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