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 unsigned long nr_pages = totalram_pages(); 748 749 /* 750 * The number of threads shall be limited such that the thread 751 * structures may only consume a small part of the available memory. 752 */ 753 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64) 754 threads = MAX_THREADS; 755 else 756 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE, 757 (u64) THREAD_SIZE * 8UL); 758 759 if (threads > max_threads_suggested) 760 threads = max_threads_suggested; 761 762 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS); 763 } 764 765 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT 766 /* Initialized by the architecture: */ 767 int arch_task_struct_size __read_mostly; 768 #endif 769 770 static void task_struct_whitelist(unsigned long *offset, unsigned long *size) 771 { 772 /* Fetch thread_struct whitelist for the architecture. */ 773 arch_thread_struct_whitelist(offset, size); 774 775 /* 776 * Handle zero-sized whitelist or empty thread_struct, otherwise 777 * adjust offset to position of thread_struct in task_struct. 778 */ 779 if (unlikely(*size == 0)) 780 *offset = 0; 781 else 782 *offset += offsetof(struct task_struct, thread); 783 } 784 785 void __init fork_init(void) 786 { 787 int i; 788 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 789 #ifndef ARCH_MIN_TASKALIGN 790 #define ARCH_MIN_TASKALIGN 0 791 #endif 792 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN); 793 unsigned long useroffset, usersize; 794 795 /* create a slab on which task_structs can be allocated */ 796 task_struct_whitelist(&useroffset, &usersize); 797 task_struct_cachep = kmem_cache_create_usercopy("task_struct", 798 arch_task_struct_size, align, 799 SLAB_PANIC|SLAB_ACCOUNT, 800 useroffset, usersize, NULL); 801 #endif 802 803 /* do the arch specific task caches init */ 804 arch_task_cache_init(); 805 806 set_max_threads(MAX_THREADS); 807 808 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 809 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 810 init_task.signal->rlim[RLIMIT_SIGPENDING] = 811 init_task.signal->rlim[RLIMIT_NPROC]; 812 813 for (i = 0; i < UCOUNT_COUNTS; i++) { 814 init_user_ns.ucount_max[i] = max_threads/2; 815 } 816 817 #ifdef CONFIG_VMAP_STACK 818 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache", 819 NULL, free_vm_stack_cache); 820 #endif 821 822 lockdep_init_task(&init_task); 823 } 824 825 int __weak arch_dup_task_struct(struct task_struct *dst, 826 struct task_struct *src) 827 { 828 *dst = *src; 829 return 0; 830 } 831 832 void set_task_stack_end_magic(struct task_struct *tsk) 833 { 834 unsigned long *stackend; 835 836 stackend = end_of_stack(tsk); 837 *stackend = STACK_END_MAGIC; /* for overflow detection */ 838 } 839 840 static struct task_struct *dup_task_struct(struct task_struct *orig, int node) 841 { 842 struct task_struct *tsk; 843 unsigned long *stack; 844 struct vm_struct *stack_vm_area __maybe_unused; 845 int err; 846 847 if (node == NUMA_NO_NODE) 848 node = tsk_fork_get_node(orig); 849 tsk = alloc_task_struct_node(node); 850 if (!tsk) 851 return NULL; 852 853 stack = alloc_thread_stack_node(tsk, node); 854 if (!stack) 855 goto free_tsk; 856 857 if (memcg_charge_kernel_stack(tsk)) 858 goto free_stack; 859 860 stack_vm_area = task_stack_vm_area(tsk); 861 862 err = arch_dup_task_struct(tsk, orig); 863 864 /* 865 * arch_dup_task_struct() clobbers the stack-related fields. Make 866 * sure they're properly initialized before using any stack-related 867 * functions again. 868 */ 869 tsk->stack = stack; 870 #ifdef CONFIG_VMAP_STACK 871 tsk->stack_vm_area = stack_vm_area; 872 #endif 873 #ifdef CONFIG_THREAD_INFO_IN_TASK 874 atomic_set(&tsk->stack_refcount, 1); 875 #endif 876 877 if (err) 878 goto free_stack; 879 880 #ifdef CONFIG_SECCOMP 881 /* 882 * We must handle setting up seccomp filters once we're under 883 * the sighand lock in case orig has changed between now and 884 * then. Until then, filter must be NULL to avoid messing up 885 * the usage counts on the error path calling free_task. 886 */ 887 tsk->seccomp.filter = NULL; 888 #endif 889 890 setup_thread_stack(tsk, orig); 891 clear_user_return_notifier(tsk); 892 clear_tsk_need_resched(tsk); 893 set_task_stack_end_magic(tsk); 894 895 #ifdef CONFIG_STACKPROTECTOR 896 tsk->stack_canary = get_random_canary(); 897 #endif 898 899 /* 900 * One for us, one for whoever does the "release_task()" (usually 901 * parent) 902 */ 903 atomic_set(&tsk->usage, 2); 904 #ifdef CONFIG_BLK_DEV_IO_TRACE 905 tsk->btrace_seq = 0; 906 #endif 907 tsk->splice_pipe = NULL; 908 tsk->task_frag.page = NULL; 909 tsk->wake_q.next = NULL; 910 911 account_kernel_stack(tsk, 1); 912 913 kcov_task_init(tsk); 914 915 #ifdef CONFIG_FAULT_INJECTION 916 tsk->fail_nth = 0; 917 #endif 918 919 #ifdef CONFIG_BLK_CGROUP 920 tsk->throttle_queue = NULL; 921 tsk->use_memdelay = 0; 922 #endif 923 924 #ifdef CONFIG_MEMCG 925 tsk->active_memcg = NULL; 926 #endif 927 return tsk; 928 929 free_stack: 930 free_thread_stack(tsk); 931 free_tsk: 932 free_task_struct(tsk); 933 return NULL; 934 } 935 936 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); 937 938 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT; 939 940 static int __init coredump_filter_setup(char *s) 941 { 942 default_dump_filter = 943 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) & 944 MMF_DUMP_FILTER_MASK; 945 return 1; 946 } 947 948 __setup("coredump_filter=", coredump_filter_setup); 949 950 #include <linux/init_task.h> 951 952 static void mm_init_aio(struct mm_struct *mm) 953 { 954 #ifdef CONFIG_AIO 955 spin_lock_init(&mm->ioctx_lock); 956 mm->ioctx_table = NULL; 957 #endif 958 } 959 960 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p) 961 { 962 #ifdef CONFIG_MEMCG 963 mm->owner = p; 964 #endif 965 } 966 967 static void mm_init_uprobes_state(struct mm_struct *mm) 968 { 969 #ifdef CONFIG_UPROBES 970 mm->uprobes_state.xol_area = NULL; 971 #endif 972 } 973 974 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p, 975 struct user_namespace *user_ns) 976 { 977 mm->mmap = NULL; 978 mm->mm_rb = RB_ROOT; 979 mm->vmacache_seqnum = 0; 980 atomic_set(&mm->mm_users, 1); 981 atomic_set(&mm->mm_count, 1); 982 init_rwsem(&mm->mmap_sem); 983 INIT_LIST_HEAD(&mm->mmlist); 984 mm->core_state = NULL; 985 mm_pgtables_bytes_init(mm); 986 mm->map_count = 0; 987 mm->locked_vm = 0; 988 mm->pinned_vm = 0; 989 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat)); 990 spin_lock_init(&mm->page_table_lock); 991 spin_lock_init(&mm->arg_lock); 992 mm_init_cpumask(mm); 993 mm_init_aio(mm); 994 mm_init_owner(mm, p); 995 RCU_INIT_POINTER(mm->exe_file, NULL); 996 mmu_notifier_mm_init(mm); 997 hmm_mm_init(mm); 998 init_tlb_flush_pending(mm); 999 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 1000 mm->pmd_huge_pte = NULL; 1001 #endif 1002 mm_init_uprobes_state(mm); 1003 1004 if (current->mm) { 1005 mm->flags = current->mm->flags & MMF_INIT_MASK; 1006 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK; 1007 } else { 1008 mm->flags = default_dump_filter; 1009 mm->def_flags = 0; 1010 } 1011 1012 if (mm_alloc_pgd(mm)) 1013 goto fail_nopgd; 1014 1015 if (init_new_context(p, mm)) 1016 goto fail_nocontext; 1017 1018 mm->user_ns = get_user_ns(user_ns); 1019 return mm; 1020 1021 fail_nocontext: 1022 mm_free_pgd(mm); 1023 fail_nopgd: 1024 free_mm(mm); 1025 return NULL; 1026 } 1027 1028 /* 1029 * Allocate and initialize an mm_struct. 1030 */ 1031 struct mm_struct *mm_alloc(void) 1032 { 1033 struct mm_struct *mm; 1034 1035 mm = allocate_mm(); 1036 if (!mm) 1037 return NULL; 1038 1039 memset(mm, 0, sizeof(*mm)); 1040 return mm_init(mm, current, current_user_ns()); 1041 } 1042 1043 static inline void __mmput(struct mm_struct *mm) 1044 { 1045 VM_BUG_ON(atomic_read(&mm->mm_users)); 1046 1047 uprobe_clear_state(mm); 1048 exit_aio(mm); 1049 ksm_exit(mm); 1050 khugepaged_exit(mm); /* must run before exit_mmap */ 1051 exit_mmap(mm); 1052 mm_put_huge_zero_page(mm); 1053 set_mm_exe_file(mm, NULL); 1054 if (!list_empty(&mm->mmlist)) { 1055 spin_lock(&mmlist_lock); 1056 list_del(&mm->mmlist); 1057 spin_unlock(&mmlist_lock); 1058 } 1059 if (mm->binfmt) 1060 module_put(mm->binfmt->module); 1061 mmdrop(mm); 1062 } 1063 1064 /* 1065 * Decrement the use count and release all resources for an mm. 1066 */ 1067 void mmput(struct mm_struct *mm) 1068 { 1069 might_sleep(); 1070 1071 if (atomic_dec_and_test(&mm->mm_users)) 1072 __mmput(mm); 1073 } 1074 EXPORT_SYMBOL_GPL(mmput); 1075 1076 #ifdef CONFIG_MMU 1077 static void mmput_async_fn(struct work_struct *work) 1078 { 1079 struct mm_struct *mm = container_of(work, struct mm_struct, 1080 async_put_work); 1081 1082 __mmput(mm); 1083 } 1084 1085 void mmput_async(struct mm_struct *mm) 1086 { 1087 if (atomic_dec_and_test(&mm->mm_users)) { 1088 INIT_WORK(&mm->async_put_work, mmput_async_fn); 1089 schedule_work(&mm->async_put_work); 1090 } 1091 } 1092 #endif 1093 1094 /** 1095 * set_mm_exe_file - change a reference to the mm's executable file 1096 * 1097 * This changes mm's executable file (shown as symlink /proc/[pid]/exe). 1098 * 1099 * Main users are mmput() and sys_execve(). Callers prevent concurrent 1100 * invocations: in mmput() nobody alive left, in execve task is single 1101 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the 1102 * mm->exe_file, but does so without using set_mm_exe_file() in order 1103 * to do avoid the need for any locks. 1104 */ 1105 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 1106 { 1107 struct file *old_exe_file; 1108 1109 /* 1110 * It is safe to dereference the exe_file without RCU as 1111 * this function is only called if nobody else can access 1112 * this mm -- see comment above for justification. 1113 */ 1114 old_exe_file = rcu_dereference_raw(mm->exe_file); 1115 1116 if (new_exe_file) 1117 get_file(new_exe_file); 1118 rcu_assign_pointer(mm->exe_file, new_exe_file); 1119 if (old_exe_file) 1120 fput(old_exe_file); 1121 } 1122 1123 /** 1124 * get_mm_exe_file - acquire a reference to the mm's executable file 1125 * 1126 * Returns %NULL if mm has no associated executable file. 1127 * User must release file via fput(). 1128 */ 1129 struct file *get_mm_exe_file(struct mm_struct *mm) 1130 { 1131 struct file *exe_file; 1132 1133 rcu_read_lock(); 1134 exe_file = rcu_dereference(mm->exe_file); 1135 if (exe_file && !get_file_rcu(exe_file)) 1136 exe_file = NULL; 1137 rcu_read_unlock(); 1138 return exe_file; 1139 } 1140 EXPORT_SYMBOL(get_mm_exe_file); 1141 1142 /** 1143 * get_task_exe_file - acquire a reference to the task's executable file 1144 * 1145 * Returns %NULL if task's mm (if any) has no associated executable file or 1146 * this is a kernel thread with borrowed mm (see the comment above get_task_mm). 1147 * User must release file via fput(). 1148 */ 1149 struct file *get_task_exe_file(struct task_struct *task) 1150 { 1151 struct file *exe_file = NULL; 1152 struct mm_struct *mm; 1153 1154 task_lock(task); 1155 mm = task->mm; 1156 if (mm) { 1157 if (!(task->flags & PF_KTHREAD)) 1158 exe_file = get_mm_exe_file(mm); 1159 } 1160 task_unlock(task); 1161 return exe_file; 1162 } 1163 EXPORT_SYMBOL(get_task_exe_file); 1164 1165 /** 1166 * get_task_mm - acquire a reference to the task's mm 1167 * 1168 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning 1169 * this kernel workthread has transiently adopted a user mm with use_mm, 1170 * to do its AIO) is not set and if so returns a reference to it, after 1171 * bumping up the use count. User must release the mm via mmput() 1172 * after use. Typically used by /proc and ptrace. 1173 */ 1174 struct mm_struct *get_task_mm(struct task_struct *task) 1175 { 1176 struct mm_struct *mm; 1177 1178 task_lock(task); 1179 mm = task->mm; 1180 if (mm) { 1181 if (task->flags & PF_KTHREAD) 1182 mm = NULL; 1183 else 1184 mmget(mm); 1185 } 1186 task_unlock(task); 1187 return mm; 1188 } 1189 EXPORT_SYMBOL_GPL(get_task_mm); 1190 1191 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode) 1192 { 1193 struct mm_struct *mm; 1194 int err; 1195 1196 err = mutex_lock_killable(&task->signal->cred_guard_mutex); 1197 if (err) 1198 return ERR_PTR(err); 1199 1200 mm = get_task_mm(task); 1201 if (mm && mm != current->mm && 1202 !ptrace_may_access(task, mode)) { 1203 mmput(mm); 1204 mm = ERR_PTR(-EACCES); 1205 } 1206 mutex_unlock(&task->signal->cred_guard_mutex); 1207 1208 return mm; 1209 } 1210 1211 static void complete_vfork_done(struct task_struct *tsk) 1212 { 1213 struct completion *vfork; 1214 1215 task_lock(tsk); 1216 vfork = tsk->vfork_done; 1217 if (likely(vfork)) { 1218 tsk->vfork_done = NULL; 1219 complete(vfork); 1220 } 1221 task_unlock(tsk); 1222 } 1223 1224 static int wait_for_vfork_done(struct task_struct *child, 1225 struct completion *vfork) 1226 { 1227 int killed; 1228 1229 freezer_do_not_count(); 1230 killed = wait_for_completion_killable(vfork); 1231 freezer_count(); 1232 1233 if (killed) { 1234 task_lock(child); 1235 child->vfork_done = NULL; 1236 task_unlock(child); 1237 } 1238 1239 put_task_struct(child); 1240 return killed; 1241 } 1242 1243 /* Please note the differences between mmput and mm_release. 1244 * mmput is called whenever we stop holding onto a mm_struct, 1245 * error success whatever. 1246 * 1247 * mm_release is called after a mm_struct has been removed 1248 * from the current process. 1249 * 1250 * This difference is important for error handling, when we 1251 * only half set up a mm_struct for a new process and need to restore 1252 * the old one. Because we mmput the new mm_struct before 1253 * restoring the old one. . . 1254 * Eric Biederman 10 January 1998 1255 */ 1256 void mm_release(struct task_struct *tsk, struct mm_struct *mm) 1257 { 1258 /* Get rid of any futexes when releasing the mm */ 1259 #ifdef CONFIG_FUTEX 1260 if (unlikely(tsk->robust_list)) { 1261 exit_robust_list(tsk); 1262 tsk->robust_list = NULL; 1263 } 1264 #ifdef CONFIG_COMPAT 1265 if (unlikely(tsk->compat_robust_list)) { 1266 compat_exit_robust_list(tsk); 1267 tsk->compat_robust_list = NULL; 1268 } 1269 #endif 1270 if (unlikely(!list_empty(&tsk->pi_state_list))) 1271 exit_pi_state_list(tsk); 1272 #endif 1273 1274 uprobe_free_utask(tsk); 1275 1276 /* Get rid of any cached register state */ 1277 deactivate_mm(tsk, mm); 1278 1279 /* 1280 * Signal userspace if we're not exiting with a core dump 1281 * because we want to leave the value intact for debugging 1282 * purposes. 1283 */ 1284 if (tsk->clear_child_tid) { 1285 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) && 1286 atomic_read(&mm->mm_users) > 1) { 1287 /* 1288 * We don't check the error code - if userspace has 1289 * not set up a proper pointer then tough luck. 1290 */ 1291 put_user(0, tsk->clear_child_tid); 1292 do_futex(tsk->clear_child_tid, FUTEX_WAKE, 1293 1, NULL, NULL, 0, 0); 1294 } 1295 tsk->clear_child_tid = NULL; 1296 } 1297 1298 /* 1299 * All done, finally we can wake up parent and return this mm to him. 1300 * Also kthread_stop() uses this completion for synchronization. 1301 */ 1302 if (tsk->vfork_done) 1303 complete_vfork_done(tsk); 1304 } 1305 1306 /* 1307 * Allocate a new mm structure and copy contents from the 1308 * mm structure of the passed in task structure. 1309 */ 1310 static struct mm_struct *dup_mm(struct task_struct *tsk) 1311 { 1312 struct mm_struct *mm, *oldmm = current->mm; 1313 int err; 1314 1315 mm = allocate_mm(); 1316 if (!mm) 1317 goto fail_nomem; 1318 1319 memcpy(mm, oldmm, sizeof(*mm)); 1320 1321 if (!mm_init(mm, tsk, mm->user_ns)) 1322 goto fail_nomem; 1323 1324 err = dup_mmap(mm, oldmm); 1325 if (err) 1326 goto free_pt; 1327 1328 mm->hiwater_rss = get_mm_rss(mm); 1329 mm->hiwater_vm = mm->total_vm; 1330 1331 if (mm->binfmt && !try_module_get(mm->binfmt->module)) 1332 goto free_pt; 1333 1334 return mm; 1335 1336 free_pt: 1337 /* don't put binfmt in mmput, we haven't got module yet */ 1338 mm->binfmt = NULL; 1339 mmput(mm); 1340 1341 fail_nomem: 1342 return NULL; 1343 } 1344 1345 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk) 1346 { 1347 struct mm_struct *mm, *oldmm; 1348 int retval; 1349 1350 tsk->min_flt = tsk->maj_flt = 0; 1351 tsk->nvcsw = tsk->nivcsw = 0; 1352 #ifdef CONFIG_DETECT_HUNG_TASK 1353 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; 1354 tsk->last_switch_time = 0; 1355 #endif 1356 1357 tsk->mm = NULL; 1358 tsk->active_mm = NULL; 1359 1360 /* 1361 * Are we cloning a kernel thread? 1362 * 1363 * We need to steal a active VM for that.. 1364 */ 1365 oldmm = current->mm; 1366 if (!oldmm) 1367 return 0; 1368 1369 /* initialize the new vmacache entries */ 1370 vmacache_flush(tsk); 1371 1372 if (clone_flags & CLONE_VM) { 1373 mmget(oldmm); 1374 mm = oldmm; 1375 goto good_mm; 1376 } 1377 1378 retval = -ENOMEM; 1379 mm = dup_mm(tsk); 1380 if (!mm) 1381 goto fail_nomem; 1382 1383 good_mm: 1384 tsk->mm = mm; 1385 tsk->active_mm = mm; 1386 return 0; 1387 1388 fail_nomem: 1389 return retval; 1390 } 1391 1392 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 1393 { 1394 struct fs_struct *fs = current->fs; 1395 if (clone_flags & CLONE_FS) { 1396 /* tsk->fs is already what we want */ 1397 spin_lock(&fs->lock); 1398 if (fs->in_exec) { 1399 spin_unlock(&fs->lock); 1400 return -EAGAIN; 1401 } 1402 fs->users++; 1403 spin_unlock(&fs->lock); 1404 return 0; 1405 } 1406 tsk->fs = copy_fs_struct(fs); 1407 if (!tsk->fs) 1408 return -ENOMEM; 1409 return 0; 1410 } 1411 1412 static int copy_files(unsigned long clone_flags, struct task_struct *tsk) 1413 { 1414 struct files_struct *oldf, *newf; 1415 int error = 0; 1416 1417 /* 1418 * A background process may not have any files ... 1419 */ 1420 oldf = current->files; 1421 if (!oldf) 1422 goto out; 1423 1424 if (clone_flags & CLONE_FILES) { 1425 atomic_inc(&oldf->count); 1426 goto out; 1427 } 1428 1429 newf = dup_fd(oldf, &error); 1430 if (!newf) 1431 goto out; 1432 1433 tsk->files = newf; 1434 error = 0; 1435 out: 1436 return error; 1437 } 1438 1439 static int copy_io(unsigned long clone_flags, struct task_struct *tsk) 1440 { 1441 #ifdef CONFIG_BLOCK 1442 struct io_context *ioc = current->io_context; 1443 struct io_context *new_ioc; 1444 1445 if (!ioc) 1446 return 0; 1447 /* 1448 * Share io context with parent, if CLONE_IO is set 1449 */ 1450 if (clone_flags & CLONE_IO) { 1451 ioc_task_link(ioc); 1452 tsk->io_context = ioc; 1453 } else if (ioprio_valid(ioc->ioprio)) { 1454 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE); 1455 if (unlikely(!new_ioc)) 1456 return -ENOMEM; 1457 1458 new_ioc->ioprio = ioc->ioprio; 1459 put_io_context(new_ioc); 1460 } 1461 #endif 1462 return 0; 1463 } 1464 1465 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk) 1466 { 1467 struct sighand_struct *sig; 1468 1469 if (clone_flags & CLONE_SIGHAND) { 1470 atomic_inc(¤t->sighand->count); 1471 return 0; 1472 } 1473 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 1474 rcu_assign_pointer(tsk->sighand, sig); 1475 if (!sig) 1476 return -ENOMEM; 1477 1478 atomic_set(&sig->count, 1); 1479 spin_lock_irq(¤t->sighand->siglock); 1480 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 1481 spin_unlock_irq(¤t->sighand->siglock); 1482 return 0; 1483 } 1484 1485 void __cleanup_sighand(struct sighand_struct *sighand) 1486 { 1487 if (atomic_dec_and_test(&sighand->count)) { 1488 signalfd_cleanup(sighand); 1489 /* 1490 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it 1491 * without an RCU grace period, see __lock_task_sighand(). 1492 */ 1493 kmem_cache_free(sighand_cachep, sighand); 1494 } 1495 } 1496 1497 #ifdef CONFIG_POSIX_TIMERS 1498 /* 1499 * Initialize POSIX timer handling for a thread group. 1500 */ 1501 static void posix_cpu_timers_init_group(struct signal_struct *sig) 1502 { 1503 unsigned long cpu_limit; 1504 1505 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); 1506 if (cpu_limit != RLIM_INFINITY) { 1507 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC; 1508 sig->cputimer.running = true; 1509 } 1510 1511 /* The timer lists. */ 1512 INIT_LIST_HEAD(&sig->cpu_timers[0]); 1513 INIT_LIST_HEAD(&sig->cpu_timers[1]); 1514 INIT_LIST_HEAD(&sig->cpu_timers[2]); 1515 } 1516 #else 1517 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { } 1518 #endif 1519 1520 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk) 1521 { 1522 struct signal_struct *sig; 1523 1524 if (clone_flags & CLONE_THREAD) 1525 return 0; 1526 1527 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL); 1528 tsk->signal = sig; 1529 if (!sig) 1530 return -ENOMEM; 1531 1532 sig->nr_threads = 1; 1533 atomic_set(&sig->live, 1); 1534 atomic_set(&sig->sigcnt, 1); 1535 1536 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */ 1537 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node); 1538 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head); 1539 1540 init_waitqueue_head(&sig->wait_chldexit); 1541 sig->curr_target = tsk; 1542 init_sigpending(&sig->shared_pending); 1543 INIT_HLIST_HEAD(&sig->multiprocess); 1544 seqlock_init(&sig->stats_lock); 1545 prev_cputime_init(&sig->prev_cputime); 1546 1547 #ifdef CONFIG_POSIX_TIMERS 1548 INIT_LIST_HEAD(&sig->posix_timers); 1549 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1550 sig->real_timer.function = it_real_fn; 1551 #endif 1552 1553 task_lock(current->group_leader); 1554 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 1555 task_unlock(current->group_leader); 1556 1557 posix_cpu_timers_init_group(sig); 1558 1559 tty_audit_fork(sig); 1560 sched_autogroup_fork(sig); 1561 1562 sig->oom_score_adj = current->signal->oom_score_adj; 1563 sig->oom_score_adj_min = current->signal->oom_score_adj_min; 1564 1565 mutex_init(&sig->cred_guard_mutex); 1566 1567 return 0; 1568 } 1569 1570 static void copy_seccomp(struct task_struct *p) 1571 { 1572 #ifdef CONFIG_SECCOMP 1573 /* 1574 * Must be called with sighand->lock held, which is common to 1575 * all threads in the group. Holding cred_guard_mutex is not 1576 * needed because this new task is not yet running and cannot 1577 * be racing exec. 1578 */ 1579 assert_spin_locked(¤t->sighand->siglock); 1580 1581 /* Ref-count the new filter user, and assign it. */ 1582 get_seccomp_filter(current); 1583 p->seccomp = current->seccomp; 1584 1585 /* 1586 * Explicitly enable no_new_privs here in case it got set 1587 * between the task_struct being duplicated and holding the 1588 * sighand lock. The seccomp state and nnp must be in sync. 1589 */ 1590 if (task_no_new_privs(current)) 1591 task_set_no_new_privs(p); 1592 1593 /* 1594 * If the parent gained a seccomp mode after copying thread 1595 * flags and between before we held the sighand lock, we have 1596 * to manually enable the seccomp thread flag here. 1597 */ 1598 if (p->seccomp.mode != SECCOMP_MODE_DISABLED) 1599 set_tsk_thread_flag(p, TIF_SECCOMP); 1600 #endif 1601 } 1602 1603 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr) 1604 { 1605 current->clear_child_tid = tidptr; 1606 1607 return task_pid_vnr(current); 1608 } 1609 1610 static void rt_mutex_init_task(struct task_struct *p) 1611 { 1612 raw_spin_lock_init(&p->pi_lock); 1613 #ifdef CONFIG_RT_MUTEXES 1614 p->pi_waiters = RB_ROOT_CACHED; 1615 p->pi_top_task = NULL; 1616 p->pi_blocked_on = NULL; 1617 #endif 1618 } 1619 1620 #ifdef CONFIG_POSIX_TIMERS 1621 /* 1622 * Initialize POSIX timer handling for a single task. 1623 */ 1624 static void posix_cpu_timers_init(struct task_struct *tsk) 1625 { 1626 tsk->cputime_expires.prof_exp = 0; 1627 tsk->cputime_expires.virt_exp = 0; 1628 tsk->cputime_expires.sched_exp = 0; 1629 INIT_LIST_HEAD(&tsk->cpu_timers[0]); 1630 INIT_LIST_HEAD(&tsk->cpu_timers[1]); 1631 INIT_LIST_HEAD(&tsk->cpu_timers[2]); 1632 } 1633 #else 1634 static inline void posix_cpu_timers_init(struct task_struct *tsk) { } 1635 #endif 1636 1637 static inline void init_task_pid_links(struct task_struct *task) 1638 { 1639 enum pid_type type; 1640 1641 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 1642 INIT_HLIST_NODE(&task->pid_links[type]); 1643 } 1644 } 1645 1646 static inline void 1647 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid) 1648 { 1649 if (type == PIDTYPE_PID) 1650 task->thread_pid = pid; 1651 else 1652 task->signal->pids[type] = pid; 1653 } 1654 1655 static inline void rcu_copy_process(struct task_struct *p) 1656 { 1657 #ifdef CONFIG_PREEMPT_RCU 1658 p->rcu_read_lock_nesting = 0; 1659 p->rcu_read_unlock_special.s = 0; 1660 p->rcu_blocked_node = NULL; 1661 INIT_LIST_HEAD(&p->rcu_node_entry); 1662 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 1663 #ifdef CONFIG_TASKS_RCU 1664 p->rcu_tasks_holdout = false; 1665 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list); 1666 p->rcu_tasks_idle_cpu = -1; 1667 #endif /* #ifdef CONFIG_TASKS_RCU */ 1668 } 1669 1670 /* 1671 * This creates a new process as a copy of the old one, 1672 * but does not actually start it yet. 1673 * 1674 * It copies the registers, and all the appropriate 1675 * parts of the process environment (as per the clone 1676 * flags). The actual kick-off is left to the caller. 1677 */ 1678 static __latent_entropy struct task_struct *copy_process( 1679 unsigned long clone_flags, 1680 unsigned long stack_start, 1681 unsigned long stack_size, 1682 int __user *child_tidptr, 1683 struct pid *pid, 1684 int trace, 1685 unsigned long tls, 1686 int node) 1687 { 1688 int retval; 1689 struct task_struct *p; 1690 struct multiprocess_signals delayed; 1691 1692 /* 1693 * Don't allow sharing the root directory with processes in a different 1694 * namespace 1695 */ 1696 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 1697 return ERR_PTR(-EINVAL); 1698 1699 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS)) 1700 return ERR_PTR(-EINVAL); 1701 1702 /* 1703 * Thread groups must share signals as well, and detached threads 1704 * can only be started up within the thread group. 1705 */ 1706 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 1707 return ERR_PTR(-EINVAL); 1708 1709 /* 1710 * Shared signal handlers imply shared VM. By way of the above, 1711 * thread groups also imply shared VM. Blocking this case allows 1712 * for various simplifications in other code. 1713 */ 1714 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 1715 return ERR_PTR(-EINVAL); 1716 1717 /* 1718 * Siblings of global init remain as zombies on exit since they are 1719 * not reaped by their parent (swapper). To solve this and to avoid 1720 * multi-rooted process trees, prevent global and container-inits 1721 * from creating siblings. 1722 */ 1723 if ((clone_flags & CLONE_PARENT) && 1724 current->signal->flags & SIGNAL_UNKILLABLE) 1725 return ERR_PTR(-EINVAL); 1726 1727 /* 1728 * If the new process will be in a different pid or user namespace 1729 * do not allow it to share a thread group with the forking task. 1730 */ 1731 if (clone_flags & CLONE_THREAD) { 1732 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) || 1733 (task_active_pid_ns(current) != 1734 current->nsproxy->pid_ns_for_children)) 1735 return ERR_PTR(-EINVAL); 1736 } 1737 1738 /* 1739 * Force any signals received before this point to be delivered 1740 * before the fork happens. Collect up signals sent to multiple 1741 * processes that happen during the fork and delay them so that 1742 * they appear to happen after the fork. 1743 */ 1744 sigemptyset(&delayed.signal); 1745 INIT_HLIST_NODE(&delayed.node); 1746 1747 spin_lock_irq(¤t->sighand->siglock); 1748 if (!(clone_flags & CLONE_THREAD)) 1749 hlist_add_head(&delayed.node, ¤t->signal->multiprocess); 1750 recalc_sigpending(); 1751 spin_unlock_irq(¤t->sighand->siglock); 1752 retval = -ERESTARTNOINTR; 1753 if (signal_pending(current)) 1754 goto fork_out; 1755 1756 retval = -ENOMEM; 1757 p = dup_task_struct(current, node); 1758 if (!p) 1759 goto fork_out; 1760 1761 /* 1762 * This _must_ happen before we call free_task(), i.e. before we jump 1763 * to any of the bad_fork_* labels. This is to avoid freeing 1764 * p->set_child_tid which is (ab)used as a kthread's data pointer for 1765 * kernel threads (PF_KTHREAD). 1766 */ 1767 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; 1768 /* 1769 * Clear TID on mm_release()? 1770 */ 1771 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL; 1772 1773 ftrace_graph_init_task(p); 1774 1775 rt_mutex_init_task(p); 1776 1777 #ifdef CONFIG_PROVE_LOCKING 1778 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); 1779 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 1780 #endif 1781 retval = -EAGAIN; 1782 if (atomic_read(&p->real_cred->user->processes) >= 1783 task_rlimit(p, RLIMIT_NPROC)) { 1784 if (p->real_cred->user != INIT_USER && 1785 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) 1786 goto bad_fork_free; 1787 } 1788 current->flags &= ~PF_NPROC_EXCEEDED; 1789 1790 retval = copy_creds(p, clone_flags); 1791 if (retval < 0) 1792 goto bad_fork_free; 1793 1794 /* 1795 * If multiple threads are within copy_process(), then this check 1796 * triggers too late. This doesn't hurt, the check is only there 1797 * to stop root fork bombs. 1798 */ 1799 retval = -EAGAIN; 1800 if (nr_threads >= max_threads) 1801 goto bad_fork_cleanup_count; 1802 1803 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */ 1804 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE); 1805 p->flags |= PF_FORKNOEXEC; 1806 INIT_LIST_HEAD(&p->children); 1807 INIT_LIST_HEAD(&p->sibling); 1808 rcu_copy_process(p); 1809 p->vfork_done = NULL; 1810 spin_lock_init(&p->alloc_lock); 1811 1812 init_sigpending(&p->pending); 1813 1814 p->utime = p->stime = p->gtime = 0; 1815 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 1816 p->utimescaled = p->stimescaled = 0; 1817 #endif 1818 prev_cputime_init(&p->prev_cputime); 1819 1820 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN 1821 seqcount_init(&p->vtime.seqcount); 1822 p->vtime.starttime = 0; 1823 p->vtime.state = VTIME_INACTIVE; 1824 #endif 1825 1826 #if defined(SPLIT_RSS_COUNTING) 1827 memset(&p->rss_stat, 0, sizeof(p->rss_stat)); 1828 #endif 1829 1830 p->default_timer_slack_ns = current->timer_slack_ns; 1831 1832 #ifdef CONFIG_PSI 1833 p->psi_flags = 0; 1834 #endif 1835 1836 task_io_accounting_init(&p->ioac); 1837 acct_clear_integrals(p); 1838 1839 posix_cpu_timers_init(p); 1840 1841 p->start_time = ktime_get_ns(); 1842 p->real_start_time = ktime_get_boot_ns(); 1843 p->io_context = NULL; 1844 audit_set_context(p, NULL); 1845 cgroup_fork(p); 1846 #ifdef CONFIG_NUMA 1847 p->mempolicy = mpol_dup(p->mempolicy); 1848 if (IS_ERR(p->mempolicy)) { 1849 retval = PTR_ERR(p->mempolicy); 1850 p->mempolicy = NULL; 1851 goto bad_fork_cleanup_threadgroup_lock; 1852 } 1853 #endif 1854 #ifdef CONFIG_CPUSETS 1855 p->cpuset_mem_spread_rotor = NUMA_NO_NODE; 1856 p->cpuset_slab_spread_rotor = NUMA_NO_NODE; 1857 seqcount_init(&p->mems_allowed_seq); 1858 #endif 1859 #ifdef CONFIG_TRACE_IRQFLAGS 1860 p->irq_events = 0; 1861 p->hardirqs_enabled = 0; 1862 p->hardirq_enable_ip = 0; 1863 p->hardirq_enable_event = 0; 1864 p->hardirq_disable_ip = _THIS_IP_; 1865 p->hardirq_disable_event = 0; 1866 p->softirqs_enabled = 1; 1867 p->softirq_enable_ip = _THIS_IP_; 1868 p->softirq_enable_event = 0; 1869 p->softirq_disable_ip = 0; 1870 p->softirq_disable_event = 0; 1871 p->hardirq_context = 0; 1872 p->softirq_context = 0; 1873 #endif 1874 1875 p->pagefault_disabled = 0; 1876 1877 #ifdef CONFIG_LOCKDEP 1878 p->lockdep_depth = 0; /* no locks held yet */ 1879 p->curr_chain_key = 0; 1880 p->lockdep_recursion = 0; 1881 lockdep_init_task(p); 1882 #endif 1883 1884 #ifdef CONFIG_DEBUG_MUTEXES 1885 p->blocked_on = NULL; /* not blocked yet */ 1886 #endif 1887 #ifdef CONFIG_BCACHE 1888 p->sequential_io = 0; 1889 p->sequential_io_avg = 0; 1890 #endif 1891 1892 /* Perform scheduler related setup. Assign this task to a CPU. */ 1893 retval = sched_fork(clone_flags, p); 1894 if (retval) 1895 goto bad_fork_cleanup_policy; 1896 1897 retval = perf_event_init_task(p); 1898 if (retval) 1899 goto bad_fork_cleanup_policy; 1900 retval = audit_alloc(p); 1901 if (retval) 1902 goto bad_fork_cleanup_perf; 1903 /* copy all the process information */ 1904 shm_init_task(p); 1905 retval = security_task_alloc(p, clone_flags); 1906 if (retval) 1907 goto bad_fork_cleanup_audit; 1908 retval = copy_semundo(clone_flags, p); 1909 if (retval) 1910 goto bad_fork_cleanup_security; 1911 retval = copy_files(clone_flags, p); 1912 if (retval) 1913 goto bad_fork_cleanup_semundo; 1914 retval = copy_fs(clone_flags, p); 1915 if (retval) 1916 goto bad_fork_cleanup_files; 1917 retval = copy_sighand(clone_flags, p); 1918 if (retval) 1919 goto bad_fork_cleanup_fs; 1920 retval = copy_signal(clone_flags, p); 1921 if (retval) 1922 goto bad_fork_cleanup_sighand; 1923 retval = copy_mm(clone_flags, p); 1924 if (retval) 1925 goto bad_fork_cleanup_signal; 1926 retval = copy_namespaces(clone_flags, p); 1927 if (retval) 1928 goto bad_fork_cleanup_mm; 1929 retval = copy_io(clone_flags, p); 1930 if (retval) 1931 goto bad_fork_cleanup_namespaces; 1932 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls); 1933 if (retval) 1934 goto bad_fork_cleanup_io; 1935 1936 stackleak_task_init(p); 1937 1938 if (pid != &init_struct_pid) { 1939 pid = alloc_pid(p->nsproxy->pid_ns_for_children); 1940 if (IS_ERR(pid)) { 1941 retval = PTR_ERR(pid); 1942 goto bad_fork_cleanup_thread; 1943 } 1944 } 1945 1946 #ifdef CONFIG_BLOCK 1947 p->plug = NULL; 1948 #endif 1949 #ifdef CONFIG_FUTEX 1950 p->robust_list = NULL; 1951 #ifdef CONFIG_COMPAT 1952 p->compat_robust_list = NULL; 1953 #endif 1954 INIT_LIST_HEAD(&p->pi_state_list); 1955 p->pi_state_cache = NULL; 1956 #endif 1957 /* 1958 * sigaltstack should be cleared when sharing the same VM 1959 */ 1960 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM) 1961 sas_ss_reset(p); 1962 1963 /* 1964 * Syscall tracing and stepping should be turned off in the 1965 * child regardless of CLONE_PTRACE. 1966 */ 1967 user_disable_single_step(p); 1968 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); 1969 #ifdef TIF_SYSCALL_EMU 1970 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); 1971 #endif 1972 clear_all_latency_tracing(p); 1973 1974 /* ok, now we should be set up.. */ 1975 p->pid = pid_nr(pid); 1976 if (clone_flags & CLONE_THREAD) { 1977 p->exit_signal = -1; 1978 p->group_leader = current->group_leader; 1979 p->tgid = current->tgid; 1980 } else { 1981 if (clone_flags & CLONE_PARENT) 1982 p->exit_signal = current->group_leader->exit_signal; 1983 else 1984 p->exit_signal = (clone_flags & CSIGNAL); 1985 p->group_leader = p; 1986 p->tgid = p->pid; 1987 } 1988 1989 p->nr_dirtied = 0; 1990 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10); 1991 p->dirty_paused_when = 0; 1992 1993 p->pdeath_signal = 0; 1994 INIT_LIST_HEAD(&p->thread_group); 1995 p->task_works = NULL; 1996 1997 cgroup_threadgroup_change_begin(current); 1998 /* 1999 * Ensure that the cgroup subsystem policies allow the new process to be 2000 * forked. It should be noted the the new process's css_set can be changed 2001 * between here and cgroup_post_fork() if an organisation operation is in 2002 * progress. 2003 */ 2004 retval = cgroup_can_fork(p); 2005 if (retval) 2006 goto bad_fork_free_pid; 2007 2008 /* 2009 * Make it visible to the rest of the system, but dont wake it up yet. 2010 * Need tasklist lock for parent etc handling! 2011 */ 2012 write_lock_irq(&tasklist_lock); 2013 2014 /* CLONE_PARENT re-uses the old parent */ 2015 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { 2016 p->real_parent = current->real_parent; 2017 p->parent_exec_id = current->parent_exec_id; 2018 } else { 2019 p->real_parent = current; 2020 p->parent_exec_id = current->self_exec_id; 2021 } 2022 2023 klp_copy_process(p); 2024 2025 spin_lock(¤t->sighand->siglock); 2026 2027 /* 2028 * Copy seccomp details explicitly here, in case they were changed 2029 * before holding sighand lock. 2030 */ 2031 copy_seccomp(p); 2032 2033 rseq_fork(p, clone_flags); 2034 2035 /* Don't start children in a dying pid namespace */ 2036 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) { 2037 retval = -ENOMEM; 2038 goto bad_fork_cancel_cgroup; 2039 } 2040 2041 /* Let kill terminate clone/fork in the middle */ 2042 if (fatal_signal_pending(current)) { 2043 retval = -EINTR; 2044 goto bad_fork_cancel_cgroup; 2045 } 2046 2047 2048 init_task_pid_links(p); 2049 if (likely(p->pid)) { 2050 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace); 2051 2052 init_task_pid(p, PIDTYPE_PID, pid); 2053 if (thread_group_leader(p)) { 2054 init_task_pid(p, PIDTYPE_TGID, pid); 2055 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current)); 2056 init_task_pid(p, PIDTYPE_SID, task_session(current)); 2057 2058 if (is_child_reaper(pid)) { 2059 ns_of_pid(pid)->child_reaper = p; 2060 p->signal->flags |= SIGNAL_UNKILLABLE; 2061 } 2062 p->signal->shared_pending.signal = delayed.signal; 2063 p->signal->tty = tty_kref_get(current->signal->tty); 2064 /* 2065 * Inherit has_child_subreaper flag under the same 2066 * tasklist_lock with adding child to the process tree 2067 * for propagate_has_child_subreaper optimization. 2068 */ 2069 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper || 2070 p->real_parent->signal->is_child_subreaper; 2071 list_add_tail(&p->sibling, &p->real_parent->children); 2072 list_add_tail_rcu(&p->tasks, &init_task.tasks); 2073 attach_pid(p, PIDTYPE_TGID); 2074 attach_pid(p, PIDTYPE_PGID); 2075 attach_pid(p, PIDTYPE_SID); 2076 __this_cpu_inc(process_counts); 2077 } else { 2078 current->signal->nr_threads++; 2079 atomic_inc(¤t->signal->live); 2080 atomic_inc(¤t->signal->sigcnt); 2081 task_join_group_stop(p); 2082 list_add_tail_rcu(&p->thread_group, 2083 &p->group_leader->thread_group); 2084 list_add_tail_rcu(&p->thread_node, 2085 &p->signal->thread_head); 2086 } 2087 attach_pid(p, PIDTYPE_PID); 2088 nr_threads++; 2089 } 2090 total_forks++; 2091 hlist_del_init(&delayed.node); 2092 spin_unlock(¤t->sighand->siglock); 2093 syscall_tracepoint_update(p); 2094 write_unlock_irq(&tasklist_lock); 2095 2096 proc_fork_connector(p); 2097 cgroup_post_fork(p); 2098 cgroup_threadgroup_change_end(current); 2099 perf_event_fork(p); 2100 2101 trace_task_newtask(p, clone_flags); 2102 uprobe_copy_process(p, clone_flags); 2103 2104 return p; 2105 2106 bad_fork_cancel_cgroup: 2107 spin_unlock(¤t->sighand->siglock); 2108 write_unlock_irq(&tasklist_lock); 2109 cgroup_cancel_fork(p); 2110 bad_fork_free_pid: 2111 cgroup_threadgroup_change_end(current); 2112 if (pid != &init_struct_pid) 2113 free_pid(pid); 2114 bad_fork_cleanup_thread: 2115 exit_thread(p); 2116 bad_fork_cleanup_io: 2117 if (p->io_context) 2118 exit_io_context(p); 2119 bad_fork_cleanup_namespaces: 2120 exit_task_namespaces(p); 2121 bad_fork_cleanup_mm: 2122 if (p->mm) 2123 mmput(p->mm); 2124 bad_fork_cleanup_signal: 2125 if (!(clone_flags & CLONE_THREAD)) 2126 free_signal_struct(p->signal); 2127 bad_fork_cleanup_sighand: 2128 __cleanup_sighand(p->sighand); 2129 bad_fork_cleanup_fs: 2130 exit_fs(p); /* blocking */ 2131 bad_fork_cleanup_files: 2132 exit_files(p); /* blocking */ 2133 bad_fork_cleanup_semundo: 2134 exit_sem(p); 2135 bad_fork_cleanup_security: 2136 security_task_free(p); 2137 bad_fork_cleanup_audit: 2138 audit_free(p); 2139 bad_fork_cleanup_perf: 2140 perf_event_free_task(p); 2141 bad_fork_cleanup_policy: 2142 lockdep_free_task(p); 2143 #ifdef CONFIG_NUMA 2144 mpol_put(p->mempolicy); 2145 bad_fork_cleanup_threadgroup_lock: 2146 #endif 2147 delayacct_tsk_free(p); 2148 bad_fork_cleanup_count: 2149 atomic_dec(&p->cred->user->processes); 2150 exit_creds(p); 2151 bad_fork_free: 2152 p->state = TASK_DEAD; 2153 put_task_stack(p); 2154 free_task(p); 2155 fork_out: 2156 spin_lock_irq(¤t->sighand->siglock); 2157 hlist_del_init(&delayed.node); 2158 spin_unlock_irq(¤t->sighand->siglock); 2159 return ERR_PTR(retval); 2160 } 2161 2162 static inline void init_idle_pids(struct task_struct *idle) 2163 { 2164 enum pid_type type; 2165 2166 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 2167 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */ 2168 init_task_pid(idle, type, &init_struct_pid); 2169 } 2170 } 2171 2172 struct task_struct *fork_idle(int cpu) 2173 { 2174 struct task_struct *task; 2175 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0, 2176 cpu_to_node(cpu)); 2177 if (!IS_ERR(task)) { 2178 init_idle_pids(task); 2179 init_idle(task, cpu); 2180 } 2181 2182 return task; 2183 } 2184 2185 /* 2186 * Ok, this is the main fork-routine. 2187 * 2188 * It copies the process, and if successful kick-starts 2189 * it and waits for it to finish using the VM if required. 2190 */ 2191 long _do_fork(unsigned long clone_flags, 2192 unsigned long stack_start, 2193 unsigned long stack_size, 2194 int __user *parent_tidptr, 2195 int __user *child_tidptr, 2196 unsigned long tls) 2197 { 2198 struct completion vfork; 2199 struct pid *pid; 2200 struct task_struct *p; 2201 int trace = 0; 2202 long nr; 2203 2204 /* 2205 * Determine whether and which event to report to ptracer. When 2206 * called from kernel_thread or CLONE_UNTRACED is explicitly 2207 * requested, no event is reported; otherwise, report if the event 2208 * for the type of forking is enabled. 2209 */ 2210 if (!(clone_flags & CLONE_UNTRACED)) { 2211 if (clone_flags & CLONE_VFORK) 2212 trace = PTRACE_EVENT_VFORK; 2213 else if ((clone_flags & CSIGNAL) != SIGCHLD) 2214 trace = PTRACE_EVENT_CLONE; 2215 else 2216 trace = PTRACE_EVENT_FORK; 2217 2218 if (likely(!ptrace_event_enabled(current, trace))) 2219 trace = 0; 2220 } 2221 2222 p = copy_process(clone_flags, stack_start, stack_size, 2223 child_tidptr, NULL, trace, tls, NUMA_NO_NODE); 2224 add_latent_entropy(); 2225 2226 if (IS_ERR(p)) 2227 return PTR_ERR(p); 2228 2229 /* 2230 * Do this prior waking up the new thread - the thread pointer 2231 * might get invalid after that point, if the thread exits quickly. 2232 */ 2233 trace_sched_process_fork(current, p); 2234 2235 pid = get_task_pid(p, PIDTYPE_PID); 2236 nr = pid_vnr(pid); 2237 2238 if (clone_flags & CLONE_PARENT_SETTID) 2239 put_user(nr, parent_tidptr); 2240 2241 if (clone_flags & CLONE_VFORK) { 2242 p->vfork_done = &vfork; 2243 init_completion(&vfork); 2244 get_task_struct(p); 2245 } 2246 2247 wake_up_new_task(p); 2248 2249 /* forking complete and child started to run, tell ptracer */ 2250 if (unlikely(trace)) 2251 ptrace_event_pid(trace, pid); 2252 2253 if (clone_flags & CLONE_VFORK) { 2254 if (!wait_for_vfork_done(p, &vfork)) 2255 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid); 2256 } 2257 2258 put_pid(pid); 2259 return nr; 2260 } 2261 2262 #ifndef CONFIG_HAVE_COPY_THREAD_TLS 2263 /* For compatibility with architectures that call do_fork directly rather than 2264 * using the syscall entry points below. */ 2265 long do_fork(unsigned long clone_flags, 2266 unsigned long stack_start, 2267 unsigned long stack_size, 2268 int __user *parent_tidptr, 2269 int __user *child_tidptr) 2270 { 2271 return _do_fork(clone_flags, stack_start, stack_size, 2272 parent_tidptr, child_tidptr, 0); 2273 } 2274 #endif 2275 2276 /* 2277 * Create a kernel thread. 2278 */ 2279 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags) 2280 { 2281 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn, 2282 (unsigned long)arg, NULL, NULL, 0); 2283 } 2284 2285 #ifdef __ARCH_WANT_SYS_FORK 2286 SYSCALL_DEFINE0(fork) 2287 { 2288 #ifdef CONFIG_MMU 2289 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0); 2290 #else 2291 /* can not support in nommu mode */ 2292 return -EINVAL; 2293 #endif 2294 } 2295 #endif 2296 2297 #ifdef __ARCH_WANT_SYS_VFORK 2298 SYSCALL_DEFINE0(vfork) 2299 { 2300 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0, 2301 0, NULL, NULL, 0); 2302 } 2303 #endif 2304 2305 #ifdef __ARCH_WANT_SYS_CLONE 2306 #ifdef CONFIG_CLONE_BACKWARDS 2307 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2308 int __user *, parent_tidptr, 2309 unsigned long, tls, 2310 int __user *, child_tidptr) 2311 #elif defined(CONFIG_CLONE_BACKWARDS2) 2312 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags, 2313 int __user *, parent_tidptr, 2314 int __user *, child_tidptr, 2315 unsigned long, tls) 2316 #elif defined(CONFIG_CLONE_BACKWARDS3) 2317 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp, 2318 int, stack_size, 2319 int __user *, parent_tidptr, 2320 int __user *, child_tidptr, 2321 unsigned long, tls) 2322 #else 2323 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2324 int __user *, parent_tidptr, 2325 int __user *, child_tidptr, 2326 unsigned long, tls) 2327 #endif 2328 { 2329 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls); 2330 } 2331 #endif 2332 2333 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data) 2334 { 2335 struct task_struct *leader, *parent, *child; 2336 int res; 2337 2338 read_lock(&tasklist_lock); 2339 leader = top = top->group_leader; 2340 down: 2341 for_each_thread(leader, parent) { 2342 list_for_each_entry(child, &parent->children, sibling) { 2343 res = visitor(child, data); 2344 if (res) { 2345 if (res < 0) 2346 goto out; 2347 leader = child; 2348 goto down; 2349 } 2350 up: 2351 ; 2352 } 2353 } 2354 2355 if (leader != top) { 2356 child = leader; 2357 parent = child->real_parent; 2358 leader = parent->group_leader; 2359 goto up; 2360 } 2361 out: 2362 read_unlock(&tasklist_lock); 2363 } 2364 2365 #ifndef ARCH_MIN_MMSTRUCT_ALIGN 2366 #define ARCH_MIN_MMSTRUCT_ALIGN 0 2367 #endif 2368 2369 static void sighand_ctor(void *data) 2370 { 2371 struct sighand_struct *sighand = data; 2372 2373 spin_lock_init(&sighand->siglock); 2374 init_waitqueue_head(&sighand->signalfd_wqh); 2375 } 2376 2377 void __init proc_caches_init(void) 2378 { 2379 unsigned int mm_size; 2380 2381 sighand_cachep = kmem_cache_create("sighand_cache", 2382 sizeof(struct sighand_struct), 0, 2383 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU| 2384 SLAB_ACCOUNT, sighand_ctor); 2385 signal_cachep = kmem_cache_create("signal_cache", 2386 sizeof(struct signal_struct), 0, 2387 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2388 NULL); 2389 files_cachep = kmem_cache_create("files_cache", 2390 sizeof(struct files_struct), 0, 2391 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2392 NULL); 2393 fs_cachep = kmem_cache_create("fs_cache", 2394 sizeof(struct fs_struct), 0, 2395 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2396 NULL); 2397 2398 /* 2399 * The mm_cpumask is located at the end of mm_struct, and is 2400 * dynamically sized based on the maximum CPU number this system 2401 * can have, taking hotplug into account (nr_cpu_ids). 2402 */ 2403 mm_size = sizeof(struct mm_struct) + cpumask_size(); 2404 2405 mm_cachep = kmem_cache_create_usercopy("mm_struct", 2406 mm_size, ARCH_MIN_MMSTRUCT_ALIGN, 2407 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2408 offsetof(struct mm_struct, saved_auxv), 2409 sizeof_field(struct mm_struct, saved_auxv), 2410 NULL); 2411 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT); 2412 mmap_init(); 2413 nsproxy_cache_init(); 2414 } 2415 2416 /* 2417 * Check constraints on flags passed to the unshare system call. 2418 */ 2419 static int check_unshare_flags(unsigned long unshare_flags) 2420 { 2421 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND| 2422 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM| 2423 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET| 2424 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP)) 2425 return -EINVAL; 2426 /* 2427 * Not implemented, but pretend it works if there is nothing 2428 * to unshare. Note that unsharing the address space or the 2429 * signal handlers also need to unshare the signal queues (aka 2430 * CLONE_THREAD). 2431 */ 2432 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) { 2433 if (!thread_group_empty(current)) 2434 return -EINVAL; 2435 } 2436 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) { 2437 if (atomic_read(¤t->sighand->count) > 1) 2438 return -EINVAL; 2439 } 2440 if (unshare_flags & CLONE_VM) { 2441 if (!current_is_single_threaded()) 2442 return -EINVAL; 2443 } 2444 2445 return 0; 2446 } 2447 2448 /* 2449 * Unshare the filesystem structure if it is being shared 2450 */ 2451 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) 2452 { 2453 struct fs_struct *fs = current->fs; 2454 2455 if (!(unshare_flags & CLONE_FS) || !fs) 2456 return 0; 2457 2458 /* don't need lock here; in the worst case we'll do useless copy */ 2459 if (fs->users == 1) 2460 return 0; 2461 2462 *new_fsp = copy_fs_struct(fs); 2463 if (!*new_fsp) 2464 return -ENOMEM; 2465 2466 return 0; 2467 } 2468 2469 /* 2470 * Unshare file descriptor table if it is being shared 2471 */ 2472 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp) 2473 { 2474 struct files_struct *fd = current->files; 2475 int error = 0; 2476 2477 if ((unshare_flags & CLONE_FILES) && 2478 (fd && atomic_read(&fd->count) > 1)) { 2479 *new_fdp = dup_fd(fd, &error); 2480 if (!*new_fdp) 2481 return error; 2482 } 2483 2484 return 0; 2485 } 2486 2487 /* 2488 * unshare allows a process to 'unshare' part of the process 2489 * context which was originally shared using clone. copy_* 2490 * functions used by do_fork() cannot be used here directly 2491 * because they modify an inactive task_struct that is being 2492 * constructed. Here we are modifying the current, active, 2493 * task_struct. 2494 */ 2495 int ksys_unshare(unsigned long unshare_flags) 2496 { 2497 struct fs_struct *fs, *new_fs = NULL; 2498 struct files_struct *fd, *new_fd = NULL; 2499 struct cred *new_cred = NULL; 2500 struct nsproxy *new_nsproxy = NULL; 2501 int do_sysvsem = 0; 2502 int err; 2503 2504 /* 2505 * If unsharing a user namespace must also unshare the thread group 2506 * and unshare the filesystem root and working directories. 2507 */ 2508 if (unshare_flags & CLONE_NEWUSER) 2509 unshare_flags |= CLONE_THREAD | CLONE_FS; 2510 /* 2511 * If unsharing vm, must also unshare signal handlers. 2512 */ 2513 if (unshare_flags & CLONE_VM) 2514 unshare_flags |= CLONE_SIGHAND; 2515 /* 2516 * If unsharing a signal handlers, must also unshare the signal queues. 2517 */ 2518 if (unshare_flags & CLONE_SIGHAND) 2519 unshare_flags |= CLONE_THREAD; 2520 /* 2521 * If unsharing namespace, must also unshare filesystem information. 2522 */ 2523 if (unshare_flags & CLONE_NEWNS) 2524 unshare_flags |= CLONE_FS; 2525 2526 err = check_unshare_flags(unshare_flags); 2527 if (err) 2528 goto bad_unshare_out; 2529 /* 2530 * CLONE_NEWIPC must also detach from the undolist: after switching 2531 * to a new ipc namespace, the semaphore arrays from the old 2532 * namespace are unreachable. 2533 */ 2534 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) 2535 do_sysvsem = 1; 2536 err = unshare_fs(unshare_flags, &new_fs); 2537 if (err) 2538 goto bad_unshare_out; 2539 err = unshare_fd(unshare_flags, &new_fd); 2540 if (err) 2541 goto bad_unshare_cleanup_fs; 2542 err = unshare_userns(unshare_flags, &new_cred); 2543 if (err) 2544 goto bad_unshare_cleanup_fd; 2545 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, 2546 new_cred, new_fs); 2547 if (err) 2548 goto bad_unshare_cleanup_cred; 2549 2550 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) { 2551 if (do_sysvsem) { 2552 /* 2553 * CLONE_SYSVSEM is equivalent to sys_exit(). 2554 */ 2555 exit_sem(current); 2556 } 2557 if (unshare_flags & CLONE_NEWIPC) { 2558 /* Orphan segments in old ns (see sem above). */ 2559 exit_shm(current); 2560 shm_init_task(current); 2561 } 2562 2563 if (new_nsproxy) 2564 switch_task_namespaces(current, new_nsproxy); 2565 2566 task_lock(current); 2567 2568 if (new_fs) { 2569 fs = current->fs; 2570 spin_lock(&fs->lock); 2571 current->fs = new_fs; 2572 if (--fs->users) 2573 new_fs = NULL; 2574 else 2575 new_fs = fs; 2576 spin_unlock(&fs->lock); 2577 } 2578 2579 if (new_fd) { 2580 fd = current->files; 2581 current->files = new_fd; 2582 new_fd = fd; 2583 } 2584 2585 task_unlock(current); 2586 2587 if (new_cred) { 2588 /* Install the new user namespace */ 2589 commit_creds(new_cred); 2590 new_cred = NULL; 2591 } 2592 } 2593 2594 perf_event_namespaces(current); 2595 2596 bad_unshare_cleanup_cred: 2597 if (new_cred) 2598 put_cred(new_cred); 2599 bad_unshare_cleanup_fd: 2600 if (new_fd) 2601 put_files_struct(new_fd); 2602 2603 bad_unshare_cleanup_fs: 2604 if (new_fs) 2605 free_fs_struct(new_fs); 2606 2607 bad_unshare_out: 2608 return err; 2609 } 2610 2611 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags) 2612 { 2613 return ksys_unshare(unshare_flags); 2614 } 2615 2616 /* 2617 * Helper to unshare the files of the current task. 2618 * We don't want to expose copy_files internals to 2619 * the exec layer of the kernel. 2620 */ 2621 2622 int unshare_files(struct files_struct **displaced) 2623 { 2624 struct task_struct *task = current; 2625 struct files_struct *copy = NULL; 2626 int error; 2627 2628 error = unshare_fd(CLONE_FILES, ©); 2629 if (error || !copy) { 2630 *displaced = NULL; 2631 return error; 2632 } 2633 *displaced = task->files; 2634 task_lock(task); 2635 task->files = copy; 2636 task_unlock(task); 2637 return 0; 2638 } 2639 2640 int sysctl_max_threads(struct ctl_table *table, int write, 2641 void __user *buffer, size_t *lenp, loff_t *ppos) 2642 { 2643 struct ctl_table t; 2644 int ret; 2645 int threads = max_threads; 2646 int min = MIN_THREADS; 2647 int max = MAX_THREADS; 2648 2649 t = *table; 2650 t.data = &threads; 2651 t.extra1 = &min; 2652 t.extra2 = &max; 2653 2654 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); 2655 if (ret || !write) 2656 return ret; 2657 2658 set_max_threads(threads); 2659 2660 return 0; 2661 } 2662