 304 void thread_stack_cache_init(void) 305 { 306 thread_stack_cache = kmem_cache_create_usercopy("thread_stack", 307 THREAD_SIZE, THREAD_SIZE, 0, 0, 308 THREAD_SIZE, NULL); 309 BUG_ON(thread_stack_cache == NULL); 310 } 311 # endif 312 #endif 313 314 /* SLAB cache for signal_struct structures (tsk->signal) */ 315 static struct kmem_cache *signal_cachep; 316 317 /* SLAB cache for sighand_struct structures (tsk->sighand) */ 318 struct kmem_cache *sighand_cachep; 319 320 /* SLAB cache for files_struct structures (tsk->files) */ 321 struct kmem_cache *files_cachep; 322 323 /* SLAB cache for fs_struct structures (tsk->fs) */ 324 struct kmem_cache *fs_cachep; 325 326 /* SLAB cache for vm_area_struct structures */ 327 static struct kmem_cache *vm_area_cachep; 328 329 /* SLAB cache for mm_struct structures (tsk->mm) */ 330 static struct kmem_cache *mm_cachep; 331 332 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm) 333 { 334 struct vm_area_struct *vma; 335 336 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 337 if (vma) 338 vma_init(vma, mm); 339 return vma; 340 } 341 342 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig) 343 { 344 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 345 346 if (new) { 347 *new = *orig; 348 INIT_LIST_HEAD(&new->anon_vma_chain); 349 } 350 return new; 351 } 352 353 void vm_area_free(struct vm_area_struct *vma) 354 { 355 kmem_cache_free(vm_area_cachep, vma); 356 } 357 358 static void account_kernel_stack(struct task_struct *tsk, int account) 359 { 360 void *stack = task_stack_page(tsk); 361 struct vm_struct *vm = task_stack_vm_area(tsk); 362 363 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0); 364 365 if (vm) { 366 int i; 367 368 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE); 369 370 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 371 mod_zone_page_state(page_zone(vm->pages[i]), 372 NR_KERNEL_STACK_KB, 373 PAGE_SIZE / 1024 * account); 374 } 375 } else { 376 /* 377 * All stack pages are in the same zone and belong to the 378 * same memcg. 379 */ 380 struct page *first_page = virt_to_page(stack); 381 382 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB, 383 THREAD_SIZE / 1024 * account); 384 385 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB, 386 account * (THREAD_SIZE / 1024)); 387 } 388 } 389 390 static int memcg_charge_kernel_stack(struct task_struct *tsk) 391 { 392 #ifdef CONFIG_VMAP_STACK 393 struct vm_struct *vm = task_stack_vm_area(tsk); 394 int ret; 395 396 if (vm) { 397 int i; 398 399 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 400 /* 401 * If memcg_kmem_charge() fails, page->mem_cgroup 402 * pointer is NULL, and both memcg_kmem_uncharge() 403 * and mod_memcg_page_state() in free_thread_stack() 404 * will ignore this page. So it's safe. 405 */ 406 ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0); 407 if (ret) 408 return ret; 409 410 mod_memcg_page_state(vm->pages[i], 411 MEMCG_KERNEL_STACK_KB, 412 PAGE_SIZE / 1024); 413 } 414 } 415 #endif 416 return 0; 417 } 418 419 static void release_task_stack(struct task_struct *tsk) 420 { 421 if (WARN_ON(tsk->state != TASK_DEAD)) 422 return; /* Better to leak the stack than to free prematurely */ 423 424 account_kernel_stack(tsk, -1); 425 arch_release_thread_stack(tsk->stack); 426 free_thread_stack(tsk); 427 tsk->stack = NULL; 428 #ifdef CONFIG_VMAP_STACK 429 tsk->stack_vm_area = NULL; 430 #endif 431 } 432 433 #ifdef CONFIG_THREAD_INFO_IN_TASK 434 void put_task_stack(struct task_struct *tsk) 435 { 436 if (atomic_dec_and_test(&tsk->stack_refcount)) 437 release_task_stack(tsk); 438 } 439 #endif 440 441 void free_task(struct task_struct *tsk) 442 { 443 #ifndef CONFIG_THREAD_INFO_IN_TASK 444 /* 445 * The task is finally done with both the stack and thread_info, 446 * so free both. 447 */ 448 release_task_stack(tsk); 449 #else 450 /* 451 * If the task had a separate stack allocation, it should be gone 452 * by now. 453 */ 454 WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0); 455 #endif 456 rt_mutex_debug_task_free(tsk); 457 ftrace_graph_exit_task(tsk); 458 put_seccomp_filter(tsk); 459 arch_release_task_struct(tsk); 460 if (tsk->flags & PF_KTHREAD) 461 free_kthread_struct(tsk); 462 free_task_struct(tsk); 463 } 464 EXPORT_SYMBOL(free_task); 465 466 #ifdef CONFIG_MMU 467 static __latent_entropy int dup_mmap(struct mm_struct *mm, 468 struct mm_struct *oldmm) 469 { 470 struct vm_area_struct *mpnt, *tmp, *prev, **pprev; 471 struct rb_node **rb_link, *rb_parent; 472 int retval; 473 unsigned long charge; 474 LIST_HEAD(uf); 475 476 uprobe_start_dup_mmap(); 477 if (down_write_killable(&oldmm->mmap_sem)) { 478 retval = -EINTR; 479 goto fail_uprobe_end; 480 } 481 flush_cache_dup_mm(oldmm); 482 uprobe_dup_mmap(oldmm, mm); 483 /* 484 * Not linked in yet - no deadlock potential: 485 */ 486 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING); 487 488 /* No ordering required: file already has been exposed. */ 489 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm)); 490 491 mm->total_vm = oldmm->total_vm; 492 mm->data_vm = oldmm->data_vm; 493 mm->exec_vm = oldmm->exec_vm; 494 mm->stack_vm = oldmm->stack_vm; 495 496 rb_link = &mm->mm_rb.rb_node; 497 rb_parent = NULL; 498 pprev = &mm->mmap; 499 retval = ksm_fork(mm, oldmm); 500 if (retval) 501 goto out; 502 retval = khugepaged_fork(mm, oldmm); 503 if (retval) 504 goto out; 505 506 prev = NULL; 507 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { 508 struct file *file; 509 510 if (mpnt->vm_flags & VM_DONTCOPY) { 511 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt)); 512 continue; 513 } 514 charge = 0; 515 /* 516 * Don't duplicate many vmas if we've been oom-killed (for 517 * example) 518 */ 519 if (fatal_signal_pending(current)) { 520 retval = -EINTR; 521 goto out; 522 } 523 if (mpnt->vm_flags & VM_ACCOUNT) { 524 unsigned long len = vma_pages(mpnt); 525 526 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */ 527 goto fail_nomem; 528 charge = len; 529 } 530 tmp = vm_area_dup(mpnt); 531 if (!tmp) 532 goto fail_nomem; 533 retval = vma_dup_policy(mpnt, tmp); 534 if (retval) 535 goto fail_nomem_policy; 536 tmp->vm_mm = mm; 537 retval = dup_userfaultfd(tmp, &uf); 538 if (retval) 539 goto fail_nomem_anon_vma_fork; 540 if (tmp->vm_flags & VM_WIPEONFORK) { 541 /* VM_WIPEONFORK gets a clean slate in the child. */ 542 tmp->anon_vma = NULL; 543 if (anon_vma_prepare(tmp)) 544 goto fail_nomem_anon_vma_fork; 545 } else if (anon_vma_fork(tmp, mpnt)) 546 goto fail_nomem_anon_vma_fork; 547 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT); 548 tmp->vm_next = tmp->vm_prev = NULL; 549 file = tmp->vm_file; 550 if (file) { 551 struct inode *inode = file_inode(file); 552 struct address_space *mapping = file->f_mapping; 553 554 get_file(file); 555 if (tmp->vm_flags & VM_DENYWRITE) 556 atomic_dec(&inode->i_writecount); 557 i_mmap_lock_write(mapping); 558 if (tmp->vm_flags & VM_SHARED) 559 atomic_inc(&mapping->i_mmap_writable); 560 flush_dcache_mmap_lock(mapping); 561 /* insert tmp into the share list, just after mpnt */ 562 vma_interval_tree_insert_after(tmp, mpnt, 563 &mapping->i_mmap); 564 flush_dcache_mmap_unlock(mapping); 565 i_mmap_unlock_write(mapping); 566 } 567 568 /* 569 * Clear hugetlb-related page reserves for children. This only 570 * affects MAP_PRIVATE mappings. Faults generated by the child 571 * are not guaranteed to succeed, even if read-only 572 */ 573 if (is_vm_hugetlb_page(tmp)) 574 reset_vma_resv_huge_pages(tmp); 575 576 /* 577 * Link in the new vma and copy the page table entries. 578 */ 579 *pprev = tmp; 580 pprev = &tmp->vm_next; 581 tmp->vm_prev = prev; 582 prev = tmp; 583 584 __vma_link_rb(mm, tmp, rb_link, rb_parent); 585 rb_link = &tmp->vm_rb.rb_right; 586 rb_parent = &tmp->vm_rb; 587 588 mm->map_count++; 589 if (!(tmp->vm_flags & VM_WIPEONFORK)) 590 retval = copy_page_range(mm, oldmm, mpnt); 591 592 if (tmp->vm_ops && tmp->vm_ops->open) 593 tmp->vm_ops->open(tmp); 594 595 if (retval) 596 goto out; 597 } 598 /* a new mm has just been created */ 599 retval = arch_dup_mmap(oldmm, mm); 600 out: 601 up_write(&mm->mmap_sem); 602 flush_tlb_mm(oldmm); 603 up_write(&oldmm->mmap_sem); 604 dup_userfaultfd_complete(&uf); 605 fail_uprobe_end: 606 uprobe_end_dup_mmap(); 607 return retval; 608 fail_nomem_anon_vma_fork: 609 mpol_put(vma_policy(tmp)); 610 fail_nomem_policy: 611 vm_area_free(tmp); 612 fail_nomem: 613 retval = -ENOMEM; 614 vm_unacct_memory(charge); 615 goto out; 616 } 617 618 static inline int mm_alloc_pgd(struct mm_struct *mm) 619 { 620 mm->pgd = pgd_alloc(mm); 621 if (unlikely(!mm->pgd)) 622 return -ENOMEM; 623 return 0; 624 } 625 626 static inline void mm_free_pgd(struct mm_struct *mm) 627 { 628 pgd_free(mm, mm->pgd); 629 } 630 #else 631 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) 632 { 633 down_write(&oldmm->mmap_sem); 634 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm)); 635 up_write(&oldmm->mmap_sem); 636 return 0; 637 } 638 #define mm_alloc_pgd(mm) (0) 639 #define mm_free_pgd(mm) 640 #endif /* CONFIG_MMU */ 641 642 static void check_mm(struct mm_struct *mm) 643 { 644 int i; 645 646 for (i = 0; i < NR_MM_COUNTERS; i++) { 647 long x = atomic_long_read(&mm->rss_stat.count[i]); 648 649 if (unlikely(x)) 650 printk(KERN_ALERT "BUG: Bad rss-counter state " 651 "mm:%p idx:%d val:%ld\n", mm, i, x); 652 } 653 654 if (mm_pgtables_bytes(mm)) 655 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n", 656 mm_pgtables_bytes(mm)); 657 658 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 659 VM_BUG_ON_MM(mm->pmd_huge_pte, mm); 660 #endif 661 } 662 663 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL)) 664 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) 665 666 /* 667 * Called when the last reference to the mm 668 * is dropped: either by a lazy thread or by 669 * mmput. Free the page directory and the mm. 670 */ 671 void __mmdrop(struct mm_struct *mm) 672 { 673 BUG_ON(mm == &init_mm); 674 WARN_ON_ONCE(mm == current->mm); 675 WARN_ON_ONCE(mm == current->active_mm); 676 mm_free_pgd(mm); 677 destroy_context(mm); 678 hmm_mm_destroy(mm); 679 mmu_notifier_mm_destroy(mm); 680 check_mm(mm); 681 put_user_ns(mm->user_ns); 682 free_mm(mm); 683 } 684 EXPORT_SYMBOL_GPL(__mmdrop); 685 686 static void mmdrop_async_fn(struct work_struct *work) 687 { 688 struct mm_struct *mm; 689 690 mm = container_of(work, struct mm_struct, async_put_work); 691 __mmdrop(mm); 692 } 693 694 static void mmdrop_async(struct mm_struct *mm) 695 { 696 if (unlikely(atomic_dec_and_test(&mm->mm_count))) { 697 INIT_WORK(&mm->async_put_work, mmdrop_async_fn); 698 schedule_work(&mm->async_put_work); 699 } 700 } 701 702 static inline void free_signal_struct(struct signal_struct *sig) 703 { 704 taskstats_tgid_free(sig); 705 sched_autogroup_exit(sig); 706 /* 707 * __mmdrop is not safe to call from softirq context on x86 due to 708 * pgd_dtor so postpone it to the async context 709 */ 710 if (sig->oom_mm) 711 mmdrop_async(sig->oom_mm); 712 kmem_cache_free(signal_cachep, sig); 713 } 714 715 static inline void put_signal_struct(struct signal_struct *sig) 716 { 717 if (atomic_dec_and_test(&sig->sigcnt)) 718 free_signal_struct(sig); 719 } 720 721 void __put_task_struct(struct task_struct *tsk) 722 { 723 WARN_ON(!tsk->exit_state); 724 WARN_ON(atomic_read(&tsk->usage)); 725 WARN_ON(tsk == current); 726 727 cgroup_free(tsk); 728 task_numa_free(tsk); 729 security_task_free(tsk); 730 exit_creds(tsk); 731 delayacct_tsk_free(tsk); 732 put_signal_struct(tsk->signal); 733 734 if (!profile_handoff_task(tsk)) 735 free_task(tsk); 736 } 737 EXPORT_SYMBOL_GPL(__put_task_struct); 738 739 void __init __weak arch_task_cache_init(void) { } 740 741 /* 742 * set_max_threads 743 */ 744 static void set_max_threads(unsigned int max_threads_suggested) 745 { 746 u64 threads; 747 748 /* 749 * The number of threads shall be limited such that the thread 750 * structures may only consume a small part of the available memory. 751 */ 752 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64) 753 threads = MAX_THREADS; 754 else 755 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE, 756 (u64) THREAD_SIZE * 8UL); 757 758 if (threads > max_threads_suggested) 759 threads = max_threads_suggested; 760 761 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS); 762 } 763 764 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT 765 /* Initialized by the architecture: */ 766 int arch_task_struct_size __read_mostly; 767 #endif 768 769 static void task_struct_whitelist(unsigned long *offset, unsigned long *size) 770 { 771 /* Fetch thread_struct whitelist for the architecture. */ 772 arch_thread_struct_whitelist(offset, size); 773 774 /* 775 * Handle zero-sized whitelist or empty thread_struct, otherwise 776 * adjust offset to position of thread_struct in task_struct. 777 */ 778 if (unlikely(*size == 0)) 779 *offset = 0; 780 else 781 *offset += offsetof(struct task_struct, thread); 782 } 783 784 void __init fork_init(void) 785 { 786 int i; 787 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 788 #ifndef ARCH_MIN_TASKALIGN 789 #define ARCH_MIN_TASKALIGN 0 790 #endif 791 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN); 792 unsigned long useroffset, usersize; 793 794 /* create a slab on which task_structs can be allocated */ 795 task_struct_whitelist(&useroffset, &usersize); 796 task_struct_cachep = kmem_cache_create_usercopy("task_struct", 797 arch_task_struct_size, align, 798 SLAB_PANIC|SLAB_ACCOUNT, 799 useroffset, usersize, NULL); 800 #endif 801 802 /* do the arch specific task caches init */ 803 arch_task_cache_init(); 804 805 set_max_threads(MAX_THREADS); 806 807 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 808 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 809 init_task.signal->rlim[RLIMIT_SIGPENDING] = 810 init_task.signal->rlim[RLIMIT_NPROC]; 811 812 for (i = 0; i < UCOUNT_COUNTS; i++) { 813 init_user_ns.ucount_max[i] = max_threads/2; 814 } 815 816 #ifdef CONFIG_VMAP_STACK 817 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache", 818 NULL, free_vm_stack_cache); 819 #endif 820 821 lockdep_init_task(&init_task); 822 } 823 824 int __weak arch_dup_task_struct(struct task_struct *dst, 825 struct task_struct *src) 826 { 827 *dst = *src; 828 return 0; 829 } 830 831 void set_task_stack_end_magic(struct task_struct *tsk) 832 { 833 unsigned long *stackend; 834 835 stackend = end_of_stack(tsk); 836 *stackend = STACK_END_MAGIC; /* for overflow detection */ 837 } 838 839 static struct task_struct *dup_task_struct(struct task_struct *orig, int node) 840 { 841 struct task_struct *tsk; 842 unsigned long *stack; 843 struct vm_struct *stack_vm_area; 844 int err; 845 846 if (node == NUMA_NO_NODE) 847 node = tsk_fork_get_node(orig); 848 tsk = alloc_task_struct_node(node); 849 if (!tsk) 850 return NULL; 851 852 stack = alloc_thread_stack_node(tsk, node); 853 if (!stack) 854 goto free_tsk; 855 856 if (memcg_charge_kernel_stack(tsk)) 857 goto free_stack; 858 859 stack_vm_area = task_stack_vm_area(tsk); 860 861 err = arch_dup_task_struct(tsk, orig); 862 863 /* 864 * arch_dup_task_struct() clobbers the stack-related fields. Make 865 * sure they're properly initialized before using any stack-related 866 * functions again. 867 */ 868 tsk->stack = stack; 869 #ifdef CONFIG_VMAP_STACK 870 tsk->stack_vm_area = stack_vm_area; 871 #endif 872 #ifdef CONFIG_THREAD_INFO_IN_TASK 873 atomic_set(&tsk->stack_refcount, 1); 874 #endif 875 876 if (err) 877 goto free_stack; 878 879 #ifdef CONFIG_SECCOMP 880 /* 881 * We must handle setting up seccomp filters once we're under 882 * the sighand lock in case orig has changed between now and 883 * then. Until then, filter must be NULL to avoid messing up 884 * the usage counts on the error path calling free_task. 885 */ 886 tsk->seccomp.filter = NULL; 887 #endif 888 889 setup_thread_stack(tsk, orig); 890 clear_user_return_notifier(tsk); 891 clear_tsk_need_resched(tsk); 892 set_task_stack_end_magic(tsk); 893 894 #ifdef CONFIG_STACKPROTECTOR 895 tsk->stack_canary = get_random_canary(); 896 #endif 897 898 /* 899 * One for us, one for whoever does the "release_task()" (usually 900 * parent) 901 */ 902 atomic_set(&tsk->usage, 2); 903 #ifdef CONFIG_BLK_DEV_IO_TRACE 904 tsk->btrace_seq = 0; 905 #endif 906 tsk->splice_pipe = NULL; 907 tsk->task_frag.page = NULL; 908 tsk->wake_q.next = NULL; 909 910 account_kernel_stack(tsk, 1); 911 912 kcov_task_init(tsk); 913 914 #ifdef CONFIG_FAULT_INJECTION 915 tsk->fail_nth = 0; 916 #endif 917 918 #ifdef CONFIG_BLK_CGROUP 919 tsk->throttle_queue = NULL; 920 tsk->use_memdelay = 0; 921 #endif 922 923 #ifdef CONFIG_MEMCG 924 tsk->active_memcg = NULL; 925 #endif 926 return tsk; 927 928 free_stack: 929 free_thread_stack(tsk); 930 free_tsk: 931 free_task_struct(tsk); 932 return NULL; 933 } 934 935 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); 936 937 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT; 938 939 static int __init coredump_filter_setup(char *s) 940 { 941 default_dump_filter = 942 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) & 943 MMF_DUMP_FILTER_MASK; 944 return 1; 945 } 946 947 __setup("coredump_filter=", coredump_filter_setup); 948 949 #include <linux/init_task.h> 950 951 static void mm_init_aio(struct mm_struct *mm) 952 { 953 #ifdef CONFIG_AIO 954 spin_lock_init(&mm->ioctx_lock); 955 mm->ioctx_table = NULL; 956 #endif 957 } 958 959 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p) 960 { 961 #ifdef CONFIG_MEMCG 962 mm->owner = p; 963 #endif 964 } 965 966 static void mm_init_uprobes_state(struct mm_struct *mm) 967 { 968 #ifdef CONFIG_UPROBES 969 mm->uprobes_state.xol_area = NULL; 970 #endif 971 } 972 973 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p, 974 struct user_namespace *user_ns) 975 { 976 mm->mmap = NULL; 977 mm->mm_rb = RB_ROOT; 978 mm->vmacache_seqnum = 0; 979 atomic_set(&mm->mm_users, 1); 980 atomic_set(&mm->mm_count, 1); 981 init_rwsem(&mm->mmap_sem); 982 INIT_LIST_HEAD(&mm->mmlist); 983 mm->core_state = NULL; 984 mm_pgtables_bytes_init(mm); 985 mm->map_count = 0; 986 mm->locked_vm = 0; 987 mm->pinned_vm = 0; 988 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat)); 989 spin_lock_init(&mm->page_table_lock); 990 spin_lock_init(&mm->arg_lock); 991 mm_init_cpumask(mm); 992 mm_init_aio(mm); 993 mm_init_owner(mm, p); 994 RCU_INIT_POINTER(mm->exe_file, NULL); 995 mmu_notifier_mm_init(mm); 996 hmm_mm_init(mm); 997 init_tlb_flush_pending(mm); 998 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 999 mm->pmd_huge_pte = NULL; 1000 #endif 1001 mm_init_uprobes_state(mm); 1002 1003 if (current->mm) { 1004 mm->flags = current->mm->flags & MMF_INIT_MASK; 1005 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK; 1006 } else { 1007 mm->flags = default_dump_filter; 1008 mm->def_flags = 0; 1009 } 1010 1011 if (mm_alloc_pgd(mm)) 1012 goto fail_nopgd; 1013 1014 if (init_new_context(p, mm)) 1015 goto fail_nocontext; 1016 1017 mm->user_ns = get_user_ns(user_ns); 1018 return mm; 1019 1020 fail_nocontext: 1021 mm_free_pgd(mm); 1022 fail_nopgd: 1023 free_mm(mm); 1024 return NULL; 1025 } 1026 1027 /* 1028 * Allocate and initialize an mm_struct. 1029 */ 1030 struct mm_struct *mm_alloc(void) 1031 { 1032 struct mm_struct *mm; 1033 1034 mm = allocate_mm(); 1035 if (!mm) 1036 return NULL; 1037 1038 memset(mm, 0, sizeof(*mm)); 1039 return mm_init(mm, current, current_user_ns()); 1040 } 1041 1042 static inline void __mmput(struct mm_struct *mm) 1043 { 1044 VM_BUG_ON(atomic_read(&mm->mm_users)); 1045 1046 uprobe_clear_state(mm); 1047 exit_aio(mm); 1048 ksm_exit(mm); 1049 khugepaged_exit(mm); /* must run before exit_mmap */ 1050 exit_mmap(mm); 1051 mm_put_huge_zero_page(mm); 1052 set_mm_exe_file(mm, NULL); 1053 if (!list_empty(&mm->mmlist)) { 1054 spin_lock(&mmlist_lock); 1055 list_del(&mm->mmlist); 1056 spin_unlock(&mmlist_lock); 1057 } 1058 if (mm->binfmt) 1059 module_put(mm->binfmt->module); 1060 mmdrop(mm); 1061 } 1062 1063 /* 1064 * Decrement the use count and release all resources for an mm. 1065 */ 1066 void mmput(struct mm_struct *mm) 1067 { 1068 might_sleep(); 1069 1070 if (atomic_dec_and_test(&mm->mm_users)) 1071 __mmput(mm); 1072 } 1073 EXPORT_SYMBOL_GPL(mmput); 1074 1075 #ifdef CONFIG_MMU 1076 static void mmput_async_fn(struct work_struct *work) 1077 { 1078 struct mm_struct *mm = container_of(work, struct mm_struct, 1079 async_put_work); 1080 1081 __mmput(mm); 1082 } 1083 1084 void mmput_async(struct mm_struct *mm) 1085 { 1086 if (atomic_dec_and_test(&mm->mm_users)) { 1087 INIT_WORK(&mm->async_put_work, mmput_async_fn); 1088 schedule_work(&mm->async_put_work); 1089 } 1090 } 1091 #endif 1092 1093 /** 1094 * set_mm_exe_file - change a reference to the mm's executable file 1095 * 1096 * This changes mm's executable file (shown as symlink /proc/[pid]/exe). 1097 * 1098 * Main users are mmput() and sys_execve(). Callers prevent concurrent 1099 * invocations: in mmput() nobody alive left, in execve task is single 1100 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the 1101 * mm->exe_file, but does so without using set_mm_exe_file() in order 1102 * to do avoid the need for any locks. 1103 */ 1104 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 1105 { 1106 struct file *old_exe_file; 1107 1108 /* 1109 * It is safe to dereference the exe_file without RCU as 1110 * this function is only called if nobody else can access 1111 * this mm -- see comment above for justification. 1112 */ 1113 old_exe_file = rcu_dereference_raw(mm->exe_file); 1114 1115 if (new_exe_file) 1116 get_file(new_exe_file); 1117 rcu_assign_pointer(mm->exe_file, new_exe_file); 1118 if (old_exe_file) 1119 fput(old_exe_file); 1120 } 1121 1122 /** 1123 * get_mm_exe_file - acquire a reference to the mm's executable file 1124 * 1125 * Returns %NULL if mm has no associated executable file. 1126 * User must release file via fput(). 1127 */ 1128 struct file *get_mm_exe_file(struct mm_struct *mm) 1129 { 1130 struct file *exe_file; 1131 1132 rcu_read_lock(); 1133 exe_file = rcu_dereference(mm->exe_file); 1134 if (exe_file && !get_file_rcu(exe_file)) 1135 exe_file = NULL; 1136 rcu_read_unlock(); 1137 return exe_file; 1138 } 1139 EXPORT_SYMBOL(get_mm_exe_file); 1140 1141 /** 1142 * get_task_exe_file - acquire a reference to the task's executable file 1143 * 1144 * Returns %NULL if task's mm (if any) has no associated executable file or 1145 * this is a kernel thread with borrowed mm (see the comment above get_task_mm). 1146 * User must release file via fput(). 1147 */ 1148 struct file *get_task_exe_file(struct task_struct *task) 1149 { 1150 struct file *exe_file = NULL; 1151 struct mm_struct *mm; 1152 1153 task_lock(task); 1154 mm = task->mm; 1155 if (mm) { 1156 if (!(task->flags & PF_KTHREAD)) 1157 exe_file = get_mm_exe_file(mm); 1158 } 1159 task_unlock(task); 1160 return exe_file; 1161 } 1162 EXPORT_SYMBOL(get_task_exe_file); 1163 1164 /** 1165 * get_task_mm - acquire a reference to the task's mm 1166 * 1167 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning 1168 * this kernel workthread has transiently adopted a user mm with use_mm, 1169 * to do its AIO) is not set and if so returns a reference to it, after 1170 * bumping up the use count. User must release the mm via mmput() 1171 * after use. Typically used by /proc and ptrace. 1172 */ 1173 struct mm_struct *get_task_mm(struct task_struct *task) 1174 { 1175 struct mm_struct *mm; 1176 1177 task_lock(task); 1178 mm = task->mm; 1179 if (mm) { 1180 if (task->flags & PF_KTHREAD) 1181 mm = NULL; 1182 else 1183 mmget(mm); 1184 } 1185 task_unlock(task); 1186 return mm; 1187 } 1188 EXPORT_SYMBOL_GPL(get_task_mm); 1189 1190 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode) 1191 { 1192 struct mm_struct *mm; 1193 int err; 1194 1195 err = mutex_lock_killable(&task->signal->cred_guard_mutex); 1196 if (err) 1197 return ERR_PTR(err); 1198 1199 mm = get_task_mm(task); 1200 if (mm && mm != current->mm && 1201 !ptrace_may_access(task, mode)) { 1202 mmput(mm); 1203 mm = ERR_PTR(-EACCES); 1204 } 1205 mutex_unlock(&task->signal->cred_guard_mutex); 1206 1207 return mm; 1208 } 1209 1210 static void complete_vfork_done(struct task_struct *tsk) 1211 { 1212 struct completion *vfork; 1213 1214 task_lock(tsk); 1215 vfork = tsk->vfork_done; 1216 if (likely(vfork)) { 1217 tsk->vfork_done = NULL; 1218 complete(vfork); 1219 } 1220 task_unlock(tsk); 1221 } 1222 1223 static int wait_for_vfork_done(struct task_struct *child, 1224 struct completion *vfork) 1225 { 1226 int killed; 1227 1228 freezer_do_not_count(); 1229 killed = wait_for_completion_killable(vfork); 1230 freezer_count(); 1231 1232 if (killed) { 1233 task_lock(child); 1234 child->vfork_done = NULL; 1235 task_unlock(child); 1236 } 1237 1238 put_task_struct(child); 1239 return killed; 1240 } 1241 1242 /* Please note the differences between mmput and mm_release. 1243 * mmput is called whenever we stop holding onto a mm_struct, 1244 * error success whatever. 1245 * 1246 * mm_release is called after a mm_struct has been removed 1247 * from the current process. 1248 * 1249 * This difference is important for error handling, when we 1250 * only half set up a mm_struct for a new process and need to restore 1251 * the old one. Because we mmput the new mm_struct before 1252 * restoring the old one. . . 1253 * Eric Biederman 10 January 1998 1254 */ 1255 void mm_release(struct task_struct *tsk, struct mm_struct *mm) 1256 { 1257 /* Get rid of any futexes when releasing the mm */ 1258 #ifdef CONFIG_FUTEX 1259 if (unlikely(tsk->robust_list)) { 1260 exit_robust_list(tsk); 1261 tsk->robust_list = NULL; 1262 } 1263 #ifdef CONFIG_COMPAT 1264 if (unlikely(tsk->compat_robust_list)) { 1265 compat_exit_robust_list(tsk); 1266 tsk->compat_robust_list = NULL; 1267 } 1268 #endif 1269 if (unlikely(!list_empty(&tsk->pi_state_list))) 1270 exit_pi_state_list(tsk); 1271 #endif 1272 1273 uprobe_free_utask(tsk); 1274 1275 /* Get rid of any cached register state */ 1276 deactivate_mm(tsk, mm); 1277 1278 /* 1279 * Signal userspace if we're not exiting with a core dump 1280 * because we want to leave the value intact for debugging 1281 * purposes. 1282 */ 1283 if (tsk->clear_child_tid) { 1284 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) && 1285 atomic_read(&mm->mm_users) > 1) { 1286 /* 1287 * We don't check the error code - if userspace has 1288 * not set up a proper pointer then tough luck. 1289 */ 1290 put_user(0, tsk->clear_child_tid); 1291 do_futex(tsk->clear_child_tid, FUTEX_WAKE, 1292 1, NULL, NULL, 0, 0); 1293 } 1294 tsk->clear_child_tid = NULL; 1295 } 1296 1297 /* 1298 * All done, finally we can wake up parent and return this mm to him. 1299 * Also kthread_stop() uses this completion for synchronization. 1300 */ 1301 if (tsk->vfork_done) 1302 complete_vfork_done(tsk); 1303 } 1304 1305 /* 1306 * Allocate a new mm structure and copy contents from the 1307 * mm structure of the passed in task structure. 1308 */ 1309 static struct mm_struct *dup_mm(struct task_struct *tsk) 1310 { 1311 struct mm_struct *mm, *oldmm = current->mm; 1312 int err; 1313 1314 mm = allocate_mm(); 1315 if (!mm) 1316 goto fail_nomem; 1317 1318 memcpy(mm, oldmm, sizeof(*mm)); 1319 1320 if (!mm_init(mm, tsk, mm->user_ns)) 1321 goto fail_nomem; 1322 1323 err = dup_mmap(mm, oldmm); 1324 if (err) 1325 goto free_pt; 1326 1327 mm->hiwater_rss = get_mm_rss(mm); 1328 mm->hiwater_vm = mm->total_vm; 1329 1330 if (mm->binfmt && !try_module_get(mm->binfmt->module)) 1331 goto free_pt; 1332 1333 return mm; 1334 1335 free_pt: 1336 /* don't put binfmt in mmput, we haven't got module yet */ 1337 mm->binfmt = NULL; 1338 mmput(mm); 1339 1340 fail_nomem: 1341 return NULL; 1342 } 1343 1344 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk) 1345 { 1346 struct mm_struct *mm, *oldmm; 1347 int retval; 1348 1349 tsk->min_flt = tsk->maj_flt = 0; 1350 tsk->nvcsw = tsk->nivcsw = 0; 1351 #ifdef CONFIG_DETECT_HUNG_TASK 1352 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; 1353 tsk->last_switch_time = 0; 1354 #endif 1355 1356 tsk->mm = NULL; 1357 tsk->active_mm = NULL; 1358 1359 /* 1360 * Are we cloning a kernel thread? 1361 * 1362 * We need to steal a active VM for that.. 1363 */ 1364 oldmm = current->mm; 1365 if (!oldmm) 1366 return 0; 1367 1368 /* initialize the new vmacache entries */ 1369 vmacache_flush(tsk); 1370 1371 if (clone_flags & CLONE_VM) { 1372 mmget(oldmm); 1373 mm = oldmm; 1374 goto good_mm; 1375 } 1376 1377 retval = -ENOMEM; 1378 mm = dup_mm(tsk); 1379 if (!mm) 1380 goto fail_nomem; 1381 1382 good_mm: 1383 tsk->mm = mm; 1384 tsk->active_mm = mm; 1385 return 0; 1386 1387 fail_nomem: 1388 return retval; 1389 } 1390 1391 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 1392 { 1393 struct fs_struct *fs = current->fs; 1394 if (clone_flags & CLONE_FS) { 1395 /* tsk->fs is already what we want */ 1396 spin_lock(&fs->lock); 1397 if (fs->in_exec) { 1398 spin_unlock(&fs->lock); 1399 return -EAGAIN; 1400 } 1401 fs->users++; 1402 spin_unlock(&fs->lock); 1403 return 0; 1404 } 1405 tsk->fs = copy_fs_struct(fs); 1406 if (!tsk->fs) 1407 return -ENOMEM; 1408 return 0; 1409 } 1410 1411 static int copy_files(unsigned long clone_flags, struct task_struct *tsk) 1412 { 1413 struct files_struct *oldf, *newf; 1414 int error = 0; 1415 1416 /* 1417 * A background process may not have any files ... 1418 */ 1419 oldf = current->files; 1420 if (!oldf) 1421 goto out; 1422 1423 if (clone_flags & CLONE_FILES) { 1424 atomic_inc(&oldf->count); 1425 goto out; 1426 } 1427 1428 newf = dup_fd(oldf, &error); 1429 if (!newf) 1430 goto out; 1431 1432 tsk->files = newf; 1433 error = 0; 1434 out: 1435 return error; 1436 } 1437 1438 static int copy_io(unsigned long clone_flags, struct task_struct *tsk) 1439 { 1440 #ifdef CONFIG_BLOCK 1441 struct io_context *ioc = current->io_context; 1442 struct io_context *new_ioc; 1443 1444 if (!ioc) 1445 return 0; 1446 /* 1447 * Share io context with parent, if CLONE_IO is set 1448 */ 1449 if (clone_flags & CLONE_IO) { 1450 ioc_task_link(ioc); 1451 tsk->io_context = ioc; 1452 } else if (ioprio_valid(ioc->ioprio)) { 1453 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE); 1454 if (unlikely(!new_ioc)) 1455 return -ENOMEM; 1456 1457 new_ioc->ioprio = ioc->ioprio; 1458 put_io_context(new_ioc); 1459 } 1460 #endif 1461 return 0; 1462 } 1463 1464 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk) 1465 { 1466 struct sighand_struct *sig; 1467 1468 if (clone_flags & CLONE_SIGHAND) { 1469 atomic_inc(¤t->sighand->count); 1470 return 0; 1471 } 1472 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 1473 rcu_assign_pointer(tsk->sighand, sig); 1474 if (!sig) 1475 return -ENOMEM; 1476 1477 atomic_set(&sig->count, 1); 1478 spin_lock_irq(¤t->sighand->siglock); 1479 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 1480 spin_unlock_irq(¤t->sighand->siglock); 1481 return 0; 1482 } 1483 1484 void __cleanup_sighand(struct sighand_struct *sighand) 1485 { 1486 if (atomic_dec_and_test(&sighand->count)) { 1487 signalfd_cleanup(sighand); 1488 /* 1489 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it 1490 * without an RCU grace period, see __lock_task_sighand(). 1491 */ 1492 kmem_cache_free(sighand_cachep, sighand); 1493 } 1494 } 1495 1496 #ifdef CONFIG_POSIX_TIMERS 1497 /* 1498 * Initialize POSIX timer handling for a thread group. 1499 */ 1500 static void posix_cpu_timers_init_group(struct signal_struct *sig) 1501 { 1502 unsigned long cpu_limit; 1503 1504 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); 1505 if (cpu_limit != RLIM_INFINITY) { 1506 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC; 1507 sig->cputimer.running = true; 1508 } 1509 1510 /* The timer lists. */ 1511 INIT_LIST_HEAD(&sig->cpu_timers[0]); 1512 INIT_LIST_HEAD(&sig->cpu_timers[1]); 1513 INIT_LIST_HEAD(&sig->cpu_timers[2]); 1514 } 1515 #else 1516 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { } 1517 #endif 1518 1519 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk) 1520 { 1521 struct signal_struct *sig; 1522 1523 if (clone_flags & CLONE_THREAD) 1524 return 0; 1525 1526 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL); 1527 tsk->signal = sig; 1528 if (!sig) 1529 return -ENOMEM; 1530 1531 sig->nr_threads = 1; 1532 atomic_set(&sig->live, 1); 1533 atomic_set(&sig->sigcnt, 1); 1534 1535 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */ 1536 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node); 1537 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head); 1538 1539 init_waitqueue_head(&sig->wait_chldexit); 1540 sig->curr_target = tsk; 1541 init_sigpending(&sig->shared_pending); 1542 INIT_HLIST_HEAD(&sig->multiprocess); 1543 seqlock_init(&sig->stats_lock); 1544 prev_cputime_init(&sig->prev_cputime); 1545 1546 #ifdef CONFIG_POSIX_TIMERS 1547 INIT_LIST_HEAD(&sig->posix_timers); 1548 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1549 sig->real_timer.function = it_real_fn; 1550 #endif 1551 1552 task_lock(current->group_leader); 1553 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 1554 task_unlock(current->group_leader); 1555 1556 posix_cpu_timers_init_group(sig); 1557 1558 tty_audit_fork(sig); 1559 sched_autogroup_fork(sig); 1560 1561 sig->oom_score_adj = current->signal->oom_score_adj; 1562 sig->oom_score_adj_min = current->signal->oom_score_adj_min; 1563 1564 mutex_init(&sig->cred_guard_mutex); 1565 1566 return 0; 1567 } 1568 1569 static void copy_seccomp(struct task_struct *p) 1570 { 1571 #ifdef CONFIG_SECCOMP 1572 /* 1573 * Must be called with sighand->lock held, which is common to 1574 * all threads in the group. Holding cred_guard_mutex is not 1575 * needed because this new task is not yet running and cannot 1576 * be racing exec. 1577 */ 1578 assert_spin_locked(¤t->sighand->siglock); 1579 1580 /* Ref-count the new filter user, and assign it. */ 1581 get_seccomp_filter(current); 1582 p->seccomp = current->seccomp; 1583 1584 /* 1585 * Explicitly enable no_new_privs here in case it got set 1586 * between the task_struct being duplicated and holding the 1587 * sighand lock. The seccomp state and nnp must be in sync. 1588 */ 1589 if (task_no_new_privs(current)) 1590 task_set_no_new_privs(p); 1591 1592 /* 1593 * If the parent gained a seccomp mode after copying thread 1594 * flags and between before we held the sighand lock, we have 1595 * to manually enable the seccomp thread flag here. 1596 */ 1597 if (p->seccomp.mode != SECCOMP_MODE_DISABLED) 1598 set_tsk_thread_flag(p, TIF_SECCOMP); 1599 #endif 1600 } 1601 1602 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr) 1603 { 1604 current->clear_child_tid = tidptr; 1605 1606 return task_pid_vnr(current); 1607 } 1608 1609 static void rt_mutex_init_task(struct task_struct *p) 1610 { 1611 raw_spin_lock_init(&p->pi_lock); 1612 #ifdef CONFIG_RT_MUTEXES 1613 p->pi_waiters = RB_ROOT_CACHED; 1614 p->pi_top_task = NULL; 1615 p->pi_blocked_on = NULL; 1616 #endif 1617 } 1618 1619 #ifdef CONFIG_POSIX_TIMERS 1620 /* 1621 * Initialize POSIX timer handling for a single task. 1622 */ 1623 static void posix_cpu_timers_init(struct task_struct *tsk) 1624 { 1625 tsk->cputime_expires.prof_exp = 0; 1626 tsk->cputime_expires.virt_exp = 0; 1627 tsk->cputime_expires.sched_exp = 0; 1628 INIT_LIST_HEAD(&tsk->cpu_timers[0]); 1629 INIT_LIST_HEAD(&tsk->cpu_timers[1]); 1630 INIT_LIST_HEAD(&tsk->cpu_timers[2]); 1631 } 1632 #else 1633 static inline void posix_cpu_timers_init(struct task_struct *tsk) { } 1634 #endif 1635 1636 static inline void init_task_pid_links(struct task_struct *task) 1637 { 1638 enum pid_type type; 1639 1640 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 1641 INIT_HLIST_NODE(&task->pid_links[type]); 1642 } 1643 } 1644 1645 static inline void 1646 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid) 1647 { 1648 if (type == PIDTYPE_PID) 1649 task->thread_pid = pid; 1650 else 1651 task->signal->pids[type] = pid; 1652 } 1653 1654 static inline void rcu_copy_process(struct task_struct *p) 1655 { 1656 #ifdef CONFIG_PREEMPT_RCU 1657 p->rcu_read_lock_nesting = 0; 1658 p->rcu_read_unlock_special.s = 0; 1659 p->rcu_blocked_node = NULL; 1660 INIT_LIST_HEAD(&p->rcu_node_entry); 1661 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 1662 #ifdef CONFIG_TASKS_RCU 1663 p->rcu_tasks_holdout = false; 1664 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list); 1665 p->rcu_tasks_idle_cpu = -1; 1666 #endif /* #ifdef CONFIG_TASKS_RCU */ 1667 } 1668 1669 /* 1670 * This creates a new process as a copy of the old one, 1671 * but does not actually start it yet. 1672 * 1673 * It copies the registers, and all the appropriate 1674 * parts of the process environment (as per the clone 1675 * flags). The actual kick-off is left to the caller. 1676 */ 1677 static __latent_entropy struct task_struct *copy_process( 1678 unsigned long clone_flags, 1679 unsigned long stack_start, 1680 unsigned long stack_size, 1681 int __user *child_tidptr, 1682 struct pid *pid, 1683 int trace, 1684 unsigned long tls, 1685 int node) 1686 { 1687 int retval; 1688 struct task_struct *p; 1689 struct multiprocess_signals delayed; 1690 1691 /* 1692 * Don't allow sharing the root directory with processes in a different 1693 * namespace 1694 */ 1695 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 1696 return ERR_PTR(-EINVAL); 1697 1698 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS)) 1699 return ERR_PTR(-EINVAL); 1700 1701 /* 1702 * Thread groups must share signals as well, and detached threads 1703 * can only be started up within the thread group. 1704 */ 1705 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 1706 return ERR_PTR(-EINVAL); 1707 1708 /* 1709 * Shared signal handlers imply shared VM. By way of the above, 1710 * thread groups also imply shared VM. Blocking this case allows 1711 * for various simplifications in other code. 1712 */ 1713 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 1714 return ERR_PTR(-EINVAL); 1715 1716 /* 1717 * Siblings of global init remain as zombies on exit since they are 1718 * not reaped by their parent (swapper). To solve this and to avoid 1719 * multi-rooted process trees, prevent global and container-inits 1720 * from creating siblings. 1721 */ 1722 if ((clone_flags & CLONE_PARENT) && 1723 current->signal->flags & SIGNAL_UNKILLABLE) 1724 return ERR_PTR(-EINVAL); 1725 1726 /* 1727 * If the new process will be in a different pid or user namespace 1728 * do not allow it to share a thread group with the forking task. 1729 */ 1730 if (clone_flags & CLONE_THREAD) { 1731 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) || 1732 (task_active_pid_ns(current) != 1733 current->nsproxy->pid_ns_for_children)) 1734 return ERR_PTR(-EINVAL); 1735 } 1736 1737 /* 1738 * Force any signals received before this point to be delivered 1739 * before the fork happens. Collect up signals sent to multiple 1740 * processes that happen during the fork and delay them so that 1741 * they appear to happen after the fork. 1742 */ 1743 sigemptyset(&delayed.signal); 1744 INIT_HLIST_NODE(&delayed.node); 1745 1746 spin_lock_irq(¤t->sighand->siglock); 1747 if (!(clone_flags & CLONE_THREAD)) 1748 hlist_add_head(&delayed.node, ¤t->signal->multiprocess); 1749 recalc_sigpending(); 1750 spin_unlock_irq(¤t->sighand->siglock); 1751 retval = -ERESTARTNOINTR; 1752 if (signal_pending(current)) 1753 goto fork_out; 1754 1755 retval = -ENOMEM; 1756 p = dup_task_struct(current, node); 1757 if (!p) 1758 goto fork_out; 1759 1760 /* 1761 * This _must_ happen before we call free_task(), i.e. before we jump 1762 * to any of the bad_fork_* labels. This is to avoid freeing 1763 * p->set_child_tid which is (ab)used as a kthread's data pointer for 1764 * kernel threads (PF_KTHREAD). 1765 */ 1766 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; 1767 /* 1768 * Clear TID on mm_release()? 1769 */ 1770 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL; 1771 1772 ftrace_graph_init_task(p); 1773 1774 rt_mutex_init_task(p); 1775 1776 #ifdef CONFIG_PROVE_LOCKING 1777 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); 1778 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 1779 #endif 1780 retval = -EAGAIN; 1781 if (atomic_read(&p->real_cred->user->processes) >= 1782 task_rlimit(p, RLIMIT_NPROC)) { 1783 if (p->real_cred->user != INIT_USER && 1784 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) 1785 goto bad_fork_free; 1786 } 1787 current->flags &= ~PF_NPROC_EXCEEDED; 1788 1789 retval = copy_creds(p, clone_flags); 1790 if (retval < 0) 1791 goto bad_fork_free; 1792 1793 /* 1794 * If multiple threads are within copy_process(), then this check 1795 * triggers too late. This doesn't hurt, the check is only there 1796 * to stop root fork bombs. 1797 */ 1798 retval = -EAGAIN; 1799 if (nr_threads >= max_threads) 1800 goto bad_fork_cleanup_count; 1801 1802 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */ 1803 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE); 1804 p->flags |= PF_FORKNOEXEC; 1805 INIT_LIST_HEAD(&p->children); 1806 INIT_LIST_HEAD(&p->sibling); 1807 rcu_copy_process(p); 1808 p->vfork_done = NULL; 1809 spin_lock_init(&p->alloc_lock); 1810 1811 init_sigpending(&p->pending); 1812 1813 p->utime = p->stime = p->gtime = 0; 1814 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 1815 p->utimescaled = p->stimescaled = 0; 1816 #endif 1817 prev_cputime_init(&p->prev_cputime); 1818 1819 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN 1820 seqcount_init(&p->vtime.seqcount); 1821 p->vtime.starttime = 0; 1822 p->vtime.state = VTIME_INACTIVE; 1823 #endif 1824 1825 #if defined(SPLIT_RSS_COUNTING) 1826 memset(&p->rss_stat, 0, sizeof(p->rss_stat)); 1827 #endif 1828 1829 p->default_timer_slack_ns = current->timer_slack_ns; 1830 1831 #ifdef CONFIG_PSI 1832 p->psi_flags = 0; 1833 #endif 1834 1835 task_io_accounting_init(&p->ioac); 1836 acct_clear_integrals(p); 1837 1838 posix_cpu_timers_init(p); 1839 1840 p->start_time = ktime_get_ns(); 1841 p->real_start_time = ktime_get_boot_ns(); 1842 p->io_context = NULL; 1843 audit_set_context(p, NULL); 1844 cgroup_fork(p); 1845 #ifdef CONFIG_NUMA 1846 p->mempolicy = mpol_dup(p->mempolicy); 1847 if (IS_ERR(p->mempolicy)) { 1848 retval = PTR_ERR(p->mempolicy); 1849 p->mempolicy = NULL; 1850 goto bad_fork_cleanup_threadgroup_lock; 1851 } 1852 #endif 1853 #ifdef CONFIG_CPUSETS 1854 p->cpuset_mem_spread_rotor = NUMA_NO_NODE; 1855 p->cpuset_slab_spread_rotor = NUMA_NO_NODE; 1856 seqcount_init(&p->mems_allowed_seq); 1857 #endif 1858 #ifdef CONFIG_TRACE_IRQFLAGS 1859 p->irq_events = 0; 1860 p->hardirqs_enabled = 0; 1861 p->hardirq_enable_ip = 0; 1862 p->hardirq_enable_event = 0; 1863 p->hardirq_disable_ip = _THIS_IP_; 1864 p->hardirq_disable_event = 0; 1865 p->softirqs_enabled = 1; 1866 p->softirq_enable_ip = _THIS_IP_; 1867 p->softirq_enable_event = 0; 1868 p->softirq_disable_ip = 0; 1869 p->softirq_disable_event = 0; 1870 p->hardirq_context = 0; 1871 p->softirq_context = 0; 1872 #endif 1873 1874 p->pagefault_disabled = 0; 1875 1876 #ifdef CONFIG_LOCKDEP 1877 p->lockdep_depth = 0; /* no locks held yet */ 1878 p->curr_chain_key = 0; 1879 p->lockdep_recursion = 0; 1880 lockdep_init_task(p); 1881 #endif 1882 1883 #ifdef CONFIG_DEBUG_MUTEXES 1884 p->blocked_on = NULL; /* not blocked yet */ 1885 #endif 1886 #ifdef CONFIG_BCACHE 1887 p->sequential_io = 0; 1888 p->sequential_io_avg = 0; 1889 #endif 1890 1891 /* Perform scheduler related setup. Assign this task to a CPU. */ 1892 retval = sched_fork(clone_flags, p); 1893 if (retval) 1894 goto bad_fork_cleanup_policy; 1895 1896 retval = perf_event_init_task(p); 1897 if (retval) 1898 goto bad_fork_cleanup_policy; 1899 retval = audit_alloc(p); 1900 if (retval) 1901 goto bad_fork_cleanup_perf; 1902 /* copy all the process information */ 1903 shm_init_task(p); 1904 retval = security_task_alloc(p, clone_flags); 1905 if (retval) 1906 goto bad_fork_cleanup_audit; 1907 retval = copy_semundo(clone_flags, p); 1908 if (retval) 1909 goto bad_fork_cleanup_security; 1910 retval = copy_files(clone_flags, p); 1911 if (retval) 1912 goto bad_fork_cleanup_semundo; 1913 retval = copy_fs(clone_flags, p); 1914 if (retval) 1915 goto bad_fork_cleanup_files; 1916 retval = copy_sighand(clone_flags, p); 1917 if (retval) 1918 goto bad_fork_cleanup_fs; 1919 retval = copy_signal(clone_flags, p); 1920 if (retval) 1921 goto bad_fork_cleanup_sighand; 1922 retval = copy_mm(clone_flags, p); 1923 if (retval) 1924 goto bad_fork_cleanup_signal; 1925 retval = copy_namespaces(clone_flags, p); 1926 if (retval) 1927 goto bad_fork_cleanup_mm; 1928 retval = copy_io(clone_flags, p); 1929 if (retval) 1930 goto bad_fork_cleanup_namespaces; 1931 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls); 1932 if (retval) 1933 goto bad_fork_cleanup_io; 1934 1935 stackleak_task_init(p); 1936 1937 if (pid != &init_struct_pid) { 1938 pid = alloc_pid(p->nsproxy->pid_ns_for_children); 1939 if (IS_ERR(pid)) { 1940 retval = PTR_ERR(pid); 1941 goto bad_fork_cleanup_thread; 1942 } 1943 } 1944 1945 #ifdef CONFIG_BLOCK 1946 p->plug = NULL; 1947 #endif 1948 #ifdef CONFIG_FUTEX 1949 p->robust_list = NULL; 1950 #ifdef CONFIG_COMPAT 1951 p->compat_robust_list = NULL; 1952 #endif 1953 INIT_LIST_HEAD(&p->pi_state_list); 1954 p->pi_state_cache = NULL; 1955 #endif 1956 /* 1957 * sigaltstack should be cleared when sharing the same VM 1958 */ 1959 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM) 1960 sas_ss_reset(p); 1961 1962 /* 1963 * Syscall tracing and stepping should be turned off in the 1964 * child regardless of CLONE_PTRACE. 1965 */ 1966 user_disable_single_step(p); 1967 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); 1968 #ifdef TIF_SYSCALL_EMU 1969 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); 1970 #endif 1971 clear_all_latency_tracing(p); 1972 1973 /* ok, now we should be set up.. */ 1974 p->pid = pid_nr(pid); 1975 if (clone_flags & CLONE_THREAD) { 1976 p->exit_signal = -1; 1977 p->group_leader = current->group_leader; 1978 p->tgid = current->tgid; 1979 } else { 1980 if (clone_flags & CLONE_PARENT) 1981 p->exit_signal = current->group_leader->exit_signal; 1982 else 1983 p->exit_signal = (clone_flags & CSIGNAL); 1984 p->group_leader = p; 1985 p->tgid = p->pid; 1986 } 1987 1988 p->nr_dirtied = 0; 1989 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10); 1990 p->dirty_paused_when = 0; 1991 1992 p->pdeath_signal = 0; 1993 INIT_LIST_HEAD(&p->thread_group); 1994 p->task_works = NULL; 1995 1996 cgroup_threadgroup_change_begin(current); 1997 /* 1998 * Ensure that the cgroup subsystem policies allow the new process to be 1999 * forked. It should be noted the the new process's css_set can be changed 2000 * between here and cgroup_post_fork() if an organisation operation is in 2001 * progress. 2002 */ 2003 retval = cgroup_can_fork(p); 2004 if (retval) 2005 goto bad_fork_free_pid; 2006 2007 /* 2008 * Make it visible to the rest of the system, but dont wake it up yet. 2009 * Need tasklist lock for parent etc handling! 2010 */ 2011 write_lock_irq(&tasklist_lock); 2012 2013 /* CLONE_PARENT re-uses the old parent */ 2014 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { 2015 p->real_parent = current->real_parent; 2016 p->parent_exec_id = current->parent_exec_id; 2017 } else { 2018 p->real_parent = current; 2019 p->parent_exec_id = current->self_exec_id; 2020 } 2021 2022 klp_copy_process(p); 2023 2024 spin_lock(¤t->sighand->siglock); 2025 2026 /* 2027 * Copy seccomp details explicitly here, in case they were changed 2028 * before holding sighand lock. 2029 */ 2030 copy_seccomp(p); 2031 2032 rseq_fork(p, clone_flags); 2033 2034 /* Don't start children in a dying pid namespace */ 2035 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) { 2036 retval = -ENOMEM; 2037 goto bad_fork_cancel_cgroup; 2038 } 2039 2040 /* Let kill terminate clone/fork in the middle */ 2041 if (fatal_signal_pending(current)) { 2042 retval = -EINTR; 2043 goto bad_fork_cancel_cgroup; 2044 } 2045 2046 2047 init_task_pid_links(p); 2048 if (likely(p->pid)) { 2049 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace); 2050 2051 init_task_pid(p, PIDTYPE_PID, pid); 2052 if (thread_group_leader(p)) { 2053 init_task_pid(p, PIDTYPE_TGID, pid); 2054 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current)); 2055 init_task_pid(p, PIDTYPE_SID, task_session(current)); 2056 2057 if (is_child_reaper(pid)) { 2058 ns_of_pid(pid)->child_reaper = p; 2059 p->signal->flags |= SIGNAL_UNKILLABLE; 2060 } 2061 p->signal->shared_pending.signal = delayed.signal; 2062 p->signal->tty = tty_kref_get(current->signal->tty); 2063 /* 2064 * Inherit has_child_subreaper flag under the same 2065 * tasklist_lock with adding child to the process tree 2066 * for propagate_has_child_subreaper optimization. 2067 */ 2068 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper || 2069 p->real_parent->signal->is_child_subreaper; 2070 list_add_tail(&p->sibling, &p->real_parent->children); 2071 list_add_tail_rcu(&p->tasks, &init_task.tasks); 2072 attach_pid(p, PIDTYPE_TGID); 2073 attach_pid(p, PIDTYPE_PGID); 2074 attach_pid(p, PIDTYPE_SID); 2075 __this_cpu_inc(process_counts); 2076 } else { 2077 current->signal->nr_threads++; 2078 atomic_inc(¤t->signal->live); 2079 atomic_inc(¤t->signal->sigcnt); 2080 task_join_group_stop(p); 2081 list_add_tail_rcu(&p->thread_group, 2082 &p->group_leader->thread_group); 2083 list_add_tail_rcu(&p->thread_node, 2084 &p->signal->thread_head); 2085 } 2086 attach_pid(p, PIDTYPE_PID); 2087 nr_threads++; 2088 } 2089 total_forks++; 2090 hlist_del_init(&delayed.node); 2091 spin_unlock(¤t->sighand->siglock); 2092 syscall_tracepoint_update(p); 2093 write_unlock_irq(&tasklist_lock); 2094 2095 proc_fork_connector(p); 2096 cgroup_post_fork(p); 2097 cgroup_threadgroup_change_end(current); 2098 perf_event_fork(p); 2099 2100 trace_task_newtask(p, clone_flags); 2101 uprobe_copy_process(p, clone_flags); 2102 2103 return p; 2104 2105 bad_fork_cancel_cgroup: 2106 spin_unlock(¤t->sighand->siglock); 2107 write_unlock_irq(&tasklist_lock); 2108 cgroup_cancel_fork(p); 2109 bad_fork_free_pid: 2110 cgroup_threadgroup_change_end(current); 2111 if (pid != &init_struct_pid) 2112 free_pid(pid); 2113 bad_fork_cleanup_thread: 2114 exit_thread(p); 2115 bad_fork_cleanup_io: 2116 if (p->io_context) 2117 exit_io_context(p); 2118 bad_fork_cleanup_namespaces: 2119 exit_task_namespaces(p); 2120 bad_fork_cleanup_mm: 2121 if (p->mm) 2122 mmput(p->mm); 2123 bad_fork_cleanup_signal: 2124 if (!(clone_flags & CLONE_THREAD)) 2125 free_signal_struct(p->signal); 2126 bad_fork_cleanup_sighand: 2127 __cleanup_sighand(p->sighand); 2128 bad_fork_cleanup_fs: 2129 exit_fs(p); /* blocking */ 2130 bad_fork_cleanup_files: 2131 exit_files(p); /* blocking */ 2132 bad_fork_cleanup_semundo: 2133 exit_sem(p); 2134 bad_fork_cleanup_security: 2135 security_task_free(p); 2136 bad_fork_cleanup_audit: 2137 audit_free(p); 2138 bad_fork_cleanup_perf: 2139 perf_event_free_task(p); 2140 bad_fork_cleanup_policy: 2141 lockdep_free_task(p); 2142 #ifdef CONFIG_NUMA 2143 mpol_put(p->mempolicy); 2144 bad_fork_cleanup_threadgroup_lock: 2145 #endif 2146 delayacct_tsk_free(p); 2147 bad_fork_cleanup_count: 2148 atomic_dec(&p->cred->user->processes); 2149 exit_creds(p); 2150 bad_fork_free: 2151 p->state = TASK_DEAD; 2152 put_task_stack(p); 2153 free_task(p); 2154 fork_out: 2155 spin_lock_irq(¤t->sighand->siglock); 2156 hlist_del_init(&delayed.node); 2157 spin_unlock_irq(¤t->sighand->siglock); 2158 return ERR_PTR(retval); 2159 } 2160 2161 static inline void init_idle_pids(struct task_struct *idle) 2162 { 2163 enum pid_type type; 2164 2165 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 2166 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */ 2167 init_task_pid(idle, type, &init_struct_pid); 2168 } 2169 } 2170 2171 struct task_struct *fork_idle(int cpu) 2172 { 2173 struct task_struct *task; 2174 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0, 2175 cpu_to_node(cpu)); 2176 if (!IS_ERR(task)) { 2177 init_idle_pids(task); 2178 init_idle(task, cpu); 2179 } 2180 2181 return task; 2182 } 2183 2184 /* 2185 * Ok, this is the main fork-routine. 2186 * 2187 * It copies the process, and if successful kick-starts 2188 * it and waits for it to finish using the VM if required. 2189 */ 2190 long _do_fork(unsigned long clone_flags, 2191 unsigned long stack_start, 2192 unsigned long stack_size, 2193 int __user *parent_tidptr, 2194 int __user *child_tidptr, 2195 unsigned long tls) 2196 { 2197 struct completion vfork; 2198 struct pid *pid; 2199 struct task_struct *p; 2200 int trace = 0; 2201 long nr; 2202 2203 /* 2204 * Determine whether and which event to report to ptracer. When 2205 * called from kernel_thread or CLONE_UNTRACED is explicitly 2206 * requested, no event is reported; otherwise, report if the event 2207 * for the type of forking is enabled. 2208 */ 2209 if (!(clone_flags & CLONE_UNTRACED)) { 2210 if (clone_flags & CLONE_VFORK) 2211 trace = PTRACE_EVENT_VFORK; 2212 else if ((clone_flags & CSIGNAL) != SIGCHLD) 2213 trace = PTRACE_EVENT_CLONE; 2214 else 2215 trace = PTRACE_EVENT_FORK; 2216 2217 if (likely(!ptrace_event_enabled(current, trace))) 2218 trace = 0; 2219 } 2220 2221 p = copy_process(clone_flags, stack_start, stack_size, 2222 child_tidptr, NULL, trace, tls, NUMA_NO_NODE); 2223 add_latent_entropy(); 2224 2225 if (IS_ERR(p)) 2226 return PTR_ERR(p); 2227 2228 /* 2229 * Do this prior waking up the new thread - the thread pointer 2230 * might get invalid after that point, if the thread exits quickly. 2231 */ 2232 trace_sched_process_fork(current, p); 2233 2234 pid = get_task_pid(p, PIDTYPE_PID); 2235 nr = pid_vnr(pid); 2236 2237 if (clone_flags & CLONE_PARENT_SETTID) 2238 put_user(nr, parent_tidptr); 2239 2240 if (clone_flags & CLONE_VFORK) { 2241 p->vfork_done = &vfork; 2242 init_completion(&vfork); 2243 get_task_struct(p); 2244 } 2245 2246 wake_up_new_task(p); 2247 2248 /* forking complete and child started to run, tell ptracer */ 2249 if (unlikely(trace)) 2250 ptrace_event_pid(trace, pid); 2251 2252 if (clone_flags & CLONE_VFORK) { 2253 if (!wait_for_vfork_done(p, &vfork)) 2254 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid); 2255 } 2256 2257 put_pid(pid); 2258 return nr; 2259 } 2260 2261 #ifndef CONFIG_HAVE_COPY_THREAD_TLS 2262 /* For compatibility with architectures that call do_fork directly rather than 2263 * using the syscall entry points below. */ 2264 long do_fork(unsigned long clone_flags, 2265 unsigned long stack_start, 2266 unsigned long stack_size, 2267 int __user *parent_tidptr, 2268 int __user *child_tidptr) 2269 { 2270 return _do_fork(clone_flags, stack_start, stack_size, 2271 parent_tidptr, child_tidptr, 0); 2272 } 2273 #endif 2274 2275 /* 2276 * Create a kernel thread. 2277 */ 2278 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags) 2279 { 2280 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn, 2281 (unsigned long)arg, NULL, NULL, 0); 2282 } 2283 2284 #ifdef __ARCH_WANT_SYS_FORK 2285 SYSCALL_DEFINE0(fork) 2286 { 2287 #ifdef CONFIG_MMU 2288 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0); 2289 #else 2290 /* can not support in nommu mode */ 2291 return -EINVAL; 2292 #endif 2293 } 2294 #endif 2295 2296 #ifdef __ARCH_WANT_SYS_VFORK 2297 SYSCALL_DEFINE0(vfork) 2298 { 2299 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0, 2300 0, NULL, NULL, 0); 2301 } 2302 #endif 2303 2304 #ifdef __ARCH_WANT_SYS_CLONE 2305 #ifdef CONFIG_CLONE_BACKWARDS 2306 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2307 int __user *, parent_tidptr, 2308 unsigned long, tls, 2309 int __user *, child_tidptr) 2310 #elif defined(CONFIG_CLONE_BACKWARDS2) 2311 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags, 2312 int __user *, parent_tidptr, 2313 int __user *, child_tidptr, 2314 unsigned long, tls) 2315 #elif defined(CONFIG_CLONE_BACKWARDS3) 2316 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp, 2317 int, stack_size, 2318 int __user *, parent_tidptr, 2319 int __user *, child_tidptr, 2320 unsigned long, tls) 2321 #else 2322 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2323 int __user *, parent_tidptr, 2324 int __user *, child_tidptr, 2325 unsigned long, tls) 2326 #endif 2327 { 2328 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls); 2329 } 2330 #endif 2331 2332 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data) 2333 { 2334 struct task_struct *leader, *parent, *child; 2335 int res; 2336 2337 read_lock(&tasklist_lock); 2338 leader = top = top->group_leader; 2339 down: 2340 for_each_thread(leader, parent) { 2341 list_for_each_entry(child, &parent->children, sibling) { 2342 res = visitor(child, data); 2343 if (res) { 2344 if (res < 0) 2345 goto out; 2346 leader = child; 2347 goto down; 2348 } 2349 up: 2350 ; 2351 } 2352 } 2353 2354 if (leader != top) { 2355 child = leader; 2356 parent = child->real_parent; 2357 leader = parent->group_leader; 2358 goto up; 2359 } 2360 out: 2361 read_unlock(&tasklist_lock); 2362 } 2363 2364 #ifndef ARCH_MIN_MMSTRUCT_ALIGN 2365 #define ARCH_MIN_MMSTRUCT_ALIGN 0 2366 #endif 2367 2368 static void sighand_ctor(void *data) 2369 { 2370 struct sighand_struct *sighand = data; 2371 2372 spin_lock_init(&sighand->siglock); 2373 init_waitqueue_head(&sighand->signalfd_wqh); 2374 } 2375 2376 void __init proc_caches_init(void) 2377 { 2378 unsigned int mm_size; 2379 2380 sighand_cachep = kmem_cache_create("sighand_cache", 2381 sizeof(struct sighand_struct), 0, 2382 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU| 2383 SLAB_ACCOUNT, sighand_ctor); 2384 signal_cachep = kmem_cache_create("signal_cache", 2385 sizeof(struct signal_struct), 0, 2386 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2387 NULL); 2388 files_cachep = kmem_cache_create("files_cache", 2389 sizeof(struct files_struct), 0, 2390 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2391 NULL); 2392 fs_cachep = kmem_cache_create("fs_cache", 2393 sizeof(struct fs_struct), 0, 2394 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2395 NULL); 2396 2397 /* 2398 * The mm_cpumask is located at the end of mm_struct, and is 2399 * dynamically sized based on the maximum CPU number this system 2400 * can have, taking hotplug into account (nr_cpu_ids). 2401 */ 2402 mm_size = sizeof(struct mm_struct) + cpumask_size(); 2403 2404 mm_cachep = kmem_cache_create_usercopy("mm_struct", 2405 mm_size, ARCH_MIN_MMSTRUCT_ALIGN, 2406 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2407 offsetof(struct mm_struct, saved_auxv), 2408 sizeof_field(struct mm_struct, saved_auxv), 2409 NULL); 2410 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT); 2411 mmap_init(); 2412 nsproxy_cache_init(); 2413 } 2414 2415 /* 2416 * Check constraints on flags passed to the unshare system call. 2417 */ 2418 static int check_unshare_flags(unsigned long unshare_flags) 2419 { 2420 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND| 2421 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM| 2422 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET| 2423 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP)) 2424 return -EINVAL; 2425 /* 2426 * Not implemented, but pretend it works if there is nothing 2427 * to unshare. Note that unsharing the address space or the 2428 * signal handlers also need to unshare the signal queues (aka 2429 * CLONE_THREAD). 2430 */ 2431 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) { 2432 if (!thread_group_empty(current)) 2433 return -EINVAL; 2434 } 2435 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) { 2436 if (atomic_read(¤t->sighand->count) > 1) 2437 return -EINVAL; 2438 } 2439 if (unshare_flags & CLONE_VM) { 2440 if (!current_is_single_threaded()) 2441 return -EINVAL; 2442 } 2443 2444 return 0; 2445 } 2446 2447 /* 2448 * Unshare the filesystem structure if it is being shared 2449 */ 2450 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) 2451 { 2452 struct fs_struct *fs = current->fs; 2453 2454 if (!(unshare_flags & CLONE_FS) || !fs) 2455 return 0; 2456 2457 /* don't need lock here; in the worst case we'll do useless copy */ 2458 if (fs->users == 1) 2459 return 0; 2460 2461 *new_fsp = copy_fs_struct(fs); 2462 if (!*new_fsp) 2463 return -ENOMEM; 2464 2465 return 0; 2466 } 2467 2468 /* 2469 * Unshare file descriptor table if it is being shared 2470 */ 2471 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp) 2472 { 2473 struct files_struct *fd = current->files; 2474 int error = 0; 2475 2476 if ((unshare_flags & CLONE_FILES) && 2477 (fd && atomic_read(&fd->count) > 1)) { 2478 *new_fdp = dup_fd(fd, &error); 2479 if (!*new_fdp) 2480 return error; 2481 } 2482 2483 return 0; 2484 } 2485 2486 /* 2487 * unshare allows a process to 'unshare' part of the process 2488 * context which was originally shared using clone. copy_* 2489 * functions used by do_fork() cannot be used here directly 2490 * because they modify an inactive task_struct that is being 2491 * constructed. Here we are modifying the current, active, 2492 * task_struct. 2493 */ 2494 int ksys_unshare(unsigned long unshare_flags) 2495 { 2496 struct fs_struct *fs, *new_fs = NULL; 2497 struct files_struct *fd, *new_fd = NULL; 2498 struct cred *new_cred = NULL; 2499 struct nsproxy *new_nsproxy = NULL; 2500 int do_sysvsem = 0; 2501 int err; 2502 2503 /* 2504 * If unsharing a user namespace must also unshare the thread group 2505 * and unshare the filesystem root and working directories. 2506 */ 2507 if (unshare_flags & CLONE_NEWUSER) 2508 unshare_flags |= CLONE_THREAD | CLONE_FS; 2509 /* 2510 * If unsharing vm, must also unshare signal handlers. 2511 */ 2512 if (unshare_flags & CLONE_VM) 2513 unshare_flags |= CLONE_SIGHAND; 2514 /* 2515 * If unsharing a signal handlers, must also unshare the signal queues. 2516 */ 2517 if (unshare_flags & CLONE_SIGHAND) 2518 unshare_flags |= CLONE_THREAD; 2519 /* 2520 * If unsharing namespace, must also unshare filesystem information. 2521 */ 2522 if (unshare_flags & CLONE_NEWNS) 2523 unshare_flags |= CLONE_FS; 2524 2525 err = check_unshare_flags(unshare_flags); 2526 if (err) 2527 goto bad_unshare_out; 2528 /* 2529 * CLONE_NEWIPC must also detach from the undolist: after switching 2530 * to a new ipc namespace, the semaphore arrays from the old 2531 * namespace are unreachable. 2532 */ 2533 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) 2534 do_sysvsem = 1; 2535 err = unshare_fs(unshare_flags, &new_fs); 2536 if (err) 2537 goto bad_unshare_out; 2538 err = unshare_fd(unshare_flags, &new_fd); 2539 if (err) 2540 goto bad_unshare_cleanup_fs; 2541 err = unshare_userns(unshare_flags, &new_cred); 2542 if (err) 2543 goto bad_unshare_cleanup_fd; 2544 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, 2545 new_cred, new_fs); 2546 if (err) 2547 goto bad_unshare_cleanup_cred; 2548 2549 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) { 2550 if (do_sysvsem) { 2551 /* 2552 * CLONE_SYSVSEM is equivalent to sys_exit(). 2553 */ 2554 exit_sem(current); 2555 } 2556 if (unshare_flags & CLONE_NEWIPC) { 2557 /* Orphan segments in old ns (see sem above). */ 2558 exit_shm(current); 2559 shm_init_task(current); 2560 } 2561 2562 if (new_nsproxy) 2563 switch_task_namespaces(current, new_nsproxy); 2564 2565 task_lock(current); 2566 2567 if (new_fs) { 2568 fs = current->fs; 2569 spin_lock(&fs->lock); 2570 current->fs = new_fs; 2571 if (--fs->users) 2572 new_fs = NULL; 2573 else 2574 new_fs = fs; 2575 spin_unlock(&fs->lock); 2576 } 2577 2578 if (new_fd) { 2579 fd = current->files; 2580 current->files = new_fd; 2581 new_fd = fd; 2582 } 2583 2584 task_unlock(current); 2585 2586 if (new_cred) { 2587 /* Install the new user namespace */ 2588 commit_creds(new_cred); 2589 new_cred = NULL; 2590 } 2591 } 2592 2593 perf_event_namespaces(current); 2594 2595 bad_unshare_cleanup_cred: 2596 if (new_cred) 2597 put_cred(new_cred); 2598 bad_unshare_cleanup_fd: 2599 if (new_fd) 2600 put_files_struct(new_fd); 2601 2602 bad_unshare_cleanup_fs: 2603 if (new_fs) 2604 free_fs_struct(new_fs); 2605 2606 bad_unshare_out: 2607 return err; 2608 } 2609 2610 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags) 2611 { 2612 return ksys_unshare(unshare_flags); 2613 } 2614 2615 /* 2616 * Helper to unshare the files of the current task. 2617 * We don't want to expose copy_files internals to 2618 * the exec layer of the kernel. 2619 */ 2620 2621 int unshare_files(struct files_struct **displaced) 2622 { 2623 struct task_struct *task = current; 2624 struct files_struct *copy = NULL; 2625 int error; 2626 2627 error = unshare_fd(CLONE_FILES, ©); 2628 if (error || !copy) { 2629 *displaced = NULL; 2630 return error; 2631 } 2632 *displaced = task->files; 2633 task_lock(task); 2634 task->files = copy; 2635 task_unlock(task); 2636 return 0; 2637 } 2638 2639 int sysctl_max_threads(struct ctl_table *table, int write, 2640 void __user *buffer, size_t *lenp, loff_t *ppos) 2641 { 2642 struct ctl_table t; 2643 int ret; 2644 int threads = max_threads; 2645 int min = MIN_THREADS; 2646 int max = MAX_THREADS; 2647 2648 t = *table; 2649 t.data = &threads; 2650 t.extra1 = &min; 2651 t.extra2 = &max; 2652 2653 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); 2654 if (ret || !write) 2655 return ret; 2656 2657 set_max_threads(threads); 2658 2659 return 0; 2660 } 2661