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/perf_event.h> 81 #include <linux/posix-timers.h> 82 #include <linux/user-return-notifier.h> 83 #include <linux/oom.h> 84 #include <linux/khugepaged.h> 85 #include <linux/signalfd.h> 86 #include <linux/uprobes.h> 87 #include <linux/aio.h> 88 #include <linux/compiler.h> 89 #include <linux/sysctl.h> 90 #include <linux/kcov.h> 91 #include <linux/livepatch.h> 92 #include <linux/thread_info.h> 93 #include <linux/stackleak.h> 94 95 #include <asm/pgtable.h> 96 #include <asm/pgalloc.h> 97 #include <linux/uaccess.h> 98 #include <asm/mmu_context.h> 99 #include <asm/cacheflush.h> 100 #include <asm/tlbflush.h> 101 102 #include <trace/events/sched.h> 103 104 #define CREATE_TRACE_POINTS 105 #include <trace/events/task.h> 106 107 /* 108 * Minimum number of threads to boot the kernel 109 */ 110 #define MIN_THREADS 20 111 112 /* 113 * Maximum number of threads 114 */ 115 #define MAX_THREADS FUTEX_TID_MASK 116 117 /* 118 * Protected counters by write_lock_irq(&tasklist_lock) 119 */ 120 unsigned long total_forks; /* Handle normal Linux uptimes. */ 121 int nr_threads; /* The idle threads do not count.. */ 122 123 int max_threads; /* tunable limit on nr_threads */ 124 125 DEFINE_PER_CPU(unsigned long, process_counts) = 0; 126 127 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */ 128 129 #ifdef CONFIG_PROVE_RCU 130 int lockdep_tasklist_lock_is_held(void) 131 { 132 return lockdep_is_held(&tasklist_lock); 133 } 134 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held); 135 #endif /* #ifdef CONFIG_PROVE_RCU */ 136 137 int nr_processes(void) 138 { 139 int cpu; 140 int total = 0; 141 142 for_each_possible_cpu(cpu) 143 total += per_cpu(process_counts, cpu); 144 145 return total; 146 } 147 148 void __weak arch_release_task_struct(struct task_struct *tsk) 149 { 150 } 151 152 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 153 static struct kmem_cache *task_struct_cachep; 154 155 static inline struct task_struct *alloc_task_struct_node(int node) 156 { 157 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node); 158 } 159 160 static inline void free_task_struct(struct task_struct *tsk) 161 { 162 kmem_cache_free(task_struct_cachep, tsk); 163 } 164 #endif 165 166 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR 167 168 /* 169 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a 170 * kmemcache based allocator. 171 */ 172 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) 173 174 #ifdef CONFIG_VMAP_STACK 175 /* 176 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB 177 * flush. Try to minimize the number of calls by caching stacks. 178 */ 179 #define NR_CACHED_STACKS 2 180 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]); 181 182 static int free_vm_stack_cache(unsigned int cpu) 183 { 184 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu); 185 int i; 186 187 for (i = 0; i < NR_CACHED_STACKS; i++) { 188 struct vm_struct *vm_stack = cached_vm_stacks[i]; 189 190 if (!vm_stack) 191 continue; 192 193 vfree(vm_stack->addr); 194 cached_vm_stacks[i] = NULL; 195 } 196 197 return 0; 198 } 199 #endif 200 201 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node) 202 { 203 #ifdef CONFIG_VMAP_STACK 204 void *stack; 205 int i; 206 207 for (i = 0; i < NR_CACHED_STACKS; i++) { 208 struct vm_struct *s; 209 210 s = this_cpu_xchg(cached_stacks[i], NULL); 211 212 if (!s) 213 continue; 214 215 /* Clear stale pointers from reused stack. */ 216 memset(s->addr, 0, THREAD_SIZE); 217 218 tsk->stack_vm_area = s; 219 tsk->stack = s->addr; 220 return s->addr; 221 } 222 223 /* 224 * Allocated stacks are cached and later reused by new threads, 225 * so memcg accounting is performed manually on assigning/releasing 226 * stacks to tasks. Drop __GFP_ACCOUNT. 227 */ 228 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN, 229 VMALLOC_START, VMALLOC_END, 230 THREADINFO_GFP & ~__GFP_ACCOUNT, 231 PAGE_KERNEL, 232 0, node, __builtin_return_address(0)); 233 234 /* 235 * We can't call find_vm_area() in interrupt context, and 236 * free_thread_stack() can be called in interrupt context, 237 * so cache the vm_struct. 238 */ 239 if (stack) { 240 tsk->stack_vm_area = find_vm_area(stack); 241 tsk->stack = stack; 242 } 243 return stack; 244 #else 245 struct page *page = alloc_pages_node(node, THREADINFO_GFP, 246 THREAD_SIZE_ORDER); 247 248 return page ? page_address(page) : NULL; 249 #endif 250 } 251 252 static inline void free_thread_stack(struct task_struct *tsk) 253 { 254 #ifdef CONFIG_VMAP_STACK 255 struct vm_struct *vm = task_stack_vm_area(tsk); 256 257 if (vm) { 258 int i; 259 260 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 261 mod_memcg_page_state(vm->pages[i], 262 MEMCG_KERNEL_STACK_KB, 263 -(int)(PAGE_SIZE / 1024)); 264 265 memcg_kmem_uncharge(vm->pages[i], 0); 266 } 267 268 for (i = 0; i < NR_CACHED_STACKS; i++) { 269 if (this_cpu_cmpxchg(cached_stacks[i], 270 NULL, tsk->stack_vm_area) != NULL) 271 continue; 272 273 return; 274 } 275 276 vfree_atomic(tsk->stack); 277 return; 278 } 279 #endif 280 281 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER); 282 } 283 # else 284 static struct kmem_cache *thread_stack_cache; 285 286 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, 287 int node) 288 { 289 unsigned long *stack; 290 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node); 291 tsk->stack = stack; 292 return stack; 293 } 294 295 static void free_thread_stack(struct task_struct *tsk) 296 { 297 kmem_cache_free(thread_stack_cache, tsk->stack); 298 } 299 300 void thread_stack_cache_init(void) 301 { 302 thread_stack_cache = kmem_cache_create_usercopy("thread_stack", 303 THREAD_SIZE, THREAD_SIZE, 0, 0, 304 THREAD_SIZE, NULL); 305 BUG_ON(thread_stack_cache == NULL); 306 } 307 # endif 308 #endif 309 310 /* SLAB cache for signal_struct structures (tsk->signal) */ 311 static struct kmem_cache *signal_cachep; 312 313 /* SLAB cache for sighand_struct structures (tsk->sighand) */ 314 struct kmem_cache *sighand_cachep; 315 316 /* SLAB cache for files_struct structures (tsk->files) */ 317 struct kmem_cache *files_cachep; 318 319 /* SLAB cache for fs_struct structures (tsk->fs) */ 320 struct kmem_cache *fs_cachep; 321 322 /* SLAB cache for vm_area_struct structures */ 323 static struct kmem_cache *vm_area_cachep; 324 325 /* SLAB cache for mm_struct structures (tsk->mm) */ 326 static struct kmem_cache *mm_cachep; 327 328 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm) 329 { 330 struct vm_area_struct *vma; 331 332 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 333 if (vma) 334 vma_init(vma, mm); 335 return vma; 336 } 337 338 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig) 339 { 340 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 341 342 if (new) { 343 *new = *orig; 344 INIT_LIST_HEAD(&new->anon_vma_chain); 345 } 346 return new; 347 } 348 349 void vm_area_free(struct vm_area_struct *vma) 350 { 351 kmem_cache_free(vm_area_cachep, vma); 352 } 353 354 static void account_kernel_stack(struct task_struct *tsk, int account) 355 { 356 void *stack = task_stack_page(tsk); 357 struct vm_struct *vm = task_stack_vm_area(tsk); 358 359 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0); 360 361 if (vm) { 362 int i; 363 364 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE); 365 366 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 367 mod_zone_page_state(page_zone(vm->pages[i]), 368 NR_KERNEL_STACK_KB, 369 PAGE_SIZE / 1024 * account); 370 } 371 } else { 372 /* 373 * All stack pages are in the same zone and belong to the 374 * same memcg. 375 */ 376 struct page *first_page = virt_to_page(stack); 377 378 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB, 379 THREAD_SIZE / 1024 * account); 380 381 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB, 382 account * (THREAD_SIZE / 1024)); 383 } 384 } 385 386 static int memcg_charge_kernel_stack(struct task_struct *tsk) 387 { 388 #ifdef CONFIG_VMAP_STACK 389 struct vm_struct *vm = task_stack_vm_area(tsk); 390 int ret; 391 392 if (vm) { 393 int i; 394 395 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 396 /* 397 * If memcg_kmem_charge() fails, page->mem_cgroup 398 * pointer is NULL, and both memcg_kmem_uncharge() 399 * and mod_memcg_page_state() in free_thread_stack() 400 * will ignore this page. So it's safe. 401 */ 402 ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0); 403 if (ret) 404 return ret; 405 406 mod_memcg_page_state(vm->pages[i], 407 MEMCG_KERNEL_STACK_KB, 408 PAGE_SIZE / 1024); 409 } 410 } 411 #endif 412 return 0; 413 } 414 415 static void release_task_stack(struct task_struct *tsk) 416 { 417 if (WARN_ON(tsk->state != TASK_DEAD)) 418 return; /* Better to leak the stack than to free prematurely */ 419 420 account_kernel_stack(tsk, -1); 421 free_thread_stack(tsk); 422 tsk->stack = NULL; 423 #ifdef CONFIG_VMAP_STACK 424 tsk->stack_vm_area = NULL; 425 #endif 426 } 427 428 #ifdef CONFIG_THREAD_INFO_IN_TASK 429 void put_task_stack(struct task_struct *tsk) 430 { 431 if (refcount_dec_and_test(&tsk->stack_refcount)) 432 release_task_stack(tsk); 433 } 434 #endif 435 436 void free_task(struct task_struct *tsk) 437 { 438 #ifndef CONFIG_THREAD_INFO_IN_TASK 439 /* 440 * The task is finally done with both the stack and thread_info, 441 * so free both. 442 */ 443 release_task_stack(tsk); 444 #else 445 /* 446 * If the task had a separate stack allocation, it should be gone 447 * by now. 448 */ 449 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0); 450 #endif 451 rt_mutex_debug_task_free(tsk); 452 ftrace_graph_exit_task(tsk); 453 put_seccomp_filter(tsk); 454 arch_release_task_struct(tsk); 455 if (tsk->flags & PF_KTHREAD) 456 free_kthread_struct(tsk); 457 free_task_struct(tsk); 458 } 459 EXPORT_SYMBOL(free_task); 460 461 #ifdef CONFIG_MMU 462 static __latent_entropy int dup_mmap(struct mm_struct *mm, 463 struct mm_struct *oldmm) 464 { 465 struct vm_area_struct *mpnt, *tmp, *prev, **pprev; 466 struct rb_node **rb_link, *rb_parent; 467 int retval; 468 unsigned long charge; 469 LIST_HEAD(uf); 470 471 uprobe_start_dup_mmap(); 472 if (down_write_killable(&oldmm->mmap_sem)) { 473 retval = -EINTR; 474 goto fail_uprobe_end; 475 } 476 flush_cache_dup_mm(oldmm); 477 uprobe_dup_mmap(oldmm, mm); 478 /* 479 * Not linked in yet - no deadlock potential: 480 */ 481 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING); 482 483 /* No ordering required: file already has been exposed. */ 484 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm)); 485 486 mm->total_vm = oldmm->total_vm; 487 mm->data_vm = oldmm->data_vm; 488 mm->exec_vm = oldmm->exec_vm; 489 mm->stack_vm = oldmm->stack_vm; 490 491 rb_link = &mm->mm_rb.rb_node; 492 rb_parent = NULL; 493 pprev = &mm->mmap; 494 retval = ksm_fork(mm, oldmm); 495 if (retval) 496 goto out; 497 retval = khugepaged_fork(mm, oldmm); 498 if (retval) 499 goto out; 500 501 prev = NULL; 502 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { 503 struct file *file; 504 505 if (mpnt->vm_flags & VM_DONTCOPY) { 506 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt)); 507 continue; 508 } 509 charge = 0; 510 /* 511 * Don't duplicate many vmas if we've been oom-killed (for 512 * example) 513 */ 514 if (fatal_signal_pending(current)) { 515 retval = -EINTR; 516 goto out; 517 } 518 if (mpnt->vm_flags & VM_ACCOUNT) { 519 unsigned long len = vma_pages(mpnt); 520 521 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */ 522 goto fail_nomem; 523 charge = len; 524 } 525 tmp = vm_area_dup(mpnt); 526 if (!tmp) 527 goto fail_nomem; 528 retval = vma_dup_policy(mpnt, tmp); 529 if (retval) 530 goto fail_nomem_policy; 531 tmp->vm_mm = mm; 532 retval = dup_userfaultfd(tmp, &uf); 533 if (retval) 534 goto fail_nomem_anon_vma_fork; 535 if (tmp->vm_flags & VM_WIPEONFORK) { 536 /* VM_WIPEONFORK gets a clean slate in the child. */ 537 tmp->anon_vma = NULL; 538 if (anon_vma_prepare(tmp)) 539 goto fail_nomem_anon_vma_fork; 540 } else if (anon_vma_fork(tmp, mpnt)) 541 goto fail_nomem_anon_vma_fork; 542 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT); 543 tmp->vm_next = tmp->vm_prev = NULL; 544 file = tmp->vm_file; 545 if (file) { 546 struct inode *inode = file_inode(file); 547 struct address_space *mapping = file->f_mapping; 548 549 get_file(file); 550 if (tmp->vm_flags & VM_DENYWRITE) 551 atomic_dec(&inode->i_writecount); 552 i_mmap_lock_write(mapping); 553 if (tmp->vm_flags & VM_SHARED) 554 atomic_inc(&mapping->i_mmap_writable); 555 flush_dcache_mmap_lock(mapping); 556 /* insert tmp into the share list, just after mpnt */ 557 vma_interval_tree_insert_after(tmp, mpnt, 558 &mapping->i_mmap); 559 flush_dcache_mmap_unlock(mapping); 560 i_mmap_unlock_write(mapping); 561 } 562 563 /* 564 * Clear hugetlb-related page reserves for children. This only 565 * affects MAP_PRIVATE mappings. Faults generated by the child 566 * are not guaranteed to succeed, even if read-only 567 */ 568 if (is_vm_hugetlb_page(tmp)) 569 reset_vma_resv_huge_pages(tmp); 570 571 /* 572 * Link in the new vma and copy the page table entries. 573 */ 574 *pprev = tmp; 575 pprev = &tmp->vm_next; 576 tmp->vm_prev = prev; 577 prev = tmp; 578 579 __vma_link_rb(mm, tmp, rb_link, rb_parent); 580 rb_link = &tmp->vm_rb.rb_right; 581 rb_parent = &tmp->vm_rb; 582 583 mm->map_count++; 584 if (!(tmp->vm_flags & VM_WIPEONFORK)) 585 retval = copy_page_range(mm, oldmm, mpnt); 586 587 if (tmp->vm_ops && tmp->vm_ops->open) 588 tmp->vm_ops->open(tmp); 589 590 if (retval) 591 goto out; 592 } 593 /* a new mm has just been created */ 594 retval = arch_dup_mmap(oldmm, mm); 595 out: 596 up_write(&mm->mmap_sem); 597 flush_tlb_mm(oldmm); 598 up_write(&oldmm->mmap_sem); 599 dup_userfaultfd_complete(&uf); 600 fail_uprobe_end: 601 uprobe_end_dup_mmap(); 602 return retval; 603 fail_nomem_anon_vma_fork: 604 mpol_put(vma_policy(tmp)); 605 fail_nomem_policy: 606 vm_area_free(tmp); 607 fail_nomem: 608 retval = -ENOMEM; 609 vm_unacct_memory(charge); 610 goto out; 611 } 612 613 static inline int mm_alloc_pgd(struct mm_struct *mm) 614 { 615 mm->pgd = pgd_alloc(mm); 616 if (unlikely(!mm->pgd)) 617 return -ENOMEM; 618 return 0; 619 } 620 621 static inline void mm_free_pgd(struct mm_struct *mm) 622 { 623 pgd_free(mm, mm->pgd); 624 } 625 #else 626 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) 627 { 628 down_write(&oldmm->mmap_sem); 629 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm)); 630 up_write(&oldmm->mmap_sem); 631 return 0; 632 } 633 #define mm_alloc_pgd(mm) (0) 634 #define mm_free_pgd(mm) 635 #endif /* CONFIG_MMU */ 636 637 static void check_mm(struct mm_struct *mm) 638 { 639 int i; 640 641 for (i = 0; i < NR_MM_COUNTERS; i++) { 642 long x = atomic_long_read(&mm->rss_stat.count[i]); 643 644 if (unlikely(x)) 645 printk(KERN_ALERT "BUG: Bad rss-counter state " 646 "mm:%p idx:%d val:%ld\n", mm, i, x); 647 } 648 649 if (mm_pgtables_bytes(mm)) 650 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n", 651 mm_pgtables_bytes(mm)); 652 653 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 654 VM_BUG_ON_MM(mm->pmd_huge_pte, mm); 655 #endif 656 } 657 658 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL)) 659 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) 660 661 /* 662 * Called when the last reference to the mm 663 * is dropped: either by a lazy thread or by 664 * mmput. Free the page directory and the mm. 665 */ 666 void __mmdrop(struct mm_struct *mm) 667 { 668 BUG_ON(mm == &init_mm); 669 WARN_ON_ONCE(mm == current->mm); 670 WARN_ON_ONCE(mm == current->active_mm); 671 mm_free_pgd(mm); 672 destroy_context(mm); 673 hmm_mm_destroy(mm); 674 mmu_notifier_mm_destroy(mm); 675 check_mm(mm); 676 put_user_ns(mm->user_ns); 677 free_mm(mm); 678 } 679 EXPORT_SYMBOL_GPL(__mmdrop); 680 681 static void mmdrop_async_fn(struct work_struct *work) 682 { 683 struct mm_struct *mm; 684 685 mm = container_of(work, struct mm_struct, async_put_work); 686 __mmdrop(mm); 687 } 688 689 static void mmdrop_async(struct mm_struct *mm) 690 { 691 if (unlikely(atomic_dec_and_test(&mm->mm_count))) { 692 INIT_WORK(&mm->async_put_work, mmdrop_async_fn); 693 schedule_work(&mm->async_put_work); 694 } 695 } 696 697 static inline void free_signal_struct(struct signal_struct *sig) 698 { 699 taskstats_tgid_free(sig); 700 sched_autogroup_exit(sig); 701 /* 702 * __mmdrop is not safe to call from softirq context on x86 due to 703 * pgd_dtor so postpone it to the async context 704 */ 705 if (sig->oom_mm) 706 mmdrop_async(sig->oom_mm); 707 kmem_cache_free(signal_cachep, sig); 708 } 709 710 static inline void put_signal_struct(struct signal_struct *sig) 711 { 712 if (refcount_dec_and_test(&sig->sigcnt)) 713 free_signal_struct(sig); 714 } 715 716 void __put_task_struct(struct task_struct *tsk) 717 { 718 WARN_ON(!tsk->exit_state); 719 WARN_ON(refcount_read(&tsk->usage)); 720 WARN_ON(tsk == current); 721 722 cgroup_free(tsk); 723 task_numa_free(tsk); 724 security_task_free(tsk); 725 exit_creds(tsk); 726 delayacct_tsk_free(tsk); 727 put_signal_struct(tsk->signal); 728 729 if (!profile_handoff_task(tsk)) 730 free_task(tsk); 731 } 732 EXPORT_SYMBOL_GPL(__put_task_struct); 733 734 void __init __weak arch_task_cache_init(void) { } 735 736 /* 737 * set_max_threads 738 */ 739 static void set_max_threads(unsigned int max_threads_suggested) 740 { 741 u64 threads; 742 unsigned long nr_pages = totalram_pages(); 743 744 /* 745 * The number of threads shall be limited such that the thread 746 * structures may only consume a small part of the available memory. 747 */ 748 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64) 749 threads = MAX_THREADS; 750 else 751 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE, 752 (u64) THREAD_SIZE * 8UL); 753 754 if (threads > max_threads_suggested) 755 threads = max_threads_suggested; 756 757 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS); 758 } 759 760 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT 761 /* Initialized by the architecture: */ 762 int arch_task_struct_size __read_mostly; 763 #endif 764 765 static void task_struct_whitelist(unsigned long *offset, unsigned long *size) 766 { 767 /* Fetch thread_struct whitelist for the architecture. */ 768 arch_thread_struct_whitelist(offset, size); 769 770 /* 771 * Handle zero-sized whitelist or empty thread_struct, otherwise 772 * adjust offset to position of thread_struct in task_struct. 773 */ 774 if (unlikely(*size == 0)) 775 *offset = 0; 776 else 777 *offset += offsetof(struct task_struct, thread); 778 } 779 780 void __init fork_init(void) 781 { 782 int i; 783 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 784 #ifndef ARCH_MIN_TASKALIGN 785 #define ARCH_MIN_TASKALIGN 0 786 #endif 787 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN); 788 unsigned long useroffset, usersize; 789 790 /* create a slab on which task_structs can be allocated */ 791 task_struct_whitelist(&useroffset, &usersize); 792 task_struct_cachep = kmem_cache_create_usercopy("task_struct", 793 arch_task_struct_size, align, 794 SLAB_PANIC|SLAB_ACCOUNT, 795 useroffset, usersize, NULL); 796 #endif 797 798 /* do the arch specific task caches init */ 799 arch_task_cache_init(); 800 801 set_max_threads(MAX_THREADS); 802 803 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 804 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 805 init_task.signal->rlim[RLIMIT_SIGPENDING] = 806 init_task.signal->rlim[RLIMIT_NPROC]; 807 808 for (i = 0; i < UCOUNT_COUNTS; i++) { 809 init_user_ns.ucount_max[i] = max_threads/2; 810 } 811 812 #ifdef CONFIG_VMAP_STACK 813 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache", 814 NULL, free_vm_stack_cache); 815 #endif 816 817 lockdep_init_task(&init_task); 818 uprobes_init(); 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 atomic64_set(&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 * dup_mm() - duplicates an existing mm structure 1304 * @tsk: the task_struct with which the new mm will be associated. 1305 * @oldmm: the mm to duplicate. 1306 * 1307 * Allocates a new mm structure and duplicates the provided @oldmm structure 1308 * content into it. 1309 * 1310 * Return: the duplicated mm or NULL on failure. 1311 */ 1312 static struct mm_struct *dup_mm(struct task_struct *tsk, 1313 struct mm_struct *oldmm) 1314 { 1315 struct mm_struct *mm; 1316 int err; 1317 1318 mm = allocate_mm(); 1319 if (!mm) 1320 goto fail_nomem; 1321 1322 memcpy(mm, oldmm, sizeof(*mm)); 1323 1324 if (!mm_init(mm, tsk, mm->user_ns)) 1325 goto fail_nomem; 1326 1327 err = dup_mmap(mm, oldmm); 1328 if (err) 1329 goto free_pt; 1330 1331 mm->hiwater_rss = get_mm_rss(mm); 1332 mm->hiwater_vm = mm->total_vm; 1333 1334 if (mm->binfmt && !try_module_get(mm->binfmt->module)) 1335 goto free_pt; 1336 1337 return mm; 1338 1339 free_pt: 1340 /* don't put binfmt in mmput, we haven't got module yet */ 1341 mm->binfmt = NULL; 1342 mmput(mm); 1343 1344 fail_nomem: 1345 return NULL; 1346 } 1347 1348 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk) 1349 { 1350 struct mm_struct *mm, *oldmm; 1351 int retval; 1352 1353 tsk->min_flt = tsk->maj_flt = 0; 1354 tsk->nvcsw = tsk->nivcsw = 0; 1355 #ifdef CONFIG_DETECT_HUNG_TASK 1356 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; 1357 tsk->last_switch_time = 0; 1358 #endif 1359 1360 tsk->mm = NULL; 1361 tsk->active_mm = NULL; 1362 1363 /* 1364 * Are we cloning a kernel thread? 1365 * 1366 * We need to steal a active VM for that.. 1367 */ 1368 oldmm = current->mm; 1369 if (!oldmm) 1370 return 0; 1371 1372 /* initialize the new vmacache entries */ 1373 vmacache_flush(tsk); 1374 1375 if (clone_flags & CLONE_VM) { 1376 mmget(oldmm); 1377 mm = oldmm; 1378 goto good_mm; 1379 } 1380 1381 retval = -ENOMEM; 1382 mm = dup_mm(tsk, current->mm); 1383 if (!mm) 1384 goto fail_nomem; 1385 1386 good_mm: 1387 tsk->mm = mm; 1388 tsk->active_mm = mm; 1389 return 0; 1390 1391 fail_nomem: 1392 return retval; 1393 } 1394 1395 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 1396 { 1397 struct fs_struct *fs = current->fs; 1398 if (clone_flags & CLONE_FS) { 1399 /* tsk->fs is already what we want */ 1400 spin_lock(&fs->lock); 1401 if (fs->in_exec) { 1402 spin_unlock(&fs->lock); 1403 return -EAGAIN; 1404 } 1405 fs->users++; 1406 spin_unlock(&fs->lock); 1407 return 0; 1408 } 1409 tsk->fs = copy_fs_struct(fs); 1410 if (!tsk->fs) 1411 return -ENOMEM; 1412 return 0; 1413 } 1414 1415 static int copy_files(unsigned long clone_flags, struct task_struct *tsk) 1416 { 1417 struct files_struct *oldf, *newf; 1418 int error = 0; 1419 1420 /* 1421 * A background process may not have any files ... 1422 */ 1423 oldf = current->files; 1424 if (!oldf) 1425 goto out; 1426 1427 if (clone_flags & CLONE_FILES) { 1428 atomic_inc(&oldf->count); 1429 goto out; 1430 } 1431 1432 newf = dup_fd(oldf, &error); 1433 if (!newf) 1434 goto out; 1435 1436 tsk->files = newf; 1437 error = 0; 1438 out: 1439 return error; 1440 } 1441 1442 static int copy_io(unsigned long clone_flags, struct task_struct *tsk) 1443 { 1444 #ifdef CONFIG_BLOCK 1445 struct io_context *ioc = current->io_context; 1446 struct io_context *new_ioc; 1447 1448 if (!ioc) 1449 return 0; 1450 /* 1451 * Share io context with parent, if CLONE_IO is set 1452 */ 1453 if (clone_flags & CLONE_IO) { 1454 ioc_task_link(ioc); 1455 tsk->io_context = ioc; 1456 } else if (ioprio_valid(ioc->ioprio)) { 1457 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE); 1458 if (unlikely(!new_ioc)) 1459 return -ENOMEM; 1460 1461 new_ioc->ioprio = ioc->ioprio; 1462 put_io_context(new_ioc); 1463 } 1464 #endif 1465 return 0; 1466 } 1467 1468 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk) 1469 { 1470 struct sighand_struct *sig; 1471 1472 if (clone_flags & CLONE_SIGHAND) { 1473 refcount_inc(¤t->sighand->count); 1474 return 0; 1475 } 1476 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 1477 rcu_assign_pointer(tsk->sighand, sig); 1478 if (!sig) 1479 return -ENOMEM; 1480 1481 refcount_set(&sig->count, 1); 1482 spin_lock_irq(¤t->sighand->siglock); 1483 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 1484 spin_unlock_irq(¤t->sighand->siglock); 1485 return 0; 1486 } 1487 1488 void __cleanup_sighand(struct sighand_struct *sighand) 1489 { 1490 if (refcount_dec_and_test(&sighand->count)) { 1491 signalfd_cleanup(sighand); 1492 /* 1493 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it 1494 * without an RCU grace period, see __lock_task_sighand(). 1495 */ 1496 kmem_cache_free(sighand_cachep, sighand); 1497 } 1498 } 1499 1500 #ifdef CONFIG_POSIX_TIMERS 1501 /* 1502 * Initialize POSIX timer handling for a thread group. 1503 */ 1504 static void posix_cpu_timers_init_group(struct signal_struct *sig) 1505 { 1506 unsigned long cpu_limit; 1507 1508 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); 1509 if (cpu_limit != RLIM_INFINITY) { 1510 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC; 1511 sig->cputimer.running = true; 1512 } 1513 1514 /* The timer lists. */ 1515 INIT_LIST_HEAD(&sig->cpu_timers[0]); 1516 INIT_LIST_HEAD(&sig->cpu_timers[1]); 1517 INIT_LIST_HEAD(&sig->cpu_timers[2]); 1518 } 1519 #else 1520 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { } 1521 #endif 1522 1523 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk) 1524 { 1525 struct signal_struct *sig; 1526 1527 if (clone_flags & CLONE_THREAD) 1528 return 0; 1529 1530 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL); 1531 tsk->signal = sig; 1532 if (!sig) 1533 return -ENOMEM; 1534 1535 sig->nr_threads = 1; 1536 atomic_set(&sig->live, 1); 1537 refcount_set(&sig->sigcnt, 1); 1538 1539 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */ 1540 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node); 1541 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head); 1542 1543 init_waitqueue_head(&sig->wait_chldexit); 1544 sig->curr_target = tsk; 1545 init_sigpending(&sig->shared_pending); 1546 INIT_HLIST_HEAD(&sig->multiprocess); 1547 seqlock_init(&sig->stats_lock); 1548 prev_cputime_init(&sig->prev_cputime); 1549 1550 #ifdef CONFIG_POSIX_TIMERS 1551 INIT_LIST_HEAD(&sig->posix_timers); 1552 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1553 sig->real_timer.function = it_real_fn; 1554 #endif 1555 1556 task_lock(current->group_leader); 1557 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 1558 task_unlock(current->group_leader); 1559 1560 posix_cpu_timers_init_group(sig); 1561 1562 tty_audit_fork(sig); 1563 sched_autogroup_fork(sig); 1564 1565 sig->oom_score_adj = current->signal->oom_score_adj; 1566 sig->oom_score_adj_min = current->signal->oom_score_adj_min; 1567 1568 mutex_init(&sig->cred_guard_mutex); 1569 1570 return 0; 1571 } 1572 1573 static void copy_seccomp(struct task_struct *p) 1574 { 1575 #ifdef CONFIG_SECCOMP 1576 /* 1577 * Must be called with sighand->lock held, which is common to 1578 * all threads in the group. Holding cred_guard_mutex is not 1579 * needed because this new task is not yet running and cannot 1580 * be racing exec. 1581 */ 1582 assert_spin_locked(¤t->sighand->siglock); 1583 1584 /* Ref-count the new filter user, and assign it. */ 1585 get_seccomp_filter(current); 1586 p->seccomp = current->seccomp; 1587 1588 /* 1589 * Explicitly enable no_new_privs here in case it got set 1590 * between the task_struct being duplicated and holding the 1591 * sighand lock. The seccomp state and nnp must be in sync. 1592 */ 1593 if (task_no_new_privs(current)) 1594 task_set_no_new_privs(p); 1595 1596 /* 1597 * If the parent gained a seccomp mode after copying thread 1598 * flags and between before we held the sighand lock, we have 1599 * to manually enable the seccomp thread flag here. 1600 */ 1601 if (p->seccomp.mode != SECCOMP_MODE_DISABLED) 1602 set_tsk_thread_flag(p, TIF_SECCOMP); 1603 #endif 1604 } 1605 1606 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr) 1607 { 1608 current->clear_child_tid = tidptr; 1609 1610 return task_pid_vnr(current); 1611 } 1612 1613 static void rt_mutex_init_task(struct task_struct *p) 1614 { 1615 raw_spin_lock_init(&p->pi_lock); 1616 #ifdef CONFIG_RT_MUTEXES 1617 p->pi_waiters = RB_ROOT_CACHED; 1618 p->pi_top_task = NULL; 1619 p->pi_blocked_on = NULL; 1620 #endif 1621 } 1622 1623 #ifdef CONFIG_POSIX_TIMERS 1624 /* 1625 * Initialize POSIX timer handling for a single task. 1626 */ 1627 static void posix_cpu_timers_init(struct task_struct *tsk) 1628 { 1629 tsk->cputime_expires.prof_exp = 0; 1630 tsk->cputime_expires.virt_exp = 0; 1631 tsk->cputime_expires.sched_exp = 0; 1632 INIT_LIST_HEAD(&tsk->cpu_timers[0]); 1633 INIT_LIST_HEAD(&tsk->cpu_timers[1]); 1634 INIT_LIST_HEAD(&tsk->cpu_timers[2]); 1635 } 1636 #else 1637 static inline void posix_cpu_timers_init(struct task_struct *tsk) { } 1638 #endif 1639 1640 static inline void init_task_pid_links(struct task_struct *task) 1641 { 1642 enum pid_type type; 1643 1644 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 1645 INIT_HLIST_NODE(&task->pid_links[type]); 1646 } 1647 } 1648 1649 static inline void 1650 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid) 1651 { 1652 if (type == PIDTYPE_PID) 1653 task->thread_pid = pid; 1654 else 1655 task->signal->pids[type] = pid; 1656 } 1657 1658 static inline void rcu_copy_process(struct task_struct *p) 1659 { 1660 #ifdef CONFIG_PREEMPT_RCU 1661 p->rcu_read_lock_nesting = 0; 1662 p->rcu_read_unlock_special.s = 0; 1663 p->rcu_blocked_node = NULL; 1664 INIT_LIST_HEAD(&p->rcu_node_entry); 1665 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 1666 #ifdef CONFIG_TASKS_RCU 1667 p->rcu_tasks_holdout = false; 1668 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list); 1669 p->rcu_tasks_idle_cpu = -1; 1670 #endif /* #ifdef CONFIG_TASKS_RCU */ 1671 } 1672 1673 /* 1674 * This creates a new process as a copy of the old one, 1675 * but does not actually start it yet. 1676 * 1677 * It copies the registers, and all the appropriate 1678 * parts of the process environment (as per the clone 1679 * flags). The actual kick-off is left to the caller. 1680 */ 1681 static __latent_entropy struct task_struct *copy_process( 1682 unsigned long clone_flags, 1683 unsigned long stack_start, 1684 unsigned long stack_size, 1685 int __user *child_tidptr, 1686 struct pid *pid, 1687 int trace, 1688 unsigned long tls, 1689 int node) 1690 { 1691 int retval; 1692 struct task_struct *p; 1693 struct multiprocess_signals delayed; 1694 1695 /* 1696 * Don't allow sharing the root directory with processes in a different 1697 * namespace 1698 */ 1699 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 1700 return ERR_PTR(-EINVAL); 1701 1702 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS)) 1703 return ERR_PTR(-EINVAL); 1704 1705 /* 1706 * Thread groups must share signals as well, and detached threads 1707 * can only be started up within the thread group. 1708 */ 1709 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 1710 return ERR_PTR(-EINVAL); 1711 1712 /* 1713 * Shared signal handlers imply shared VM. By way of the above, 1714 * thread groups also imply shared VM. Blocking this case allows 1715 * for various simplifications in other code. 1716 */ 1717 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 1718 return ERR_PTR(-EINVAL); 1719 1720 /* 1721 * Siblings of global init remain as zombies on exit since they are 1722 * not reaped by their parent (swapper). To solve this and to avoid 1723 * multi-rooted process trees, prevent global and container-inits 1724 * from creating siblings. 1725 */ 1726 if ((clone_flags & CLONE_PARENT) && 1727 current->signal->flags & SIGNAL_UNKILLABLE) 1728 return ERR_PTR(-EINVAL); 1729 1730 /* 1731 * If the new process will be in a different pid or user namespace 1732 * do not allow it to share a thread group with the forking task. 1733 */ 1734 if (clone_flags & CLONE_THREAD) { 1735 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) || 1736 (task_active_pid_ns(current) != 1737 current->nsproxy->pid_ns_for_children)) 1738 return ERR_PTR(-EINVAL); 1739 } 1740 1741 /* 1742 * Force any signals received before this point to be delivered 1743 * before the fork happens. Collect up signals sent to multiple 1744 * processes that happen during the fork and delay them so that 1745 * they appear to happen after the fork. 1746 */ 1747 sigemptyset(&delayed.signal); 1748 INIT_HLIST_NODE(&delayed.node); 1749 1750 spin_lock_irq(¤t->sighand->siglock); 1751 if (!(clone_flags & CLONE_THREAD)) 1752 hlist_add_head(&delayed.node, ¤t->signal->multiprocess); 1753 recalc_sigpending(); 1754 spin_unlock_irq(¤t->sighand->siglock); 1755 retval = -ERESTARTNOINTR; 1756 if (signal_pending(current)) 1757 goto fork_out; 1758 1759 retval = -ENOMEM; 1760 p = dup_task_struct(current, node); 1761 if (!p) 1762 goto fork_out; 1763 1764 /* 1765 * This _must_ happen before we call free_task(), i.e. before we jump 1766 * to any of the bad_fork_* labels. This is to avoid freeing 1767 * p->set_child_tid which is (ab)used as a kthread's data pointer for 1768 * kernel threads (PF_KTHREAD). 1769 */ 1770 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; 1771 /* 1772 * Clear TID on mm_release()? 1773 */ 1774 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL; 1775 1776 ftrace_graph_init_task(p); 1777 1778 rt_mutex_init_task(p); 1779 1780 #ifdef CONFIG_PROVE_LOCKING 1781 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); 1782 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 1783 #endif 1784 retval = -EAGAIN; 1785 if (atomic_read(&p->real_cred->user->processes) >= 1786 task_rlimit(p, RLIMIT_NPROC)) { 1787 if (p->real_cred->user != INIT_USER && 1788 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) 1789 goto bad_fork_free; 1790 } 1791 current->flags &= ~PF_NPROC_EXCEEDED; 1792 1793 retval = copy_creds(p, clone_flags); 1794 if (retval < 0) 1795 goto bad_fork_free; 1796 1797 /* 1798 * If multiple threads are within copy_process(), then this check 1799 * triggers too late. This doesn't hurt, the check is only there 1800 * to stop root fork bombs. 1801 */ 1802 retval = -EAGAIN; 1803 if (nr_threads >= max_threads) 1804 goto bad_fork_cleanup_count; 1805 1806 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */ 1807 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE); 1808 p->flags |= PF_FORKNOEXEC; 1809 INIT_LIST_HEAD(&p->children); 1810 INIT_LIST_HEAD(&p->sibling); 1811 rcu_copy_process(p); 1812 p->vfork_done = NULL; 1813 spin_lock_init(&p->alloc_lock); 1814 1815 init_sigpending(&p->pending); 1816 1817 p->utime = p->stime = p->gtime = 0; 1818 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 1819 p->utimescaled = p->stimescaled = 0; 1820 #endif 1821 prev_cputime_init(&p->prev_cputime); 1822 1823 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN 1824 seqcount_init(&p->vtime.seqcount); 1825 p->vtime.starttime = 0; 1826 p->vtime.state = VTIME_INACTIVE; 1827 #endif 1828 1829 #if defined(SPLIT_RSS_COUNTING) 1830 memset(&p->rss_stat, 0, sizeof(p->rss_stat)); 1831 #endif 1832 1833 p->default_timer_slack_ns = current->timer_slack_ns; 1834 1835 #ifdef CONFIG_PSI 1836 p->psi_flags = 0; 1837 #endif 1838 1839 task_io_accounting_init(&p->ioac); 1840 acct_clear_integrals(p); 1841 1842 posix_cpu_timers_init(p); 1843 1844 p->io_context = NULL; 1845 audit_set_context(p, NULL); 1846 cgroup_fork(p); 1847 #ifdef CONFIG_NUMA 1848 p->mempolicy = mpol_dup(p->mempolicy); 1849 if (IS_ERR(p->mempolicy)) { 1850 retval = PTR_ERR(p->mempolicy); 1851 p->mempolicy = NULL; 1852 goto bad_fork_cleanup_threadgroup_lock; 1853 } 1854 #endif 1855 #ifdef CONFIG_CPUSETS 1856 p->cpuset_mem_spread_rotor = NUMA_NO_NODE; 1857 p->cpuset_slab_spread_rotor = NUMA_NO_NODE; 1858 seqcount_init(&p->mems_allowed_seq); 1859 #endif 1860 #ifdef CONFIG_TRACE_IRQFLAGS 1861 p->irq_events = 0; 1862 p->hardirqs_enabled = 0; 1863 p->hardirq_enable_ip = 0; 1864 p->hardirq_enable_event = 0; 1865 p->hardirq_disable_ip = _THIS_IP_; 1866 p->hardirq_disable_event = 0; 1867 p->softirqs_enabled = 1; 1868 p->softirq_enable_ip = _THIS_IP_; 1869 p->softirq_enable_event = 0; 1870 p->softirq_disable_ip = 0; 1871 p->softirq_disable_event = 0; 1872 p->hardirq_context = 0; 1873 p->softirq_context = 0; 1874 #endif 1875 1876 p->pagefault_disabled = 0; 1877 1878 #ifdef CONFIG_LOCKDEP 1879 p->lockdep_depth = 0; /* no locks held yet */ 1880 p->curr_chain_key = 0; 1881 p->lockdep_recursion = 0; 1882 lockdep_init_task(p); 1883 #endif 1884 1885 #ifdef CONFIG_DEBUG_MUTEXES 1886 p->blocked_on = NULL; /* not blocked yet */ 1887 #endif 1888 #ifdef CONFIG_BCACHE 1889 p->sequential_io = 0; 1890 p->sequential_io_avg = 0; 1891 #endif 1892 1893 /* Perform scheduler related setup. Assign this task to a CPU. */ 1894 retval = sched_fork(clone_flags, p); 1895 if (retval) 1896 goto bad_fork_cleanup_policy; 1897 1898 retval = perf_event_init_task(p); 1899 if (retval) 1900 goto bad_fork_cleanup_policy; 1901 retval = audit_alloc(p); 1902 if (retval) 1903 goto bad_fork_cleanup_perf; 1904 /* copy all the process information */ 1905 shm_init_task(p); 1906 retval = security_task_alloc(p, clone_flags); 1907 if (retval) 1908 goto bad_fork_cleanup_audit; 1909 retval = copy_semundo(clone_flags, p); 1910 if (retval) 1911 goto bad_fork_cleanup_security; 1912 retval = copy_files(clone_flags, p); 1913 if (retval) 1914 goto bad_fork_cleanup_semundo; 1915 retval = copy_fs(clone_flags, p); 1916 if (retval) 1917 goto bad_fork_cleanup_files; 1918 retval = copy_sighand(clone_flags, p); 1919 if (retval) 1920 goto bad_fork_cleanup_fs; 1921 retval = copy_signal(clone_flags, p); 1922 if (retval) 1923 goto bad_fork_cleanup_sighand; 1924 retval = copy_mm(clone_flags, p); 1925 if (retval) 1926 goto bad_fork_cleanup_signal; 1927 retval = copy_namespaces(clone_flags, p); 1928 if (retval) 1929 goto bad_fork_cleanup_mm; 1930 retval = copy_io(clone_flags, p); 1931 if (retval) 1932 goto bad_fork_cleanup_namespaces; 1933 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls); 1934 if (retval) 1935 goto bad_fork_cleanup_io; 1936 1937 stackleak_task_init(p); 1938 1939 if (pid != &init_struct_pid) { 1940 pid = alloc_pid(p->nsproxy->pid_ns_for_children); 1941 if (IS_ERR(pid)) { 1942 retval = PTR_ERR(pid); 1943 goto bad_fork_cleanup_thread; 1944 } 1945 } 1946 1947 #ifdef CONFIG_BLOCK 1948 p->plug = NULL; 1949 #endif 1950 #ifdef CONFIG_FUTEX 1951 p->robust_list = NULL; 1952 #ifdef CONFIG_COMPAT 1953 p->compat_robust_list = NULL; 1954 #endif 1955 INIT_LIST_HEAD(&p->pi_state_list); 1956 p->pi_state_cache = NULL; 1957 #endif 1958 /* 1959 * sigaltstack should be cleared when sharing the same VM 1960 */ 1961 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM) 1962 sas_ss_reset(p); 1963 1964 /* 1965 * Syscall tracing and stepping should be turned off in the 1966 * child regardless of CLONE_PTRACE. 1967 */ 1968 user_disable_single_step(p); 1969 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); 1970 #ifdef TIF_SYSCALL_EMU 1971 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); 1972 #endif 1973 clear_all_latency_tracing(p); 1974 1975 /* ok, now we should be set up.. */ 1976 p->pid = pid_nr(pid); 1977 if (clone_flags & CLONE_THREAD) { 1978 p->exit_signal = -1; 1979 p->group_leader = current->group_leader; 1980 p->tgid = current->tgid; 1981 } else { 1982 if (clone_flags & CLONE_PARENT) 1983 p->exit_signal = current->group_leader->exit_signal; 1984 else 1985 p->exit_signal = (clone_flags & CSIGNAL); 1986 p->group_leader = p; 1987 p->tgid = p->pid; 1988 } 1989 1990 p->nr_dirtied = 0; 1991 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10); 1992 p->dirty_paused_when = 0; 1993 1994 p->pdeath_signal = 0; 1995 INIT_LIST_HEAD(&p->thread_group); 1996 p->task_works = NULL; 1997 1998 cgroup_threadgroup_change_begin(current); 1999 /* 2000 * Ensure that the cgroup subsystem policies allow the new process to be 2001 * forked. It should be noted the the new process's css_set can be changed 2002 * between here and cgroup_post_fork() if an organisation operation is in 2003 * progress. 2004 */ 2005 retval = cgroup_can_fork(p); 2006 if (retval) 2007 goto bad_fork_free_pid; 2008 2009 /* 2010 * From this point on we must avoid any synchronous user-space 2011 * communication until we take the tasklist-lock. In particular, we do 2012 * not want user-space to be able to predict the process start-time by 2013 * stalling fork(2) after we recorded the start_time but before it is 2014 * visible to the system. 2015 */ 2016 2017 p->start_time = ktime_get_ns(); 2018 p->real_start_time = ktime_get_boot_ns(); 2019 2020 /* 2021 * Make it visible to the rest of the system, but dont wake it up yet. 2022 * Need tasklist lock for parent etc handling! 2023 */ 2024 write_lock_irq(&tasklist_lock); 2025 2026 /* CLONE_PARENT re-uses the old parent */ 2027 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { 2028 p->real_parent = current->real_parent; 2029 p->parent_exec_id = current->parent_exec_id; 2030 } else { 2031 p->real_parent = current; 2032 p->parent_exec_id = current->self_exec_id; 2033 } 2034 2035 klp_copy_process(p); 2036 2037 spin_lock(¤t->sighand->siglock); 2038 2039 /* 2040 * Copy seccomp details explicitly here, in case they were changed 2041 * before holding sighand lock. 2042 */ 2043 copy_seccomp(p); 2044 2045 rseq_fork(p, clone_flags); 2046 2047 /* Don't start children in a dying pid namespace */ 2048 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) { 2049 retval = -ENOMEM; 2050 goto bad_fork_cancel_cgroup; 2051 } 2052 2053 /* Let kill terminate clone/fork in the middle */ 2054 if (fatal_signal_pending(current)) { 2055 retval = -EINTR; 2056 goto bad_fork_cancel_cgroup; 2057 } 2058 2059 2060 init_task_pid_links(p); 2061 if (likely(p->pid)) { 2062 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace); 2063 2064 init_task_pid(p, PIDTYPE_PID, pid); 2065 if (thread_group_leader(p)) { 2066 init_task_pid(p, PIDTYPE_TGID, pid); 2067 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current)); 2068 init_task_pid(p, PIDTYPE_SID, task_session(current)); 2069 2070 if (is_child_reaper(pid)) { 2071 ns_of_pid(pid)->child_reaper = p; 2072 p->signal->flags |= SIGNAL_UNKILLABLE; 2073 } 2074 p->signal->shared_pending.signal = delayed.signal; 2075 p->signal->tty = tty_kref_get(current->signal->tty); 2076 /* 2077 * Inherit has_child_subreaper flag under the same 2078 * tasklist_lock with adding child to the process tree 2079 * for propagate_has_child_subreaper optimization. 2080 */ 2081 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper || 2082 p->real_parent->signal->is_child_subreaper; 2083 list_add_tail(&p->sibling, &p->real_parent->children); 2084 list_add_tail_rcu(&p->tasks, &init_task.tasks); 2085 attach_pid(p, PIDTYPE_TGID); 2086 attach_pid(p, PIDTYPE_PGID); 2087 attach_pid(p, PIDTYPE_SID); 2088 __this_cpu_inc(process_counts); 2089 } else { 2090 current->signal->nr_threads++; 2091 atomic_inc(¤t->signal->live); 2092 refcount_inc(¤t->signal->sigcnt); 2093 task_join_group_stop(p); 2094 list_add_tail_rcu(&p->thread_group, 2095 &p->group_leader->thread_group); 2096 list_add_tail_rcu(&p->thread_node, 2097 &p->signal->thread_head); 2098 } 2099 attach_pid(p, PIDTYPE_PID); 2100 nr_threads++; 2101 } 2102 total_forks++; 2103 hlist_del_init(&delayed.node); 2104 spin_unlock(¤t->sighand->siglock); 2105 syscall_tracepoint_update(p); 2106 write_unlock_irq(&tasklist_lock); 2107 2108 proc_fork_connector(p); 2109 cgroup_post_fork(p); 2110 cgroup_threadgroup_change_end(current); 2111 perf_event_fork(p); 2112 2113 trace_task_newtask(p, clone_flags); 2114 uprobe_copy_process(p, clone_flags); 2115 2116 return p; 2117 2118 bad_fork_cancel_cgroup: 2119 spin_unlock(¤t->sighand->siglock); 2120 write_unlock_irq(&tasklist_lock); 2121 cgroup_cancel_fork(p); 2122 bad_fork_free_pid: 2123 cgroup_threadgroup_change_end(current); 2124 if (pid != &init_struct_pid) 2125 free_pid(pid); 2126 bad_fork_cleanup_thread: 2127 exit_thread(p); 2128 bad_fork_cleanup_io: 2129 if (p->io_context) 2130 exit_io_context(p); 2131 bad_fork_cleanup_namespaces: 2132 exit_task_namespaces(p); 2133 bad_fork_cleanup_mm: 2134 if (p->mm) 2135 mmput(p->mm); 2136 bad_fork_cleanup_signal: 2137 if (!(clone_flags & CLONE_THREAD)) 2138 free_signal_struct(p->signal); 2139 bad_fork_cleanup_sighand: 2140 __cleanup_sighand(p->sighand); 2141 bad_fork_cleanup_fs: 2142 exit_fs(p); /* blocking */ 2143 bad_fork_cleanup_files: 2144 exit_files(p); /* blocking */ 2145 bad_fork_cleanup_semundo: 2146 exit_sem(p); 2147 bad_fork_cleanup_security: 2148 security_task_free(p); 2149 bad_fork_cleanup_audit: 2150 audit_free(p); 2151 bad_fork_cleanup_perf: 2152 perf_event_free_task(p); 2153 bad_fork_cleanup_policy: 2154 lockdep_free_task(p); 2155 #ifdef CONFIG_NUMA 2156 mpol_put(p->mempolicy); 2157 bad_fork_cleanup_threadgroup_lock: 2158 #endif 2159 delayacct_tsk_free(p); 2160 bad_fork_cleanup_count: 2161 atomic_dec(&p->cred->user->processes); 2162 exit_creds(p); 2163 bad_fork_free: 2164 p->state = TASK_DEAD; 2165 put_task_stack(p); 2166 free_task(p); 2167 fork_out: 2168 spin_lock_irq(¤t->sighand->siglock); 2169 hlist_del_init(&delayed.node); 2170 spin_unlock_irq(¤t->sighand->siglock); 2171 return ERR_PTR(retval); 2172 } 2173 2174 static inline void init_idle_pids(struct task_struct *idle) 2175 { 2176 enum pid_type type; 2177 2178 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 2179 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */ 2180 init_task_pid(idle, type, &init_struct_pid); 2181 } 2182 } 2183 2184 struct task_struct *fork_idle(int cpu) 2185 { 2186 struct task_struct *task; 2187 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0, 2188 cpu_to_node(cpu)); 2189 if (!IS_ERR(task)) { 2190 init_idle_pids(task); 2191 init_idle(task, cpu); 2192 } 2193 2194 return task; 2195 } 2196 2197 struct mm_struct *copy_init_mm(void) 2198 { 2199 return dup_mm(NULL, &init_mm); 2200 } 2201 2202 /* 2203 * Ok, this is the main fork-routine. 2204 * 2205 * It copies the process, and if successful kick-starts 2206 * it and waits for it to finish using the VM if required. 2207 */ 2208 long _do_fork(unsigned long clone_flags, 2209 unsigned long stack_start, 2210 unsigned long stack_size, 2211 int __user *parent_tidptr, 2212 int __user *child_tidptr, 2213 unsigned long tls) 2214 { 2215 struct completion vfork; 2216 struct pid *pid; 2217 struct task_struct *p; 2218 int trace = 0; 2219 long nr; 2220 2221 /* 2222 * Determine whether and which event to report to ptracer. When 2223 * called from kernel_thread or CLONE_UNTRACED is explicitly 2224 * requested, no event is reported; otherwise, report if the event 2225 * for the type of forking is enabled. 2226 */ 2227 if (!(clone_flags & CLONE_UNTRACED)) { 2228 if (clone_flags & CLONE_VFORK) 2229 trace = PTRACE_EVENT_VFORK; 2230 else if ((clone_flags & CSIGNAL) != SIGCHLD) 2231 trace = PTRACE_EVENT_CLONE; 2232 else 2233 trace = PTRACE_EVENT_FORK; 2234 2235 if (likely(!ptrace_event_enabled(current, trace))) 2236 trace = 0; 2237 } 2238 2239 p = copy_process(clone_flags, stack_start, stack_size, 2240 child_tidptr, NULL, trace, tls, NUMA_NO_NODE); 2241 add_latent_entropy(); 2242 2243 if (IS_ERR(p)) 2244 return PTR_ERR(p); 2245 2246 /* 2247 * Do this prior waking up the new thread - the thread pointer 2248 * might get invalid after that point, if the thread exits quickly. 2249 */ 2250 trace_sched_process_fork(current, p); 2251 2252 pid = get_task_pid(p, PIDTYPE_PID); 2253 nr = pid_vnr(pid); 2254 2255 if (clone_flags & CLONE_PARENT_SETTID) 2256 put_user(nr, parent_tidptr); 2257 2258 if (clone_flags & CLONE_VFORK) { 2259 p->vfork_done = &vfork; 2260 init_completion(&vfork); 2261 get_task_struct(p); 2262 } 2263 2264 wake_up_new_task(p); 2265 2266 /* forking complete and child started to run, tell ptracer */ 2267 if (unlikely(trace)) 2268 ptrace_event_pid(trace, pid); 2269 2270 if (clone_flags & CLONE_VFORK) { 2271 if (!wait_for_vfork_done(p, &vfork)) 2272 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid); 2273 } 2274 2275 put_pid(pid); 2276 return nr; 2277 } 2278 2279 #ifndef CONFIG_HAVE_COPY_THREAD_TLS 2280 /* For compatibility with architectures that call do_fork directly rather than 2281 * using the syscall entry points below. */ 2282 long do_fork(unsigned long clone_flags, 2283 unsigned long stack_start, 2284 unsigned long stack_size, 2285 int __user *parent_tidptr, 2286 int __user *child_tidptr) 2287 { 2288 return _do_fork(clone_flags, stack_start, stack_size, 2289 parent_tidptr, child_tidptr, 0); 2290 } 2291 #endif 2292 2293 /* 2294 * Create a kernel thread. 2295 */ 2296 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags) 2297 { 2298 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn, 2299 (unsigned long)arg, NULL, NULL, 0); 2300 } 2301 2302 #ifdef __ARCH_WANT_SYS_FORK 2303 SYSCALL_DEFINE0(fork) 2304 { 2305 #ifdef CONFIG_MMU 2306 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0); 2307 #else 2308 /* can not support in nommu mode */ 2309 return -EINVAL; 2310 #endif 2311 } 2312 #endif 2313 2314 #ifdef __ARCH_WANT_SYS_VFORK 2315 SYSCALL_DEFINE0(vfork) 2316 { 2317 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0, 2318 0, NULL, NULL, 0); 2319 } 2320 #endif 2321 2322 #ifdef __ARCH_WANT_SYS_CLONE 2323 #ifdef CONFIG_CLONE_BACKWARDS 2324 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2325 int __user *, parent_tidptr, 2326 unsigned long, tls, 2327 int __user *, child_tidptr) 2328 #elif defined(CONFIG_CLONE_BACKWARDS2) 2329 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags, 2330 int __user *, parent_tidptr, 2331 int __user *, child_tidptr, 2332 unsigned long, tls) 2333 #elif defined(CONFIG_CLONE_BACKWARDS3) 2334 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp, 2335 int, stack_size, 2336 int __user *, parent_tidptr, 2337 int __user *, child_tidptr, 2338 unsigned long, tls) 2339 #else 2340 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2341 int __user *, parent_tidptr, 2342 int __user *, child_tidptr, 2343 unsigned long, tls) 2344 #endif 2345 { 2346 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls); 2347 } 2348 #endif 2349 2350 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data) 2351 { 2352 struct task_struct *leader, *parent, *child; 2353 int res; 2354 2355 read_lock(&tasklist_lock); 2356 leader = top = top->group_leader; 2357 down: 2358 for_each_thread(leader, parent) { 2359 list_for_each_entry(child, &parent->children, sibling) { 2360 res = visitor(child, data); 2361 if (res) { 2362 if (res < 0) 2363 goto out; 2364 leader = child; 2365 goto down; 2366 } 2367 up: 2368 ; 2369 } 2370 } 2371 2372 if (leader != top) { 2373 child = leader; 2374 parent = child->real_parent; 2375 leader = parent->group_leader; 2376 goto up; 2377 } 2378 out: 2379 read_unlock(&tasklist_lock); 2380 } 2381 2382 #ifndef ARCH_MIN_MMSTRUCT_ALIGN 2383 #define ARCH_MIN_MMSTRUCT_ALIGN 0 2384 #endif 2385 2386 static void sighand_ctor(void *data) 2387 { 2388 struct sighand_struct *sighand = data; 2389 2390 spin_lock_init(&sighand->siglock); 2391 init_waitqueue_head(&sighand->signalfd_wqh); 2392 } 2393 2394 void __init proc_caches_init(void) 2395 { 2396 unsigned int mm_size; 2397 2398 sighand_cachep = kmem_cache_create("sighand_cache", 2399 sizeof(struct sighand_struct), 0, 2400 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU| 2401 SLAB_ACCOUNT, sighand_ctor); 2402 signal_cachep = kmem_cache_create("signal_cache", 2403 sizeof(struct signal_struct), 0, 2404 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2405 NULL); 2406 files_cachep = kmem_cache_create("files_cache", 2407 sizeof(struct files_struct), 0, 2408 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2409 NULL); 2410 fs_cachep = kmem_cache_create("fs_cache", 2411 sizeof(struct fs_struct), 0, 2412 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2413 NULL); 2414 2415 /* 2416 * The mm_cpumask is located at the end of mm_struct, and is 2417 * dynamically sized based on the maximum CPU number this system 2418 * can have, taking hotplug into account (nr_cpu_ids). 2419 */ 2420 mm_size = sizeof(struct mm_struct) + cpumask_size(); 2421 2422 mm_cachep = kmem_cache_create_usercopy("mm_struct", 2423 mm_size, ARCH_MIN_MMSTRUCT_ALIGN, 2424 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2425 offsetof(struct mm_struct, saved_auxv), 2426 sizeof_field(struct mm_struct, saved_auxv), 2427 NULL); 2428 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT); 2429 mmap_init(); 2430 nsproxy_cache_init(); 2431 } 2432 2433 /* 2434 * Check constraints on flags passed to the unshare system call. 2435 */ 2436 static int check_unshare_flags(unsigned long unshare_flags) 2437 { 2438 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND| 2439 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM| 2440 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET| 2441 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP)) 2442 return -EINVAL; 2443 /* 2444 * Not implemented, but pretend it works if there is nothing 2445 * to unshare. Note that unsharing the address space or the 2446 * signal handlers also need to unshare the signal queues (aka 2447 * CLONE_THREAD). 2448 */ 2449 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) { 2450 if (!thread_group_empty(current)) 2451 return -EINVAL; 2452 } 2453 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) { 2454 if (refcount_read(¤t->sighand->count) > 1) 2455 return -EINVAL; 2456 } 2457 if (unshare_flags & CLONE_VM) { 2458 if (!current_is_single_threaded()) 2459 return -EINVAL; 2460 } 2461 2462 return 0; 2463 } 2464 2465 /* 2466 * Unshare the filesystem structure if it is being shared 2467 */ 2468 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) 2469 { 2470 struct fs_struct *fs = current->fs; 2471 2472 if (!(unshare_flags & CLONE_FS) || !fs) 2473 return 0; 2474 2475 /* don't need lock here; in the worst case we'll do useless copy */ 2476 if (fs->users == 1) 2477 return 0; 2478 2479 *new_fsp = copy_fs_struct(fs); 2480 if (!*new_fsp) 2481 return -ENOMEM; 2482 2483 return 0; 2484 } 2485 2486 /* 2487 * Unshare file descriptor table if it is being shared 2488 */ 2489 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp) 2490 { 2491 struct files_struct *fd = current->files; 2492 int error = 0; 2493 2494 if ((unshare_flags & CLONE_FILES) && 2495 (fd && atomic_read(&fd->count) > 1)) { 2496 *new_fdp = dup_fd(fd, &error); 2497 if (!*new_fdp) 2498 return error; 2499 } 2500 2501 return 0; 2502 } 2503 2504 /* 2505 * unshare allows a process to 'unshare' part of the process 2506 * context which was originally shared using clone. copy_* 2507 * functions used by do_fork() cannot be used here directly 2508 * because they modify an inactive task_struct that is being 2509 * constructed. Here we are modifying the current, active, 2510 * task_struct. 2511 */ 2512 int ksys_unshare(unsigned long unshare_flags) 2513 { 2514 struct fs_struct *fs, *new_fs = NULL; 2515 struct files_struct *fd, *new_fd = NULL; 2516 struct cred *new_cred = NULL; 2517 struct nsproxy *new_nsproxy = NULL; 2518 int do_sysvsem = 0; 2519 int err; 2520 2521 /* 2522 * If unsharing a user namespace must also unshare the thread group 2523 * and unshare the filesystem root and working directories. 2524 */ 2525 if (unshare_flags & CLONE_NEWUSER) 2526 unshare_flags |= CLONE_THREAD | CLONE_FS; 2527 /* 2528 * If unsharing vm, must also unshare signal handlers. 2529 */ 2530 if (unshare_flags & CLONE_VM) 2531 unshare_flags |= CLONE_SIGHAND; 2532 /* 2533 * If unsharing a signal handlers, must also unshare the signal queues. 2534 */ 2535 if (unshare_flags & CLONE_SIGHAND) 2536 unshare_flags |= CLONE_THREAD; 2537 /* 2538 * If unsharing namespace, must also unshare filesystem information. 2539 */ 2540 if (unshare_flags & CLONE_NEWNS) 2541 unshare_flags |= CLONE_FS; 2542 2543 err = check_unshare_flags(unshare_flags); 2544 if (err) 2545 goto bad_unshare_out; 2546 /* 2547 * CLONE_NEWIPC must also detach from the undolist: after switching 2548 * to a new ipc namespace, the semaphore arrays from the old 2549 * namespace are unreachable. 2550 */ 2551 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) 2552 do_sysvsem = 1; 2553 err = unshare_fs(unshare_flags, &new_fs); 2554 if (err) 2555 goto bad_unshare_out; 2556 err = unshare_fd(unshare_flags, &new_fd); 2557 if (err) 2558 goto bad_unshare_cleanup_fs; 2559 err = unshare_userns(unshare_flags, &new_cred); 2560 if (err) 2561 goto bad_unshare_cleanup_fd; 2562 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, 2563 new_cred, new_fs); 2564 if (err) 2565 goto bad_unshare_cleanup_cred; 2566 2567 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) { 2568 if (do_sysvsem) { 2569 /* 2570 * CLONE_SYSVSEM is equivalent to sys_exit(). 2571 */ 2572 exit_sem(current); 2573 } 2574 if (unshare_flags & CLONE_NEWIPC) { 2575 /* Orphan segments in old ns (see sem above). */ 2576 exit_shm(current); 2577 shm_init_task(current); 2578 } 2579 2580 if (new_nsproxy) 2581 switch_task_namespaces(current, new_nsproxy); 2582 2583 task_lock(current); 2584 2585 if (new_fs) { 2586 fs = current->fs; 2587 spin_lock(&fs->lock); 2588 current->fs = new_fs; 2589 if (--fs->users) 2590 new_fs = NULL; 2591 else 2592 new_fs = fs; 2593 spin_unlock(&fs->lock); 2594 } 2595 2596 if (new_fd) { 2597 fd = current->files; 2598 current->files = new_fd; 2599 new_fd = fd; 2600 } 2601 2602 task_unlock(current); 2603 2604 if (new_cred) { 2605 /* Install the new user namespace */ 2606 commit_creds(new_cred); 2607 new_cred = NULL; 2608 } 2609 } 2610 2611 perf_event_namespaces(current); 2612 2613 bad_unshare_cleanup_cred: 2614 if (new_cred) 2615 put_cred(new_cred); 2616 bad_unshare_cleanup_fd: 2617 if (new_fd) 2618 put_files_struct(new_fd); 2619 2620 bad_unshare_cleanup_fs: 2621 if (new_fs) 2622 free_fs_struct(new_fs); 2623 2624 bad_unshare_out: 2625 return err; 2626 } 2627 2628 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags) 2629 { 2630 return ksys_unshare(unshare_flags); 2631 } 2632 2633 /* 2634 * Helper to unshare the files of the current task. 2635 * We don't want to expose copy_files internals to 2636 * the exec layer of the kernel. 2637 */ 2638 2639 int unshare_files(struct files_struct **displaced) 2640 { 2641 struct task_struct *task = current; 2642 struct files_struct *copy = NULL; 2643 int error; 2644 2645 error = unshare_fd(CLONE_FILES, ©); 2646 if (error || !copy) { 2647 *displaced = NULL; 2648 return error; 2649 } 2650 *displaced = task->files; 2651 task_lock(task); 2652 task->files = copy; 2653 task_unlock(task); 2654 return 0; 2655 } 2656 2657 int sysctl_max_threads(struct ctl_table *table, int write, 2658 void __user *buffer, size_t *lenp, loff_t *ppos) 2659 { 2660 struct ctl_table t; 2661 int ret; 2662 int threads = max_threads; 2663 int min = MIN_THREADS; 2664 int max = MAX_THREADS; 2665 2666 t = *table; 2667 t.data = &threads; 2668 t.extra1 = &min; 2669 t.extra2 = &max; 2670 2671 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); 2672 if (ret || !write) 2673 return ret; 2674 2675 set_max_threads(threads); 2676 2677 return 0; 2678 } 2679