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