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 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR 168 169 /* 170 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a 171 * kmemcache based allocator. 172 */ 173 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) 174 175 #ifdef CONFIG_VMAP_STACK 176 /* 177 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB 178 * flush. Try to minimize the number of calls by caching stacks. 179 */ 180 #define NR_CACHED_STACKS 2 181 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]); 182 183 static int free_vm_stack_cache(unsigned int cpu) 184 { 185 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu); 186 int i; 187 188 for (i = 0; i < NR_CACHED_STACKS; i++) { 189 struct vm_struct *vm_stack = cached_vm_stacks[i]; 190 191 if (!vm_stack) 192 continue; 193 194 vfree(vm_stack->addr); 195 cached_vm_stacks[i] = NULL; 196 } 197 198 return 0; 199 } 200 #endif 201 202 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node) 203 { 204 #ifdef CONFIG_VMAP_STACK 205 void *stack; 206 int i; 207 208 for (i = 0; i < NR_CACHED_STACKS; i++) { 209 struct vm_struct *s; 210 211 s = this_cpu_xchg(cached_stacks[i], NULL); 212 213 if (!s) 214 continue; 215 216 /* Clear stale pointers from reused stack. */ 217 memset(s->addr, 0, THREAD_SIZE); 218 219 tsk->stack_vm_area = s; 220 tsk->stack = s->addr; 221 return s->addr; 222 } 223 224 /* 225 * Allocated stacks are cached and later reused by new threads, 226 * so memcg accounting is performed manually on assigning/releasing 227 * stacks to tasks. Drop __GFP_ACCOUNT. 228 */ 229 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN, 230 VMALLOC_START, VMALLOC_END, 231 THREADINFO_GFP & ~__GFP_ACCOUNT, 232 PAGE_KERNEL, 233 0, node, __builtin_return_address(0)); 234 235 /* 236 * We can't call find_vm_area() in interrupt context, and 237 * free_thread_stack() can be called in interrupt context, 238 * so cache the vm_struct. 239 */ 240 if (stack) { 241 tsk->stack_vm_area = find_vm_area(stack); 242 tsk->stack = stack; 243 } 244 return stack; 245 #else 246 struct page *page = alloc_pages_node(node, THREADINFO_GFP, 247 THREAD_SIZE_ORDER); 248 249 return page ? page_address(page) : NULL; 250 #endif 251 } 252 253 static inline void free_thread_stack(struct task_struct *tsk) 254 { 255 #ifdef CONFIG_VMAP_STACK 256 struct vm_struct *vm = task_stack_vm_area(tsk); 257 258 if (vm) { 259 int i; 260 261 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 262 mod_memcg_page_state(vm->pages[i], 263 MEMCG_KERNEL_STACK_KB, 264 -(int)(PAGE_SIZE / 1024)); 265 266 memcg_kmem_uncharge(vm->pages[i], 0); 267 } 268 269 for (i = 0; i < NR_CACHED_STACKS; i++) { 270 if (this_cpu_cmpxchg(cached_stacks[i], 271 NULL, tsk->stack_vm_area) != NULL) 272 continue; 273 274 return; 275 } 276 277 vfree_atomic(tsk->stack); 278 return; 279 } 280 #endif 281 282 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER); 283 } 284 # else 285 static struct kmem_cache *thread_stack_cache; 286 287 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, 288 int node) 289 { 290 unsigned long *stack; 291 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node); 292 tsk->stack = stack; 293 return stack; 294 } 295 296 static void free_thread_stack(struct task_struct *tsk) 297 { 298 kmem_cache_free(thread_stack_cache, tsk->stack); 299 } 300 301 void thread_stack_cache_init(void) 302 { 303 thread_stack_cache = kmem_cache_create_usercopy("thread_stack", 304 THREAD_SIZE, THREAD_SIZE, 0, 0, 305 THREAD_SIZE, NULL); 306 BUG_ON(thread_stack_cache == NULL); 307 } 308 # endif 309 #endif 310 311 /* SLAB cache for signal_struct structures (tsk->signal) */ 312 static struct kmem_cache *signal_cachep; 313 314 /* SLAB cache for sighand_struct structures (tsk->sighand) */ 315 struct kmem_cache *sighand_cachep; 316 317 /* SLAB cache for files_struct structures (tsk->files) */ 318 struct kmem_cache *files_cachep; 319 320 /* SLAB cache for fs_struct structures (tsk->fs) */ 321 struct kmem_cache *fs_cachep; 322 323 /* SLAB cache for vm_area_struct structures */ 324 static struct kmem_cache *vm_area_cachep; 325 326 /* SLAB cache for mm_struct structures (tsk->mm) */ 327 static struct kmem_cache *mm_cachep; 328 329 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm) 330 { 331 struct vm_area_struct *vma; 332 333 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 334 if (vma) 335 vma_init(vma, mm); 336 return vma; 337 } 338 339 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig) 340 { 341 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 342 343 if (new) { 344 *new = *orig; 345 INIT_LIST_HEAD(&new->anon_vma_chain); 346 } 347 return new; 348 } 349 350 void vm_area_free(struct vm_area_struct *vma) 351 { 352 kmem_cache_free(vm_area_cachep, vma); 353 } 354 355 static void account_kernel_stack(struct task_struct *tsk, int account) 356 { 357 void *stack = task_stack_page(tsk); 358 struct vm_struct *vm = task_stack_vm_area(tsk); 359 360 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0); 361 362 if (vm) { 363 int i; 364 365 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE); 366 367 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 368 mod_zone_page_state(page_zone(vm->pages[i]), 369 NR_KERNEL_STACK_KB, 370 PAGE_SIZE / 1024 * account); 371 } 372 } else { 373 /* 374 * All stack pages are in the same zone and belong to the 375 * same memcg. 376 */ 377 struct page *first_page = virt_to_page(stack); 378 379 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB, 380 THREAD_SIZE / 1024 * account); 381 382 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB, 383 account * (THREAD_SIZE / 1024)); 384 } 385 } 386 387 static int memcg_charge_kernel_stack(struct task_struct *tsk) 388 { 389 #ifdef CONFIG_VMAP_STACK 390 struct vm_struct *vm = task_stack_vm_area(tsk); 391 int ret; 392 393 if (vm) { 394 int i; 395 396 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 397 /* 398 * If memcg_kmem_charge() fails, page->mem_cgroup 399 * pointer is NULL, and both memcg_kmem_uncharge() 400 * and mod_memcg_page_state() in free_thread_stack() 401 * will ignore this page. So it's safe. 402 */ 403 ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0); 404 if (ret) 405 return ret; 406 407 mod_memcg_page_state(vm->pages[i], 408 MEMCG_KERNEL_STACK_KB, 409 PAGE_SIZE / 1024); 410 } 411 } 412 #endif 413 return 0; 414 } 415 416 static void release_task_stack(struct task_struct *tsk) 417 { 418 if (WARN_ON(tsk->state != TASK_DEAD)) 419 return; /* Better to leak the stack than to free prematurely */ 420 421 account_kernel_stack(tsk, -1); 422 free_thread_stack(tsk); 423 tsk->stack = NULL; 424 #ifdef CONFIG_VMAP_STACK 425 tsk->stack_vm_area = NULL; 426 #endif 427 } 428 429 #ifdef CONFIG_THREAD_INFO_IN_TASK 430 void put_task_stack(struct task_struct *tsk) 431 { 432 if (refcount_dec_and_test(&tsk->stack_refcount)) 433 release_task_stack(tsk); 434 } 435 #endif 436 437 void free_task(struct task_struct *tsk) 438 { 439 #ifndef CONFIG_THREAD_INFO_IN_TASK 440 /* 441 * The task is finally done with both the stack and thread_info, 442 * so free both. 443 */ 444 release_task_stack(tsk); 445 #else 446 /* 447 * If the task had a separate stack allocation, it should be gone 448 * by now. 449 */ 450 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0); 451 #endif 452 rt_mutex_debug_task_free(tsk); 453 ftrace_graph_exit_task(tsk); 454 put_seccomp_filter(tsk); 455 arch_release_task_struct(tsk); 456 if (tsk->flags & PF_KTHREAD) 457 free_kthread_struct(tsk); 458 free_task_struct(tsk); 459 } 460 EXPORT_SYMBOL(free_task); 461 462 #ifdef CONFIG_MMU 463 static __latent_entropy int dup_mmap(struct mm_struct *mm, 464 struct mm_struct *oldmm) 465 { 466 struct vm_area_struct *mpnt, *tmp, *prev, **pprev; 467 struct rb_node **rb_link, *rb_parent; 468 int retval; 469 unsigned long charge; 470 LIST_HEAD(uf); 471 472 uprobe_start_dup_mmap(); 473 if (down_write_killable(&oldmm->mmap_sem)) { 474 retval = -EINTR; 475 goto fail_uprobe_end; 476 } 477 flush_cache_dup_mm(oldmm); 478 uprobe_dup_mmap(oldmm, mm); 479 /* 480 * Not linked in yet - no deadlock potential: 481 */ 482 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING); 483 484 /* No ordering required: file already has been exposed. */ 485 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm)); 486 487 mm->total_vm = oldmm->total_vm; 488 mm->data_vm = oldmm->data_vm; 489 mm->exec_vm = oldmm->exec_vm; 490 mm->stack_vm = oldmm->stack_vm; 491 492 rb_link = &mm->mm_rb.rb_node; 493 rb_parent = NULL; 494 pprev = &mm->mmap; 495 retval = ksm_fork(mm, oldmm); 496 if (retval) 497 goto out; 498 retval = khugepaged_fork(mm, oldmm); 499 if (retval) 500 goto out; 501 502 prev = NULL; 503 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { 504 struct file *file; 505 506 if (mpnt->vm_flags & VM_DONTCOPY) { 507 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt)); 508 continue; 509 } 510 charge = 0; 511 /* 512 * Don't duplicate many vmas if we've been oom-killed (for 513 * example) 514 */ 515 if (fatal_signal_pending(current)) { 516 retval = -EINTR; 517 goto out; 518 } 519 if (mpnt->vm_flags & VM_ACCOUNT) { 520 unsigned long len = vma_pages(mpnt); 521 522 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */ 523 goto fail_nomem; 524 charge = len; 525 } 526 tmp = vm_area_dup(mpnt); 527 if (!tmp) 528 goto fail_nomem; 529 retval = vma_dup_policy(mpnt, tmp); 530 if (retval) 531 goto fail_nomem_policy; 532 tmp->vm_mm = mm; 533 retval = dup_userfaultfd(tmp, &uf); 534 if (retval) 535 goto fail_nomem_anon_vma_fork; 536 if (tmp->vm_flags & VM_WIPEONFORK) { 537 /* VM_WIPEONFORK gets a clean slate in the child. */ 538 tmp->anon_vma = NULL; 539 if (anon_vma_prepare(tmp)) 540 goto fail_nomem_anon_vma_fork; 541 } else if (anon_vma_fork(tmp, mpnt)) 542 goto fail_nomem_anon_vma_fork; 543 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT); 544 tmp->vm_next = tmp->vm_prev = NULL; 545 file = tmp->vm_file; 546 if (file) { 547 struct inode *inode = file_inode(file); 548 struct address_space *mapping = file->f_mapping; 549 550 get_file(file); 551 if (tmp->vm_flags & VM_DENYWRITE) 552 atomic_dec(&inode->i_writecount); 553 i_mmap_lock_write(mapping); 554 if (tmp->vm_flags & VM_SHARED) 555 atomic_inc(&mapping->i_mmap_writable); 556 flush_dcache_mmap_lock(mapping); 557 /* insert tmp into the share list, just after mpnt */ 558 vma_interval_tree_insert_after(tmp, mpnt, 559 &mapping->i_mmap); 560 flush_dcache_mmap_unlock(mapping); 561 i_mmap_unlock_write(mapping); 562 } 563 564 /* 565 * Clear hugetlb-related page reserves for children. This only 566 * affects MAP_PRIVATE mappings. Faults generated by the child 567 * are not guaranteed to succeed, even if read-only 568 */ 569 if (is_vm_hugetlb_page(tmp)) 570 reset_vma_resv_huge_pages(tmp); 571 572 /* 573 * Link in the new vma and copy the page table entries. 574 */ 575 *pprev = tmp; 576 pprev = &tmp->vm_next; 577 tmp->vm_prev = prev; 578 prev = tmp; 579 580 __vma_link_rb(mm, tmp, rb_link, rb_parent); 581 rb_link = &tmp->vm_rb.rb_right; 582 rb_parent = &tmp->vm_rb; 583 584 mm->map_count++; 585 if (!(tmp->vm_flags & VM_WIPEONFORK)) 586 retval = copy_page_range(mm, oldmm, mpnt); 587 588 if (tmp->vm_ops && tmp->vm_ops->open) 589 tmp->vm_ops->open(tmp); 590 591 if (retval) 592 goto out; 593 } 594 /* a new mm has just been created */ 595 retval = arch_dup_mmap(oldmm, mm); 596 out: 597 up_write(&mm->mmap_sem); 598 flush_tlb_mm(oldmm); 599 up_write(&oldmm->mmap_sem); 600 dup_userfaultfd_complete(&uf); 601 fail_uprobe_end: 602 uprobe_end_dup_mmap(); 603 return retval; 604 fail_nomem_anon_vma_fork: 605 mpol_put(vma_policy(tmp)); 606 fail_nomem_policy: 607 vm_area_free(tmp); 608 fail_nomem: 609 retval = -ENOMEM; 610 vm_unacct_memory(charge); 611 goto out; 612 } 613 614 static inline int mm_alloc_pgd(struct mm_struct *mm) 615 { 616 mm->pgd = pgd_alloc(mm); 617 if (unlikely(!mm->pgd)) 618 return -ENOMEM; 619 return 0; 620 } 621 622 static inline void mm_free_pgd(struct mm_struct *mm) 623 { 624 pgd_free(mm, mm->pgd); 625 } 626 #else 627 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) 628 { 629 down_write(&oldmm->mmap_sem); 630 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm)); 631 up_write(&oldmm->mmap_sem); 632 return 0; 633 } 634 #define mm_alloc_pgd(mm) (0) 635 #define mm_free_pgd(mm) 636 #endif /* CONFIG_MMU */ 637 638 static void check_mm(struct mm_struct *mm) 639 { 640 int i; 641 642 for (i = 0; i < NR_MM_COUNTERS; i++) { 643 long x = atomic_long_read(&mm->rss_stat.count[i]); 644 645 if (unlikely(x)) 646 printk(KERN_ALERT "BUG: Bad rss-counter state " 647 "mm:%p idx:%d val:%ld\n", mm, i, x); 648 } 649 650 if (mm_pgtables_bytes(mm)) 651 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n", 652 mm_pgtables_bytes(mm)); 653 654 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 655 VM_BUG_ON_MM(mm->pmd_huge_pte, mm); 656 #endif 657 } 658 659 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL)) 660 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) 661 662 /* 663 * Called when the last reference to the mm 664 * is dropped: either by a lazy thread or by 665 * mmput. Free the page directory and the mm. 666 */ 667 void __mmdrop(struct mm_struct *mm) 668 { 669 BUG_ON(mm == &init_mm); 670 WARN_ON_ONCE(mm == current->mm); 671 WARN_ON_ONCE(mm == current->active_mm); 672 mm_free_pgd(mm); 673 destroy_context(mm); 674 hmm_mm_destroy(mm); 675 mmu_notifier_mm_destroy(mm); 676 check_mm(mm); 677 put_user_ns(mm->user_ns); 678 free_mm(mm); 679 } 680 EXPORT_SYMBOL_GPL(__mmdrop); 681 682 static void mmdrop_async_fn(struct work_struct *work) 683 { 684 struct mm_struct *mm; 685 686 mm = container_of(work, struct mm_struct, async_put_work); 687 __mmdrop(mm); 688 } 689 690 static void mmdrop_async(struct mm_struct *mm) 691 { 692 if (unlikely(atomic_dec_and_test(&mm->mm_count))) { 693 INIT_WORK(&mm->async_put_work, mmdrop_async_fn); 694 schedule_work(&mm->async_put_work); 695 } 696 } 697 698 static inline void free_signal_struct(struct signal_struct *sig) 699 { 700 taskstats_tgid_free(sig); 701 sched_autogroup_exit(sig); 702 /* 703 * __mmdrop is not safe to call from softirq context on x86 due to 704 * pgd_dtor so postpone it to the async context 705 */ 706 if (sig->oom_mm) 707 mmdrop_async(sig->oom_mm); 708 kmem_cache_free(signal_cachep, sig); 709 } 710 711 static inline void put_signal_struct(struct signal_struct *sig) 712 { 713 if (refcount_dec_and_test(&sig->sigcnt)) 714 free_signal_struct(sig); 715 } 716 717 void __put_task_struct(struct task_struct *tsk) 718 { 719 WARN_ON(!tsk->exit_state); 720 WARN_ON(refcount_read(&tsk->usage)); 721 WARN_ON(tsk == current); 722 723 cgroup_free(tsk); 724 task_numa_free(tsk); 725 security_task_free(tsk); 726 exit_creds(tsk); 727 delayacct_tsk_free(tsk); 728 put_signal_struct(tsk->signal); 729 730 if (!profile_handoff_task(tsk)) 731 free_task(tsk); 732 } 733 EXPORT_SYMBOL_GPL(__put_task_struct); 734 735 void __init __weak arch_task_cache_init(void) { } 736 737 /* 738 * set_max_threads 739 */ 740 static void set_max_threads(unsigned int max_threads_suggested) 741 { 742 u64 threads; 743 unsigned long nr_pages = totalram_pages(); 744 745 /* 746 * The number of threads shall be limited such that the thread 747 * structures may only consume a small part of the available memory. 748 */ 749 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64) 750 threads = MAX_THREADS; 751 else 752 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE, 753 (u64) THREAD_SIZE * 8UL); 754 755 if (threads > max_threads_suggested) 756 threads = max_threads_suggested; 757 758 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS); 759 } 760 761 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT 762 /* Initialized by the architecture: */ 763 int arch_task_struct_size __read_mostly; 764 #endif 765 766 static void task_struct_whitelist(unsigned long *offset, unsigned long *size) 767 { 768 /* Fetch thread_struct whitelist for the architecture. */ 769 arch_thread_struct_whitelist(offset, size); 770 771 /* 772 * Handle zero-sized whitelist or empty thread_struct, otherwise 773 * adjust offset to position of thread_struct in task_struct. 774 */ 775 if (unlikely(*size == 0)) 776 *offset = 0; 777 else 778 *offset += offsetof(struct task_struct, thread); 779 } 780 781 void __init fork_init(void) 782 { 783 int i; 784 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 785 #ifndef ARCH_MIN_TASKALIGN 786 #define ARCH_MIN_TASKALIGN 0 787 #endif 788 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN); 789 unsigned long useroffset, usersize; 790 791 /* create a slab on which task_structs can be allocated */ 792 task_struct_whitelist(&useroffset, &usersize); 793 task_struct_cachep = kmem_cache_create_usercopy("task_struct", 794 arch_task_struct_size, align, 795 SLAB_PANIC|SLAB_ACCOUNT, 796 useroffset, usersize, NULL); 797 #endif 798 799 /* do the arch specific task caches init */ 800 arch_task_cache_init(); 801 802 set_max_threads(MAX_THREADS); 803 804 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 805 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 806 init_task.signal->rlim[RLIMIT_SIGPENDING] = 807 init_task.signal->rlim[RLIMIT_NPROC]; 808 809 for (i = 0; i < UCOUNT_COUNTS; i++) { 810 init_user_ns.ucount_max[i] = max_threads/2; 811 } 812 813 #ifdef CONFIG_VMAP_STACK 814 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache", 815 NULL, free_vm_stack_cache); 816 #endif 817 818 lockdep_init_task(&init_task); 819 } 820 821 int __weak arch_dup_task_struct(struct task_struct *dst, 822 struct task_struct *src) 823 { 824 *dst = *src; 825 return 0; 826 } 827 828 void set_task_stack_end_magic(struct task_struct *tsk) 829 { 830 unsigned long *stackend; 831 832 stackend = end_of_stack(tsk); 833 *stackend = STACK_END_MAGIC; /* for overflow detection */ 834 } 835 836 static struct task_struct *dup_task_struct(struct task_struct *orig, int node) 837 { 838 struct task_struct *tsk; 839 unsigned long *stack; 840 struct vm_struct *stack_vm_area __maybe_unused; 841 int err; 842 843 if (node == NUMA_NO_NODE) 844 node = tsk_fork_get_node(orig); 845 tsk = alloc_task_struct_node(node); 846 if (!tsk) 847 return NULL; 848 849 stack = alloc_thread_stack_node(tsk, node); 850 if (!stack) 851 goto free_tsk; 852 853 if (memcg_charge_kernel_stack(tsk)) 854 goto free_stack; 855 856 stack_vm_area = task_stack_vm_area(tsk); 857 858 err = arch_dup_task_struct(tsk, orig); 859 860 /* 861 * arch_dup_task_struct() clobbers the stack-related fields. Make 862 * sure they're properly initialized before using any stack-related 863 * functions again. 864 */ 865 tsk->stack = stack; 866 #ifdef CONFIG_VMAP_STACK 867 tsk->stack_vm_area = stack_vm_area; 868 #endif 869 #ifdef CONFIG_THREAD_INFO_IN_TASK 870 refcount_set(&tsk->stack_refcount, 1); 871 #endif 872 873 if (err) 874 goto free_stack; 875 876 #ifdef CONFIG_SECCOMP 877 /* 878 * We must handle setting up seccomp filters once we're under 879 * the sighand lock in case orig has changed between now and 880 * then. Until then, filter must be NULL to avoid messing up 881 * the usage counts on the error path calling free_task. 882 */ 883 tsk->seccomp.filter = NULL; 884 #endif 885 886 setup_thread_stack(tsk, orig); 887 clear_user_return_notifier(tsk); 888 clear_tsk_need_resched(tsk); 889 set_task_stack_end_magic(tsk); 890 891 #ifdef CONFIG_STACKPROTECTOR 892 tsk->stack_canary = get_random_canary(); 893 #endif 894 895 /* 896 * One for us, one for whoever does the "release_task()" (usually 897 * parent) 898 */ 899 refcount_set(&tsk->usage, 2); 900 #ifdef CONFIG_BLK_DEV_IO_TRACE 901 tsk->btrace_seq = 0; 902 #endif 903 tsk->splice_pipe = NULL; 904 tsk->task_frag.page = NULL; 905 tsk->wake_q.next = NULL; 906 907 account_kernel_stack(tsk, 1); 908 909 kcov_task_init(tsk); 910 911 #ifdef CONFIG_FAULT_INJECTION 912 tsk->fail_nth = 0; 913 #endif 914 915 #ifdef CONFIG_BLK_CGROUP 916 tsk->throttle_queue = NULL; 917 tsk->use_memdelay = 0; 918 #endif 919 920 #ifdef CONFIG_MEMCG 921 tsk->active_memcg = NULL; 922 #endif 923 return tsk; 924 925 free_stack: 926 free_thread_stack(tsk); 927 free_tsk: 928 free_task_struct(tsk); 929 return NULL; 930 } 931 932 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); 933 934 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT; 935 936 static int __init coredump_filter_setup(char *s) 937 { 938 default_dump_filter = 939 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) & 940 MMF_DUMP_FILTER_MASK; 941 return 1; 942 } 943 944 __setup("coredump_filter=", coredump_filter_setup); 945 946 #include <linux/init_task.h> 947 948 static void mm_init_aio(struct mm_struct *mm) 949 { 950 #ifdef CONFIG_AIO 951 spin_lock_init(&mm->ioctx_lock); 952 mm->ioctx_table = NULL; 953 #endif 954 } 955 956 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p) 957 { 958 #ifdef CONFIG_MEMCG 959 mm->owner = p; 960 #endif 961 } 962 963 static void mm_init_uprobes_state(struct mm_struct *mm) 964 { 965 #ifdef CONFIG_UPROBES 966 mm->uprobes_state.xol_area = NULL; 967 #endif 968 } 969 970 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p, 971 struct user_namespace *user_ns) 972 { 973 mm->mmap = NULL; 974 mm->mm_rb = RB_ROOT; 975 mm->vmacache_seqnum = 0; 976 atomic_set(&mm->mm_users, 1); 977 atomic_set(&mm->mm_count, 1); 978 init_rwsem(&mm->mmap_sem); 979 INIT_LIST_HEAD(&mm->mmlist); 980 mm->core_state = NULL; 981 mm_pgtables_bytes_init(mm); 982 mm->map_count = 0; 983 mm->locked_vm = 0; 984 mm->pinned_vm = 0; 985 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat)); 986 spin_lock_init(&mm->page_table_lock); 987 spin_lock_init(&mm->arg_lock); 988 mm_init_cpumask(mm); 989 mm_init_aio(mm); 990 mm_init_owner(mm, p); 991 RCU_INIT_POINTER(mm->exe_file, NULL); 992 mmu_notifier_mm_init(mm); 993 hmm_mm_init(mm); 994 init_tlb_flush_pending(mm); 995 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 996 mm->pmd_huge_pte = NULL; 997 #endif 998 mm_init_uprobes_state(mm); 999 1000 if (current->mm) { 1001 mm->flags = current->mm->flags & MMF_INIT_MASK; 1002 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK; 1003 } else { 1004 mm->flags = default_dump_filter; 1005 mm->def_flags = 0; 1006 } 1007 1008 if (mm_alloc_pgd(mm)) 1009 goto fail_nopgd; 1010 1011 if (init_new_context(p, mm)) 1012 goto fail_nocontext; 1013 1014 mm->user_ns = get_user_ns(user_ns); 1015 return mm; 1016 1017 fail_nocontext: 1018 mm_free_pgd(mm); 1019 fail_nopgd: 1020 free_mm(mm); 1021 return NULL; 1022 } 1023 1024 /* 1025 * Allocate and initialize an mm_struct. 1026 */ 1027 struct mm_struct *mm_alloc(void) 1028 { 1029 struct mm_struct *mm; 1030 1031 mm = allocate_mm(); 1032 if (!mm) 1033 return NULL; 1034 1035 memset(mm, 0, sizeof(*mm)); 1036 return mm_init(mm, current, current_user_ns()); 1037 } 1038 1039 static inline void __mmput(struct mm_struct *mm) 1040 { 1041 VM_BUG_ON(atomic_read(&mm->mm_users)); 1042 1043 uprobe_clear_state(mm); 1044 exit_aio(mm); 1045 ksm_exit(mm); 1046 khugepaged_exit(mm); /* must run before exit_mmap */ 1047 exit_mmap(mm); 1048 mm_put_huge_zero_page(mm); 1049 set_mm_exe_file(mm, NULL); 1050 if (!list_empty(&mm->mmlist)) { 1051 spin_lock(&mmlist_lock); 1052 list_del(&mm->mmlist); 1053 spin_unlock(&mmlist_lock); 1054 } 1055 if (mm->binfmt) 1056 module_put(mm->binfmt->module); 1057 mmdrop(mm); 1058 } 1059 1060 /* 1061 * Decrement the use count and release all resources for an mm. 1062 */ 1063 void mmput(struct mm_struct *mm) 1064 { 1065 might_sleep(); 1066 1067 if (atomic_dec_and_test(&mm->mm_users)) 1068 __mmput(mm); 1069 } 1070 EXPORT_SYMBOL_GPL(mmput); 1071 1072 #ifdef CONFIG_MMU 1073 static void mmput_async_fn(struct work_struct *work) 1074 { 1075 struct mm_struct *mm = container_of(work, struct mm_struct, 1076 async_put_work); 1077 1078 __mmput(mm); 1079 } 1080 1081 void mmput_async(struct mm_struct *mm) 1082 { 1083 if (atomic_dec_and_test(&mm->mm_users)) { 1084 INIT_WORK(&mm->async_put_work, mmput_async_fn); 1085 schedule_work(&mm->async_put_work); 1086 } 1087 } 1088 #endif 1089 1090 /** 1091 * set_mm_exe_file - change a reference to the mm's executable file 1092 * 1093 * This changes mm's executable file (shown as symlink /proc/[pid]/exe). 1094 * 1095 * Main users are mmput() and sys_execve(). Callers prevent concurrent 1096 * invocations: in mmput() nobody alive left, in execve task is single 1097 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the 1098 * mm->exe_file, but does so without using set_mm_exe_file() in order 1099 * to do avoid the need for any locks. 1100 */ 1101 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 1102 { 1103 struct file *old_exe_file; 1104 1105 /* 1106 * It is safe to dereference the exe_file without RCU as 1107 * this function is only called if nobody else can access 1108 * this mm -- see comment above for justification. 1109 */ 1110 old_exe_file = rcu_dereference_raw(mm->exe_file); 1111 1112 if (new_exe_file) 1113 get_file(new_exe_file); 1114 rcu_assign_pointer(mm->exe_file, new_exe_file); 1115 if (old_exe_file) 1116 fput(old_exe_file); 1117 } 1118 1119 /** 1120 * get_mm_exe_file - acquire a reference to the mm's executable file 1121 * 1122 * Returns %NULL if mm has no associated executable file. 1123 * User must release file via fput(). 1124 */ 1125 struct file *get_mm_exe_file(struct mm_struct *mm) 1126 { 1127 struct file *exe_file; 1128 1129 rcu_read_lock(); 1130 exe_file = rcu_dereference(mm->exe_file); 1131 if (exe_file && !get_file_rcu(exe_file)) 1132 exe_file = NULL; 1133 rcu_read_unlock(); 1134 return exe_file; 1135 } 1136 EXPORT_SYMBOL(get_mm_exe_file); 1137 1138 /** 1139 * get_task_exe_file - acquire a reference to the task's executable file 1140 * 1141 * Returns %NULL if task's mm (if any) has no associated executable file or 1142 * this is a kernel thread with borrowed mm (see the comment above get_task_mm). 1143 * User must release file via fput(). 1144 */ 1145 struct file *get_task_exe_file(struct task_struct *task) 1146 { 1147 struct file *exe_file = NULL; 1148 struct mm_struct *mm; 1149 1150 task_lock(task); 1151 mm = task->mm; 1152 if (mm) { 1153 if (!(task->flags & PF_KTHREAD)) 1154 exe_file = get_mm_exe_file(mm); 1155 } 1156 task_unlock(task); 1157 return exe_file; 1158 } 1159 EXPORT_SYMBOL(get_task_exe_file); 1160 1161 /** 1162 * get_task_mm - acquire a reference to the task's mm 1163 * 1164 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning 1165 * this kernel workthread has transiently adopted a user mm with use_mm, 1166 * to do its AIO) is not set and if so returns a reference to it, after 1167 * bumping up the use count. User must release the mm via mmput() 1168 * after use. Typically used by /proc and ptrace. 1169 */ 1170 struct mm_struct *get_task_mm(struct task_struct *task) 1171 { 1172 struct mm_struct *mm; 1173 1174 task_lock(task); 1175 mm = task->mm; 1176 if (mm) { 1177 if (task->flags & PF_KTHREAD) 1178 mm = NULL; 1179 else 1180 mmget(mm); 1181 } 1182 task_unlock(task); 1183 return mm; 1184 } 1185 EXPORT_SYMBOL_GPL(get_task_mm); 1186 1187 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode) 1188 { 1189 struct mm_struct *mm; 1190 int err; 1191 1192 err = mutex_lock_killable(&task->signal->cred_guard_mutex); 1193 if (err) 1194 return ERR_PTR(err); 1195 1196 mm = get_task_mm(task); 1197 if (mm && mm != current->mm && 1198 !ptrace_may_access(task, mode)) { 1199 mmput(mm); 1200 mm = ERR_PTR(-EACCES); 1201 } 1202 mutex_unlock(&task->signal->cred_guard_mutex); 1203 1204 return mm; 1205 } 1206 1207 static void complete_vfork_done(struct task_struct *tsk) 1208 { 1209 struct completion *vfork; 1210 1211 task_lock(tsk); 1212 vfork = tsk->vfork_done; 1213 if (likely(vfork)) { 1214 tsk->vfork_done = NULL; 1215 complete(vfork); 1216 } 1217 task_unlock(tsk); 1218 } 1219 1220 static int wait_for_vfork_done(struct task_struct *child, 1221 struct completion *vfork) 1222 { 1223 int killed; 1224 1225 freezer_do_not_count(); 1226 killed = wait_for_completion_killable(vfork); 1227 freezer_count(); 1228 1229 if (killed) { 1230 task_lock(child); 1231 child->vfork_done = NULL; 1232 task_unlock(child); 1233 } 1234 1235 put_task_struct(child); 1236 return killed; 1237 } 1238 1239 /* Please note the differences between mmput and mm_release. 1240 * mmput is called whenever we stop holding onto a mm_struct, 1241 * error success whatever. 1242 * 1243 * mm_release is called after a mm_struct has been removed 1244 * from the current process. 1245 * 1246 * This difference is important for error handling, when we 1247 * only half set up a mm_struct for a new process and need to restore 1248 * the old one. Because we mmput the new mm_struct before 1249 * restoring the old one. . . 1250 * Eric Biederman 10 January 1998 1251 */ 1252 void mm_release(struct task_struct *tsk, struct mm_struct *mm) 1253 { 1254 /* Get rid of any futexes when releasing the mm */ 1255 #ifdef CONFIG_FUTEX 1256 if (unlikely(tsk->robust_list)) { 1257 exit_robust_list(tsk); 1258 tsk->robust_list = NULL; 1259 } 1260 #ifdef CONFIG_COMPAT 1261 if (unlikely(tsk->compat_robust_list)) { 1262 compat_exit_robust_list(tsk); 1263 tsk->compat_robust_list = NULL; 1264 } 1265 #endif 1266 if (unlikely(!list_empty(&tsk->pi_state_list))) 1267 exit_pi_state_list(tsk); 1268 #endif 1269 1270 uprobe_free_utask(tsk); 1271 1272 /* Get rid of any cached register state */ 1273 deactivate_mm(tsk, mm); 1274 1275 /* 1276 * Signal userspace if we're not exiting with a core dump 1277 * because we want to leave the value intact for debugging 1278 * purposes. 1279 */ 1280 if (tsk->clear_child_tid) { 1281 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) && 1282 atomic_read(&mm->mm_users) > 1) { 1283 /* 1284 * We don't check the error code - if userspace has 1285 * not set up a proper pointer then tough luck. 1286 */ 1287 put_user(0, tsk->clear_child_tid); 1288 do_futex(tsk->clear_child_tid, FUTEX_WAKE, 1289 1, NULL, NULL, 0, 0); 1290 } 1291 tsk->clear_child_tid = NULL; 1292 } 1293 1294 /* 1295 * All done, finally we can wake up parent and return this mm to him. 1296 * Also kthread_stop() uses this completion for synchronization. 1297 */ 1298 if (tsk->vfork_done) 1299 complete_vfork_done(tsk); 1300 } 1301 1302 /* 1303 * Allocate a new mm structure and copy contents from the 1304 * mm structure of the passed in task structure. 1305 */ 1306 static struct mm_struct *dup_mm(struct task_struct *tsk) 1307 { 1308 struct mm_struct *mm, *oldmm = current->mm; 1309 int err; 1310 1311 mm = allocate_mm(); 1312 if (!mm) 1313 goto fail_nomem; 1314 1315 memcpy(mm, oldmm, sizeof(*mm)); 1316 1317 if (!mm_init(mm, tsk, mm->user_ns)) 1318 goto fail_nomem; 1319 1320 err = dup_mmap(mm, oldmm); 1321 if (err) 1322 goto free_pt; 1323 1324 mm->hiwater_rss = get_mm_rss(mm); 1325 mm->hiwater_vm = mm->total_vm; 1326 1327 if (mm->binfmt && !try_module_get(mm->binfmt->module)) 1328 goto free_pt; 1329 1330 return mm; 1331 1332 free_pt: 1333 /* don't put binfmt in mmput, we haven't got module yet */ 1334 mm->binfmt = NULL; 1335 mmput(mm); 1336 1337 fail_nomem: 1338 return NULL; 1339 } 1340 1341 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk) 1342 { 1343 struct mm_struct *mm, *oldmm; 1344 int retval; 1345 1346 tsk->min_flt = tsk->maj_flt = 0; 1347 tsk->nvcsw = tsk->nivcsw = 0; 1348 #ifdef CONFIG_DETECT_HUNG_TASK 1349 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; 1350 tsk->last_switch_time = 0; 1351 #endif 1352 1353 tsk->mm = NULL; 1354 tsk->active_mm = NULL; 1355 1356 /* 1357 * Are we cloning a kernel thread? 1358 * 1359 * We need to steal a active VM for that.. 1360 */ 1361 oldmm = current->mm; 1362 if (!oldmm) 1363 return 0; 1364 1365 /* initialize the new vmacache entries */ 1366 vmacache_flush(tsk); 1367 1368 if (clone_flags & CLONE_VM) { 1369 mmget(oldmm); 1370 mm = oldmm; 1371 goto good_mm; 1372 } 1373 1374 retval = -ENOMEM; 1375 mm = dup_mm(tsk); 1376 if (!mm) 1377 goto fail_nomem; 1378 1379 good_mm: 1380 tsk->mm = mm; 1381 tsk->active_mm = mm; 1382 return 0; 1383 1384 fail_nomem: 1385 return retval; 1386 } 1387 1388 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 1389 { 1390 struct fs_struct *fs = current->fs; 1391 if (clone_flags & CLONE_FS) { 1392 /* tsk->fs is already what we want */ 1393 spin_lock(&fs->lock); 1394 if (fs->in_exec) { 1395 spin_unlock(&fs->lock); 1396 return -EAGAIN; 1397 } 1398 fs->users++; 1399 spin_unlock(&fs->lock); 1400 return 0; 1401 } 1402 tsk->fs = copy_fs_struct(fs); 1403 if (!tsk->fs) 1404 return -ENOMEM; 1405 return 0; 1406 } 1407 1408 static int copy_files(unsigned long clone_flags, struct task_struct *tsk) 1409 { 1410 struct files_struct *oldf, *newf; 1411 int error = 0; 1412 1413 /* 1414 * A background process may not have any files ... 1415 */ 1416 oldf = current->files; 1417 if (!oldf) 1418 goto out; 1419 1420 if (clone_flags & CLONE_FILES) { 1421 atomic_inc(&oldf->count); 1422 goto out; 1423 } 1424 1425 newf = dup_fd(oldf, &error); 1426 if (!newf) 1427 goto out; 1428 1429 tsk->files = newf; 1430 error = 0; 1431 out: 1432 return error; 1433 } 1434 1435 static int copy_io(unsigned long clone_flags, struct task_struct *tsk) 1436 { 1437 #ifdef CONFIG_BLOCK 1438 struct io_context *ioc = current->io_context; 1439 struct io_context *new_ioc; 1440 1441 if (!ioc) 1442 return 0; 1443 /* 1444 * Share io context with parent, if CLONE_IO is set 1445 */ 1446 if (clone_flags & CLONE_IO) { 1447 ioc_task_link(ioc); 1448 tsk->io_context = ioc; 1449 } else if (ioprio_valid(ioc->ioprio)) { 1450 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE); 1451 if (unlikely(!new_ioc)) 1452 return -ENOMEM; 1453 1454 new_ioc->ioprio = ioc->ioprio; 1455 put_io_context(new_ioc); 1456 } 1457 #endif 1458 return 0; 1459 } 1460 1461 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk) 1462 { 1463 struct sighand_struct *sig; 1464 1465 if (clone_flags & CLONE_SIGHAND) { 1466 refcount_inc(¤t->sighand->count); 1467 return 0; 1468 } 1469 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 1470 rcu_assign_pointer(tsk->sighand, sig); 1471 if (!sig) 1472 return -ENOMEM; 1473 1474 refcount_set(&sig->count, 1); 1475 spin_lock_irq(¤t->sighand->siglock); 1476 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 1477 spin_unlock_irq(¤t->sighand->siglock); 1478 return 0; 1479 } 1480 1481 void __cleanup_sighand(struct sighand_struct *sighand) 1482 { 1483 if (refcount_dec_and_test(&sighand->count)) { 1484 signalfd_cleanup(sighand); 1485 /* 1486 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it 1487 * without an RCU grace period, see __lock_task_sighand(). 1488 */ 1489 kmem_cache_free(sighand_cachep, sighand); 1490 } 1491 } 1492 1493 #ifdef CONFIG_POSIX_TIMERS 1494 /* 1495 * Initialize POSIX timer handling for a thread group. 1496 */ 1497 static void posix_cpu_timers_init_group(struct signal_struct *sig) 1498 { 1499 unsigned long cpu_limit; 1500 1501 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); 1502 if (cpu_limit != RLIM_INFINITY) { 1503 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC; 1504 sig->cputimer.running = true; 1505 } 1506 1507 /* The timer lists. */ 1508 INIT_LIST_HEAD(&sig->cpu_timers[0]); 1509 INIT_LIST_HEAD(&sig->cpu_timers[1]); 1510 INIT_LIST_HEAD(&sig->cpu_timers[2]); 1511 } 1512 #else 1513 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { } 1514 #endif 1515 1516 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk) 1517 { 1518 struct signal_struct *sig; 1519 1520 if (clone_flags & CLONE_THREAD) 1521 return 0; 1522 1523 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL); 1524 tsk->signal = sig; 1525 if (!sig) 1526 return -ENOMEM; 1527 1528 sig->nr_threads = 1; 1529 atomic_set(&sig->live, 1); 1530 refcount_set(&sig->sigcnt, 1); 1531 1532 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */ 1533 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node); 1534 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head); 1535 1536 init_waitqueue_head(&sig->wait_chldexit); 1537 sig->curr_target = tsk; 1538 init_sigpending(&sig->shared_pending); 1539 INIT_HLIST_HEAD(&sig->multiprocess); 1540 seqlock_init(&sig->stats_lock); 1541 prev_cputime_init(&sig->prev_cputime); 1542 1543 #ifdef CONFIG_POSIX_TIMERS 1544 INIT_LIST_HEAD(&sig->posix_timers); 1545 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1546 sig->real_timer.function = it_real_fn; 1547 #endif 1548 1549 task_lock(current->group_leader); 1550 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 1551 task_unlock(current->group_leader); 1552 1553 posix_cpu_timers_init_group(sig); 1554 1555 tty_audit_fork(sig); 1556 sched_autogroup_fork(sig); 1557 1558 sig->oom_score_adj = current->signal->oom_score_adj; 1559 sig->oom_score_adj_min = current->signal->oom_score_adj_min; 1560 1561 mutex_init(&sig->cred_guard_mutex); 1562 1563 return 0; 1564 } 1565 1566 static void copy_seccomp(struct task_struct *p) 1567 { 1568 #ifdef CONFIG_SECCOMP 1569 /* 1570 * Must be called with sighand->lock held, which is common to 1571 * all threads in the group. Holding cred_guard_mutex is not 1572 * needed because this new task is not yet running and cannot 1573 * be racing exec. 1574 */ 1575 assert_spin_locked(¤t->sighand->siglock); 1576 1577 /* Ref-count the new filter user, and assign it. */ 1578 get_seccomp_filter(current); 1579 p->seccomp = current->seccomp; 1580 1581 /* 1582 * Explicitly enable no_new_privs here in case it got set 1583 * between the task_struct being duplicated and holding the 1584 * sighand lock. The seccomp state and nnp must be in sync. 1585 */ 1586 if (task_no_new_privs(current)) 1587 task_set_no_new_privs(p); 1588 1589 /* 1590 * If the parent gained a seccomp mode after copying thread 1591 * flags and between before we held the sighand lock, we have 1592 * to manually enable the seccomp thread flag here. 1593 */ 1594 if (p->seccomp.mode != SECCOMP_MODE_DISABLED) 1595 set_tsk_thread_flag(p, TIF_SECCOMP); 1596 #endif 1597 } 1598 1599 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr) 1600 { 1601 current->clear_child_tid = tidptr; 1602 1603 return task_pid_vnr(current); 1604 } 1605 1606 static void rt_mutex_init_task(struct task_struct *p) 1607 { 1608 raw_spin_lock_init(&p->pi_lock); 1609 #ifdef CONFIG_RT_MUTEXES 1610 p->pi_waiters = RB_ROOT_CACHED; 1611 p->pi_top_task = NULL; 1612 p->pi_blocked_on = NULL; 1613 #endif 1614 } 1615 1616 #ifdef CONFIG_POSIX_TIMERS 1617 /* 1618 * Initialize POSIX timer handling for a single task. 1619 */ 1620 static void posix_cpu_timers_init(struct task_struct *tsk) 1621 { 1622 tsk->cputime_expires.prof_exp = 0; 1623 tsk->cputime_expires.virt_exp = 0; 1624 tsk->cputime_expires.sched_exp = 0; 1625 INIT_LIST_HEAD(&tsk->cpu_timers[0]); 1626 INIT_LIST_HEAD(&tsk->cpu_timers[1]); 1627 INIT_LIST_HEAD(&tsk->cpu_timers[2]); 1628 } 1629 #else 1630 static inline void posix_cpu_timers_init(struct task_struct *tsk) { } 1631 #endif 1632 1633 static inline void init_task_pid_links(struct task_struct *task) 1634 { 1635 enum pid_type type; 1636 1637 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 1638 INIT_HLIST_NODE(&task->pid_links[type]); 1639 } 1640 } 1641 1642 static inline void 1643 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid) 1644 { 1645 if (type == PIDTYPE_PID) 1646 task->thread_pid = pid; 1647 else 1648 task->signal->pids[type] = pid; 1649 } 1650 1651 static inline void rcu_copy_process(struct task_struct *p) 1652 { 1653 #ifdef CONFIG_PREEMPT_RCU 1654 p->rcu_read_lock_nesting = 0; 1655 p->rcu_read_unlock_special.s = 0; 1656 p->rcu_blocked_node = NULL; 1657 INIT_LIST_HEAD(&p->rcu_node_entry); 1658 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 1659 #ifdef CONFIG_TASKS_RCU 1660 p->rcu_tasks_holdout = false; 1661 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list); 1662 p->rcu_tasks_idle_cpu = -1; 1663 #endif /* #ifdef CONFIG_TASKS_RCU */ 1664 } 1665 1666 /* 1667 * This creates a new process as a copy of the old one, 1668 * but does not actually start it yet. 1669 * 1670 * It copies the registers, and all the appropriate 1671 * parts of the process environment (as per the clone 1672 * flags). The actual kick-off is left to the caller. 1673 */ 1674 static __latent_entropy struct task_struct *copy_process( 1675 unsigned long clone_flags, 1676 unsigned long stack_start, 1677 unsigned long stack_size, 1678 int __user *child_tidptr, 1679 struct pid *pid, 1680 int trace, 1681 unsigned long tls, 1682 int node) 1683 { 1684 int retval; 1685 struct task_struct *p; 1686 struct multiprocess_signals delayed; 1687 1688 /* 1689 * Don't allow sharing the root directory with processes in a different 1690 * namespace 1691 */ 1692 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 1693 return ERR_PTR(-EINVAL); 1694 1695 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS)) 1696 return ERR_PTR(-EINVAL); 1697 1698 /* 1699 * Thread groups must share signals as well, and detached threads 1700 * can only be started up within the thread group. 1701 */ 1702 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 1703 return ERR_PTR(-EINVAL); 1704 1705 /* 1706 * Shared signal handlers imply shared VM. By way of the above, 1707 * thread groups also imply shared VM. Blocking this case allows 1708 * for various simplifications in other code. 1709 */ 1710 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 1711 return ERR_PTR(-EINVAL); 1712 1713 /* 1714 * Siblings of global init remain as zombies on exit since they are 1715 * not reaped by their parent (swapper). To solve this and to avoid 1716 * multi-rooted process trees, prevent global and container-inits 1717 * from creating siblings. 1718 */ 1719 if ((clone_flags & CLONE_PARENT) && 1720 current->signal->flags & SIGNAL_UNKILLABLE) 1721 return ERR_PTR(-EINVAL); 1722 1723 /* 1724 * If the new process will be in a different pid or user namespace 1725 * do not allow it to share a thread group with the forking task. 1726 */ 1727 if (clone_flags & CLONE_THREAD) { 1728 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) || 1729 (task_active_pid_ns(current) != 1730 current->nsproxy->pid_ns_for_children)) 1731 return ERR_PTR(-EINVAL); 1732 } 1733 1734 /* 1735 * Force any signals received before this point to be delivered 1736 * before the fork happens. Collect up signals sent to multiple 1737 * processes that happen during the fork and delay them so that 1738 * they appear to happen after the fork. 1739 */ 1740 sigemptyset(&delayed.signal); 1741 INIT_HLIST_NODE(&delayed.node); 1742 1743 spin_lock_irq(¤t->sighand->siglock); 1744 if (!(clone_flags & CLONE_THREAD)) 1745 hlist_add_head(&delayed.node, ¤t->signal->multiprocess); 1746 recalc_sigpending(); 1747 spin_unlock_irq(¤t->sighand->siglock); 1748 retval = -ERESTARTNOINTR; 1749 if (signal_pending(current)) 1750 goto fork_out; 1751 1752 retval = -ENOMEM; 1753 p = dup_task_struct(current, node); 1754 if (!p) 1755 goto fork_out; 1756 1757 /* 1758 * This _must_ happen before we call free_task(), i.e. before we jump 1759 * to any of the bad_fork_* labels. This is to avoid freeing 1760 * p->set_child_tid which is (ab)used as a kthread's data pointer for 1761 * kernel threads (PF_KTHREAD). 1762 */ 1763 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; 1764 /* 1765 * Clear TID on mm_release()? 1766 */ 1767 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL; 1768 1769 ftrace_graph_init_task(p); 1770 1771 rt_mutex_init_task(p); 1772 1773 #ifdef CONFIG_PROVE_LOCKING 1774 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); 1775 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 1776 #endif 1777 retval = -EAGAIN; 1778 if (atomic_read(&p->real_cred->user->processes) >= 1779 task_rlimit(p, RLIMIT_NPROC)) { 1780 if (p->real_cred->user != INIT_USER && 1781 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) 1782 goto bad_fork_free; 1783 } 1784 current->flags &= ~PF_NPROC_EXCEEDED; 1785 1786 retval = copy_creds(p, clone_flags); 1787 if (retval < 0) 1788 goto bad_fork_free; 1789 1790 /* 1791 * If multiple threads are within copy_process(), then this check 1792 * triggers too late. This doesn't hurt, the check is only there 1793 * to stop root fork bombs. 1794 */ 1795 retval = -EAGAIN; 1796 if (nr_threads >= max_threads) 1797 goto bad_fork_cleanup_count; 1798 1799 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */ 1800 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE); 1801 p->flags |= PF_FORKNOEXEC; 1802 INIT_LIST_HEAD(&p->children); 1803 INIT_LIST_HEAD(&p->sibling); 1804 rcu_copy_process(p); 1805 p->vfork_done = NULL; 1806 spin_lock_init(&p->alloc_lock); 1807 1808 init_sigpending(&p->pending); 1809 1810 p->utime = p->stime = p->gtime = 0; 1811 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 1812 p->utimescaled = p->stimescaled = 0; 1813 #endif 1814 prev_cputime_init(&p->prev_cputime); 1815 1816 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN 1817 seqcount_init(&p->vtime.seqcount); 1818 p->vtime.starttime = 0; 1819 p->vtime.state = VTIME_INACTIVE; 1820 #endif 1821 1822 #if defined(SPLIT_RSS_COUNTING) 1823 memset(&p->rss_stat, 0, sizeof(p->rss_stat)); 1824 #endif 1825 1826 p->default_timer_slack_ns = current->timer_slack_ns; 1827 1828 #ifdef CONFIG_PSI 1829 p->psi_flags = 0; 1830 #endif 1831 1832 task_io_accounting_init(&p->ioac); 1833 acct_clear_integrals(p); 1834 1835 posix_cpu_timers_init(p); 1836 1837 p->io_context = NULL; 1838 audit_set_context(p, NULL); 1839 cgroup_fork(p); 1840 #ifdef CONFIG_NUMA 1841 p->mempolicy = mpol_dup(p->mempolicy); 1842 if (IS_ERR(p->mempolicy)) { 1843 retval = PTR_ERR(p->mempolicy); 1844 p->mempolicy = NULL; 1845 goto bad_fork_cleanup_threadgroup_lock; 1846 } 1847 #endif 1848 #ifdef CONFIG_CPUSETS 1849 p->cpuset_mem_spread_rotor = NUMA_NO_NODE; 1850 p->cpuset_slab_spread_rotor = NUMA_NO_NODE; 1851 seqcount_init(&p->mems_allowed_seq); 1852 #endif 1853 #ifdef CONFIG_TRACE_IRQFLAGS 1854 p->irq_events = 0; 1855 p->hardirqs_enabled = 0; 1856 p->hardirq_enable_ip = 0; 1857 p->hardirq_enable_event = 0; 1858 p->hardirq_disable_ip = _THIS_IP_; 1859 p->hardirq_disable_event = 0; 1860 p->softirqs_enabled = 1; 1861 p->softirq_enable_ip = _THIS_IP_; 1862 p->softirq_enable_event = 0; 1863 p->softirq_disable_ip = 0; 1864 p->softirq_disable_event = 0; 1865 p->hardirq_context = 0; 1866 p->softirq_context = 0; 1867 #endif 1868 1869 p->pagefault_disabled = 0; 1870 1871 #ifdef CONFIG_LOCKDEP 1872 p->lockdep_depth = 0; /* no locks held yet */ 1873 p->curr_chain_key = 0; 1874 p->lockdep_recursion = 0; 1875 lockdep_init_task(p); 1876 #endif 1877 1878 #ifdef CONFIG_DEBUG_MUTEXES 1879 p->blocked_on = NULL; /* not blocked yet */ 1880 #endif 1881 #ifdef CONFIG_BCACHE 1882 p->sequential_io = 0; 1883 p->sequential_io_avg = 0; 1884 #endif 1885 1886 /* Perform scheduler related setup. Assign this task to a CPU. */ 1887 retval = sched_fork(clone_flags, p); 1888 if (retval) 1889 goto bad_fork_cleanup_policy; 1890 1891 retval = perf_event_init_task(p); 1892 if (retval) 1893 goto bad_fork_cleanup_policy; 1894 retval = audit_alloc(p); 1895 if (retval) 1896 goto bad_fork_cleanup_perf; 1897 /* copy all the process information */ 1898 shm_init_task(p); 1899 retval = security_task_alloc(p, clone_flags); 1900 if (retval) 1901 goto bad_fork_cleanup_audit; 1902 retval = copy_semundo(clone_flags, p); 1903 if (retval) 1904 goto bad_fork_cleanup_security; 1905 retval = copy_files(clone_flags, p); 1906 if (retval) 1907 goto bad_fork_cleanup_semundo; 1908 retval = copy_fs(clone_flags, p); 1909 if (retval) 1910 goto bad_fork_cleanup_files; 1911 retval = copy_sighand(clone_flags, p); 1912 if (retval) 1913 goto bad_fork_cleanup_fs; 1914 retval = copy_signal(clone_flags, p); 1915 if (retval) 1916 goto bad_fork_cleanup_sighand; 1917 retval = copy_mm(clone_flags, p); 1918 if (retval) 1919 goto bad_fork_cleanup_signal; 1920 retval = copy_namespaces(clone_flags, p); 1921 if (retval) 1922 goto bad_fork_cleanup_mm; 1923 retval = copy_io(clone_flags, p); 1924 if (retval) 1925 goto bad_fork_cleanup_namespaces; 1926 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls); 1927 if (retval) 1928 goto bad_fork_cleanup_io; 1929 1930 stackleak_task_init(p); 1931 1932 if (pid != &init_struct_pid) { 1933 pid = alloc_pid(p->nsproxy->pid_ns_for_children); 1934 if (IS_ERR(pid)) { 1935 retval = PTR_ERR(pid); 1936 goto bad_fork_cleanup_thread; 1937 } 1938 } 1939 1940 #ifdef CONFIG_BLOCK 1941 p->plug = NULL; 1942 #endif 1943 #ifdef CONFIG_FUTEX 1944 p->robust_list = NULL; 1945 #ifdef CONFIG_COMPAT 1946 p->compat_robust_list = NULL; 1947 #endif 1948 INIT_LIST_HEAD(&p->pi_state_list); 1949 p->pi_state_cache = NULL; 1950 #endif 1951 /* 1952 * sigaltstack should be cleared when sharing the same VM 1953 */ 1954 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM) 1955 sas_ss_reset(p); 1956 1957 /* 1958 * Syscall tracing and stepping should be turned off in the 1959 * child regardless of CLONE_PTRACE. 1960 */ 1961 user_disable_single_step(p); 1962 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); 1963 #ifdef TIF_SYSCALL_EMU 1964 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); 1965 #endif 1966 clear_all_latency_tracing(p); 1967 1968 /* ok, now we should be set up.. */ 1969 p->pid = pid_nr(pid); 1970 if (clone_flags & CLONE_THREAD) { 1971 p->exit_signal = -1; 1972 p->group_leader = current->group_leader; 1973 p->tgid = current->tgid; 1974 } else { 1975 if (clone_flags & CLONE_PARENT) 1976 p->exit_signal = current->group_leader->exit_signal; 1977 else 1978 p->exit_signal = (clone_flags & CSIGNAL); 1979 p->group_leader = p; 1980 p->tgid = p->pid; 1981 } 1982 1983 p->nr_dirtied = 0; 1984 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10); 1985 p->dirty_paused_when = 0; 1986 1987 p->pdeath_signal = 0; 1988 INIT_LIST_HEAD(&p->thread_group); 1989 p->task_works = NULL; 1990 1991 cgroup_threadgroup_change_begin(current); 1992 /* 1993 * Ensure that the cgroup subsystem policies allow the new process to be 1994 * forked. It should be noted the the new process's css_set can be changed 1995 * between here and cgroup_post_fork() if an organisation operation is in 1996 * progress. 1997 */ 1998 retval = cgroup_can_fork(p); 1999 if (retval) 2000 goto bad_fork_free_pid; 2001 2002 /* 2003 * From this point on we must avoid any synchronous user-space 2004 * communication until we take the tasklist-lock. In particular, we do 2005 * not want user-space to be able to predict the process start-time by 2006 * stalling fork(2) after we recorded the start_time but before it is 2007 * visible to the system. 2008 */ 2009 2010 p->start_time = ktime_get_ns(); 2011 p->real_start_time = ktime_get_boot_ns(); 2012 2013 /* 2014 * Make it visible to the rest of the system, but dont wake it up yet. 2015 * Need tasklist lock for parent etc handling! 2016 */ 2017 write_lock_irq(&tasklist_lock); 2018 2019 /* CLONE_PARENT re-uses the old parent */ 2020 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { 2021 p->real_parent = current->real_parent; 2022 p->parent_exec_id = current->parent_exec_id; 2023 } else { 2024 p->real_parent = current; 2025 p->parent_exec_id = current->self_exec_id; 2026 } 2027 2028 klp_copy_process(p); 2029 2030 spin_lock(¤t->sighand->siglock); 2031 2032 /* 2033 * Copy seccomp details explicitly here, in case they were changed 2034 * before holding sighand lock. 2035 */ 2036 copy_seccomp(p); 2037 2038 rseq_fork(p, clone_flags); 2039 2040 /* Don't start children in a dying pid namespace */ 2041 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) { 2042 retval = -ENOMEM; 2043 goto bad_fork_cancel_cgroup; 2044 } 2045 2046 /* Let kill terminate clone/fork in the middle */ 2047 if (fatal_signal_pending(current)) { 2048 retval = -EINTR; 2049 goto bad_fork_cancel_cgroup; 2050 } 2051 2052 2053 init_task_pid_links(p); 2054 if (likely(p->pid)) { 2055 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace); 2056 2057 init_task_pid(p, PIDTYPE_PID, pid); 2058 if (thread_group_leader(p)) { 2059 init_task_pid(p, PIDTYPE_TGID, pid); 2060 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current)); 2061 init_task_pid(p, PIDTYPE_SID, task_session(current)); 2062 2063 if (is_child_reaper(pid)) { 2064 ns_of_pid(pid)->child_reaper = p; 2065 p->signal->flags |= SIGNAL_UNKILLABLE; 2066 } 2067 p->signal->shared_pending.signal = delayed.signal; 2068 p->signal->tty = tty_kref_get(current->signal->tty); 2069 /* 2070 * Inherit has_child_subreaper flag under the same 2071 * tasklist_lock with adding child to the process tree 2072 * for propagate_has_child_subreaper optimization. 2073 */ 2074 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper || 2075 p->real_parent->signal->is_child_subreaper; 2076 list_add_tail(&p->sibling, &p->real_parent->children); 2077 list_add_tail_rcu(&p->tasks, &init_task.tasks); 2078 attach_pid(p, PIDTYPE_TGID); 2079 attach_pid(p, PIDTYPE_PGID); 2080 attach_pid(p, PIDTYPE_SID); 2081 __this_cpu_inc(process_counts); 2082 } else { 2083 current->signal->nr_threads++; 2084 atomic_inc(¤t->signal->live); 2085 refcount_inc(¤t->signal->sigcnt); 2086 task_join_group_stop(p); 2087 list_add_tail_rcu(&p->thread_group, 2088 &p->group_leader->thread_group); 2089 list_add_tail_rcu(&p->thread_node, 2090 &p->signal->thread_head); 2091 } 2092 attach_pid(p, PIDTYPE_PID); 2093 nr_threads++; 2094 } 2095 total_forks++; 2096 hlist_del_init(&delayed.node); 2097 spin_unlock(¤t->sighand->siglock); 2098 syscall_tracepoint_update(p); 2099 write_unlock_irq(&tasklist_lock); 2100 2101 proc_fork_connector(p); 2102 cgroup_post_fork(p); 2103 cgroup_threadgroup_change_end(current); 2104 perf_event_fork(p); 2105 2106 trace_task_newtask(p, clone_flags); 2107 uprobe_copy_process(p, clone_flags); 2108 2109 return p; 2110 2111 bad_fork_cancel_cgroup: 2112 spin_unlock(¤t->sighand->siglock); 2113 write_unlock_irq(&tasklist_lock); 2114 cgroup_cancel_fork(p); 2115 bad_fork_free_pid: 2116 cgroup_threadgroup_change_end(current); 2117 if (pid != &init_struct_pid) 2118 free_pid(pid); 2119 bad_fork_cleanup_thread: 2120 exit_thread(p); 2121 bad_fork_cleanup_io: 2122 if (p->io_context) 2123 exit_io_context(p); 2124 bad_fork_cleanup_namespaces: 2125 exit_task_namespaces(p); 2126 bad_fork_cleanup_mm: 2127 if (p->mm) 2128 mmput(p->mm); 2129 bad_fork_cleanup_signal: 2130 if (!(clone_flags & CLONE_THREAD)) 2131 free_signal_struct(p->signal); 2132 bad_fork_cleanup_sighand: 2133 __cleanup_sighand(p->sighand); 2134 bad_fork_cleanup_fs: 2135 exit_fs(p); /* blocking */ 2136 bad_fork_cleanup_files: 2137 exit_files(p); /* blocking */ 2138 bad_fork_cleanup_semundo: 2139 exit_sem(p); 2140 bad_fork_cleanup_security: 2141 security_task_free(p); 2142 bad_fork_cleanup_audit: 2143 audit_free(p); 2144 bad_fork_cleanup_perf: 2145 perf_event_free_task(p); 2146 bad_fork_cleanup_policy: 2147 lockdep_free_task(p); 2148 #ifdef CONFIG_NUMA 2149 mpol_put(p->mempolicy); 2150 bad_fork_cleanup_threadgroup_lock: 2151 #endif 2152 delayacct_tsk_free(p); 2153 bad_fork_cleanup_count: 2154 atomic_dec(&p->cred->user->processes); 2155 exit_creds(p); 2156 bad_fork_free: 2157 p->state = TASK_DEAD; 2158 put_task_stack(p); 2159 free_task(p); 2160 fork_out: 2161 spin_lock_irq(¤t->sighand->siglock); 2162 hlist_del_init(&delayed.node); 2163 spin_unlock_irq(¤t->sighand->siglock); 2164 return ERR_PTR(retval); 2165 } 2166 2167 static inline void init_idle_pids(struct task_struct *idle) 2168 { 2169 enum pid_type type; 2170 2171 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 2172 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */ 2173 init_task_pid(idle, type, &init_struct_pid); 2174 } 2175 } 2176 2177 struct task_struct *fork_idle(int cpu) 2178 { 2179 struct task_struct *task; 2180 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0, 2181 cpu_to_node(cpu)); 2182 if (!IS_ERR(task)) { 2183 init_idle_pids(task); 2184 init_idle(task, cpu); 2185 } 2186 2187 return task; 2188 } 2189 2190 /* 2191 * Ok, this is the main fork-routine. 2192 * 2193 * It copies the process, and if successful kick-starts 2194 * it and waits for it to finish using the VM if required. 2195 */ 2196 long _do_fork(unsigned long clone_flags, 2197 unsigned long stack_start, 2198 unsigned long stack_size, 2199 int __user *parent_tidptr, 2200 int __user *child_tidptr, 2201 unsigned long tls) 2202 { 2203 struct completion vfork; 2204 struct pid *pid; 2205 struct task_struct *p; 2206 int trace = 0; 2207 long nr; 2208 2209 /* 2210 * Determine whether and which event to report to ptracer. When 2211 * called from kernel_thread or CLONE_UNTRACED is explicitly 2212 * requested, no event is reported; otherwise, report if the event 2213 * for the type of forking is enabled. 2214 */ 2215 if (!(clone_flags & CLONE_UNTRACED)) { 2216 if (clone_flags & CLONE_VFORK) 2217 trace = PTRACE_EVENT_VFORK; 2218 else if ((clone_flags & CSIGNAL) != SIGCHLD) 2219 trace = PTRACE_EVENT_CLONE; 2220 else 2221 trace = PTRACE_EVENT_FORK; 2222 2223 if (likely(!ptrace_event_enabled(current, trace))) 2224 trace = 0; 2225 } 2226 2227 p = copy_process(clone_flags, stack_start, stack_size, 2228 child_tidptr, NULL, trace, tls, NUMA_NO_NODE); 2229 add_latent_entropy(); 2230 2231 if (IS_ERR(p)) 2232 return PTR_ERR(p); 2233 2234 /* 2235 * Do this prior waking up the new thread - the thread pointer 2236 * might get invalid after that point, if the thread exits quickly. 2237 */ 2238 trace_sched_process_fork(current, p); 2239 2240 pid = get_task_pid(p, PIDTYPE_PID); 2241 nr = pid_vnr(pid); 2242 2243 if (clone_flags & CLONE_PARENT_SETTID) 2244 put_user(nr, parent_tidptr); 2245 2246 if (clone_flags & CLONE_VFORK) { 2247 p->vfork_done = &vfork; 2248 init_completion(&vfork); 2249 get_task_struct(p); 2250 } 2251 2252 wake_up_new_task(p); 2253 2254 /* forking complete and child started to run, tell ptracer */ 2255 if (unlikely(trace)) 2256 ptrace_event_pid(trace, pid); 2257 2258 if (clone_flags & CLONE_VFORK) { 2259 if (!wait_for_vfork_done(p, &vfork)) 2260 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid); 2261 } 2262 2263 put_pid(pid); 2264 return nr; 2265 } 2266 2267 #ifndef CONFIG_HAVE_COPY_THREAD_TLS 2268 /* For compatibility with architectures that call do_fork directly rather than 2269 * using the syscall entry points below. */ 2270 long do_fork(unsigned long clone_flags, 2271 unsigned long stack_start, 2272 unsigned long stack_size, 2273 int __user *parent_tidptr, 2274 int __user *child_tidptr) 2275 { 2276 return _do_fork(clone_flags, stack_start, stack_size, 2277 parent_tidptr, child_tidptr, 0); 2278 } 2279 #endif 2280 2281 /* 2282 * Create a kernel thread. 2283 */ 2284 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags) 2285 { 2286 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn, 2287 (unsigned long)arg, NULL, NULL, 0); 2288 } 2289 2290 #ifdef __ARCH_WANT_SYS_FORK 2291 SYSCALL_DEFINE0(fork) 2292 { 2293 #ifdef CONFIG_MMU 2294 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0); 2295 #else 2296 /* can not support in nommu mode */ 2297 return -EINVAL; 2298 #endif 2299 } 2300 #endif 2301 2302 #ifdef __ARCH_WANT_SYS_VFORK 2303 SYSCALL_DEFINE0(vfork) 2304 { 2305 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0, 2306 0, NULL, NULL, 0); 2307 } 2308 #endif 2309 2310 #ifdef __ARCH_WANT_SYS_CLONE 2311 #ifdef CONFIG_CLONE_BACKWARDS 2312 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2313 int __user *, parent_tidptr, 2314 unsigned long, tls, 2315 int __user *, child_tidptr) 2316 #elif defined(CONFIG_CLONE_BACKWARDS2) 2317 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags, 2318 int __user *, parent_tidptr, 2319 int __user *, child_tidptr, 2320 unsigned long, tls) 2321 #elif defined(CONFIG_CLONE_BACKWARDS3) 2322 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp, 2323 int, stack_size, 2324 int __user *, parent_tidptr, 2325 int __user *, child_tidptr, 2326 unsigned long, tls) 2327 #else 2328 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2329 int __user *, parent_tidptr, 2330 int __user *, child_tidptr, 2331 unsigned long, tls) 2332 #endif 2333 { 2334 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls); 2335 } 2336 #endif 2337 2338 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data) 2339 { 2340 struct task_struct *leader, *parent, *child; 2341 int res; 2342 2343 read_lock(&tasklist_lock); 2344 leader = top = top->group_leader; 2345 down: 2346 for_each_thread(leader, parent) { 2347 list_for_each_entry(child, &parent->children, sibling) { 2348 res = visitor(child, data); 2349 if (res) { 2350 if (res < 0) 2351 goto out; 2352 leader = child; 2353 goto down; 2354 } 2355 up: 2356 ; 2357 } 2358 } 2359 2360 if (leader != top) { 2361 child = leader; 2362 parent = child->real_parent; 2363 leader = parent->group_leader; 2364 goto up; 2365 } 2366 out: 2367 read_unlock(&tasklist_lock); 2368 } 2369 2370 #ifndef ARCH_MIN_MMSTRUCT_ALIGN 2371 #define ARCH_MIN_MMSTRUCT_ALIGN 0 2372 #endif 2373 2374 static void sighand_ctor(void *data) 2375 { 2376 struct sighand_struct *sighand = data; 2377 2378 spin_lock_init(&sighand->siglock); 2379 init_waitqueue_head(&sighand->signalfd_wqh); 2380 } 2381 2382 void __init proc_caches_init(void) 2383 { 2384 unsigned int mm_size; 2385 2386 sighand_cachep = kmem_cache_create("sighand_cache", 2387 sizeof(struct sighand_struct), 0, 2388 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU| 2389 SLAB_ACCOUNT, sighand_ctor); 2390 signal_cachep = kmem_cache_create("signal_cache", 2391 sizeof(struct signal_struct), 0, 2392 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2393 NULL); 2394 files_cachep = kmem_cache_create("files_cache", 2395 sizeof(struct files_struct), 0, 2396 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2397 NULL); 2398 fs_cachep = kmem_cache_create("fs_cache", 2399 sizeof(struct fs_struct), 0, 2400 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2401 NULL); 2402 2403 /* 2404 * The mm_cpumask is located at the end of mm_struct, and is 2405 * dynamically sized based on the maximum CPU number this system 2406 * can have, taking hotplug into account (nr_cpu_ids). 2407 */ 2408 mm_size = sizeof(struct mm_struct) + cpumask_size(); 2409 2410 mm_cachep = kmem_cache_create_usercopy("mm_struct", 2411 mm_size, ARCH_MIN_MMSTRUCT_ALIGN, 2412 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2413 offsetof(struct mm_struct, saved_auxv), 2414 sizeof_field(struct mm_struct, saved_auxv), 2415 NULL); 2416 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT); 2417 mmap_init(); 2418 nsproxy_cache_init(); 2419 } 2420 2421 /* 2422 * Check constraints on flags passed to the unshare system call. 2423 */ 2424 static int check_unshare_flags(unsigned long unshare_flags) 2425 { 2426 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND| 2427 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM| 2428 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET| 2429 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP)) 2430 return -EINVAL; 2431 /* 2432 * Not implemented, but pretend it works if there is nothing 2433 * to unshare. Note that unsharing the address space or the 2434 * signal handlers also need to unshare the signal queues (aka 2435 * CLONE_THREAD). 2436 */ 2437 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) { 2438 if (!thread_group_empty(current)) 2439 return -EINVAL; 2440 } 2441 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) { 2442 if (refcount_read(¤t->sighand->count) > 1) 2443 return -EINVAL; 2444 } 2445 if (unshare_flags & CLONE_VM) { 2446 if (!current_is_single_threaded()) 2447 return -EINVAL; 2448 } 2449 2450 return 0; 2451 } 2452 2453 /* 2454 * Unshare the filesystem structure if it is being shared 2455 */ 2456 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) 2457 { 2458 struct fs_struct *fs = current->fs; 2459 2460 if (!(unshare_flags & CLONE_FS) || !fs) 2461 return 0; 2462 2463 /* don't need lock here; in the worst case we'll do useless copy */ 2464 if (fs->users == 1) 2465 return 0; 2466 2467 *new_fsp = copy_fs_struct(fs); 2468 if (!*new_fsp) 2469 return -ENOMEM; 2470 2471 return 0; 2472 } 2473 2474 /* 2475 * Unshare file descriptor table if it is being shared 2476 */ 2477 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp) 2478 { 2479 struct files_struct *fd = current->files; 2480 int error = 0; 2481 2482 if ((unshare_flags & CLONE_FILES) && 2483 (fd && atomic_read(&fd->count) > 1)) { 2484 *new_fdp = dup_fd(fd, &error); 2485 if (!*new_fdp) 2486 return error; 2487 } 2488 2489 return 0; 2490 } 2491 2492 /* 2493 * unshare allows a process to 'unshare' part of the process 2494 * context which was originally shared using clone. copy_* 2495 * functions used by do_fork() cannot be used here directly 2496 * because they modify an inactive task_struct that is being 2497 * constructed. Here we are modifying the current, active, 2498 * task_struct. 2499 */ 2500 int ksys_unshare(unsigned long unshare_flags) 2501 { 2502 struct fs_struct *fs, *new_fs = NULL; 2503 struct files_struct *fd, *new_fd = NULL; 2504 struct cred *new_cred = NULL; 2505 struct nsproxy *new_nsproxy = NULL; 2506 int do_sysvsem = 0; 2507 int err; 2508 2509 /* 2510 * If unsharing a user namespace must also unshare the thread group 2511 * and unshare the filesystem root and working directories. 2512 */ 2513 if (unshare_flags & CLONE_NEWUSER) 2514 unshare_flags |= CLONE_THREAD | CLONE_FS; 2515 /* 2516 * If unsharing vm, must also unshare signal handlers. 2517 */ 2518 if (unshare_flags & CLONE_VM) 2519 unshare_flags |= CLONE_SIGHAND; 2520 /* 2521 * If unsharing a signal handlers, must also unshare the signal queues. 2522 */ 2523 if (unshare_flags & CLONE_SIGHAND) 2524 unshare_flags |= CLONE_THREAD; 2525 /* 2526 * If unsharing namespace, must also unshare filesystem information. 2527 */ 2528 if (unshare_flags & CLONE_NEWNS) 2529 unshare_flags |= CLONE_FS; 2530 2531 err = check_unshare_flags(unshare_flags); 2532 if (err) 2533 goto bad_unshare_out; 2534 /* 2535 * CLONE_NEWIPC must also detach from the undolist: after switching 2536 * to a new ipc namespace, the semaphore arrays from the old 2537 * namespace are unreachable. 2538 */ 2539 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) 2540 do_sysvsem = 1; 2541 err = unshare_fs(unshare_flags, &new_fs); 2542 if (err) 2543 goto bad_unshare_out; 2544 err = unshare_fd(unshare_flags, &new_fd); 2545 if (err) 2546 goto bad_unshare_cleanup_fs; 2547 err = unshare_userns(unshare_flags, &new_cred); 2548 if (err) 2549 goto bad_unshare_cleanup_fd; 2550 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, 2551 new_cred, new_fs); 2552 if (err) 2553 goto bad_unshare_cleanup_cred; 2554 2555 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) { 2556 if (do_sysvsem) { 2557 /* 2558 * CLONE_SYSVSEM is equivalent to sys_exit(). 2559 */ 2560 exit_sem(current); 2561 } 2562 if (unshare_flags & CLONE_NEWIPC) { 2563 /* Orphan segments in old ns (see sem above). */ 2564 exit_shm(current); 2565 shm_init_task(current); 2566 } 2567 2568 if (new_nsproxy) 2569 switch_task_namespaces(current, new_nsproxy); 2570 2571 task_lock(current); 2572 2573 if (new_fs) { 2574 fs = current->fs; 2575 spin_lock(&fs->lock); 2576 current->fs = new_fs; 2577 if (--fs->users) 2578 new_fs = NULL; 2579 else 2580 new_fs = fs; 2581 spin_unlock(&fs->lock); 2582 } 2583 2584 if (new_fd) { 2585 fd = current->files; 2586 current->files = new_fd; 2587 new_fd = fd; 2588 } 2589 2590 task_unlock(current); 2591 2592 if (new_cred) { 2593 /* Install the new user namespace */ 2594 commit_creds(new_cred); 2595 new_cred = NULL; 2596 } 2597 } 2598 2599 perf_event_namespaces(current); 2600 2601 bad_unshare_cleanup_cred: 2602 if (new_cred) 2603 put_cred(new_cred); 2604 bad_unshare_cleanup_fd: 2605 if (new_fd) 2606 put_files_struct(new_fd); 2607 2608 bad_unshare_cleanup_fs: 2609 if (new_fs) 2610 free_fs_struct(new_fs); 2611 2612 bad_unshare_out: 2613 return err; 2614 } 2615 2616 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags) 2617 { 2618 return ksys_unshare(unshare_flags); 2619 } 2620 2621 /* 2622 * Helper to unshare the files of the current task. 2623 * We don't want to expose copy_files internals to 2624 * the exec layer of the kernel. 2625 */ 2626 2627 int unshare_files(struct files_struct **displaced) 2628 { 2629 struct task_struct *task = current; 2630 struct files_struct *copy = NULL; 2631 int error; 2632 2633 error = unshare_fd(CLONE_FILES, ©); 2634 if (error || !copy) { 2635 *displaced = NULL; 2636 return error; 2637 } 2638 *displaced = task->files; 2639 task_lock(task); 2640 task->files = copy; 2641 task_unlock(task); 2642 return 0; 2643 } 2644 2645 int sysctl_max_threads(struct ctl_table *table, int write, 2646 void __user *buffer, size_t *lenp, loff_t *ppos) 2647 { 2648 struct ctl_table t; 2649 int ret; 2650 int threads = max_threads; 2651 int min = MIN_THREADS; 2652 int max = MAX_THREADS; 2653 2654 t = *table; 2655 t.data = &threads; 2656 t.extra1 = &min; 2657 t.extra2 = &max; 2658 2659 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); 2660 if (ret || !write) 2661 return ret; 2662 2663 set_max_threads(threads); 2664 2665 return 0; 2666 } 2667