1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/kernel/fork.c 4 * 5 * Copyright (C) 1991, 1992 Linus Torvalds 6 */ 7 8 /* 9 * 'fork.c' contains the help-routines for the 'fork' system call 10 * (see also entry.S and others). 11 * Fork is rather simple, once you get the hang of it, but the memory 12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()' 13 */ 14 15 #include <linux/anon_inodes.h> 16 #include <linux/slab.h> 17 #include <linux/sched/autogroup.h> 18 #include <linux/sched/mm.h> 19 #include <linux/sched/coredump.h> 20 #include <linux/sched/user.h> 21 #include <linux/sched/numa_balancing.h> 22 #include <linux/sched/stat.h> 23 #include <linux/sched/task.h> 24 #include <linux/sched/task_stack.h> 25 #include <linux/sched/cputime.h> 26 #include <linux/seq_file.h> 27 #include <linux/rtmutex.h> 28 #include <linux/init.h> 29 #include <linux/unistd.h> 30 #include <linux/module.h> 31 #include <linux/vmalloc.h> 32 #include <linux/completion.h> 33 #include <linux/personality.h> 34 #include <linux/mempolicy.h> 35 #include <linux/sem.h> 36 #include <linux/file.h> 37 #include <linux/fdtable.h> 38 #include <linux/iocontext.h> 39 #include <linux/key.h> 40 #include <linux/binfmts.h> 41 #include <linux/mman.h> 42 #include <linux/mmu_notifier.h> 43 #include <linux/hmm.h> 44 #include <linux/fs.h> 45 #include <linux/mm.h> 46 #include <linux/vmacache.h> 47 #include <linux/nsproxy.h> 48 #include <linux/capability.h> 49 #include <linux/cpu.h> 50 #include <linux/cgroup.h> 51 #include <linux/security.h> 52 #include <linux/hugetlb.h> 53 #include <linux/seccomp.h> 54 #include <linux/swap.h> 55 #include <linux/syscalls.h> 56 #include <linux/jiffies.h> 57 #include <linux/futex.h> 58 #include <linux/compat.h> 59 #include <linux/kthread.h> 60 #include <linux/task_io_accounting_ops.h> 61 #include <linux/rcupdate.h> 62 #include <linux/ptrace.h> 63 #include <linux/mount.h> 64 #include <linux/audit.h> 65 #include <linux/memcontrol.h> 66 #include <linux/ftrace.h> 67 #include <linux/proc_fs.h> 68 #include <linux/profile.h> 69 #include <linux/rmap.h> 70 #include <linux/ksm.h> 71 #include <linux/acct.h> 72 #include <linux/userfaultfd_k.h> 73 #include <linux/tsacct_kern.h> 74 #include <linux/cn_proc.h> 75 #include <linux/freezer.h> 76 #include <linux/delayacct.h> 77 #include <linux/taskstats_kern.h> 78 #include <linux/random.h> 79 #include <linux/tty.h> 80 #include <linux/blkdev.h> 81 #include <linux/fs_struct.h> 82 #include <linux/magic.h> 83 #include <linux/perf_event.h> 84 #include <linux/posix-timers.h> 85 #include <linux/user-return-notifier.h> 86 #include <linux/oom.h> 87 #include <linux/khugepaged.h> 88 #include <linux/signalfd.h> 89 #include <linux/uprobes.h> 90 #include <linux/aio.h> 91 #include <linux/compiler.h> 92 #include <linux/sysctl.h> 93 #include <linux/kcov.h> 94 #include <linux/livepatch.h> 95 #include <linux/thread_info.h> 96 #include <linux/stackleak.h> 97 98 #include <asm/pgtable.h> 99 #include <asm/pgalloc.h> 100 #include <linux/uaccess.h> 101 #include <asm/mmu_context.h> 102 #include <asm/cacheflush.h> 103 #include <asm/tlbflush.h> 104 105 #include <trace/events/sched.h> 106 107 #define CREATE_TRACE_POINTS 108 #include <trace/events/task.h> 109 110 /* 111 * Minimum number of threads to boot the kernel 112 */ 113 #define MIN_THREADS 20 114 115 /* 116 * Maximum number of threads 117 */ 118 #define MAX_THREADS FUTEX_TID_MASK 119 120 /* 121 * Protected counters by write_lock_irq(&tasklist_lock) 122 */ 123 unsigned long total_forks; /* Handle normal Linux uptimes. */ 124 int nr_threads; /* The idle threads do not count.. */ 125 126 static int max_threads; /* tunable limit on nr_threads */ 127 128 DEFINE_PER_CPU(unsigned long, process_counts) = 0; 129 130 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */ 131 132 #ifdef CONFIG_PROVE_RCU 133 int lockdep_tasklist_lock_is_held(void) 134 { 135 return lockdep_is_held(&tasklist_lock); 136 } 137 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held); 138 #endif /* #ifdef CONFIG_PROVE_RCU */ 139 140 int nr_processes(void) 141 { 142 int cpu; 143 int total = 0; 144 145 for_each_possible_cpu(cpu) 146 total += per_cpu(process_counts, cpu); 147 148 return total; 149 } 150 151 void __weak arch_release_task_struct(struct task_struct *tsk) 152 { 153 } 154 155 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 156 static struct kmem_cache *task_struct_cachep; 157 158 static inline struct task_struct *alloc_task_struct_node(int node) 159 { 160 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node); 161 } 162 163 static inline void free_task_struct(struct task_struct *tsk) 164 { 165 kmem_cache_free(task_struct_cachep, tsk); 166 } 167 #endif 168 169 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR 170 171 /* 172 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a 173 * kmemcache based allocator. 174 */ 175 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) 176 177 #ifdef CONFIG_VMAP_STACK 178 /* 179 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB 180 * flush. Try to minimize the number of calls by caching stacks. 181 */ 182 #define NR_CACHED_STACKS 2 183 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]); 184 185 static int free_vm_stack_cache(unsigned int cpu) 186 { 187 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu); 188 int i; 189 190 for (i = 0; i < NR_CACHED_STACKS; i++) { 191 struct vm_struct *vm_stack = cached_vm_stacks[i]; 192 193 if (!vm_stack) 194 continue; 195 196 vfree(vm_stack->addr); 197 cached_vm_stacks[i] = NULL; 198 } 199 200 return 0; 201 } 202 #endif 203 204 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node) 205 { 206 #ifdef CONFIG_VMAP_STACK 207 void *stack; 208 int i; 209 210 for (i = 0; i < NR_CACHED_STACKS; i++) { 211 struct vm_struct *s; 212 213 s = this_cpu_xchg(cached_stacks[i], NULL); 214 215 if (!s) 216 continue; 217 218 /* Clear stale pointers from reused stack. */ 219 memset(s->addr, 0, THREAD_SIZE); 220 221 tsk->stack_vm_area = s; 222 tsk->stack = s->addr; 223 return s->addr; 224 } 225 226 /* 227 * Allocated stacks are cached and later reused by new threads, 228 * so memcg accounting is performed manually on assigning/releasing 229 * stacks to tasks. Drop __GFP_ACCOUNT. 230 */ 231 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN, 232 VMALLOC_START, VMALLOC_END, 233 THREADINFO_GFP & ~__GFP_ACCOUNT, 234 PAGE_KERNEL, 235 0, node, __builtin_return_address(0)); 236 237 /* 238 * We can't call find_vm_area() in interrupt context, and 239 * free_thread_stack() can be called in interrupt context, 240 * so cache the vm_struct. 241 */ 242 if (stack) { 243 tsk->stack_vm_area = find_vm_area(stack); 244 tsk->stack = stack; 245 } 246 return stack; 247 #else 248 struct page *page = alloc_pages_node(node, THREADINFO_GFP, 249 THREAD_SIZE_ORDER); 250 251 if (likely(page)) { 252 tsk->stack = page_address(page); 253 return tsk->stack; 254 } 255 return NULL; 256 #endif 257 } 258 259 static inline void free_thread_stack(struct task_struct *tsk) 260 { 261 #ifdef CONFIG_VMAP_STACK 262 struct vm_struct *vm = task_stack_vm_area(tsk); 263 264 if (vm) { 265 int i; 266 267 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 268 mod_memcg_page_state(vm->pages[i], 269 MEMCG_KERNEL_STACK_KB, 270 -(int)(PAGE_SIZE / 1024)); 271 272 memcg_kmem_uncharge(vm->pages[i], 0); 273 } 274 275 for (i = 0; i < NR_CACHED_STACKS; i++) { 276 if (this_cpu_cmpxchg(cached_stacks[i], 277 NULL, tsk->stack_vm_area) != NULL) 278 continue; 279 280 return; 281 } 282 283 vfree_atomic(tsk->stack); 284 return; 285 } 286 #endif 287 288 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER); 289 } 290 # else 291 static struct kmem_cache *thread_stack_cache; 292 293 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, 294 int node) 295 { 296 unsigned long *stack; 297 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node); 298 tsk->stack = stack; 299 return stack; 300 } 301 302 static void free_thread_stack(struct task_struct *tsk) 303 { 304 kmem_cache_free(thread_stack_cache, tsk->stack); 305 } 306 307 void thread_stack_cache_init(void) 308 { 309 thread_stack_cache = kmem_cache_create_usercopy("thread_stack", 310 THREAD_SIZE, THREAD_SIZE, 0, 0, 311 THREAD_SIZE, NULL); 312 BUG_ON(thread_stack_cache == NULL); 313 } 314 # endif 315 #endif 316 317 /* SLAB cache for signal_struct structures (tsk->signal) */ 318 static struct kmem_cache *signal_cachep; 319 320 /* SLAB cache for sighand_struct structures (tsk->sighand) */ 321 struct kmem_cache *sighand_cachep; 322 323 /* SLAB cache for files_struct structures (tsk->files) */ 324 struct kmem_cache *files_cachep; 325 326 /* SLAB cache for fs_struct structures (tsk->fs) */ 327 struct kmem_cache *fs_cachep; 328 329 /* SLAB cache for vm_area_struct structures */ 330 static struct kmem_cache *vm_area_cachep; 331 332 /* SLAB cache for mm_struct structures (tsk->mm) */ 333 static struct kmem_cache *mm_cachep; 334 335 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm) 336 { 337 struct vm_area_struct *vma; 338 339 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 340 if (vma) 341 vma_init(vma, mm); 342 return vma; 343 } 344 345 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig) 346 { 347 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 348 349 if (new) { 350 *new = *orig; 351 INIT_LIST_HEAD(&new->anon_vma_chain); 352 } 353 return new; 354 } 355 356 void vm_area_free(struct vm_area_struct *vma) 357 { 358 kmem_cache_free(vm_area_cachep, vma); 359 } 360 361 static void account_kernel_stack(struct task_struct *tsk, int account) 362 { 363 void *stack = task_stack_page(tsk); 364 struct vm_struct *vm = task_stack_vm_area(tsk); 365 366 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0); 367 368 if (vm) { 369 int i; 370 371 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE); 372 373 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 374 mod_zone_page_state(page_zone(vm->pages[i]), 375 NR_KERNEL_STACK_KB, 376 PAGE_SIZE / 1024 * account); 377 } 378 } else { 379 /* 380 * All stack pages are in the same zone and belong to the 381 * same memcg. 382 */ 383 struct page *first_page = virt_to_page(stack); 384 385 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB, 386 THREAD_SIZE / 1024 * account); 387 388 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB, 389 account * (THREAD_SIZE / 1024)); 390 } 391 } 392 393 static int memcg_charge_kernel_stack(struct task_struct *tsk) 394 { 395 #ifdef CONFIG_VMAP_STACK 396 struct vm_struct *vm = task_stack_vm_area(tsk); 397 int ret; 398 399 if (vm) { 400 int i; 401 402 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 403 /* 404 * If memcg_kmem_charge() fails, page->mem_cgroup 405 * pointer is NULL, and both memcg_kmem_uncharge() 406 * and mod_memcg_page_state() in free_thread_stack() 407 * will ignore this page. So it's safe. 408 */ 409 ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0); 410 if (ret) 411 return ret; 412 413 mod_memcg_page_state(vm->pages[i], 414 MEMCG_KERNEL_STACK_KB, 415 PAGE_SIZE / 1024); 416 } 417 } 418 #endif 419 return 0; 420 } 421 422 static void release_task_stack(struct task_struct *tsk) 423 { 424 if (WARN_ON(tsk->state != TASK_DEAD)) 425 return; /* Better to leak the stack than to free prematurely */ 426 427 account_kernel_stack(tsk, -1); 428 free_thread_stack(tsk); 429 tsk->stack = NULL; 430 #ifdef CONFIG_VMAP_STACK 431 tsk->stack_vm_area = NULL; 432 #endif 433 } 434 435 #ifdef CONFIG_THREAD_INFO_IN_TASK 436 void put_task_stack(struct task_struct *tsk) 437 { 438 if (refcount_dec_and_test(&tsk->stack_refcount)) 439 release_task_stack(tsk); 440 } 441 #endif 442 443 void free_task(struct task_struct *tsk) 444 { 445 #ifndef CONFIG_THREAD_INFO_IN_TASK 446 /* 447 * The task is finally done with both the stack and thread_info, 448 * so free both. 449 */ 450 release_task_stack(tsk); 451 #else 452 /* 453 * If the task had a separate stack allocation, it should be gone 454 * by now. 455 */ 456 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0); 457 #endif 458 rt_mutex_debug_task_free(tsk); 459 ftrace_graph_exit_task(tsk); 460 put_seccomp_filter(tsk); 461 arch_release_task_struct(tsk); 462 if (tsk->flags & PF_KTHREAD) 463 free_kthread_struct(tsk); 464 free_task_struct(tsk); 465 } 466 EXPORT_SYMBOL(free_task); 467 468 #ifdef CONFIG_MMU 469 static __latent_entropy int dup_mmap(struct mm_struct *mm, 470 struct mm_struct *oldmm) 471 { 472 struct vm_area_struct *mpnt, *tmp, *prev, **pprev; 473 struct rb_node **rb_link, *rb_parent; 474 int retval; 475 unsigned long charge; 476 LIST_HEAD(uf); 477 478 uprobe_start_dup_mmap(); 479 if (down_write_killable(&oldmm->mmap_sem)) { 480 retval = -EINTR; 481 goto fail_uprobe_end; 482 } 483 flush_cache_dup_mm(oldmm); 484 uprobe_dup_mmap(oldmm, mm); 485 /* 486 * Not linked in yet - no deadlock potential: 487 */ 488 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING); 489 490 /* No ordering required: file already has been exposed. */ 491 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm)); 492 493 mm->total_vm = oldmm->total_vm; 494 mm->data_vm = oldmm->data_vm; 495 mm->exec_vm = oldmm->exec_vm; 496 mm->stack_vm = oldmm->stack_vm; 497 498 rb_link = &mm->mm_rb.rb_node; 499 rb_parent = NULL; 500 pprev = &mm->mmap; 501 retval = ksm_fork(mm, oldmm); 502 if (retval) 503 goto out; 504 retval = khugepaged_fork(mm, oldmm); 505 if (retval) 506 goto out; 507 508 prev = NULL; 509 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { 510 struct file *file; 511 512 if (mpnt->vm_flags & VM_DONTCOPY) { 513 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt)); 514 continue; 515 } 516 charge = 0; 517 /* 518 * Don't duplicate many vmas if we've been oom-killed (for 519 * example) 520 */ 521 if (fatal_signal_pending(current)) { 522 retval = -EINTR; 523 goto out; 524 } 525 if (mpnt->vm_flags & VM_ACCOUNT) { 526 unsigned long len = vma_pages(mpnt); 527 528 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */ 529 goto fail_nomem; 530 charge = len; 531 } 532 tmp = vm_area_dup(mpnt); 533 if (!tmp) 534 goto fail_nomem; 535 retval = vma_dup_policy(mpnt, tmp); 536 if (retval) 537 goto fail_nomem_policy; 538 tmp->vm_mm = mm; 539 retval = dup_userfaultfd(tmp, &uf); 540 if (retval) 541 goto fail_nomem_anon_vma_fork; 542 if (tmp->vm_flags & VM_WIPEONFORK) { 543 /* VM_WIPEONFORK gets a clean slate in the child. */ 544 tmp->anon_vma = NULL; 545 if (anon_vma_prepare(tmp)) 546 goto fail_nomem_anon_vma_fork; 547 } else if (anon_vma_fork(tmp, mpnt)) 548 goto fail_nomem_anon_vma_fork; 549 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT); 550 tmp->vm_next = tmp->vm_prev = NULL; 551 file = tmp->vm_file; 552 if (file) { 553 struct inode *inode = file_inode(file); 554 struct address_space *mapping = file->f_mapping; 555 556 get_file(file); 557 if (tmp->vm_flags & VM_DENYWRITE) 558 atomic_dec(&inode->i_writecount); 559 i_mmap_lock_write(mapping); 560 if (tmp->vm_flags & VM_SHARED) 561 atomic_inc(&mapping->i_mmap_writable); 562 flush_dcache_mmap_lock(mapping); 563 /* insert tmp into the share list, just after mpnt */ 564 vma_interval_tree_insert_after(tmp, mpnt, 565 &mapping->i_mmap); 566 flush_dcache_mmap_unlock(mapping); 567 i_mmap_unlock_write(mapping); 568 } 569 570 /* 571 * Clear hugetlb-related page reserves for children. This only 572 * affects MAP_PRIVATE mappings. Faults generated by the child 573 * are not guaranteed to succeed, even if read-only 574 */ 575 if (is_vm_hugetlb_page(tmp)) 576 reset_vma_resv_huge_pages(tmp); 577 578 /* 579 * Link in the new vma and copy the page table entries. 580 */ 581 *pprev = tmp; 582 pprev = &tmp->vm_next; 583 tmp->vm_prev = prev; 584 prev = tmp; 585 586 __vma_link_rb(mm, tmp, rb_link, rb_parent); 587 rb_link = &tmp->vm_rb.rb_right; 588 rb_parent = &tmp->vm_rb; 589 590 mm->map_count++; 591 if (!(tmp->vm_flags & VM_WIPEONFORK)) 592 retval = copy_page_range(mm, oldmm, mpnt); 593 594 if (tmp->vm_ops && tmp->vm_ops->open) 595 tmp->vm_ops->open(tmp); 596 597 if (retval) 598 goto out; 599 } 600 /* a new mm has just been created */ 601 retval = arch_dup_mmap(oldmm, mm); 602 out: 603 up_write(&mm->mmap_sem); 604 flush_tlb_mm(oldmm); 605 up_write(&oldmm->mmap_sem); 606 dup_userfaultfd_complete(&uf); 607 fail_uprobe_end: 608 uprobe_end_dup_mmap(); 609 return retval; 610 fail_nomem_anon_vma_fork: 611 mpol_put(vma_policy(tmp)); 612 fail_nomem_policy: 613 vm_area_free(tmp); 614 fail_nomem: 615 retval = -ENOMEM; 616 vm_unacct_memory(charge); 617 goto out; 618 } 619 620 static inline int mm_alloc_pgd(struct mm_struct *mm) 621 { 622 mm->pgd = pgd_alloc(mm); 623 if (unlikely(!mm->pgd)) 624 return -ENOMEM; 625 return 0; 626 } 627 628 static inline void mm_free_pgd(struct mm_struct *mm) 629 { 630 pgd_free(mm, mm->pgd); 631 } 632 #else 633 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) 634 { 635 down_write(&oldmm->mmap_sem); 636 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm)); 637 up_write(&oldmm->mmap_sem); 638 return 0; 639 } 640 #define mm_alloc_pgd(mm) (0) 641 #define mm_free_pgd(mm) 642 #endif /* CONFIG_MMU */ 643 644 static void check_mm(struct mm_struct *mm) 645 { 646 int i; 647 648 for (i = 0; i < NR_MM_COUNTERS; i++) { 649 long x = atomic_long_read(&mm->rss_stat.count[i]); 650 651 if (unlikely(x)) 652 printk(KERN_ALERT "BUG: Bad rss-counter state " 653 "mm:%p idx:%d val:%ld\n", mm, i, x); 654 } 655 656 if (mm_pgtables_bytes(mm)) 657 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n", 658 mm_pgtables_bytes(mm)); 659 660 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 661 VM_BUG_ON_MM(mm->pmd_huge_pte, mm); 662 #endif 663 } 664 665 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL)) 666 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) 667 668 /* 669 * Called when the last reference to the mm 670 * is dropped: either by a lazy thread or by 671 * mmput. Free the page directory and the mm. 672 */ 673 void __mmdrop(struct mm_struct *mm) 674 { 675 BUG_ON(mm == &init_mm); 676 WARN_ON_ONCE(mm == current->mm); 677 WARN_ON_ONCE(mm == current->active_mm); 678 mm_free_pgd(mm); 679 destroy_context(mm); 680 hmm_mm_destroy(mm); 681 mmu_notifier_mm_destroy(mm); 682 check_mm(mm); 683 put_user_ns(mm->user_ns); 684 free_mm(mm); 685 } 686 EXPORT_SYMBOL_GPL(__mmdrop); 687 688 static void mmdrop_async_fn(struct work_struct *work) 689 { 690 struct mm_struct *mm; 691 692 mm = container_of(work, struct mm_struct, async_put_work); 693 __mmdrop(mm); 694 } 695 696 static void mmdrop_async(struct mm_struct *mm) 697 { 698 if (unlikely(atomic_dec_and_test(&mm->mm_count))) { 699 INIT_WORK(&mm->async_put_work, mmdrop_async_fn); 700 schedule_work(&mm->async_put_work); 701 } 702 } 703 704 static inline void free_signal_struct(struct signal_struct *sig) 705 { 706 taskstats_tgid_free(sig); 707 sched_autogroup_exit(sig); 708 /* 709 * __mmdrop is not safe to call from softirq context on x86 due to 710 * pgd_dtor so postpone it to the async context 711 */ 712 if (sig->oom_mm) 713 mmdrop_async(sig->oom_mm); 714 kmem_cache_free(signal_cachep, sig); 715 } 716 717 static inline void put_signal_struct(struct signal_struct *sig) 718 { 719 if (refcount_dec_and_test(&sig->sigcnt)) 720 free_signal_struct(sig); 721 } 722 723 void __put_task_struct(struct task_struct *tsk) 724 { 725 WARN_ON(!tsk->exit_state); 726 WARN_ON(refcount_read(&tsk->usage)); 727 WARN_ON(tsk == current); 728 729 cgroup_free(tsk); 730 task_numa_free(tsk); 731 security_task_free(tsk); 732 exit_creds(tsk); 733 delayacct_tsk_free(tsk); 734 put_signal_struct(tsk->signal); 735 736 if (!profile_handoff_task(tsk)) 737 free_task(tsk); 738 } 739 EXPORT_SYMBOL_GPL(__put_task_struct); 740 741 void __init __weak arch_task_cache_init(void) { } 742 743 /* 744 * set_max_threads 745 */ 746 static void set_max_threads(unsigned int max_threads_suggested) 747 { 748 u64 threads; 749 unsigned long nr_pages = totalram_pages(); 750 751 /* 752 * The number of threads shall be limited such that the thread 753 * structures may only consume a small part of the available memory. 754 */ 755 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64) 756 threads = MAX_THREADS; 757 else 758 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE, 759 (u64) THREAD_SIZE * 8UL); 760 761 if (threads > max_threads_suggested) 762 threads = max_threads_suggested; 763 764 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS); 765 } 766 767 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT 768 /* Initialized by the architecture: */ 769 int arch_task_struct_size __read_mostly; 770 #endif 771 772 static void task_struct_whitelist(unsigned long *offset, unsigned long *size) 773 { 774 /* Fetch thread_struct whitelist for the architecture. */ 775 arch_thread_struct_whitelist(offset, size); 776 777 /* 778 * Handle zero-sized whitelist or empty thread_struct, otherwise 779 * adjust offset to position of thread_struct in task_struct. 780 */ 781 if (unlikely(*size == 0)) 782 *offset = 0; 783 else 784 *offset += offsetof(struct task_struct, thread); 785 } 786 787 void __init fork_init(void) 788 { 789 int i; 790 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 791 #ifndef ARCH_MIN_TASKALIGN 792 #define ARCH_MIN_TASKALIGN 0 793 #endif 794 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN); 795 unsigned long useroffset, usersize; 796 797 /* create a slab on which task_structs can be allocated */ 798 task_struct_whitelist(&useroffset, &usersize); 799 task_struct_cachep = kmem_cache_create_usercopy("task_struct", 800 arch_task_struct_size, align, 801 SLAB_PANIC|SLAB_ACCOUNT, 802 useroffset, usersize, NULL); 803 #endif 804 805 /* do the arch specific task caches init */ 806 arch_task_cache_init(); 807 808 set_max_threads(MAX_THREADS); 809 810 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 811 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 812 init_task.signal->rlim[RLIMIT_SIGPENDING] = 813 init_task.signal->rlim[RLIMIT_NPROC]; 814 815 for (i = 0; i < UCOUNT_COUNTS; i++) { 816 init_user_ns.ucount_max[i] = max_threads/2; 817 } 818 819 #ifdef CONFIG_VMAP_STACK 820 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache", 821 NULL, free_vm_stack_cache); 822 #endif 823 824 lockdep_init_task(&init_task); 825 uprobes_init(); 826 } 827 828 int __weak arch_dup_task_struct(struct task_struct *dst, 829 struct task_struct *src) 830 { 831 *dst = *src; 832 return 0; 833 } 834 835 void set_task_stack_end_magic(struct task_struct *tsk) 836 { 837 unsigned long *stackend; 838 839 stackend = end_of_stack(tsk); 840 *stackend = STACK_END_MAGIC; /* for overflow detection */ 841 } 842 843 static struct task_struct *dup_task_struct(struct task_struct *orig, int node) 844 { 845 struct task_struct *tsk; 846 unsigned long *stack; 847 struct vm_struct *stack_vm_area __maybe_unused; 848 int err; 849 850 if (node == NUMA_NO_NODE) 851 node = tsk_fork_get_node(orig); 852 tsk = alloc_task_struct_node(node); 853 if (!tsk) 854 return NULL; 855 856 stack = alloc_thread_stack_node(tsk, node); 857 if (!stack) 858 goto free_tsk; 859 860 if (memcg_charge_kernel_stack(tsk)) 861 goto free_stack; 862 863 stack_vm_area = task_stack_vm_area(tsk); 864 865 err = arch_dup_task_struct(tsk, orig); 866 867 /* 868 * arch_dup_task_struct() clobbers the stack-related fields. Make 869 * sure they're properly initialized before using any stack-related 870 * functions again. 871 */ 872 tsk->stack = stack; 873 #ifdef CONFIG_VMAP_STACK 874 tsk->stack_vm_area = stack_vm_area; 875 #endif 876 #ifdef CONFIG_THREAD_INFO_IN_TASK 877 refcount_set(&tsk->stack_refcount, 1); 878 #endif 879 880 if (err) 881 goto free_stack; 882 883 #ifdef CONFIG_SECCOMP 884 /* 885 * We must handle setting up seccomp filters once we're under 886 * the sighand lock in case orig has changed between now and 887 * then. Until then, filter must be NULL to avoid messing up 888 * the usage counts on the error path calling free_task. 889 */ 890 tsk->seccomp.filter = NULL; 891 #endif 892 893 setup_thread_stack(tsk, orig); 894 clear_user_return_notifier(tsk); 895 clear_tsk_need_resched(tsk); 896 set_task_stack_end_magic(tsk); 897 898 #ifdef CONFIG_STACKPROTECTOR 899 tsk->stack_canary = get_random_canary(); 900 #endif 901 if (orig->cpus_ptr == &orig->cpus_mask) 902 tsk->cpus_ptr = &tsk->cpus_mask; 903 904 /* 905 * One for us, one for whoever does the "release_task()" (usually 906 * parent) 907 */ 908 refcount_set(&tsk->usage, 2); 909 #ifdef CONFIG_BLK_DEV_IO_TRACE 910 tsk->btrace_seq = 0; 911 #endif 912 tsk->splice_pipe = NULL; 913 tsk->task_frag.page = NULL; 914 tsk->wake_q.next = NULL; 915 916 account_kernel_stack(tsk, 1); 917 918 kcov_task_init(tsk); 919 920 #ifdef CONFIG_FAULT_INJECTION 921 tsk->fail_nth = 0; 922 #endif 923 924 #ifdef CONFIG_BLK_CGROUP 925 tsk->throttle_queue = NULL; 926 tsk->use_memdelay = 0; 927 #endif 928 929 #ifdef CONFIG_MEMCG 930 tsk->active_memcg = NULL; 931 #endif 932 return tsk; 933 934 free_stack: 935 free_thread_stack(tsk); 936 free_tsk: 937 free_task_struct(tsk); 938 return NULL; 939 } 940 941 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); 942 943 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT; 944 945 static int __init coredump_filter_setup(char *s) 946 { 947 default_dump_filter = 948 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) & 949 MMF_DUMP_FILTER_MASK; 950 return 1; 951 } 952 953 __setup("coredump_filter=", coredump_filter_setup); 954 955 #include <linux/init_task.h> 956 957 static void mm_init_aio(struct mm_struct *mm) 958 { 959 #ifdef CONFIG_AIO 960 spin_lock_init(&mm->ioctx_lock); 961 mm->ioctx_table = NULL; 962 #endif 963 } 964 965 static __always_inline void mm_clear_owner(struct mm_struct *mm, 966 struct task_struct *p) 967 { 968 #ifdef CONFIG_MEMCG 969 if (mm->owner == p) 970 WRITE_ONCE(mm->owner, NULL); 971 #endif 972 } 973 974 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p) 975 { 976 #ifdef CONFIG_MEMCG 977 mm->owner = p; 978 #endif 979 } 980 981 static void mm_init_uprobes_state(struct mm_struct *mm) 982 { 983 #ifdef CONFIG_UPROBES 984 mm->uprobes_state.xol_area = NULL; 985 #endif 986 } 987 988 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p, 989 struct user_namespace *user_ns) 990 { 991 mm->mmap = NULL; 992 mm->mm_rb = RB_ROOT; 993 mm->vmacache_seqnum = 0; 994 atomic_set(&mm->mm_users, 1); 995 atomic_set(&mm->mm_count, 1); 996 init_rwsem(&mm->mmap_sem); 997 INIT_LIST_HEAD(&mm->mmlist); 998 mm->core_state = NULL; 999 mm_pgtables_bytes_init(mm); 1000 mm->map_count = 0; 1001 mm->locked_vm = 0; 1002 atomic64_set(&mm->pinned_vm, 0); 1003 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat)); 1004 spin_lock_init(&mm->page_table_lock); 1005 spin_lock_init(&mm->arg_lock); 1006 mm_init_cpumask(mm); 1007 mm_init_aio(mm); 1008 mm_init_owner(mm, p); 1009 RCU_INIT_POINTER(mm->exe_file, NULL); 1010 mmu_notifier_mm_init(mm); 1011 hmm_mm_init(mm); 1012 init_tlb_flush_pending(mm); 1013 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 1014 mm->pmd_huge_pte = NULL; 1015 #endif 1016 mm_init_uprobes_state(mm); 1017 1018 if (current->mm) { 1019 mm->flags = current->mm->flags & MMF_INIT_MASK; 1020 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK; 1021 } else { 1022 mm->flags = default_dump_filter; 1023 mm->def_flags = 0; 1024 } 1025 1026 if (mm_alloc_pgd(mm)) 1027 goto fail_nopgd; 1028 1029 if (init_new_context(p, mm)) 1030 goto fail_nocontext; 1031 1032 mm->user_ns = get_user_ns(user_ns); 1033 return mm; 1034 1035 fail_nocontext: 1036 mm_free_pgd(mm); 1037 fail_nopgd: 1038 free_mm(mm); 1039 return NULL; 1040 } 1041 1042 /* 1043 * Allocate and initialize an mm_struct. 1044 */ 1045 struct mm_struct *mm_alloc(void) 1046 { 1047 struct mm_struct *mm; 1048 1049 mm = allocate_mm(); 1050 if (!mm) 1051 return NULL; 1052 1053 memset(mm, 0, sizeof(*mm)); 1054 return mm_init(mm, current, current_user_ns()); 1055 } 1056 1057 static inline void __mmput(struct mm_struct *mm) 1058 { 1059 VM_BUG_ON(atomic_read(&mm->mm_users)); 1060 1061 uprobe_clear_state(mm); 1062 exit_aio(mm); 1063 ksm_exit(mm); 1064 khugepaged_exit(mm); /* must run before exit_mmap */ 1065 exit_mmap(mm); 1066 mm_put_huge_zero_page(mm); 1067 set_mm_exe_file(mm, NULL); 1068 if (!list_empty(&mm->mmlist)) { 1069 spin_lock(&mmlist_lock); 1070 list_del(&mm->mmlist); 1071 spin_unlock(&mmlist_lock); 1072 } 1073 if (mm->binfmt) 1074 module_put(mm->binfmt->module); 1075 mmdrop(mm); 1076 } 1077 1078 /* 1079 * Decrement the use count and release all resources for an mm. 1080 */ 1081 void mmput(struct mm_struct *mm) 1082 { 1083 might_sleep(); 1084 1085 if (atomic_dec_and_test(&mm->mm_users)) 1086 __mmput(mm); 1087 } 1088 EXPORT_SYMBOL_GPL(mmput); 1089 1090 #ifdef CONFIG_MMU 1091 static void mmput_async_fn(struct work_struct *work) 1092 { 1093 struct mm_struct *mm = container_of(work, struct mm_struct, 1094 async_put_work); 1095 1096 __mmput(mm); 1097 } 1098 1099 void mmput_async(struct mm_struct *mm) 1100 { 1101 if (atomic_dec_and_test(&mm->mm_users)) { 1102 INIT_WORK(&mm->async_put_work, mmput_async_fn); 1103 schedule_work(&mm->async_put_work); 1104 } 1105 } 1106 #endif 1107 1108 /** 1109 * set_mm_exe_file - change a reference to the mm's executable file 1110 * 1111 * This changes mm's executable file (shown as symlink /proc/[pid]/exe). 1112 * 1113 * Main users are mmput() and sys_execve(). Callers prevent concurrent 1114 * invocations: in mmput() nobody alive left, in execve task is single 1115 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the 1116 * mm->exe_file, but does so without using set_mm_exe_file() in order 1117 * to do avoid the need for any locks. 1118 */ 1119 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 1120 { 1121 struct file *old_exe_file; 1122 1123 /* 1124 * It is safe to dereference the exe_file without RCU as 1125 * this function is only called if nobody else can access 1126 * this mm -- see comment above for justification. 1127 */ 1128 old_exe_file = rcu_dereference_raw(mm->exe_file); 1129 1130 if (new_exe_file) 1131 get_file(new_exe_file); 1132 rcu_assign_pointer(mm->exe_file, new_exe_file); 1133 if (old_exe_file) 1134 fput(old_exe_file); 1135 } 1136 1137 /** 1138 * get_mm_exe_file - acquire a reference to the mm's executable file 1139 * 1140 * Returns %NULL if mm has no associated executable file. 1141 * User must release file via fput(). 1142 */ 1143 struct file *get_mm_exe_file(struct mm_struct *mm) 1144 { 1145 struct file *exe_file; 1146 1147 rcu_read_lock(); 1148 exe_file = rcu_dereference(mm->exe_file); 1149 if (exe_file && !get_file_rcu(exe_file)) 1150 exe_file = NULL; 1151 rcu_read_unlock(); 1152 return exe_file; 1153 } 1154 EXPORT_SYMBOL(get_mm_exe_file); 1155 1156 /** 1157 * get_task_exe_file - acquire a reference to the task's executable file 1158 * 1159 * Returns %NULL if task's mm (if any) has no associated executable file or 1160 * this is a kernel thread with borrowed mm (see the comment above get_task_mm). 1161 * User must release file via fput(). 1162 */ 1163 struct file *get_task_exe_file(struct task_struct *task) 1164 { 1165 struct file *exe_file = NULL; 1166 struct mm_struct *mm; 1167 1168 task_lock(task); 1169 mm = task->mm; 1170 if (mm) { 1171 if (!(task->flags & PF_KTHREAD)) 1172 exe_file = get_mm_exe_file(mm); 1173 } 1174 task_unlock(task); 1175 return exe_file; 1176 } 1177 EXPORT_SYMBOL(get_task_exe_file); 1178 1179 /** 1180 * get_task_mm - acquire a reference to the task's mm 1181 * 1182 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning 1183 * this kernel workthread has transiently adopted a user mm with use_mm, 1184 * to do its AIO) is not set and if so returns a reference to it, after 1185 * bumping up the use count. User must release the mm via mmput() 1186 * after use. Typically used by /proc and ptrace. 1187 */ 1188 struct mm_struct *get_task_mm(struct task_struct *task) 1189 { 1190 struct mm_struct *mm; 1191 1192 task_lock(task); 1193 mm = task->mm; 1194 if (mm) { 1195 if (task->flags & PF_KTHREAD) 1196 mm = NULL; 1197 else 1198 mmget(mm); 1199 } 1200 task_unlock(task); 1201 return mm; 1202 } 1203 EXPORT_SYMBOL_GPL(get_task_mm); 1204 1205 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode) 1206 { 1207 struct mm_struct *mm; 1208 int err; 1209 1210 err = mutex_lock_killable(&task->signal->cred_guard_mutex); 1211 if (err) 1212 return ERR_PTR(err); 1213 1214 mm = get_task_mm(task); 1215 if (mm && mm != current->mm && 1216 !ptrace_may_access(task, mode)) { 1217 mmput(mm); 1218 mm = ERR_PTR(-EACCES); 1219 } 1220 mutex_unlock(&task->signal->cred_guard_mutex); 1221 1222 return mm; 1223 } 1224 1225 static void complete_vfork_done(struct task_struct *tsk) 1226 { 1227 struct completion *vfork; 1228 1229 task_lock(tsk); 1230 vfork = tsk->vfork_done; 1231 if (likely(vfork)) { 1232 tsk->vfork_done = NULL; 1233 complete(vfork); 1234 } 1235 task_unlock(tsk); 1236 } 1237 1238 static int wait_for_vfork_done(struct task_struct *child, 1239 struct completion *vfork) 1240 { 1241 int killed; 1242 1243 freezer_do_not_count(); 1244 cgroup_enter_frozen(); 1245 killed = wait_for_completion_killable(vfork); 1246 cgroup_leave_frozen(false); 1247 freezer_count(); 1248 1249 if (killed) { 1250 task_lock(child); 1251 child->vfork_done = NULL; 1252 task_unlock(child); 1253 } 1254 1255 put_task_struct(child); 1256 return killed; 1257 } 1258 1259 /* Please note the differences between mmput and mm_release. 1260 * mmput is called whenever we stop holding onto a mm_struct, 1261 * error success whatever. 1262 * 1263 * mm_release is called after a mm_struct has been removed 1264 * from the current process. 1265 * 1266 * This difference is important for error handling, when we 1267 * only half set up a mm_struct for a new process and need to restore 1268 * the old one. Because we mmput the new mm_struct before 1269 * restoring the old one. . . 1270 * Eric Biederman 10 January 1998 1271 */ 1272 void mm_release(struct task_struct *tsk, struct mm_struct *mm) 1273 { 1274 /* Get rid of any futexes when releasing the mm */ 1275 #ifdef CONFIG_FUTEX 1276 if (unlikely(tsk->robust_list)) { 1277 exit_robust_list(tsk); 1278 tsk->robust_list = NULL; 1279 } 1280 #ifdef CONFIG_COMPAT 1281 if (unlikely(tsk->compat_robust_list)) { 1282 compat_exit_robust_list(tsk); 1283 tsk->compat_robust_list = NULL; 1284 } 1285 #endif 1286 if (unlikely(!list_empty(&tsk->pi_state_list))) 1287 exit_pi_state_list(tsk); 1288 #endif 1289 1290 uprobe_free_utask(tsk); 1291 1292 /* Get rid of any cached register state */ 1293 deactivate_mm(tsk, mm); 1294 1295 /* 1296 * Signal userspace if we're not exiting with a core dump 1297 * because we want to leave the value intact for debugging 1298 * purposes. 1299 */ 1300 if (tsk->clear_child_tid) { 1301 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) && 1302 atomic_read(&mm->mm_users) > 1) { 1303 /* 1304 * We don't check the error code - if userspace has 1305 * not set up a proper pointer then tough luck. 1306 */ 1307 put_user(0, tsk->clear_child_tid); 1308 do_futex(tsk->clear_child_tid, FUTEX_WAKE, 1309 1, NULL, NULL, 0, 0); 1310 } 1311 tsk->clear_child_tid = NULL; 1312 } 1313 1314 /* 1315 * All done, finally we can wake up parent and return this mm to him. 1316 * Also kthread_stop() uses this completion for synchronization. 1317 */ 1318 if (tsk->vfork_done) 1319 complete_vfork_done(tsk); 1320 } 1321 1322 /** 1323 * dup_mm() - duplicates an existing mm structure 1324 * @tsk: the task_struct with which the new mm will be associated. 1325 * @oldmm: the mm to duplicate. 1326 * 1327 * Allocates a new mm structure and duplicates the provided @oldmm structure 1328 * content into it. 1329 * 1330 * Return: the duplicated mm or NULL on failure. 1331 */ 1332 static struct mm_struct *dup_mm(struct task_struct *tsk, 1333 struct mm_struct *oldmm) 1334 { 1335 struct mm_struct *mm; 1336 int err; 1337 1338 mm = allocate_mm(); 1339 if (!mm) 1340 goto fail_nomem; 1341 1342 memcpy(mm, oldmm, sizeof(*mm)); 1343 1344 if (!mm_init(mm, tsk, mm->user_ns)) 1345 goto fail_nomem; 1346 1347 err = dup_mmap(mm, oldmm); 1348 if (err) 1349 goto free_pt; 1350 1351 mm->hiwater_rss = get_mm_rss(mm); 1352 mm->hiwater_vm = mm->total_vm; 1353 1354 if (mm->binfmt && !try_module_get(mm->binfmt->module)) 1355 goto free_pt; 1356 1357 return mm; 1358 1359 free_pt: 1360 /* don't put binfmt in mmput, we haven't got module yet */ 1361 mm->binfmt = NULL; 1362 mm_init_owner(mm, NULL); 1363 mmput(mm); 1364 1365 fail_nomem: 1366 return NULL; 1367 } 1368 1369 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk) 1370 { 1371 struct mm_struct *mm, *oldmm; 1372 int retval; 1373 1374 tsk->min_flt = tsk->maj_flt = 0; 1375 tsk->nvcsw = tsk->nivcsw = 0; 1376 #ifdef CONFIG_DETECT_HUNG_TASK 1377 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; 1378 tsk->last_switch_time = 0; 1379 #endif 1380 1381 tsk->mm = NULL; 1382 tsk->active_mm = NULL; 1383 1384 /* 1385 * Are we cloning a kernel thread? 1386 * 1387 * We need to steal a active VM for that.. 1388 */ 1389 oldmm = current->mm; 1390 if (!oldmm) 1391 return 0; 1392 1393 /* initialize the new vmacache entries */ 1394 vmacache_flush(tsk); 1395 1396 if (clone_flags & CLONE_VM) { 1397 mmget(oldmm); 1398 mm = oldmm; 1399 goto good_mm; 1400 } 1401 1402 retval = -ENOMEM; 1403 mm = dup_mm(tsk, current->mm); 1404 if (!mm) 1405 goto fail_nomem; 1406 1407 good_mm: 1408 tsk->mm = mm; 1409 tsk->active_mm = mm; 1410 return 0; 1411 1412 fail_nomem: 1413 return retval; 1414 } 1415 1416 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 1417 { 1418 struct fs_struct *fs = current->fs; 1419 if (clone_flags & CLONE_FS) { 1420 /* tsk->fs is already what we want */ 1421 spin_lock(&fs->lock); 1422 if (fs->in_exec) { 1423 spin_unlock(&fs->lock); 1424 return -EAGAIN; 1425 } 1426 fs->users++; 1427 spin_unlock(&fs->lock); 1428 return 0; 1429 } 1430 tsk->fs = copy_fs_struct(fs); 1431 if (!tsk->fs) 1432 return -ENOMEM; 1433 return 0; 1434 } 1435 1436 static int copy_files(unsigned long clone_flags, struct task_struct *tsk) 1437 { 1438 struct files_struct *oldf, *newf; 1439 int error = 0; 1440 1441 /* 1442 * A background process may not have any files ... 1443 */ 1444 oldf = current->files; 1445 if (!oldf) 1446 goto out; 1447 1448 if (clone_flags & CLONE_FILES) { 1449 atomic_inc(&oldf->count); 1450 goto out; 1451 } 1452 1453 newf = dup_fd(oldf, &error); 1454 if (!newf) 1455 goto out; 1456 1457 tsk->files = newf; 1458 error = 0; 1459 out: 1460 return error; 1461 } 1462 1463 static int copy_io(unsigned long clone_flags, struct task_struct *tsk) 1464 { 1465 #ifdef CONFIG_BLOCK 1466 struct io_context *ioc = current->io_context; 1467 struct io_context *new_ioc; 1468 1469 if (!ioc) 1470 return 0; 1471 /* 1472 * Share io context with parent, if CLONE_IO is set 1473 */ 1474 if (clone_flags & CLONE_IO) { 1475 ioc_task_link(ioc); 1476 tsk->io_context = ioc; 1477 } else if (ioprio_valid(ioc->ioprio)) { 1478 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE); 1479 if (unlikely(!new_ioc)) 1480 return -ENOMEM; 1481 1482 new_ioc->ioprio = ioc->ioprio; 1483 put_io_context(new_ioc); 1484 } 1485 #endif 1486 return 0; 1487 } 1488 1489 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk) 1490 { 1491 struct sighand_struct *sig; 1492 1493 if (clone_flags & CLONE_SIGHAND) { 1494 refcount_inc(¤t->sighand->count); 1495 return 0; 1496 } 1497 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 1498 rcu_assign_pointer(tsk->sighand, sig); 1499 if (!sig) 1500 return -ENOMEM; 1501 1502 refcount_set(&sig->count, 1); 1503 spin_lock_irq(¤t->sighand->siglock); 1504 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 1505 spin_unlock_irq(¤t->sighand->siglock); 1506 return 0; 1507 } 1508 1509 void __cleanup_sighand(struct sighand_struct *sighand) 1510 { 1511 if (refcount_dec_and_test(&sighand->count)) { 1512 signalfd_cleanup(sighand); 1513 /* 1514 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it 1515 * without an RCU grace period, see __lock_task_sighand(). 1516 */ 1517 kmem_cache_free(sighand_cachep, sighand); 1518 } 1519 } 1520 1521 #ifdef CONFIG_POSIX_TIMERS 1522 /* 1523 * Initialize POSIX timer handling for a thread group. 1524 */ 1525 static void posix_cpu_timers_init_group(struct signal_struct *sig) 1526 { 1527 unsigned long cpu_limit; 1528 1529 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); 1530 if (cpu_limit != RLIM_INFINITY) { 1531 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC; 1532 sig->cputimer.running = true; 1533 } 1534 1535 /* The timer lists. */ 1536 INIT_LIST_HEAD(&sig->cpu_timers[0]); 1537 INIT_LIST_HEAD(&sig->cpu_timers[1]); 1538 INIT_LIST_HEAD(&sig->cpu_timers[2]); 1539 } 1540 #else 1541 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { } 1542 #endif 1543 1544 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk) 1545 { 1546 struct signal_struct *sig; 1547 1548 if (clone_flags & CLONE_THREAD) 1549 return 0; 1550 1551 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL); 1552 tsk->signal = sig; 1553 if (!sig) 1554 return -ENOMEM; 1555 1556 sig->nr_threads = 1; 1557 atomic_set(&sig->live, 1); 1558 refcount_set(&sig->sigcnt, 1); 1559 1560 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */ 1561 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node); 1562 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head); 1563 1564 init_waitqueue_head(&sig->wait_chldexit); 1565 sig->curr_target = tsk; 1566 init_sigpending(&sig->shared_pending); 1567 INIT_HLIST_HEAD(&sig->multiprocess); 1568 seqlock_init(&sig->stats_lock); 1569 prev_cputime_init(&sig->prev_cputime); 1570 1571 #ifdef CONFIG_POSIX_TIMERS 1572 INIT_LIST_HEAD(&sig->posix_timers); 1573 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1574 sig->real_timer.function = it_real_fn; 1575 #endif 1576 1577 task_lock(current->group_leader); 1578 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 1579 task_unlock(current->group_leader); 1580 1581 posix_cpu_timers_init_group(sig); 1582 1583 tty_audit_fork(sig); 1584 sched_autogroup_fork(sig); 1585 1586 sig->oom_score_adj = current->signal->oom_score_adj; 1587 sig->oom_score_adj_min = current->signal->oom_score_adj_min; 1588 1589 mutex_init(&sig->cred_guard_mutex); 1590 1591 return 0; 1592 } 1593 1594 static void copy_seccomp(struct task_struct *p) 1595 { 1596 #ifdef CONFIG_SECCOMP 1597 /* 1598 * Must be called with sighand->lock held, which is common to 1599 * all threads in the group. Holding cred_guard_mutex is not 1600 * needed because this new task is not yet running and cannot 1601 * be racing exec. 1602 */ 1603 assert_spin_locked(¤t->sighand->siglock); 1604 1605 /* Ref-count the new filter user, and assign it. */ 1606 get_seccomp_filter(current); 1607 p->seccomp = current->seccomp; 1608 1609 /* 1610 * Explicitly enable no_new_privs here in case it got set 1611 * between the task_struct being duplicated and holding the 1612 * sighand lock. The seccomp state and nnp must be in sync. 1613 */ 1614 if (task_no_new_privs(current)) 1615 task_set_no_new_privs(p); 1616 1617 /* 1618 * If the parent gained a seccomp mode after copying thread 1619 * flags and between before we held the sighand lock, we have 1620 * to manually enable the seccomp thread flag here. 1621 */ 1622 if (p->seccomp.mode != SECCOMP_MODE_DISABLED) 1623 set_tsk_thread_flag(p, TIF_SECCOMP); 1624 #endif 1625 } 1626 1627 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr) 1628 { 1629 current->clear_child_tid = tidptr; 1630 1631 return task_pid_vnr(current); 1632 } 1633 1634 static void rt_mutex_init_task(struct task_struct *p) 1635 { 1636 raw_spin_lock_init(&p->pi_lock); 1637 #ifdef CONFIG_RT_MUTEXES 1638 p->pi_waiters = RB_ROOT_CACHED; 1639 p->pi_top_task = NULL; 1640 p->pi_blocked_on = NULL; 1641 #endif 1642 } 1643 1644 #ifdef CONFIG_POSIX_TIMERS 1645 /* 1646 * Initialize POSIX timer handling for a single task. 1647 */ 1648 static void posix_cpu_timers_init(struct task_struct *tsk) 1649 { 1650 tsk->cputime_expires.prof_exp = 0; 1651 tsk->cputime_expires.virt_exp = 0; 1652 tsk->cputime_expires.sched_exp = 0; 1653 INIT_LIST_HEAD(&tsk->cpu_timers[0]); 1654 INIT_LIST_HEAD(&tsk->cpu_timers[1]); 1655 INIT_LIST_HEAD(&tsk->cpu_timers[2]); 1656 } 1657 #else 1658 static inline void posix_cpu_timers_init(struct task_struct *tsk) { } 1659 #endif 1660 1661 static inline void init_task_pid_links(struct task_struct *task) 1662 { 1663 enum pid_type type; 1664 1665 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 1666 INIT_HLIST_NODE(&task->pid_links[type]); 1667 } 1668 } 1669 1670 static inline void 1671 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid) 1672 { 1673 if (type == PIDTYPE_PID) 1674 task->thread_pid = pid; 1675 else 1676 task->signal->pids[type] = pid; 1677 } 1678 1679 static inline void rcu_copy_process(struct task_struct *p) 1680 { 1681 #ifdef CONFIG_PREEMPT_RCU 1682 p->rcu_read_lock_nesting = 0; 1683 p->rcu_read_unlock_special.s = 0; 1684 p->rcu_blocked_node = NULL; 1685 INIT_LIST_HEAD(&p->rcu_node_entry); 1686 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 1687 #ifdef CONFIG_TASKS_RCU 1688 p->rcu_tasks_holdout = false; 1689 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list); 1690 p->rcu_tasks_idle_cpu = -1; 1691 #endif /* #ifdef CONFIG_TASKS_RCU */ 1692 } 1693 1694 static int pidfd_release(struct inode *inode, struct file *file) 1695 { 1696 struct pid *pid = file->private_data; 1697 1698 file->private_data = NULL; 1699 put_pid(pid); 1700 return 0; 1701 } 1702 1703 #ifdef CONFIG_PROC_FS 1704 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f) 1705 { 1706 struct pid_namespace *ns = proc_pid_ns(file_inode(m->file)); 1707 struct pid *pid = f->private_data; 1708 1709 seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns)); 1710 seq_putc(m, '\n'); 1711 } 1712 #endif 1713 1714 /* 1715 * Poll support for process exit notification. 1716 */ 1717 static unsigned int pidfd_poll(struct file *file, struct poll_table_struct *pts) 1718 { 1719 struct task_struct *task; 1720 struct pid *pid = file->private_data; 1721 int poll_flags = 0; 1722 1723 poll_wait(file, &pid->wait_pidfd, pts); 1724 1725 rcu_read_lock(); 1726 task = pid_task(pid, PIDTYPE_PID); 1727 /* 1728 * Inform pollers only when the whole thread group exits. 1729 * If the thread group leader exits before all other threads in the 1730 * group, then poll(2) should block, similar to the wait(2) family. 1731 */ 1732 if (!task || (task->exit_state && thread_group_empty(task))) 1733 poll_flags = POLLIN | POLLRDNORM; 1734 rcu_read_unlock(); 1735 1736 return poll_flags; 1737 } 1738 1739 const struct file_operations pidfd_fops = { 1740 .release = pidfd_release, 1741 .poll = pidfd_poll, 1742 #ifdef CONFIG_PROC_FS 1743 .show_fdinfo = pidfd_show_fdinfo, 1744 #endif 1745 }; 1746 1747 static void __delayed_free_task(struct rcu_head *rhp) 1748 { 1749 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); 1750 1751 free_task(tsk); 1752 } 1753 1754 static __always_inline void delayed_free_task(struct task_struct *tsk) 1755 { 1756 if (IS_ENABLED(CONFIG_MEMCG)) 1757 call_rcu(&tsk->rcu, __delayed_free_task); 1758 else 1759 free_task(tsk); 1760 } 1761 1762 /* 1763 * This creates a new process as a copy of the old one, 1764 * but does not actually start it yet. 1765 * 1766 * It copies the registers, and all the appropriate 1767 * parts of the process environment (as per the clone 1768 * flags). The actual kick-off is left to the caller. 1769 */ 1770 static __latent_entropy struct task_struct *copy_process( 1771 struct pid *pid, 1772 int trace, 1773 int node, 1774 struct kernel_clone_args *args) 1775 { 1776 int pidfd = -1, retval; 1777 struct task_struct *p; 1778 struct multiprocess_signals delayed; 1779 struct file *pidfile = NULL; 1780 u64 clone_flags = args->flags; 1781 1782 /* 1783 * Don't allow sharing the root directory with processes in a different 1784 * namespace 1785 */ 1786 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 1787 return ERR_PTR(-EINVAL); 1788 1789 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS)) 1790 return ERR_PTR(-EINVAL); 1791 1792 /* 1793 * Thread groups must share signals as well, and detached threads 1794 * can only be started up within the thread group. 1795 */ 1796 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 1797 return ERR_PTR(-EINVAL); 1798 1799 /* 1800 * Shared signal handlers imply shared VM. By way of the above, 1801 * thread groups also imply shared VM. Blocking this case allows 1802 * for various simplifications in other code. 1803 */ 1804 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 1805 return ERR_PTR(-EINVAL); 1806 1807 /* 1808 * Siblings of global init remain as zombies on exit since they are 1809 * not reaped by their parent (swapper). To solve this and to avoid 1810 * multi-rooted process trees, prevent global and container-inits 1811 * from creating siblings. 1812 */ 1813 if ((clone_flags & CLONE_PARENT) && 1814 current->signal->flags & SIGNAL_UNKILLABLE) 1815 return ERR_PTR(-EINVAL); 1816 1817 /* 1818 * If the new process will be in a different pid or user namespace 1819 * do not allow it to share a thread group with the forking task. 1820 */ 1821 if (clone_flags & CLONE_THREAD) { 1822 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) || 1823 (task_active_pid_ns(current) != 1824 current->nsproxy->pid_ns_for_children)) 1825 return ERR_PTR(-EINVAL); 1826 } 1827 1828 if (clone_flags & CLONE_PIDFD) { 1829 /* 1830 * - CLONE_DETACHED is blocked so that we can potentially 1831 * reuse it later for CLONE_PIDFD. 1832 * - CLONE_THREAD is blocked until someone really needs it. 1833 */ 1834 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD)) 1835 return ERR_PTR(-EINVAL); 1836 } 1837 1838 /* 1839 * Force any signals received before this point to be delivered 1840 * before the fork happens. Collect up signals sent to multiple 1841 * processes that happen during the fork and delay them so that 1842 * they appear to happen after the fork. 1843 */ 1844 sigemptyset(&delayed.signal); 1845 INIT_HLIST_NODE(&delayed.node); 1846 1847 spin_lock_irq(¤t->sighand->siglock); 1848 if (!(clone_flags & CLONE_THREAD)) 1849 hlist_add_head(&delayed.node, ¤t->signal->multiprocess); 1850 recalc_sigpending(); 1851 spin_unlock_irq(¤t->sighand->siglock); 1852 retval = -ERESTARTNOINTR; 1853 if (signal_pending(current)) 1854 goto fork_out; 1855 1856 retval = -ENOMEM; 1857 p = dup_task_struct(current, node); 1858 if (!p) 1859 goto fork_out; 1860 1861 /* 1862 * This _must_ happen before we call free_task(), i.e. before we jump 1863 * to any of the bad_fork_* labels. This is to avoid freeing 1864 * p->set_child_tid which is (ab)used as a kthread's data pointer for 1865 * kernel threads (PF_KTHREAD). 1866 */ 1867 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL; 1868 /* 1869 * Clear TID on mm_release()? 1870 */ 1871 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL; 1872 1873 ftrace_graph_init_task(p); 1874 1875 rt_mutex_init_task(p); 1876 1877 #ifdef CONFIG_PROVE_LOCKING 1878 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); 1879 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 1880 #endif 1881 retval = -EAGAIN; 1882 if (atomic_read(&p->real_cred->user->processes) >= 1883 task_rlimit(p, RLIMIT_NPROC)) { 1884 if (p->real_cred->user != INIT_USER && 1885 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) 1886 goto bad_fork_free; 1887 } 1888 current->flags &= ~PF_NPROC_EXCEEDED; 1889 1890 retval = copy_creds(p, clone_flags); 1891 if (retval < 0) 1892 goto bad_fork_free; 1893 1894 /* 1895 * If multiple threads are within copy_process(), then this check 1896 * triggers too late. This doesn't hurt, the check is only there 1897 * to stop root fork bombs. 1898 */ 1899 retval = -EAGAIN; 1900 if (nr_threads >= max_threads) 1901 goto bad_fork_cleanup_count; 1902 1903 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */ 1904 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE); 1905 p->flags |= PF_FORKNOEXEC; 1906 INIT_LIST_HEAD(&p->children); 1907 INIT_LIST_HEAD(&p->sibling); 1908 rcu_copy_process(p); 1909 p->vfork_done = NULL; 1910 spin_lock_init(&p->alloc_lock); 1911 1912 init_sigpending(&p->pending); 1913 1914 p->utime = p->stime = p->gtime = 0; 1915 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 1916 p->utimescaled = p->stimescaled = 0; 1917 #endif 1918 prev_cputime_init(&p->prev_cputime); 1919 1920 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN 1921 seqcount_init(&p->vtime.seqcount); 1922 p->vtime.starttime = 0; 1923 p->vtime.state = VTIME_INACTIVE; 1924 #endif 1925 1926 #if defined(SPLIT_RSS_COUNTING) 1927 memset(&p->rss_stat, 0, sizeof(p->rss_stat)); 1928 #endif 1929 1930 p->default_timer_slack_ns = current->timer_slack_ns; 1931 1932 #ifdef CONFIG_PSI 1933 p->psi_flags = 0; 1934 #endif 1935 1936 task_io_accounting_init(&p->ioac); 1937 acct_clear_integrals(p); 1938 1939 posix_cpu_timers_init(p); 1940 1941 p->io_context = NULL; 1942 audit_set_context(p, NULL); 1943 cgroup_fork(p); 1944 #ifdef CONFIG_NUMA 1945 p->mempolicy = mpol_dup(p->mempolicy); 1946 if (IS_ERR(p->mempolicy)) { 1947 retval = PTR_ERR(p->mempolicy); 1948 p->mempolicy = NULL; 1949 goto bad_fork_cleanup_threadgroup_lock; 1950 } 1951 #endif 1952 #ifdef CONFIG_CPUSETS 1953 p->cpuset_mem_spread_rotor = NUMA_NO_NODE; 1954 p->cpuset_slab_spread_rotor = NUMA_NO_NODE; 1955 seqcount_init(&p->mems_allowed_seq); 1956 #endif 1957 #ifdef CONFIG_TRACE_IRQFLAGS 1958 p->irq_events = 0; 1959 p->hardirqs_enabled = 0; 1960 p->hardirq_enable_ip = 0; 1961 p->hardirq_enable_event = 0; 1962 p->hardirq_disable_ip = _THIS_IP_; 1963 p->hardirq_disable_event = 0; 1964 p->softirqs_enabled = 1; 1965 p->softirq_enable_ip = _THIS_IP_; 1966 p->softirq_enable_event = 0; 1967 p->softirq_disable_ip = 0; 1968 p->softirq_disable_event = 0; 1969 p->hardirq_context = 0; 1970 p->softirq_context = 0; 1971 #endif 1972 1973 p->pagefault_disabled = 0; 1974 1975 #ifdef CONFIG_LOCKDEP 1976 lockdep_init_task(p); 1977 #endif 1978 1979 #ifdef CONFIG_DEBUG_MUTEXES 1980 p->blocked_on = NULL; /* not blocked yet */ 1981 #endif 1982 #ifdef CONFIG_BCACHE 1983 p->sequential_io = 0; 1984 p->sequential_io_avg = 0; 1985 #endif 1986 1987 /* Perform scheduler related setup. Assign this task to a CPU. */ 1988 retval = sched_fork(clone_flags, p); 1989 if (retval) 1990 goto bad_fork_cleanup_policy; 1991 1992 retval = perf_event_init_task(p); 1993 if (retval) 1994 goto bad_fork_cleanup_policy; 1995 retval = audit_alloc(p); 1996 if (retval) 1997 goto bad_fork_cleanup_perf; 1998 /* copy all the process information */ 1999 shm_init_task(p); 2000 retval = security_task_alloc(p, clone_flags); 2001 if (retval) 2002 goto bad_fork_cleanup_audit; 2003 retval = copy_semundo(clone_flags, p); 2004 if (retval) 2005 goto bad_fork_cleanup_security; 2006 retval = copy_files(clone_flags, p); 2007 if (retval) 2008 goto bad_fork_cleanup_semundo; 2009 retval = copy_fs(clone_flags, p); 2010 if (retval) 2011 goto bad_fork_cleanup_files; 2012 retval = copy_sighand(clone_flags, p); 2013 if (retval) 2014 goto bad_fork_cleanup_fs; 2015 retval = copy_signal(clone_flags, p); 2016 if (retval) 2017 goto bad_fork_cleanup_sighand; 2018 retval = copy_mm(clone_flags, p); 2019 if (retval) 2020 goto bad_fork_cleanup_signal; 2021 retval = copy_namespaces(clone_flags, p); 2022 if (retval) 2023 goto bad_fork_cleanup_mm; 2024 retval = copy_io(clone_flags, p); 2025 if (retval) 2026 goto bad_fork_cleanup_namespaces; 2027 retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p, 2028 args->tls); 2029 if (retval) 2030 goto bad_fork_cleanup_io; 2031 2032 stackleak_task_init(p); 2033 2034 if (pid != &init_struct_pid) { 2035 pid = alloc_pid(p->nsproxy->pid_ns_for_children); 2036 if (IS_ERR(pid)) { 2037 retval = PTR_ERR(pid); 2038 goto bad_fork_cleanup_thread; 2039 } 2040 } 2041 2042 /* 2043 * This has to happen after we've potentially unshared the file 2044 * descriptor table (so that the pidfd doesn't leak into the child 2045 * if the fd table isn't shared). 2046 */ 2047 if (clone_flags & CLONE_PIDFD) { 2048 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC); 2049 if (retval < 0) 2050 goto bad_fork_free_pid; 2051 2052 pidfd = retval; 2053 2054 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid, 2055 O_RDWR | O_CLOEXEC); 2056 if (IS_ERR(pidfile)) { 2057 put_unused_fd(pidfd); 2058 retval = PTR_ERR(pidfile); 2059 goto bad_fork_free_pid; 2060 } 2061 get_pid(pid); /* held by pidfile now */ 2062 2063 retval = put_user(pidfd, args->pidfd); 2064 if (retval) 2065 goto bad_fork_put_pidfd; 2066 } 2067 2068 #ifdef CONFIG_BLOCK 2069 p->plug = NULL; 2070 #endif 2071 #ifdef CONFIG_FUTEX 2072 p->robust_list = NULL; 2073 #ifdef CONFIG_COMPAT 2074 p->compat_robust_list = NULL; 2075 #endif 2076 INIT_LIST_HEAD(&p->pi_state_list); 2077 p->pi_state_cache = NULL; 2078 #endif 2079 /* 2080 * sigaltstack should be cleared when sharing the same VM 2081 */ 2082 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM) 2083 sas_ss_reset(p); 2084 2085 /* 2086 * Syscall tracing and stepping should be turned off in the 2087 * child regardless of CLONE_PTRACE. 2088 */ 2089 user_disable_single_step(p); 2090 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); 2091 #ifdef TIF_SYSCALL_EMU 2092 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); 2093 #endif 2094 clear_tsk_latency_tracing(p); 2095 2096 /* ok, now we should be set up.. */ 2097 p->pid = pid_nr(pid); 2098 if (clone_flags & CLONE_THREAD) { 2099 p->exit_signal = -1; 2100 p->group_leader = current->group_leader; 2101 p->tgid = current->tgid; 2102 } else { 2103 if (clone_flags & CLONE_PARENT) 2104 p->exit_signal = current->group_leader->exit_signal; 2105 else 2106 p->exit_signal = args->exit_signal; 2107 p->group_leader = p; 2108 p->tgid = p->pid; 2109 } 2110 2111 p->nr_dirtied = 0; 2112 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10); 2113 p->dirty_paused_when = 0; 2114 2115 p->pdeath_signal = 0; 2116 INIT_LIST_HEAD(&p->thread_group); 2117 p->task_works = NULL; 2118 2119 cgroup_threadgroup_change_begin(current); 2120 /* 2121 * Ensure that the cgroup subsystem policies allow the new process to be 2122 * forked. It should be noted the the new process's css_set can be changed 2123 * between here and cgroup_post_fork() if an organisation operation is in 2124 * progress. 2125 */ 2126 retval = cgroup_can_fork(p); 2127 if (retval) 2128 goto bad_fork_cgroup_threadgroup_change_end; 2129 2130 /* 2131 * From this point on we must avoid any synchronous user-space 2132 * communication until we take the tasklist-lock. In particular, we do 2133 * not want user-space to be able to predict the process start-time by 2134 * stalling fork(2) after we recorded the start_time but before it is 2135 * visible to the system. 2136 */ 2137 2138 p->start_time = ktime_get_ns(); 2139 p->real_start_time = ktime_get_boottime_ns(); 2140 2141 /* 2142 * Make it visible to the rest of the system, but dont wake it up yet. 2143 * Need tasklist lock for parent etc handling! 2144 */ 2145 write_lock_irq(&tasklist_lock); 2146 2147 /* CLONE_PARENT re-uses the old parent */ 2148 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { 2149 p->real_parent = current->real_parent; 2150 p->parent_exec_id = current->parent_exec_id; 2151 } else { 2152 p->real_parent = current; 2153 p->parent_exec_id = current->self_exec_id; 2154 } 2155 2156 klp_copy_process(p); 2157 2158 spin_lock(¤t->sighand->siglock); 2159 2160 /* 2161 * Copy seccomp details explicitly here, in case they were changed 2162 * before holding sighand lock. 2163 */ 2164 copy_seccomp(p); 2165 2166 rseq_fork(p, clone_flags); 2167 2168 /* Don't start children in a dying pid namespace */ 2169 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) { 2170 retval = -ENOMEM; 2171 goto bad_fork_cancel_cgroup; 2172 } 2173 2174 /* Let kill terminate clone/fork in the middle */ 2175 if (fatal_signal_pending(current)) { 2176 retval = -EINTR; 2177 goto bad_fork_cancel_cgroup; 2178 } 2179 2180 /* past the last point of failure */ 2181 if (pidfile) 2182 fd_install(pidfd, pidfile); 2183 2184 init_task_pid_links(p); 2185 if (likely(p->pid)) { 2186 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace); 2187 2188 init_task_pid(p, PIDTYPE_PID, pid); 2189 if (thread_group_leader(p)) { 2190 init_task_pid(p, PIDTYPE_TGID, pid); 2191 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current)); 2192 init_task_pid(p, PIDTYPE_SID, task_session(current)); 2193 2194 if (is_child_reaper(pid)) { 2195 ns_of_pid(pid)->child_reaper = p; 2196 p->signal->flags |= SIGNAL_UNKILLABLE; 2197 } 2198 p->signal->shared_pending.signal = delayed.signal; 2199 p->signal->tty = tty_kref_get(current->signal->tty); 2200 /* 2201 * Inherit has_child_subreaper flag under the same 2202 * tasklist_lock with adding child to the process tree 2203 * for propagate_has_child_subreaper optimization. 2204 */ 2205 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper || 2206 p->real_parent->signal->is_child_subreaper; 2207 list_add_tail(&p->sibling, &p->real_parent->children); 2208 list_add_tail_rcu(&p->tasks, &init_task.tasks); 2209 attach_pid(p, PIDTYPE_TGID); 2210 attach_pid(p, PIDTYPE_PGID); 2211 attach_pid(p, PIDTYPE_SID); 2212 __this_cpu_inc(process_counts); 2213 } else { 2214 current->signal->nr_threads++; 2215 atomic_inc(¤t->signal->live); 2216 refcount_inc(¤t->signal->sigcnt); 2217 task_join_group_stop(p); 2218 list_add_tail_rcu(&p->thread_group, 2219 &p->group_leader->thread_group); 2220 list_add_tail_rcu(&p->thread_node, 2221 &p->signal->thread_head); 2222 } 2223 attach_pid(p, PIDTYPE_PID); 2224 nr_threads++; 2225 } 2226 total_forks++; 2227 hlist_del_init(&delayed.node); 2228 spin_unlock(¤t->sighand->siglock); 2229 syscall_tracepoint_update(p); 2230 write_unlock_irq(&tasklist_lock); 2231 2232 proc_fork_connector(p); 2233 cgroup_post_fork(p); 2234 cgroup_threadgroup_change_end(current); 2235 perf_event_fork(p); 2236 2237 trace_task_newtask(p, clone_flags); 2238 uprobe_copy_process(p, clone_flags); 2239 2240 return p; 2241 2242 bad_fork_cancel_cgroup: 2243 spin_unlock(¤t->sighand->siglock); 2244 write_unlock_irq(&tasklist_lock); 2245 cgroup_cancel_fork(p); 2246 bad_fork_cgroup_threadgroup_change_end: 2247 cgroup_threadgroup_change_end(current); 2248 bad_fork_put_pidfd: 2249 if (clone_flags & CLONE_PIDFD) { 2250 fput(pidfile); 2251 put_unused_fd(pidfd); 2252 } 2253 bad_fork_free_pid: 2254 if (pid != &init_struct_pid) 2255 free_pid(pid); 2256 bad_fork_cleanup_thread: 2257 exit_thread(p); 2258 bad_fork_cleanup_io: 2259 if (p->io_context) 2260 exit_io_context(p); 2261 bad_fork_cleanup_namespaces: 2262 exit_task_namespaces(p); 2263 bad_fork_cleanup_mm: 2264 if (p->mm) { 2265 mm_clear_owner(p->mm, p); 2266 mmput(p->mm); 2267 } 2268 bad_fork_cleanup_signal: 2269 if (!(clone_flags & CLONE_THREAD)) 2270 free_signal_struct(p->signal); 2271 bad_fork_cleanup_sighand: 2272 __cleanup_sighand(p->sighand); 2273 bad_fork_cleanup_fs: 2274 exit_fs(p); /* blocking */ 2275 bad_fork_cleanup_files: 2276 exit_files(p); /* blocking */ 2277 bad_fork_cleanup_semundo: 2278 exit_sem(p); 2279 bad_fork_cleanup_security: 2280 security_task_free(p); 2281 bad_fork_cleanup_audit: 2282 audit_free(p); 2283 bad_fork_cleanup_perf: 2284 perf_event_free_task(p); 2285 bad_fork_cleanup_policy: 2286 lockdep_free_task(p); 2287 #ifdef CONFIG_NUMA 2288 mpol_put(p->mempolicy); 2289 bad_fork_cleanup_threadgroup_lock: 2290 #endif 2291 delayacct_tsk_free(p); 2292 bad_fork_cleanup_count: 2293 atomic_dec(&p->cred->user->processes); 2294 exit_creds(p); 2295 bad_fork_free: 2296 p->state = TASK_DEAD; 2297 put_task_stack(p); 2298 delayed_free_task(p); 2299 fork_out: 2300 spin_lock_irq(¤t->sighand->siglock); 2301 hlist_del_init(&delayed.node); 2302 spin_unlock_irq(¤t->sighand->siglock); 2303 return ERR_PTR(retval); 2304 } 2305 2306 static inline void init_idle_pids(struct task_struct *idle) 2307 { 2308 enum pid_type type; 2309 2310 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 2311 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */ 2312 init_task_pid(idle, type, &init_struct_pid); 2313 } 2314 } 2315 2316 struct task_struct *fork_idle(int cpu) 2317 { 2318 struct task_struct *task; 2319 struct kernel_clone_args args = { 2320 .flags = CLONE_VM, 2321 }; 2322 2323 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args); 2324 if (!IS_ERR(task)) { 2325 init_idle_pids(task); 2326 init_idle(task, cpu); 2327 } 2328 2329 return task; 2330 } 2331 2332 struct mm_struct *copy_init_mm(void) 2333 { 2334 return dup_mm(NULL, &init_mm); 2335 } 2336 2337 /* 2338 * Ok, this is the main fork-routine. 2339 * 2340 * It copies the process, and if successful kick-starts 2341 * it and waits for it to finish using the VM if required. 2342 */ 2343 long _do_fork(struct kernel_clone_args *args) 2344 { 2345 u64 clone_flags = args->flags; 2346 struct completion vfork; 2347 struct pid *pid; 2348 struct task_struct *p; 2349 int trace = 0; 2350 long nr; 2351 2352 /* 2353 * Determine whether and which event to report to ptracer. When 2354 * called from kernel_thread or CLONE_UNTRACED is explicitly 2355 * requested, no event is reported; otherwise, report if the event 2356 * for the type of forking is enabled. 2357 */ 2358 if (!(clone_flags & CLONE_UNTRACED)) { 2359 if (clone_flags & CLONE_VFORK) 2360 trace = PTRACE_EVENT_VFORK; 2361 else if (args->exit_signal != SIGCHLD) 2362 trace = PTRACE_EVENT_CLONE; 2363 else 2364 trace = PTRACE_EVENT_FORK; 2365 2366 if (likely(!ptrace_event_enabled(current, trace))) 2367 trace = 0; 2368 } 2369 2370 p = copy_process(NULL, trace, NUMA_NO_NODE, args); 2371 add_latent_entropy(); 2372 2373 if (IS_ERR(p)) 2374 return PTR_ERR(p); 2375 2376 /* 2377 * Do this prior waking up the new thread - the thread pointer 2378 * might get invalid after that point, if the thread exits quickly. 2379 */ 2380 trace_sched_process_fork(current, p); 2381 2382 pid = get_task_pid(p, PIDTYPE_PID); 2383 nr = pid_vnr(pid); 2384 2385 if (clone_flags & CLONE_PARENT_SETTID) 2386 put_user(nr, args->parent_tid); 2387 2388 if (clone_flags & CLONE_VFORK) { 2389 p->vfork_done = &vfork; 2390 init_completion(&vfork); 2391 get_task_struct(p); 2392 } 2393 2394 wake_up_new_task(p); 2395 2396 /* forking complete and child started to run, tell ptracer */ 2397 if (unlikely(trace)) 2398 ptrace_event_pid(trace, pid); 2399 2400 if (clone_flags & CLONE_VFORK) { 2401 if (!wait_for_vfork_done(p, &vfork)) 2402 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid); 2403 } 2404 2405 put_pid(pid); 2406 return nr; 2407 } 2408 2409 #ifndef CONFIG_HAVE_COPY_THREAD_TLS 2410 /* For compatibility with architectures that call do_fork directly rather than 2411 * using the syscall entry points below. */ 2412 long do_fork(unsigned long clone_flags, 2413 unsigned long stack_start, 2414 unsigned long stack_size, 2415 int __user *parent_tidptr, 2416 int __user *child_tidptr) 2417 { 2418 struct kernel_clone_args args = { 2419 .flags = (clone_flags & ~CSIGNAL), 2420 .child_tid = child_tidptr, 2421 .parent_tid = parent_tidptr, 2422 .exit_signal = (clone_flags & CSIGNAL), 2423 .stack = stack_start, 2424 .stack_size = stack_size, 2425 }; 2426 2427 return _do_fork(&args); 2428 } 2429 #endif 2430 2431 /* 2432 * Create a kernel thread. 2433 */ 2434 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags) 2435 { 2436 struct kernel_clone_args args = { 2437 .flags = ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL), 2438 .exit_signal = (flags & CSIGNAL), 2439 .stack = (unsigned long)fn, 2440 .stack_size = (unsigned long)arg, 2441 }; 2442 2443 return _do_fork(&args); 2444 } 2445 2446 #ifdef __ARCH_WANT_SYS_FORK 2447 SYSCALL_DEFINE0(fork) 2448 { 2449 #ifdef CONFIG_MMU 2450 struct kernel_clone_args args = { 2451 .exit_signal = SIGCHLD, 2452 }; 2453 2454 return _do_fork(&args); 2455 #else 2456 /* can not support in nommu mode */ 2457 return -EINVAL; 2458 #endif 2459 } 2460 #endif 2461 2462 #ifdef __ARCH_WANT_SYS_VFORK 2463 SYSCALL_DEFINE0(vfork) 2464 { 2465 struct kernel_clone_args args = { 2466 .flags = CLONE_VFORK | CLONE_VM, 2467 .exit_signal = SIGCHLD, 2468 }; 2469 2470 return _do_fork(&args); 2471 } 2472 #endif 2473 2474 #ifdef __ARCH_WANT_SYS_CLONE 2475 #ifdef CONFIG_CLONE_BACKWARDS 2476 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2477 int __user *, parent_tidptr, 2478 unsigned long, tls, 2479 int __user *, child_tidptr) 2480 #elif defined(CONFIG_CLONE_BACKWARDS2) 2481 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags, 2482 int __user *, parent_tidptr, 2483 int __user *, child_tidptr, 2484 unsigned long, tls) 2485 #elif defined(CONFIG_CLONE_BACKWARDS3) 2486 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp, 2487 int, stack_size, 2488 int __user *, parent_tidptr, 2489 int __user *, child_tidptr, 2490 unsigned long, tls) 2491 #else 2492 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2493 int __user *, parent_tidptr, 2494 int __user *, child_tidptr, 2495 unsigned long, tls) 2496 #endif 2497 { 2498 struct kernel_clone_args args = { 2499 .flags = (clone_flags & ~CSIGNAL), 2500 .pidfd = parent_tidptr, 2501 .child_tid = child_tidptr, 2502 .parent_tid = parent_tidptr, 2503 .exit_signal = (clone_flags & CSIGNAL), 2504 .stack = newsp, 2505 .tls = tls, 2506 }; 2507 2508 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */ 2509 if ((clone_flags & CLONE_PIDFD) && (clone_flags & CLONE_PARENT_SETTID)) 2510 return -EINVAL; 2511 2512 return _do_fork(&args); 2513 } 2514 #endif 2515 2516 #ifdef __ARCH_WANT_SYS_CLONE3 2517 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs, 2518 struct clone_args __user *uargs, 2519 size_t size) 2520 { 2521 struct clone_args args; 2522 2523 if (unlikely(size > PAGE_SIZE)) 2524 return -E2BIG; 2525 2526 if (unlikely(size < sizeof(struct clone_args))) 2527 return -EINVAL; 2528 2529 if (unlikely(!access_ok(uargs, size))) 2530 return -EFAULT; 2531 2532 if (size > sizeof(struct clone_args)) { 2533 unsigned char __user *addr; 2534 unsigned char __user *end; 2535 unsigned char val; 2536 2537 addr = (void __user *)uargs + sizeof(struct clone_args); 2538 end = (void __user *)uargs + size; 2539 2540 for (; addr < end; addr++) { 2541 if (get_user(val, addr)) 2542 return -EFAULT; 2543 if (val) 2544 return -E2BIG; 2545 } 2546 2547 size = sizeof(struct clone_args); 2548 } 2549 2550 if (copy_from_user(&args, uargs, size)) 2551 return -EFAULT; 2552 2553 *kargs = (struct kernel_clone_args){ 2554 .flags = args.flags, 2555 .pidfd = u64_to_user_ptr(args.pidfd), 2556 .child_tid = u64_to_user_ptr(args.child_tid), 2557 .parent_tid = u64_to_user_ptr(args.parent_tid), 2558 .exit_signal = args.exit_signal, 2559 .stack = args.stack, 2560 .stack_size = args.stack_size, 2561 .tls = args.tls, 2562 }; 2563 2564 return 0; 2565 } 2566 2567 static bool clone3_args_valid(const struct kernel_clone_args *kargs) 2568 { 2569 /* 2570 * All lower bits of the flag word are taken. 2571 * Verify that no other unknown flags are passed along. 2572 */ 2573 if (kargs->flags & ~CLONE_LEGACY_FLAGS) 2574 return false; 2575 2576 /* 2577 * - make the CLONE_DETACHED bit reuseable for clone3 2578 * - make the CSIGNAL bits reuseable for clone3 2579 */ 2580 if (kargs->flags & (CLONE_DETACHED | CSIGNAL)) 2581 return false; 2582 2583 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) && 2584 kargs->exit_signal) 2585 return false; 2586 2587 return true; 2588 } 2589 2590 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size) 2591 { 2592 int err; 2593 2594 struct kernel_clone_args kargs; 2595 2596 err = copy_clone_args_from_user(&kargs, uargs, size); 2597 if (err) 2598 return err; 2599 2600 if (!clone3_args_valid(&kargs)) 2601 return -EINVAL; 2602 2603 return _do_fork(&kargs); 2604 } 2605 #endif 2606 2607 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data) 2608 { 2609 struct task_struct *leader, *parent, *child; 2610 int res; 2611 2612 read_lock(&tasklist_lock); 2613 leader = top = top->group_leader; 2614 down: 2615 for_each_thread(leader, parent) { 2616 list_for_each_entry(child, &parent->children, sibling) { 2617 res = visitor(child, data); 2618 if (res) { 2619 if (res < 0) 2620 goto out; 2621 leader = child; 2622 goto down; 2623 } 2624 up: 2625 ; 2626 } 2627 } 2628 2629 if (leader != top) { 2630 child = leader; 2631 parent = child->real_parent; 2632 leader = parent->group_leader; 2633 goto up; 2634 } 2635 out: 2636 read_unlock(&tasklist_lock); 2637 } 2638 2639 #ifndef ARCH_MIN_MMSTRUCT_ALIGN 2640 #define ARCH_MIN_MMSTRUCT_ALIGN 0 2641 #endif 2642 2643 static void sighand_ctor(void *data) 2644 { 2645 struct sighand_struct *sighand = data; 2646 2647 spin_lock_init(&sighand->siglock); 2648 init_waitqueue_head(&sighand->signalfd_wqh); 2649 } 2650 2651 void __init proc_caches_init(void) 2652 { 2653 unsigned int mm_size; 2654 2655 sighand_cachep = kmem_cache_create("sighand_cache", 2656 sizeof(struct sighand_struct), 0, 2657 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU| 2658 SLAB_ACCOUNT, sighand_ctor); 2659 signal_cachep = kmem_cache_create("signal_cache", 2660 sizeof(struct signal_struct), 0, 2661 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2662 NULL); 2663 files_cachep = kmem_cache_create("files_cache", 2664 sizeof(struct files_struct), 0, 2665 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2666 NULL); 2667 fs_cachep = kmem_cache_create("fs_cache", 2668 sizeof(struct fs_struct), 0, 2669 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2670 NULL); 2671 2672 /* 2673 * The mm_cpumask is located at the end of mm_struct, and is 2674 * dynamically sized based on the maximum CPU number this system 2675 * can have, taking hotplug into account (nr_cpu_ids). 2676 */ 2677 mm_size = sizeof(struct mm_struct) + cpumask_size(); 2678 2679 mm_cachep = kmem_cache_create_usercopy("mm_struct", 2680 mm_size, ARCH_MIN_MMSTRUCT_ALIGN, 2681 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 2682 offsetof(struct mm_struct, saved_auxv), 2683 sizeof_field(struct mm_struct, saved_auxv), 2684 NULL); 2685 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT); 2686 mmap_init(); 2687 nsproxy_cache_init(); 2688 } 2689 2690 /* 2691 * Check constraints on flags passed to the unshare system call. 2692 */ 2693 static int check_unshare_flags(unsigned long unshare_flags) 2694 { 2695 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND| 2696 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM| 2697 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET| 2698 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP)) 2699 return -EINVAL; 2700 /* 2701 * Not implemented, but pretend it works if there is nothing 2702 * to unshare. Note that unsharing the address space or the 2703 * signal handlers also need to unshare the signal queues (aka 2704 * CLONE_THREAD). 2705 */ 2706 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) { 2707 if (!thread_group_empty(current)) 2708 return -EINVAL; 2709 } 2710 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) { 2711 if (refcount_read(¤t->sighand->count) > 1) 2712 return -EINVAL; 2713 } 2714 if (unshare_flags & CLONE_VM) { 2715 if (!current_is_single_threaded()) 2716 return -EINVAL; 2717 } 2718 2719 return 0; 2720 } 2721 2722 /* 2723 * Unshare the filesystem structure if it is being shared 2724 */ 2725 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) 2726 { 2727 struct fs_struct *fs = current->fs; 2728 2729 if (!(unshare_flags & CLONE_FS) || !fs) 2730 return 0; 2731 2732 /* don't need lock here; in the worst case we'll do useless copy */ 2733 if (fs->users == 1) 2734 return 0; 2735 2736 *new_fsp = copy_fs_struct(fs); 2737 if (!*new_fsp) 2738 return -ENOMEM; 2739 2740 return 0; 2741 } 2742 2743 /* 2744 * Unshare file descriptor table if it is being shared 2745 */ 2746 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp) 2747 { 2748 struct files_struct *fd = current->files; 2749 int error = 0; 2750 2751 if ((unshare_flags & CLONE_FILES) && 2752 (fd && atomic_read(&fd->count) > 1)) { 2753 *new_fdp = dup_fd(fd, &error); 2754 if (!*new_fdp) 2755 return error; 2756 } 2757 2758 return 0; 2759 } 2760 2761 /* 2762 * unshare allows a process to 'unshare' part of the process 2763 * context which was originally shared using clone. copy_* 2764 * functions used by do_fork() cannot be used here directly 2765 * because they modify an inactive task_struct that is being 2766 * constructed. Here we are modifying the current, active, 2767 * task_struct. 2768 */ 2769 int ksys_unshare(unsigned long unshare_flags) 2770 { 2771 struct fs_struct *fs, *new_fs = NULL; 2772 struct files_struct *fd, *new_fd = NULL; 2773 struct cred *new_cred = NULL; 2774 struct nsproxy *new_nsproxy = NULL; 2775 int do_sysvsem = 0; 2776 int err; 2777 2778 /* 2779 * If unsharing a user namespace must also unshare the thread group 2780 * and unshare the filesystem root and working directories. 2781 */ 2782 if (unshare_flags & CLONE_NEWUSER) 2783 unshare_flags |= CLONE_THREAD | CLONE_FS; 2784 /* 2785 * If unsharing vm, must also unshare signal handlers. 2786 */ 2787 if (unshare_flags & CLONE_VM) 2788 unshare_flags |= CLONE_SIGHAND; 2789 /* 2790 * If unsharing a signal handlers, must also unshare the signal queues. 2791 */ 2792 if (unshare_flags & CLONE_SIGHAND) 2793 unshare_flags |= CLONE_THREAD; 2794 /* 2795 * If unsharing namespace, must also unshare filesystem information. 2796 */ 2797 if (unshare_flags & CLONE_NEWNS) 2798 unshare_flags |= CLONE_FS; 2799 2800 err = check_unshare_flags(unshare_flags); 2801 if (err) 2802 goto bad_unshare_out; 2803 /* 2804 * CLONE_NEWIPC must also detach from the undolist: after switching 2805 * to a new ipc namespace, the semaphore arrays from the old 2806 * namespace are unreachable. 2807 */ 2808 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) 2809 do_sysvsem = 1; 2810 err = unshare_fs(unshare_flags, &new_fs); 2811 if (err) 2812 goto bad_unshare_out; 2813 err = unshare_fd(unshare_flags, &new_fd); 2814 if (err) 2815 goto bad_unshare_cleanup_fs; 2816 err = unshare_userns(unshare_flags, &new_cred); 2817 if (err) 2818 goto bad_unshare_cleanup_fd; 2819 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, 2820 new_cred, new_fs); 2821 if (err) 2822 goto bad_unshare_cleanup_cred; 2823 2824 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) { 2825 if (do_sysvsem) { 2826 /* 2827 * CLONE_SYSVSEM is equivalent to sys_exit(). 2828 */ 2829 exit_sem(current); 2830 } 2831 if (unshare_flags & CLONE_NEWIPC) { 2832 /* Orphan segments in old ns (see sem above). */ 2833 exit_shm(current); 2834 shm_init_task(current); 2835 } 2836 2837 if (new_nsproxy) 2838 switch_task_namespaces(current, new_nsproxy); 2839 2840 task_lock(current); 2841 2842 if (new_fs) { 2843 fs = current->fs; 2844 spin_lock(&fs->lock); 2845 current->fs = new_fs; 2846 if (--fs->users) 2847 new_fs = NULL; 2848 else 2849 new_fs = fs; 2850 spin_unlock(&fs->lock); 2851 } 2852 2853 if (new_fd) { 2854 fd = current->files; 2855 current->files = new_fd; 2856 new_fd = fd; 2857 } 2858 2859 task_unlock(current); 2860 2861 if (new_cred) { 2862 /* Install the new user namespace */ 2863 commit_creds(new_cred); 2864 new_cred = NULL; 2865 } 2866 } 2867 2868 perf_event_namespaces(current); 2869 2870 bad_unshare_cleanup_cred: 2871 if (new_cred) 2872 put_cred(new_cred); 2873 bad_unshare_cleanup_fd: 2874 if (new_fd) 2875 put_files_struct(new_fd); 2876 2877 bad_unshare_cleanup_fs: 2878 if (new_fs) 2879 free_fs_struct(new_fs); 2880 2881 bad_unshare_out: 2882 return err; 2883 } 2884 2885 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags) 2886 { 2887 return ksys_unshare(unshare_flags); 2888 } 2889 2890 /* 2891 * Helper to unshare the files of the current task. 2892 * We don't want to expose copy_files internals to 2893 * the exec layer of the kernel. 2894 */ 2895 2896 int unshare_files(struct files_struct **displaced) 2897 { 2898 struct task_struct *task = current; 2899 struct files_struct *copy = NULL; 2900 int error; 2901 2902 error = unshare_fd(CLONE_FILES, ©); 2903 if (error || !copy) { 2904 *displaced = NULL; 2905 return error; 2906 } 2907 *displaced = task->files; 2908 task_lock(task); 2909 task->files = copy; 2910 task_unlock(task); 2911 return 0; 2912 } 2913 2914 int sysctl_max_threads(struct ctl_table *table, int write, 2915 void __user *buffer, size_t *lenp, loff_t *ppos) 2916 { 2917 struct ctl_table t; 2918 int ret; 2919 int threads = max_threads; 2920 int min = MIN_THREADS; 2921 int max = MAX_THREADS; 2922 2923 t = *table; 2924 t.data = &threads; 2925 t.extra1 = &min; 2926 t.extra2 = &max; 2927 2928 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); 2929 if (ret || !write) 2930 return ret; 2931 2932 set_max_threads(threads); 2933 2934 return 0; 2935 } 2936