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/fs.h> 44 #include <linux/mm.h> 45 #include <linux/mm_inline.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/fs_struct.h> 81 #include <linux/magic.h> 82 #include <linux/perf_event.h> 83 #include <linux/posix-timers.h> 84 #include <linux/user-return-notifier.h> 85 #include <linux/oom.h> 86 #include <linux/khugepaged.h> 87 #include <linux/signalfd.h> 88 #include <linux/uprobes.h> 89 #include <linux/aio.h> 90 #include <linux/compiler.h> 91 #include <linux/sysctl.h> 92 #include <linux/kcov.h> 93 #include <linux/livepatch.h> 94 #include <linux/thread_info.h> 95 #include <linux/stackleak.h> 96 #include <linux/kasan.h> 97 #include <linux/scs.h> 98 #include <linux/io_uring.h> 99 #include <linux/bpf.h> 100 #include <linux/sched/mm.h> 101 102 #include <asm/pgalloc.h> 103 #include <linux/uaccess.h> 104 #include <asm/mmu_context.h> 105 #include <asm/cacheflush.h> 106 #include <asm/tlbflush.h> 107 108 #include <trace/events/sched.h> 109 110 #define CREATE_TRACE_POINTS 111 #include <trace/events/task.h> 112 113 /* 114 * Minimum number of threads to boot the kernel 115 */ 116 #define MIN_THREADS 20 117 118 /* 119 * Maximum number of threads 120 */ 121 #define MAX_THREADS FUTEX_TID_MASK 122 123 /* 124 * Protected counters by write_lock_irq(&tasklist_lock) 125 */ 126 unsigned long total_forks; /* Handle normal Linux uptimes. */ 127 int nr_threads; /* The idle threads do not count.. */ 128 129 static int max_threads; /* tunable limit on nr_threads */ 130 131 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x) 132 133 static const char * const resident_page_types[] = { 134 NAMED_ARRAY_INDEX(MM_FILEPAGES), 135 NAMED_ARRAY_INDEX(MM_ANONPAGES), 136 NAMED_ARRAY_INDEX(MM_SWAPENTS), 137 NAMED_ARRAY_INDEX(MM_SHMEMPAGES), 138 }; 139 140 DEFINE_PER_CPU(unsigned long, process_counts) = 0; 141 142 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */ 143 144 #ifdef CONFIG_PROVE_RCU 145 int lockdep_tasklist_lock_is_held(void) 146 { 147 return lockdep_is_held(&tasklist_lock); 148 } 149 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held); 150 #endif /* #ifdef CONFIG_PROVE_RCU */ 151 152 int nr_processes(void) 153 { 154 int cpu; 155 int total = 0; 156 157 for_each_possible_cpu(cpu) 158 total += per_cpu(process_counts, cpu); 159 160 return total; 161 } 162 163 void __weak arch_release_task_struct(struct task_struct *tsk) 164 { 165 } 166 167 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 168 static struct kmem_cache *task_struct_cachep; 169 170 static inline struct task_struct *alloc_task_struct_node(int node) 171 { 172 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node); 173 } 174 175 static inline void free_task_struct(struct task_struct *tsk) 176 { 177 kmem_cache_free(task_struct_cachep, tsk); 178 } 179 #endif 180 181 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR 182 183 /* 184 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a 185 * kmemcache based allocator. 186 */ 187 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) 188 189 # ifdef CONFIG_VMAP_STACK 190 /* 191 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB 192 * flush. Try to minimize the number of calls by caching stacks. 193 */ 194 #define NR_CACHED_STACKS 2 195 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]); 196 197 struct vm_stack { 198 struct rcu_head rcu; 199 struct vm_struct *stack_vm_area; 200 }; 201 202 static bool try_release_thread_stack_to_cache(struct vm_struct *vm) 203 { 204 unsigned int i; 205 206 for (i = 0; i < NR_CACHED_STACKS; i++) { 207 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL) 208 continue; 209 return true; 210 } 211 return false; 212 } 213 214 static void thread_stack_free_rcu(struct rcu_head *rh) 215 { 216 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu); 217 218 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area)) 219 return; 220 221 vfree(vm_stack); 222 } 223 224 static void thread_stack_delayed_free(struct task_struct *tsk) 225 { 226 struct vm_stack *vm_stack = tsk->stack; 227 228 vm_stack->stack_vm_area = tsk->stack_vm_area; 229 call_rcu(&vm_stack->rcu, thread_stack_free_rcu); 230 } 231 232 static int free_vm_stack_cache(unsigned int cpu) 233 { 234 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu); 235 int i; 236 237 for (i = 0; i < NR_CACHED_STACKS; i++) { 238 struct vm_struct *vm_stack = cached_vm_stacks[i]; 239 240 if (!vm_stack) 241 continue; 242 243 vfree(vm_stack->addr); 244 cached_vm_stacks[i] = NULL; 245 } 246 247 return 0; 248 } 249 250 static int memcg_charge_kernel_stack(struct vm_struct *vm) 251 { 252 int i; 253 int ret; 254 255 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0); 256 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE); 257 258 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 259 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0); 260 if (ret) 261 goto err; 262 } 263 return 0; 264 err: 265 /* 266 * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is 267 * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will 268 * ignore this page. 269 */ 270 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) 271 memcg_kmem_uncharge_page(vm->pages[i], 0); 272 return ret; 273 } 274 275 static int alloc_thread_stack_node(struct task_struct *tsk, int node) 276 { 277 struct vm_struct *vm; 278 void *stack; 279 int i; 280 281 for (i = 0; i < NR_CACHED_STACKS; i++) { 282 struct vm_struct *s; 283 284 s = this_cpu_xchg(cached_stacks[i], NULL); 285 286 if (!s) 287 continue; 288 289 /* Reset stack metadata. */ 290 kasan_unpoison_range(s->addr, THREAD_SIZE); 291 292 stack = kasan_reset_tag(s->addr); 293 294 /* Clear stale pointers from reused stack. */ 295 memset(stack, 0, THREAD_SIZE); 296 297 if (memcg_charge_kernel_stack(s)) { 298 vfree(s->addr); 299 return -ENOMEM; 300 } 301 302 tsk->stack_vm_area = s; 303 tsk->stack = stack; 304 return 0; 305 } 306 307 /* 308 * Allocated stacks are cached and later reused by new threads, 309 * so memcg accounting is performed manually on assigning/releasing 310 * stacks to tasks. Drop __GFP_ACCOUNT. 311 */ 312 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN, 313 VMALLOC_START, VMALLOC_END, 314 THREADINFO_GFP & ~__GFP_ACCOUNT, 315 PAGE_KERNEL, 316 0, node, __builtin_return_address(0)); 317 if (!stack) 318 return -ENOMEM; 319 320 vm = find_vm_area(stack); 321 if (memcg_charge_kernel_stack(vm)) { 322 vfree(stack); 323 return -ENOMEM; 324 } 325 /* 326 * We can't call find_vm_area() in interrupt context, and 327 * free_thread_stack() can be called in interrupt context, 328 * so cache the vm_struct. 329 */ 330 tsk->stack_vm_area = vm; 331 stack = kasan_reset_tag(stack); 332 tsk->stack = stack; 333 return 0; 334 } 335 336 static void free_thread_stack(struct task_struct *tsk) 337 { 338 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area)) 339 thread_stack_delayed_free(tsk); 340 341 tsk->stack = NULL; 342 tsk->stack_vm_area = NULL; 343 } 344 345 # else /* !CONFIG_VMAP_STACK */ 346 347 static void thread_stack_free_rcu(struct rcu_head *rh) 348 { 349 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER); 350 } 351 352 static void thread_stack_delayed_free(struct task_struct *tsk) 353 { 354 struct rcu_head *rh = tsk->stack; 355 356 call_rcu(rh, thread_stack_free_rcu); 357 } 358 359 static int alloc_thread_stack_node(struct task_struct *tsk, int node) 360 { 361 struct page *page = alloc_pages_node(node, THREADINFO_GFP, 362 THREAD_SIZE_ORDER); 363 364 if (likely(page)) { 365 tsk->stack = kasan_reset_tag(page_address(page)); 366 return 0; 367 } 368 return -ENOMEM; 369 } 370 371 static void free_thread_stack(struct task_struct *tsk) 372 { 373 thread_stack_delayed_free(tsk); 374 tsk->stack = NULL; 375 } 376 377 # endif /* CONFIG_VMAP_STACK */ 378 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */ 379 380 static struct kmem_cache *thread_stack_cache; 381 382 static void thread_stack_free_rcu(struct rcu_head *rh) 383 { 384 kmem_cache_free(thread_stack_cache, rh); 385 } 386 387 static void thread_stack_delayed_free(struct task_struct *tsk) 388 { 389 struct rcu_head *rh = tsk->stack; 390 391 call_rcu(rh, thread_stack_free_rcu); 392 } 393 394 static int alloc_thread_stack_node(struct task_struct *tsk, int node) 395 { 396 unsigned long *stack; 397 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node); 398 stack = kasan_reset_tag(stack); 399 tsk->stack = stack; 400 return stack ? 0 : -ENOMEM; 401 } 402 403 static void free_thread_stack(struct task_struct *tsk) 404 { 405 thread_stack_delayed_free(tsk); 406 tsk->stack = NULL; 407 } 408 409 void thread_stack_cache_init(void) 410 { 411 thread_stack_cache = kmem_cache_create_usercopy("thread_stack", 412 THREAD_SIZE, THREAD_SIZE, 0, 0, 413 THREAD_SIZE, NULL); 414 BUG_ON(thread_stack_cache == NULL); 415 } 416 417 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */ 418 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */ 419 420 static int alloc_thread_stack_node(struct task_struct *tsk, int node) 421 { 422 unsigned long *stack; 423 424 stack = arch_alloc_thread_stack_node(tsk, node); 425 tsk->stack = stack; 426 return stack ? 0 : -ENOMEM; 427 } 428 429 static void free_thread_stack(struct task_struct *tsk) 430 { 431 arch_free_thread_stack(tsk); 432 tsk->stack = NULL; 433 } 434 435 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */ 436 437 /* SLAB cache for signal_struct structures (tsk->signal) */ 438 static struct kmem_cache *signal_cachep; 439 440 /* SLAB cache for sighand_struct structures (tsk->sighand) */ 441 struct kmem_cache *sighand_cachep; 442 443 /* SLAB cache for files_struct structures (tsk->files) */ 444 struct kmem_cache *files_cachep; 445 446 /* SLAB cache for fs_struct structures (tsk->fs) */ 447 struct kmem_cache *fs_cachep; 448 449 /* SLAB cache for vm_area_struct structures */ 450 static struct kmem_cache *vm_area_cachep; 451 452 /* SLAB cache for mm_struct structures (tsk->mm) */ 453 static struct kmem_cache *mm_cachep; 454 455 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm) 456 { 457 struct vm_area_struct *vma; 458 459 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 460 if (vma) 461 vma_init(vma, mm); 462 return vma; 463 } 464 465 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig) 466 { 467 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 468 469 if (new) { 470 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags); 471 ASSERT_EXCLUSIVE_WRITER(orig->vm_file); 472 /* 473 * orig->shared.rb may be modified concurrently, but the clone 474 * will be reinitialized. 475 */ 476 *new = data_race(*orig); 477 INIT_LIST_HEAD(&new->anon_vma_chain); 478 new->vm_next = new->vm_prev = NULL; 479 dup_anon_vma_name(orig, new); 480 } 481 return new; 482 } 483 484 void vm_area_free(struct vm_area_struct *vma) 485 { 486 free_anon_vma_name(vma); 487 kmem_cache_free(vm_area_cachep, vma); 488 } 489 490 static void account_kernel_stack(struct task_struct *tsk, int account) 491 { 492 if (IS_ENABLED(CONFIG_VMAP_STACK)) { 493 struct vm_struct *vm = task_stack_vm_area(tsk); 494 int i; 495 496 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) 497 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB, 498 account * (PAGE_SIZE / 1024)); 499 } else { 500 void *stack = task_stack_page(tsk); 501 502 /* All stack pages are in the same node. */ 503 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB, 504 account * (THREAD_SIZE / 1024)); 505 } 506 } 507 508 void exit_task_stack_account(struct task_struct *tsk) 509 { 510 account_kernel_stack(tsk, -1); 511 512 if (IS_ENABLED(CONFIG_VMAP_STACK)) { 513 struct vm_struct *vm; 514 int i; 515 516 vm = task_stack_vm_area(tsk); 517 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) 518 memcg_kmem_uncharge_page(vm->pages[i], 0); 519 } 520 } 521 522 static void release_task_stack(struct task_struct *tsk) 523 { 524 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD)) 525 return; /* Better to leak the stack than to free prematurely */ 526 527 free_thread_stack(tsk); 528 } 529 530 #ifdef CONFIG_THREAD_INFO_IN_TASK 531 void put_task_stack(struct task_struct *tsk) 532 { 533 if (refcount_dec_and_test(&tsk->stack_refcount)) 534 release_task_stack(tsk); 535 } 536 #endif 537 538 void free_task(struct task_struct *tsk) 539 { 540 release_user_cpus_ptr(tsk); 541 scs_release(tsk); 542 543 #ifndef CONFIG_THREAD_INFO_IN_TASK 544 /* 545 * The task is finally done with both the stack and thread_info, 546 * so free both. 547 */ 548 release_task_stack(tsk); 549 #else 550 /* 551 * If the task had a separate stack allocation, it should be gone 552 * by now. 553 */ 554 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0); 555 #endif 556 rt_mutex_debug_task_free(tsk); 557 ftrace_graph_exit_task(tsk); 558 arch_release_task_struct(tsk); 559 if (tsk->flags & PF_KTHREAD) 560 free_kthread_struct(tsk); 561 free_task_struct(tsk); 562 } 563 EXPORT_SYMBOL(free_task); 564 565 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm) 566 { 567 struct file *exe_file; 568 569 exe_file = get_mm_exe_file(oldmm); 570 RCU_INIT_POINTER(mm->exe_file, exe_file); 571 /* 572 * We depend on the oldmm having properly denied write access to the 573 * exe_file already. 574 */ 575 if (exe_file && deny_write_access(exe_file)) 576 pr_warn_once("deny_write_access() failed in %s\n", __func__); 577 } 578 579 #ifdef CONFIG_MMU 580 static __latent_entropy int dup_mmap(struct mm_struct *mm, 581 struct mm_struct *oldmm) 582 { 583 struct vm_area_struct *mpnt, *tmp, *prev, **pprev; 584 struct rb_node **rb_link, *rb_parent; 585 int retval; 586 unsigned long charge; 587 LIST_HEAD(uf); 588 589 uprobe_start_dup_mmap(); 590 if (mmap_write_lock_killable(oldmm)) { 591 retval = -EINTR; 592 goto fail_uprobe_end; 593 } 594 flush_cache_dup_mm(oldmm); 595 uprobe_dup_mmap(oldmm, mm); 596 /* 597 * Not linked in yet - no deadlock potential: 598 */ 599 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING); 600 601 /* No ordering required: file already has been exposed. */ 602 dup_mm_exe_file(mm, oldmm); 603 604 mm->total_vm = oldmm->total_vm; 605 mm->data_vm = oldmm->data_vm; 606 mm->exec_vm = oldmm->exec_vm; 607 mm->stack_vm = oldmm->stack_vm; 608 609 rb_link = &mm->mm_rb.rb_node; 610 rb_parent = NULL; 611 pprev = &mm->mmap; 612 retval = ksm_fork(mm, oldmm); 613 if (retval) 614 goto out; 615 khugepaged_fork(mm, oldmm); 616 617 prev = NULL; 618 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { 619 struct file *file; 620 621 if (mpnt->vm_flags & VM_DONTCOPY) { 622 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt)); 623 continue; 624 } 625 charge = 0; 626 /* 627 * Don't duplicate many vmas if we've been oom-killed (for 628 * example) 629 */ 630 if (fatal_signal_pending(current)) { 631 retval = -EINTR; 632 goto out; 633 } 634 if (mpnt->vm_flags & VM_ACCOUNT) { 635 unsigned long len = vma_pages(mpnt); 636 637 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */ 638 goto fail_nomem; 639 charge = len; 640 } 641 tmp = vm_area_dup(mpnt); 642 if (!tmp) 643 goto fail_nomem; 644 retval = vma_dup_policy(mpnt, tmp); 645 if (retval) 646 goto fail_nomem_policy; 647 tmp->vm_mm = mm; 648 retval = dup_userfaultfd(tmp, &uf); 649 if (retval) 650 goto fail_nomem_anon_vma_fork; 651 if (tmp->vm_flags & VM_WIPEONFORK) { 652 /* 653 * VM_WIPEONFORK gets a clean slate in the child. 654 * Don't prepare anon_vma until fault since we don't 655 * copy page for current vma. 656 */ 657 tmp->anon_vma = NULL; 658 } else if (anon_vma_fork(tmp, mpnt)) 659 goto fail_nomem_anon_vma_fork; 660 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT); 661 file = tmp->vm_file; 662 if (file) { 663 struct address_space *mapping = file->f_mapping; 664 665 get_file(file); 666 i_mmap_lock_write(mapping); 667 if (tmp->vm_flags & VM_SHARED) 668 mapping_allow_writable(mapping); 669 flush_dcache_mmap_lock(mapping); 670 /* insert tmp into the share list, just after mpnt */ 671 vma_interval_tree_insert_after(tmp, mpnt, 672 &mapping->i_mmap); 673 flush_dcache_mmap_unlock(mapping); 674 i_mmap_unlock_write(mapping); 675 } 676 677 /* 678 * Clear hugetlb-related page reserves for children. This only 679 * affects MAP_PRIVATE mappings. Faults generated by the child 680 * are not guaranteed to succeed, even if read-only 681 */ 682 if (is_vm_hugetlb_page(tmp)) 683 reset_vma_resv_huge_pages(tmp); 684 685 /* 686 * Link in the new vma and copy the page table entries. 687 */ 688 *pprev = tmp; 689 pprev = &tmp->vm_next; 690 tmp->vm_prev = prev; 691 prev = tmp; 692 693 __vma_link_rb(mm, tmp, rb_link, rb_parent); 694 rb_link = &tmp->vm_rb.rb_right; 695 rb_parent = &tmp->vm_rb; 696 697 mm->map_count++; 698 if (!(tmp->vm_flags & VM_WIPEONFORK)) 699 retval = copy_page_range(tmp, mpnt); 700 701 if (tmp->vm_ops && tmp->vm_ops->open) 702 tmp->vm_ops->open(tmp); 703 704 if (retval) 705 goto out; 706 } 707 /* a new mm has just been created */ 708 retval = arch_dup_mmap(oldmm, mm); 709 out: 710 mmap_write_unlock(mm); 711 flush_tlb_mm(oldmm); 712 mmap_write_unlock(oldmm); 713 dup_userfaultfd_complete(&uf); 714 fail_uprobe_end: 715 uprobe_end_dup_mmap(); 716 return retval; 717 fail_nomem_anon_vma_fork: 718 mpol_put(vma_policy(tmp)); 719 fail_nomem_policy: 720 vm_area_free(tmp); 721 fail_nomem: 722 retval = -ENOMEM; 723 vm_unacct_memory(charge); 724 goto out; 725 } 726 727 static inline int mm_alloc_pgd(struct mm_struct *mm) 728 { 729 mm->pgd = pgd_alloc(mm); 730 if (unlikely(!mm->pgd)) 731 return -ENOMEM; 732 return 0; 733 } 734 735 static inline void mm_free_pgd(struct mm_struct *mm) 736 { 737 pgd_free(mm, mm->pgd); 738 } 739 #else 740 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) 741 { 742 mmap_write_lock(oldmm); 743 dup_mm_exe_file(mm, oldmm); 744 mmap_write_unlock(oldmm); 745 return 0; 746 } 747 #define mm_alloc_pgd(mm) (0) 748 #define mm_free_pgd(mm) 749 #endif /* CONFIG_MMU */ 750 751 static void check_mm(struct mm_struct *mm) 752 { 753 int i; 754 755 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS, 756 "Please make sure 'struct resident_page_types[]' is updated as well"); 757 758 for (i = 0; i < NR_MM_COUNTERS; i++) { 759 long x = atomic_long_read(&mm->rss_stat.count[i]); 760 761 if (unlikely(x)) 762 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n", 763 mm, resident_page_types[i], x); 764 } 765 766 if (mm_pgtables_bytes(mm)) 767 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n", 768 mm_pgtables_bytes(mm)); 769 770 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 771 VM_BUG_ON_MM(mm->pmd_huge_pte, mm); 772 #endif 773 } 774 775 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL)) 776 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) 777 778 /* 779 * Called when the last reference to the mm 780 * is dropped: either by a lazy thread or by 781 * mmput. Free the page directory and the mm. 782 */ 783 void __mmdrop(struct mm_struct *mm) 784 { 785 BUG_ON(mm == &init_mm); 786 WARN_ON_ONCE(mm == current->mm); 787 WARN_ON_ONCE(mm == current->active_mm); 788 mm_free_pgd(mm); 789 destroy_context(mm); 790 mmu_notifier_subscriptions_destroy(mm); 791 check_mm(mm); 792 put_user_ns(mm->user_ns); 793 mm_pasid_drop(mm); 794 free_mm(mm); 795 } 796 EXPORT_SYMBOL_GPL(__mmdrop); 797 798 static void mmdrop_async_fn(struct work_struct *work) 799 { 800 struct mm_struct *mm; 801 802 mm = container_of(work, struct mm_struct, async_put_work); 803 __mmdrop(mm); 804 } 805 806 static void mmdrop_async(struct mm_struct *mm) 807 { 808 if (unlikely(atomic_dec_and_test(&mm->mm_count))) { 809 INIT_WORK(&mm->async_put_work, mmdrop_async_fn); 810 schedule_work(&mm->async_put_work); 811 } 812 } 813 814 static inline void free_signal_struct(struct signal_struct *sig) 815 { 816 taskstats_tgid_free(sig); 817 sched_autogroup_exit(sig); 818 /* 819 * __mmdrop is not safe to call from softirq context on x86 due to 820 * pgd_dtor so postpone it to the async context 821 */ 822 if (sig->oom_mm) 823 mmdrop_async(sig->oom_mm); 824 kmem_cache_free(signal_cachep, sig); 825 } 826 827 static inline void put_signal_struct(struct signal_struct *sig) 828 { 829 if (refcount_dec_and_test(&sig->sigcnt)) 830 free_signal_struct(sig); 831 } 832 833 void __put_task_struct(struct task_struct *tsk) 834 { 835 WARN_ON(!tsk->exit_state); 836 WARN_ON(refcount_read(&tsk->usage)); 837 WARN_ON(tsk == current); 838 839 io_uring_free(tsk); 840 cgroup_free(tsk); 841 task_numa_free(tsk, true); 842 security_task_free(tsk); 843 bpf_task_storage_free(tsk); 844 exit_creds(tsk); 845 delayacct_tsk_free(tsk); 846 put_signal_struct(tsk->signal); 847 sched_core_free(tsk); 848 free_task(tsk); 849 } 850 EXPORT_SYMBOL_GPL(__put_task_struct); 851 852 void __init __weak arch_task_cache_init(void) { } 853 854 /* 855 * set_max_threads 856 */ 857 static void set_max_threads(unsigned int max_threads_suggested) 858 { 859 u64 threads; 860 unsigned long nr_pages = totalram_pages(); 861 862 /* 863 * The number of threads shall be limited such that the thread 864 * structures may only consume a small part of the available memory. 865 */ 866 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64) 867 threads = MAX_THREADS; 868 else 869 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE, 870 (u64) THREAD_SIZE * 8UL); 871 872 if (threads > max_threads_suggested) 873 threads = max_threads_suggested; 874 875 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS); 876 } 877 878 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT 879 /* Initialized by the architecture: */ 880 int arch_task_struct_size __read_mostly; 881 #endif 882 883 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 884 static void task_struct_whitelist(unsigned long *offset, unsigned long *size) 885 { 886 /* Fetch thread_struct whitelist for the architecture. */ 887 arch_thread_struct_whitelist(offset, size); 888 889 /* 890 * Handle zero-sized whitelist or empty thread_struct, otherwise 891 * adjust offset to position of thread_struct in task_struct. 892 */ 893 if (unlikely(*size == 0)) 894 *offset = 0; 895 else 896 *offset += offsetof(struct task_struct, thread); 897 } 898 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */ 899 900 void __init fork_init(void) 901 { 902 int i; 903 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR 904 #ifndef ARCH_MIN_TASKALIGN 905 #define ARCH_MIN_TASKALIGN 0 906 #endif 907 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN); 908 unsigned long useroffset, usersize; 909 910 /* create a slab on which task_structs can be allocated */ 911 task_struct_whitelist(&useroffset, &usersize); 912 task_struct_cachep = kmem_cache_create_usercopy("task_struct", 913 arch_task_struct_size, align, 914 SLAB_PANIC|SLAB_ACCOUNT, 915 useroffset, usersize, NULL); 916 #endif 917 918 /* do the arch specific task caches init */ 919 arch_task_cache_init(); 920 921 set_max_threads(MAX_THREADS); 922 923 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 924 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 925 init_task.signal->rlim[RLIMIT_SIGPENDING] = 926 init_task.signal->rlim[RLIMIT_NPROC]; 927 928 for (i = 0; i < MAX_PER_NAMESPACE_UCOUNTS; i++) 929 init_user_ns.ucount_max[i] = max_threads/2; 930 931 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY); 932 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY); 933 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY); 934 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY); 935 936 #ifdef CONFIG_VMAP_STACK 937 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache", 938 NULL, free_vm_stack_cache); 939 #endif 940 941 scs_init(); 942 943 lockdep_init_task(&init_task); 944 uprobes_init(); 945 } 946 947 int __weak arch_dup_task_struct(struct task_struct *dst, 948 struct task_struct *src) 949 { 950 *dst = *src; 951 return 0; 952 } 953 954 void set_task_stack_end_magic(struct task_struct *tsk) 955 { 956 unsigned long *stackend; 957 958 stackend = end_of_stack(tsk); 959 *stackend = STACK_END_MAGIC; /* for overflow detection */ 960 } 961 962 static struct task_struct *dup_task_struct(struct task_struct *orig, int node) 963 { 964 struct task_struct *tsk; 965 int err; 966 967 if (node == NUMA_NO_NODE) 968 node = tsk_fork_get_node(orig); 969 tsk = alloc_task_struct_node(node); 970 if (!tsk) 971 return NULL; 972 973 err = arch_dup_task_struct(tsk, orig); 974 if (err) 975 goto free_tsk; 976 977 err = alloc_thread_stack_node(tsk, node); 978 if (err) 979 goto free_tsk; 980 981 #ifdef CONFIG_THREAD_INFO_IN_TASK 982 refcount_set(&tsk->stack_refcount, 1); 983 #endif 984 account_kernel_stack(tsk, 1); 985 986 err = scs_prepare(tsk, node); 987 if (err) 988 goto free_stack; 989 990 #ifdef CONFIG_SECCOMP 991 /* 992 * We must handle setting up seccomp filters once we're under 993 * the sighand lock in case orig has changed between now and 994 * then. Until then, filter must be NULL to avoid messing up 995 * the usage counts on the error path calling free_task. 996 */ 997 tsk->seccomp.filter = NULL; 998 #endif 999 1000 setup_thread_stack(tsk, orig); 1001 clear_user_return_notifier(tsk); 1002 clear_tsk_need_resched(tsk); 1003 set_task_stack_end_magic(tsk); 1004 clear_syscall_work_syscall_user_dispatch(tsk); 1005 1006 #ifdef CONFIG_STACKPROTECTOR 1007 tsk->stack_canary = get_random_canary(); 1008 #endif 1009 if (orig->cpus_ptr == &orig->cpus_mask) 1010 tsk->cpus_ptr = &tsk->cpus_mask; 1011 dup_user_cpus_ptr(tsk, orig, node); 1012 1013 /* 1014 * One for the user space visible state that goes away when reaped. 1015 * One for the scheduler. 1016 */ 1017 refcount_set(&tsk->rcu_users, 2); 1018 /* One for the rcu users */ 1019 refcount_set(&tsk->usage, 1); 1020 #ifdef CONFIG_BLK_DEV_IO_TRACE 1021 tsk->btrace_seq = 0; 1022 #endif 1023 tsk->splice_pipe = NULL; 1024 tsk->task_frag.page = NULL; 1025 tsk->wake_q.next = NULL; 1026 tsk->worker_private = NULL; 1027 1028 kcov_task_init(tsk); 1029 kmap_local_fork(tsk); 1030 1031 #ifdef CONFIG_FAULT_INJECTION 1032 tsk->fail_nth = 0; 1033 #endif 1034 1035 #ifdef CONFIG_BLK_CGROUP 1036 tsk->throttle_queue = NULL; 1037 tsk->use_memdelay = 0; 1038 #endif 1039 1040 #ifdef CONFIG_IOMMU_SVA 1041 tsk->pasid_activated = 0; 1042 #endif 1043 1044 #ifdef CONFIG_MEMCG 1045 tsk->active_memcg = NULL; 1046 #endif 1047 1048 #ifdef CONFIG_CPU_SUP_INTEL 1049 tsk->reported_split_lock = 0; 1050 #endif 1051 1052 return tsk; 1053 1054 free_stack: 1055 exit_task_stack_account(tsk); 1056 free_thread_stack(tsk); 1057 free_tsk: 1058 free_task_struct(tsk); 1059 return NULL; 1060 } 1061 1062 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); 1063 1064 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT; 1065 1066 static int __init coredump_filter_setup(char *s) 1067 { 1068 default_dump_filter = 1069 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) & 1070 MMF_DUMP_FILTER_MASK; 1071 return 1; 1072 } 1073 1074 __setup("coredump_filter=", coredump_filter_setup); 1075 1076 #include <linux/init_task.h> 1077 1078 static void mm_init_aio(struct mm_struct *mm) 1079 { 1080 #ifdef CONFIG_AIO 1081 spin_lock_init(&mm->ioctx_lock); 1082 mm->ioctx_table = NULL; 1083 #endif 1084 } 1085 1086 static __always_inline void mm_clear_owner(struct mm_struct *mm, 1087 struct task_struct *p) 1088 { 1089 #ifdef CONFIG_MEMCG 1090 if (mm->owner == p) 1091 WRITE_ONCE(mm->owner, NULL); 1092 #endif 1093 } 1094 1095 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p) 1096 { 1097 #ifdef CONFIG_MEMCG 1098 mm->owner = p; 1099 #endif 1100 } 1101 1102 static void mm_init_uprobes_state(struct mm_struct *mm) 1103 { 1104 #ifdef CONFIG_UPROBES 1105 mm->uprobes_state.xol_area = NULL; 1106 #endif 1107 } 1108 1109 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p, 1110 struct user_namespace *user_ns) 1111 { 1112 mm->mmap = NULL; 1113 mm->mm_rb = RB_ROOT; 1114 mm->vmacache_seqnum = 0; 1115 atomic_set(&mm->mm_users, 1); 1116 atomic_set(&mm->mm_count, 1); 1117 seqcount_init(&mm->write_protect_seq); 1118 mmap_init_lock(mm); 1119 INIT_LIST_HEAD(&mm->mmlist); 1120 mm_pgtables_bytes_init(mm); 1121 mm->map_count = 0; 1122 mm->locked_vm = 0; 1123 atomic64_set(&mm->pinned_vm, 0); 1124 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat)); 1125 spin_lock_init(&mm->page_table_lock); 1126 spin_lock_init(&mm->arg_lock); 1127 mm_init_cpumask(mm); 1128 mm_init_aio(mm); 1129 mm_init_owner(mm, p); 1130 mm_pasid_init(mm); 1131 RCU_INIT_POINTER(mm->exe_file, NULL); 1132 mmu_notifier_subscriptions_init(mm); 1133 init_tlb_flush_pending(mm); 1134 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS 1135 mm->pmd_huge_pte = NULL; 1136 #endif 1137 mm_init_uprobes_state(mm); 1138 hugetlb_count_init(mm); 1139 1140 if (current->mm) { 1141 mm->flags = current->mm->flags & MMF_INIT_MASK; 1142 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK; 1143 } else { 1144 mm->flags = default_dump_filter; 1145 mm->def_flags = 0; 1146 } 1147 1148 if (mm_alloc_pgd(mm)) 1149 goto fail_nopgd; 1150 1151 if (init_new_context(p, mm)) 1152 goto fail_nocontext; 1153 1154 mm->user_ns = get_user_ns(user_ns); 1155 lru_gen_init_mm(mm); 1156 return mm; 1157 1158 fail_nocontext: 1159 mm_free_pgd(mm); 1160 fail_nopgd: 1161 free_mm(mm); 1162 return NULL; 1163 } 1164 1165 /* 1166 * Allocate and initialize an mm_struct. 1167 */ 1168 struct mm_struct *mm_alloc(void) 1169 { 1170 struct mm_struct *mm; 1171 1172 mm = allocate_mm(); 1173 if (!mm) 1174 return NULL; 1175 1176 memset(mm, 0, sizeof(*mm)); 1177 return mm_init(mm, current, current_user_ns()); 1178 } 1179 1180 static inline void __mmput(struct mm_struct *mm) 1181 { 1182 VM_BUG_ON(atomic_read(&mm->mm_users)); 1183 1184 uprobe_clear_state(mm); 1185 exit_aio(mm); 1186 ksm_exit(mm); 1187 khugepaged_exit(mm); /* must run before exit_mmap */ 1188 exit_mmap(mm); 1189 mm_put_huge_zero_page(mm); 1190 set_mm_exe_file(mm, NULL); 1191 if (!list_empty(&mm->mmlist)) { 1192 spin_lock(&mmlist_lock); 1193 list_del(&mm->mmlist); 1194 spin_unlock(&mmlist_lock); 1195 } 1196 if (mm->binfmt) 1197 module_put(mm->binfmt->module); 1198 lru_gen_del_mm(mm); 1199 mmdrop(mm); 1200 } 1201 1202 /* 1203 * Decrement the use count and release all resources for an mm. 1204 */ 1205 void mmput(struct mm_struct *mm) 1206 { 1207 might_sleep(); 1208 1209 if (atomic_dec_and_test(&mm->mm_users)) 1210 __mmput(mm); 1211 } 1212 EXPORT_SYMBOL_GPL(mmput); 1213 1214 #ifdef CONFIG_MMU 1215 static void mmput_async_fn(struct work_struct *work) 1216 { 1217 struct mm_struct *mm = container_of(work, struct mm_struct, 1218 async_put_work); 1219 1220 __mmput(mm); 1221 } 1222 1223 void mmput_async(struct mm_struct *mm) 1224 { 1225 if (atomic_dec_and_test(&mm->mm_users)) { 1226 INIT_WORK(&mm->async_put_work, mmput_async_fn); 1227 schedule_work(&mm->async_put_work); 1228 } 1229 } 1230 #endif 1231 1232 /** 1233 * set_mm_exe_file - change a reference to the mm's executable file 1234 * 1235 * This changes mm's executable file (shown as symlink /proc/[pid]/exe). 1236 * 1237 * Main users are mmput() and sys_execve(). Callers prevent concurrent 1238 * invocations: in mmput() nobody alive left, in execve task is single 1239 * threaded. 1240 * 1241 * Can only fail if new_exe_file != NULL. 1242 */ 1243 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 1244 { 1245 struct file *old_exe_file; 1246 1247 /* 1248 * It is safe to dereference the exe_file without RCU as 1249 * this function is only called if nobody else can access 1250 * this mm -- see comment above for justification. 1251 */ 1252 old_exe_file = rcu_dereference_raw(mm->exe_file); 1253 1254 if (new_exe_file) { 1255 /* 1256 * We expect the caller (i.e., sys_execve) to already denied 1257 * write access, so this is unlikely to fail. 1258 */ 1259 if (unlikely(deny_write_access(new_exe_file))) 1260 return -EACCES; 1261 get_file(new_exe_file); 1262 } 1263 rcu_assign_pointer(mm->exe_file, new_exe_file); 1264 if (old_exe_file) { 1265 allow_write_access(old_exe_file); 1266 fput(old_exe_file); 1267 } 1268 return 0; 1269 } 1270 1271 /** 1272 * replace_mm_exe_file - replace a reference to the mm's executable file 1273 * 1274 * This changes mm's executable file (shown as symlink /proc/[pid]/exe), 1275 * dealing with concurrent invocation and without grabbing the mmap lock in 1276 * write mode. 1277 * 1278 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE). 1279 */ 1280 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 1281 { 1282 struct vm_area_struct *vma; 1283 struct file *old_exe_file; 1284 int ret = 0; 1285 1286 /* Forbid mm->exe_file change if old file still mapped. */ 1287 old_exe_file = get_mm_exe_file(mm); 1288 if (old_exe_file) { 1289 mmap_read_lock(mm); 1290 for (vma = mm->mmap; vma && !ret; vma = vma->vm_next) { 1291 if (!vma->vm_file) 1292 continue; 1293 if (path_equal(&vma->vm_file->f_path, 1294 &old_exe_file->f_path)) 1295 ret = -EBUSY; 1296 } 1297 mmap_read_unlock(mm); 1298 fput(old_exe_file); 1299 if (ret) 1300 return ret; 1301 } 1302 1303 /* set the new file, lockless */ 1304 ret = deny_write_access(new_exe_file); 1305 if (ret) 1306 return -EACCES; 1307 get_file(new_exe_file); 1308 1309 old_exe_file = xchg(&mm->exe_file, new_exe_file); 1310 if (old_exe_file) { 1311 /* 1312 * Don't race with dup_mmap() getting the file and disallowing 1313 * write access while someone might open the file writable. 1314 */ 1315 mmap_read_lock(mm); 1316 allow_write_access(old_exe_file); 1317 fput(old_exe_file); 1318 mmap_read_unlock(mm); 1319 } 1320 return 0; 1321 } 1322 1323 /** 1324 * get_mm_exe_file - acquire a reference to the mm's executable file 1325 * 1326 * Returns %NULL if mm has no associated executable file. 1327 * User must release file via fput(). 1328 */ 1329 struct file *get_mm_exe_file(struct mm_struct *mm) 1330 { 1331 struct file *exe_file; 1332 1333 rcu_read_lock(); 1334 exe_file = rcu_dereference(mm->exe_file); 1335 if (exe_file && !get_file_rcu(exe_file)) 1336 exe_file = NULL; 1337 rcu_read_unlock(); 1338 return exe_file; 1339 } 1340 1341 /** 1342 * get_task_exe_file - acquire a reference to the task's executable file 1343 * 1344 * Returns %NULL if task's mm (if any) has no associated executable file or 1345 * this is a kernel thread with borrowed mm (see the comment above get_task_mm). 1346 * User must release file via fput(). 1347 */ 1348 struct file *get_task_exe_file(struct task_struct *task) 1349 { 1350 struct file *exe_file = NULL; 1351 struct mm_struct *mm; 1352 1353 task_lock(task); 1354 mm = task->mm; 1355 if (mm) { 1356 if (!(task->flags & PF_KTHREAD)) 1357 exe_file = get_mm_exe_file(mm); 1358 } 1359 task_unlock(task); 1360 return exe_file; 1361 } 1362 1363 /** 1364 * get_task_mm - acquire a reference to the task's mm 1365 * 1366 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning 1367 * this kernel workthread has transiently adopted a user mm with use_mm, 1368 * to do its AIO) is not set and if so returns a reference to it, after 1369 * bumping up the use count. User must release the mm via mmput() 1370 * after use. Typically used by /proc and ptrace. 1371 */ 1372 struct mm_struct *get_task_mm(struct task_struct *task) 1373 { 1374 struct mm_struct *mm; 1375 1376 task_lock(task); 1377 mm = task->mm; 1378 if (mm) { 1379 if (task->flags & PF_KTHREAD) 1380 mm = NULL; 1381 else 1382 mmget(mm); 1383 } 1384 task_unlock(task); 1385 return mm; 1386 } 1387 EXPORT_SYMBOL_GPL(get_task_mm); 1388 1389 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode) 1390 { 1391 struct mm_struct *mm; 1392 int err; 1393 1394 err = down_read_killable(&task->signal->exec_update_lock); 1395 if (err) 1396 return ERR_PTR(err); 1397 1398 mm = get_task_mm(task); 1399 if (mm && mm != current->mm && 1400 !ptrace_may_access(task, mode)) { 1401 mmput(mm); 1402 mm = ERR_PTR(-EACCES); 1403 } 1404 up_read(&task->signal->exec_update_lock); 1405 1406 return mm; 1407 } 1408 1409 static void complete_vfork_done(struct task_struct *tsk) 1410 { 1411 struct completion *vfork; 1412 1413 task_lock(tsk); 1414 vfork = tsk->vfork_done; 1415 if (likely(vfork)) { 1416 tsk->vfork_done = NULL; 1417 complete(vfork); 1418 } 1419 task_unlock(tsk); 1420 } 1421 1422 static int wait_for_vfork_done(struct task_struct *child, 1423 struct completion *vfork) 1424 { 1425 int killed; 1426 1427 freezer_do_not_count(); 1428 cgroup_enter_frozen(); 1429 killed = wait_for_completion_killable(vfork); 1430 cgroup_leave_frozen(false); 1431 freezer_count(); 1432 1433 if (killed) { 1434 task_lock(child); 1435 child->vfork_done = NULL; 1436 task_unlock(child); 1437 } 1438 1439 put_task_struct(child); 1440 return killed; 1441 } 1442 1443 /* Please note the differences between mmput and mm_release. 1444 * mmput is called whenever we stop holding onto a mm_struct, 1445 * error success whatever. 1446 * 1447 * mm_release is called after a mm_struct has been removed 1448 * from the current process. 1449 * 1450 * This difference is important for error handling, when we 1451 * only half set up a mm_struct for a new process and need to restore 1452 * the old one. Because we mmput the new mm_struct before 1453 * restoring the old one. . . 1454 * Eric Biederman 10 January 1998 1455 */ 1456 static void mm_release(struct task_struct *tsk, struct mm_struct *mm) 1457 { 1458 uprobe_free_utask(tsk); 1459 1460 /* Get rid of any cached register state */ 1461 deactivate_mm(tsk, mm); 1462 1463 /* 1464 * Signal userspace if we're not exiting with a core dump 1465 * because we want to leave the value intact for debugging 1466 * purposes. 1467 */ 1468 if (tsk->clear_child_tid) { 1469 if (atomic_read(&mm->mm_users) > 1) { 1470 /* 1471 * We don't check the error code - if userspace has 1472 * not set up a proper pointer then tough luck. 1473 */ 1474 put_user(0, tsk->clear_child_tid); 1475 do_futex(tsk->clear_child_tid, FUTEX_WAKE, 1476 1, NULL, NULL, 0, 0); 1477 } 1478 tsk->clear_child_tid = NULL; 1479 } 1480 1481 /* 1482 * All done, finally we can wake up parent and return this mm to him. 1483 * Also kthread_stop() uses this completion for synchronization. 1484 */ 1485 if (tsk->vfork_done) 1486 complete_vfork_done(tsk); 1487 } 1488 1489 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm) 1490 { 1491 futex_exit_release(tsk); 1492 mm_release(tsk, mm); 1493 } 1494 1495 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm) 1496 { 1497 futex_exec_release(tsk); 1498 mm_release(tsk, mm); 1499 } 1500 1501 /** 1502 * dup_mm() - duplicates an existing mm structure 1503 * @tsk: the task_struct with which the new mm will be associated. 1504 * @oldmm: the mm to duplicate. 1505 * 1506 * Allocates a new mm structure and duplicates the provided @oldmm structure 1507 * content into it. 1508 * 1509 * Return: the duplicated mm or NULL on failure. 1510 */ 1511 static struct mm_struct *dup_mm(struct task_struct *tsk, 1512 struct mm_struct *oldmm) 1513 { 1514 struct mm_struct *mm; 1515 int err; 1516 1517 mm = allocate_mm(); 1518 if (!mm) 1519 goto fail_nomem; 1520 1521 memcpy(mm, oldmm, sizeof(*mm)); 1522 1523 if (!mm_init(mm, tsk, mm->user_ns)) 1524 goto fail_nomem; 1525 1526 err = dup_mmap(mm, oldmm); 1527 if (err) 1528 goto free_pt; 1529 1530 mm->hiwater_rss = get_mm_rss(mm); 1531 mm->hiwater_vm = mm->total_vm; 1532 1533 if (mm->binfmt && !try_module_get(mm->binfmt->module)) 1534 goto free_pt; 1535 1536 return mm; 1537 1538 free_pt: 1539 /* don't put binfmt in mmput, we haven't got module yet */ 1540 mm->binfmt = NULL; 1541 mm_init_owner(mm, NULL); 1542 mmput(mm); 1543 1544 fail_nomem: 1545 return NULL; 1546 } 1547 1548 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk) 1549 { 1550 struct mm_struct *mm, *oldmm; 1551 1552 tsk->min_flt = tsk->maj_flt = 0; 1553 tsk->nvcsw = tsk->nivcsw = 0; 1554 #ifdef CONFIG_DETECT_HUNG_TASK 1555 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; 1556 tsk->last_switch_time = 0; 1557 #endif 1558 1559 tsk->mm = NULL; 1560 tsk->active_mm = NULL; 1561 1562 /* 1563 * Are we cloning a kernel thread? 1564 * 1565 * We need to steal a active VM for that.. 1566 */ 1567 oldmm = current->mm; 1568 if (!oldmm) 1569 return 0; 1570 1571 /* initialize the new vmacache entries */ 1572 vmacache_flush(tsk); 1573 1574 if (clone_flags & CLONE_VM) { 1575 mmget(oldmm); 1576 mm = oldmm; 1577 } else { 1578 mm = dup_mm(tsk, current->mm); 1579 if (!mm) 1580 return -ENOMEM; 1581 } 1582 1583 tsk->mm = mm; 1584 tsk->active_mm = mm; 1585 return 0; 1586 } 1587 1588 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 1589 { 1590 struct fs_struct *fs = current->fs; 1591 if (clone_flags & CLONE_FS) { 1592 /* tsk->fs is already what we want */ 1593 spin_lock(&fs->lock); 1594 if (fs->in_exec) { 1595 spin_unlock(&fs->lock); 1596 return -EAGAIN; 1597 } 1598 fs->users++; 1599 spin_unlock(&fs->lock); 1600 return 0; 1601 } 1602 tsk->fs = copy_fs_struct(fs); 1603 if (!tsk->fs) 1604 return -ENOMEM; 1605 return 0; 1606 } 1607 1608 static int copy_files(unsigned long clone_flags, struct task_struct *tsk) 1609 { 1610 struct files_struct *oldf, *newf; 1611 int error = 0; 1612 1613 /* 1614 * A background process may not have any files ... 1615 */ 1616 oldf = current->files; 1617 if (!oldf) 1618 goto out; 1619 1620 if (clone_flags & CLONE_FILES) { 1621 atomic_inc(&oldf->count); 1622 goto out; 1623 } 1624 1625 newf = dup_fd(oldf, NR_OPEN_MAX, &error); 1626 if (!newf) 1627 goto out; 1628 1629 tsk->files = newf; 1630 error = 0; 1631 out: 1632 return error; 1633 } 1634 1635 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk) 1636 { 1637 struct sighand_struct *sig; 1638 1639 if (clone_flags & CLONE_SIGHAND) { 1640 refcount_inc(¤t->sighand->count); 1641 return 0; 1642 } 1643 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 1644 RCU_INIT_POINTER(tsk->sighand, sig); 1645 if (!sig) 1646 return -ENOMEM; 1647 1648 refcount_set(&sig->count, 1); 1649 spin_lock_irq(¤t->sighand->siglock); 1650 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 1651 spin_unlock_irq(¤t->sighand->siglock); 1652 1653 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */ 1654 if (clone_flags & CLONE_CLEAR_SIGHAND) 1655 flush_signal_handlers(tsk, 0); 1656 1657 return 0; 1658 } 1659 1660 void __cleanup_sighand(struct sighand_struct *sighand) 1661 { 1662 if (refcount_dec_and_test(&sighand->count)) { 1663 signalfd_cleanup(sighand); 1664 /* 1665 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it 1666 * without an RCU grace period, see __lock_task_sighand(). 1667 */ 1668 kmem_cache_free(sighand_cachep, sighand); 1669 } 1670 } 1671 1672 /* 1673 * Initialize POSIX timer handling for a thread group. 1674 */ 1675 static void posix_cpu_timers_init_group(struct signal_struct *sig) 1676 { 1677 struct posix_cputimers *pct = &sig->posix_cputimers; 1678 unsigned long cpu_limit; 1679 1680 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); 1681 posix_cputimers_group_init(pct, cpu_limit); 1682 } 1683 1684 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk) 1685 { 1686 struct signal_struct *sig; 1687 1688 if (clone_flags & CLONE_THREAD) 1689 return 0; 1690 1691 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL); 1692 tsk->signal = sig; 1693 if (!sig) 1694 return -ENOMEM; 1695 1696 sig->nr_threads = 1; 1697 atomic_set(&sig->live, 1); 1698 refcount_set(&sig->sigcnt, 1); 1699 1700 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */ 1701 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node); 1702 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head); 1703 1704 init_waitqueue_head(&sig->wait_chldexit); 1705 sig->curr_target = tsk; 1706 init_sigpending(&sig->shared_pending); 1707 INIT_HLIST_HEAD(&sig->multiprocess); 1708 seqlock_init(&sig->stats_lock); 1709 prev_cputime_init(&sig->prev_cputime); 1710 1711 #ifdef CONFIG_POSIX_TIMERS 1712 INIT_LIST_HEAD(&sig->posix_timers); 1713 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1714 sig->real_timer.function = it_real_fn; 1715 #endif 1716 1717 task_lock(current->group_leader); 1718 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 1719 task_unlock(current->group_leader); 1720 1721 posix_cpu_timers_init_group(sig); 1722 1723 tty_audit_fork(sig); 1724 sched_autogroup_fork(sig); 1725 1726 sig->oom_score_adj = current->signal->oom_score_adj; 1727 sig->oom_score_adj_min = current->signal->oom_score_adj_min; 1728 1729 mutex_init(&sig->cred_guard_mutex); 1730 init_rwsem(&sig->exec_update_lock); 1731 1732 return 0; 1733 } 1734 1735 static void copy_seccomp(struct task_struct *p) 1736 { 1737 #ifdef CONFIG_SECCOMP 1738 /* 1739 * Must be called with sighand->lock held, which is common to 1740 * all threads in the group. Holding cred_guard_mutex is not 1741 * needed because this new task is not yet running and cannot 1742 * be racing exec. 1743 */ 1744 assert_spin_locked(¤t->sighand->siglock); 1745 1746 /* Ref-count the new filter user, and assign it. */ 1747 get_seccomp_filter(current); 1748 p->seccomp = current->seccomp; 1749 1750 /* 1751 * Explicitly enable no_new_privs here in case it got set 1752 * between the task_struct being duplicated and holding the 1753 * sighand lock. The seccomp state and nnp must be in sync. 1754 */ 1755 if (task_no_new_privs(current)) 1756 task_set_no_new_privs(p); 1757 1758 /* 1759 * If the parent gained a seccomp mode after copying thread 1760 * flags and between before we held the sighand lock, we have 1761 * to manually enable the seccomp thread flag here. 1762 */ 1763 if (p->seccomp.mode != SECCOMP_MODE_DISABLED) 1764 set_task_syscall_work(p, SECCOMP); 1765 #endif 1766 } 1767 1768 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr) 1769 { 1770 current->clear_child_tid = tidptr; 1771 1772 return task_pid_vnr(current); 1773 } 1774 1775 static void rt_mutex_init_task(struct task_struct *p) 1776 { 1777 raw_spin_lock_init(&p->pi_lock); 1778 #ifdef CONFIG_RT_MUTEXES 1779 p->pi_waiters = RB_ROOT_CACHED; 1780 p->pi_top_task = NULL; 1781 p->pi_blocked_on = NULL; 1782 #endif 1783 } 1784 1785 static inline void init_task_pid_links(struct task_struct *task) 1786 { 1787 enum pid_type type; 1788 1789 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) 1790 INIT_HLIST_NODE(&task->pid_links[type]); 1791 } 1792 1793 static inline void 1794 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid) 1795 { 1796 if (type == PIDTYPE_PID) 1797 task->thread_pid = pid; 1798 else 1799 task->signal->pids[type] = pid; 1800 } 1801 1802 static inline void rcu_copy_process(struct task_struct *p) 1803 { 1804 #ifdef CONFIG_PREEMPT_RCU 1805 p->rcu_read_lock_nesting = 0; 1806 p->rcu_read_unlock_special.s = 0; 1807 p->rcu_blocked_node = NULL; 1808 INIT_LIST_HEAD(&p->rcu_node_entry); 1809 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 1810 #ifdef CONFIG_TASKS_RCU 1811 p->rcu_tasks_holdout = false; 1812 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list); 1813 p->rcu_tasks_idle_cpu = -1; 1814 #endif /* #ifdef CONFIG_TASKS_RCU */ 1815 #ifdef CONFIG_TASKS_TRACE_RCU 1816 p->trc_reader_nesting = 0; 1817 p->trc_reader_special.s = 0; 1818 INIT_LIST_HEAD(&p->trc_holdout_list); 1819 INIT_LIST_HEAD(&p->trc_blkd_node); 1820 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */ 1821 } 1822 1823 struct pid *pidfd_pid(const struct file *file) 1824 { 1825 if (file->f_op == &pidfd_fops) 1826 return file->private_data; 1827 1828 return ERR_PTR(-EBADF); 1829 } 1830 1831 static int pidfd_release(struct inode *inode, struct file *file) 1832 { 1833 struct pid *pid = file->private_data; 1834 1835 file->private_data = NULL; 1836 put_pid(pid); 1837 return 0; 1838 } 1839 1840 #ifdef CONFIG_PROC_FS 1841 /** 1842 * pidfd_show_fdinfo - print information about a pidfd 1843 * @m: proc fdinfo file 1844 * @f: file referencing a pidfd 1845 * 1846 * Pid: 1847 * This function will print the pid that a given pidfd refers to in the 1848 * pid namespace of the procfs instance. 1849 * If the pid namespace of the process is not a descendant of the pid 1850 * namespace of the procfs instance 0 will be shown as its pid. This is 1851 * similar to calling getppid() on a process whose parent is outside of 1852 * its pid namespace. 1853 * 1854 * NSpid: 1855 * If pid namespaces are supported then this function will also print 1856 * the pid of a given pidfd refers to for all descendant pid namespaces 1857 * starting from the current pid namespace of the instance, i.e. the 1858 * Pid field and the first entry in the NSpid field will be identical. 1859 * If the pid namespace of the process is not a descendant of the pid 1860 * namespace of the procfs instance 0 will be shown as its first NSpid 1861 * entry and no others will be shown. 1862 * Note that this differs from the Pid and NSpid fields in 1863 * /proc/<pid>/status where Pid and NSpid are always shown relative to 1864 * the pid namespace of the procfs instance. The difference becomes 1865 * obvious when sending around a pidfd between pid namespaces from a 1866 * different branch of the tree, i.e. where no ancestral relation is 1867 * present between the pid namespaces: 1868 * - create two new pid namespaces ns1 and ns2 in the initial pid 1869 * namespace (also take care to create new mount namespaces in the 1870 * new pid namespace and mount procfs) 1871 * - create a process with a pidfd in ns1 1872 * - send pidfd from ns1 to ns2 1873 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid 1874 * have exactly one entry, which is 0 1875 */ 1876 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f) 1877 { 1878 struct pid *pid = f->private_data; 1879 struct pid_namespace *ns; 1880 pid_t nr = -1; 1881 1882 if (likely(pid_has_task(pid, PIDTYPE_PID))) { 1883 ns = proc_pid_ns(file_inode(m->file)->i_sb); 1884 nr = pid_nr_ns(pid, ns); 1885 } 1886 1887 seq_put_decimal_ll(m, "Pid:\t", nr); 1888 1889 #ifdef CONFIG_PID_NS 1890 seq_put_decimal_ll(m, "\nNSpid:\t", nr); 1891 if (nr > 0) { 1892 int i; 1893 1894 /* If nr is non-zero it means that 'pid' is valid and that 1895 * ns, i.e. the pid namespace associated with the procfs 1896 * instance, is in the pid namespace hierarchy of pid. 1897 * Start at one below the already printed level. 1898 */ 1899 for (i = ns->level + 1; i <= pid->level; i++) 1900 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr); 1901 } 1902 #endif 1903 seq_putc(m, '\n'); 1904 } 1905 #endif 1906 1907 /* 1908 * Poll support for process exit notification. 1909 */ 1910 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts) 1911 { 1912 struct pid *pid = file->private_data; 1913 __poll_t poll_flags = 0; 1914 1915 poll_wait(file, &pid->wait_pidfd, pts); 1916 1917 /* 1918 * Inform pollers only when the whole thread group exits. 1919 * If the thread group leader exits before all other threads in the 1920 * group, then poll(2) should block, similar to the wait(2) family. 1921 */ 1922 if (thread_group_exited(pid)) 1923 poll_flags = EPOLLIN | EPOLLRDNORM; 1924 1925 return poll_flags; 1926 } 1927 1928 const struct file_operations pidfd_fops = { 1929 .release = pidfd_release, 1930 .poll = pidfd_poll, 1931 #ifdef CONFIG_PROC_FS 1932 .show_fdinfo = pidfd_show_fdinfo, 1933 #endif 1934 }; 1935 1936 static void __delayed_free_task(struct rcu_head *rhp) 1937 { 1938 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); 1939 1940 free_task(tsk); 1941 } 1942 1943 static __always_inline void delayed_free_task(struct task_struct *tsk) 1944 { 1945 if (IS_ENABLED(CONFIG_MEMCG)) 1946 call_rcu(&tsk->rcu, __delayed_free_task); 1947 else 1948 free_task(tsk); 1949 } 1950 1951 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk) 1952 { 1953 /* Skip if kernel thread */ 1954 if (!tsk->mm) 1955 return; 1956 1957 /* Skip if spawning a thread or using vfork */ 1958 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM) 1959 return; 1960 1961 /* We need to synchronize with __set_oom_adj */ 1962 mutex_lock(&oom_adj_mutex); 1963 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags); 1964 /* Update the values in case they were changed after copy_signal */ 1965 tsk->signal->oom_score_adj = current->signal->oom_score_adj; 1966 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min; 1967 mutex_unlock(&oom_adj_mutex); 1968 } 1969 1970 #ifdef CONFIG_RV 1971 static void rv_task_fork(struct task_struct *p) 1972 { 1973 int i; 1974 1975 for (i = 0; i < RV_PER_TASK_MONITORS; i++) 1976 p->rv[i].da_mon.monitoring = false; 1977 } 1978 #else 1979 #define rv_task_fork(p) do {} while (0) 1980 #endif 1981 1982 /* 1983 * This creates a new process as a copy of the old one, 1984 * but does not actually start it yet. 1985 * 1986 * It copies the registers, and all the appropriate 1987 * parts of the process environment (as per the clone 1988 * flags). The actual kick-off is left to the caller. 1989 */ 1990 static __latent_entropy struct task_struct *copy_process( 1991 struct pid *pid, 1992 int trace, 1993 int node, 1994 struct kernel_clone_args *args) 1995 { 1996 int pidfd = -1, retval; 1997 struct task_struct *p; 1998 struct multiprocess_signals delayed; 1999 struct file *pidfile = NULL; 2000 const u64 clone_flags = args->flags; 2001 struct nsproxy *nsp = current->nsproxy; 2002 2003 /* 2004 * Don't allow sharing the root directory with processes in a different 2005 * namespace 2006 */ 2007 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 2008 return ERR_PTR(-EINVAL); 2009 2010 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS)) 2011 return ERR_PTR(-EINVAL); 2012 2013 /* 2014 * Thread groups must share signals as well, and detached threads 2015 * can only be started up within the thread group. 2016 */ 2017 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 2018 return ERR_PTR(-EINVAL); 2019 2020 /* 2021 * Shared signal handlers imply shared VM. By way of the above, 2022 * thread groups also imply shared VM. Blocking this case allows 2023 * for various simplifications in other code. 2024 */ 2025 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 2026 return ERR_PTR(-EINVAL); 2027 2028 /* 2029 * Siblings of global init remain as zombies on exit since they are 2030 * not reaped by their parent (swapper). To solve this and to avoid 2031 * multi-rooted process trees, prevent global and container-inits 2032 * from creating siblings. 2033 */ 2034 if ((clone_flags & CLONE_PARENT) && 2035 current->signal->flags & SIGNAL_UNKILLABLE) 2036 return ERR_PTR(-EINVAL); 2037 2038 /* 2039 * If the new process will be in a different pid or user namespace 2040 * do not allow it to share a thread group with the forking task. 2041 */ 2042 if (clone_flags & CLONE_THREAD) { 2043 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) || 2044 (task_active_pid_ns(current) != nsp->pid_ns_for_children)) 2045 return ERR_PTR(-EINVAL); 2046 } 2047 2048 /* 2049 * If the new process will be in a different time namespace 2050 * do not allow it to share VM or a thread group with the forking task. 2051 * 2052 * On vfork, the child process enters the target time namespace only 2053 * after exec. 2054 */ 2055 if ((clone_flags & (CLONE_VM | CLONE_VFORK)) == CLONE_VM) { 2056 if (nsp->time_ns != nsp->time_ns_for_children) 2057 return ERR_PTR(-EINVAL); 2058 } 2059 2060 if (clone_flags & CLONE_PIDFD) { 2061 /* 2062 * - CLONE_DETACHED is blocked so that we can potentially 2063 * reuse it later for CLONE_PIDFD. 2064 * - CLONE_THREAD is blocked until someone really needs it. 2065 */ 2066 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD)) 2067 return ERR_PTR(-EINVAL); 2068 } 2069 2070 /* 2071 * Force any signals received before this point to be delivered 2072 * before the fork happens. Collect up signals sent to multiple 2073 * processes that happen during the fork and delay them so that 2074 * they appear to happen after the fork. 2075 */ 2076 sigemptyset(&delayed.signal); 2077 INIT_HLIST_NODE(&delayed.node); 2078 2079 spin_lock_irq(¤t->sighand->siglock); 2080 if (!(clone_flags & CLONE_THREAD)) 2081 hlist_add_head(&delayed.node, ¤t->signal->multiprocess); 2082 recalc_sigpending(); 2083 spin_unlock_irq(¤t->sighand->siglock); 2084 retval = -ERESTARTNOINTR; 2085 if (task_sigpending(current)) 2086 goto fork_out; 2087 2088 retval = -ENOMEM; 2089 p = dup_task_struct(current, node); 2090 if (!p) 2091 goto fork_out; 2092 p->flags &= ~PF_KTHREAD; 2093 if (args->kthread) 2094 p->flags |= PF_KTHREAD; 2095 if (args->io_thread) { 2096 /* 2097 * Mark us an IO worker, and block any signal that isn't 2098 * fatal or STOP 2099 */ 2100 p->flags |= PF_IO_WORKER; 2101 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP)); 2102 } 2103 2104 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL; 2105 /* 2106 * Clear TID on mm_release()? 2107 */ 2108 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL; 2109 2110 ftrace_graph_init_task(p); 2111 2112 rt_mutex_init_task(p); 2113 2114 lockdep_assert_irqs_enabled(); 2115 #ifdef CONFIG_PROVE_LOCKING 2116 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 2117 #endif 2118 retval = copy_creds(p, clone_flags); 2119 if (retval < 0) 2120 goto bad_fork_free; 2121 2122 retval = -EAGAIN; 2123 if (is_ucounts_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) { 2124 if (p->real_cred->user != INIT_USER && 2125 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) 2126 goto bad_fork_cleanup_count; 2127 } 2128 current->flags &= ~PF_NPROC_EXCEEDED; 2129 2130 /* 2131 * If multiple threads are within copy_process(), then this check 2132 * triggers too late. This doesn't hurt, the check is only there 2133 * to stop root fork bombs. 2134 */ 2135 retval = -EAGAIN; 2136 if (data_race(nr_threads >= max_threads)) 2137 goto bad_fork_cleanup_count; 2138 2139 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */ 2140 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY); 2141 p->flags |= PF_FORKNOEXEC; 2142 INIT_LIST_HEAD(&p->children); 2143 INIT_LIST_HEAD(&p->sibling); 2144 rcu_copy_process(p); 2145 p->vfork_done = NULL; 2146 spin_lock_init(&p->alloc_lock); 2147 2148 init_sigpending(&p->pending); 2149 2150 p->utime = p->stime = p->gtime = 0; 2151 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 2152 p->utimescaled = p->stimescaled = 0; 2153 #endif 2154 prev_cputime_init(&p->prev_cputime); 2155 2156 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN 2157 seqcount_init(&p->vtime.seqcount); 2158 p->vtime.starttime = 0; 2159 p->vtime.state = VTIME_INACTIVE; 2160 #endif 2161 2162 #ifdef CONFIG_IO_URING 2163 p->io_uring = NULL; 2164 #endif 2165 2166 #if defined(SPLIT_RSS_COUNTING) 2167 memset(&p->rss_stat, 0, sizeof(p->rss_stat)); 2168 #endif 2169 2170 p->default_timer_slack_ns = current->timer_slack_ns; 2171 2172 #ifdef CONFIG_PSI 2173 p->psi_flags = 0; 2174 #endif 2175 2176 task_io_accounting_init(&p->ioac); 2177 acct_clear_integrals(p); 2178 2179 posix_cputimers_init(&p->posix_cputimers); 2180 2181 p->io_context = NULL; 2182 audit_set_context(p, NULL); 2183 cgroup_fork(p); 2184 if (args->kthread) { 2185 if (!set_kthread_struct(p)) 2186 goto bad_fork_cleanup_delayacct; 2187 } 2188 #ifdef CONFIG_NUMA 2189 p->mempolicy = mpol_dup(p->mempolicy); 2190 if (IS_ERR(p->mempolicy)) { 2191 retval = PTR_ERR(p->mempolicy); 2192 p->mempolicy = NULL; 2193 goto bad_fork_cleanup_delayacct; 2194 } 2195 #endif 2196 #ifdef CONFIG_CPUSETS 2197 p->cpuset_mem_spread_rotor = NUMA_NO_NODE; 2198 p->cpuset_slab_spread_rotor = NUMA_NO_NODE; 2199 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock); 2200 #endif 2201 #ifdef CONFIG_TRACE_IRQFLAGS 2202 memset(&p->irqtrace, 0, sizeof(p->irqtrace)); 2203 p->irqtrace.hardirq_disable_ip = _THIS_IP_; 2204 p->irqtrace.softirq_enable_ip = _THIS_IP_; 2205 p->softirqs_enabled = 1; 2206 p->softirq_context = 0; 2207 #endif 2208 2209 p->pagefault_disabled = 0; 2210 2211 #ifdef CONFIG_LOCKDEP 2212 lockdep_init_task(p); 2213 #endif 2214 2215 #ifdef CONFIG_DEBUG_MUTEXES 2216 p->blocked_on = NULL; /* not blocked yet */ 2217 #endif 2218 #ifdef CONFIG_BCACHE 2219 p->sequential_io = 0; 2220 p->sequential_io_avg = 0; 2221 #endif 2222 #ifdef CONFIG_BPF_SYSCALL 2223 RCU_INIT_POINTER(p->bpf_storage, NULL); 2224 p->bpf_ctx = NULL; 2225 #endif 2226 2227 /* Perform scheduler related setup. Assign this task to a CPU. */ 2228 retval = sched_fork(clone_flags, p); 2229 if (retval) 2230 goto bad_fork_cleanup_policy; 2231 2232 retval = perf_event_init_task(p, clone_flags); 2233 if (retval) 2234 goto bad_fork_cleanup_policy; 2235 retval = audit_alloc(p); 2236 if (retval) 2237 goto bad_fork_cleanup_perf; 2238 /* copy all the process information */ 2239 shm_init_task(p); 2240 retval = security_task_alloc(p, clone_flags); 2241 if (retval) 2242 goto bad_fork_cleanup_audit; 2243 retval = copy_semundo(clone_flags, p); 2244 if (retval) 2245 goto bad_fork_cleanup_security; 2246 retval = copy_files(clone_flags, p); 2247 if (retval) 2248 goto bad_fork_cleanup_semundo; 2249 retval = copy_fs(clone_flags, p); 2250 if (retval) 2251 goto bad_fork_cleanup_files; 2252 retval = copy_sighand(clone_flags, p); 2253 if (retval) 2254 goto bad_fork_cleanup_fs; 2255 retval = copy_signal(clone_flags, p); 2256 if (retval) 2257 goto bad_fork_cleanup_sighand; 2258 retval = copy_mm(clone_flags, p); 2259 if (retval) 2260 goto bad_fork_cleanup_signal; 2261 retval = copy_namespaces(clone_flags, p); 2262 if (retval) 2263 goto bad_fork_cleanup_mm; 2264 retval = copy_io(clone_flags, p); 2265 if (retval) 2266 goto bad_fork_cleanup_namespaces; 2267 retval = copy_thread(p, args); 2268 if (retval) 2269 goto bad_fork_cleanup_io; 2270 2271 stackleak_task_init(p); 2272 2273 if (pid != &init_struct_pid) { 2274 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid, 2275 args->set_tid_size); 2276 if (IS_ERR(pid)) { 2277 retval = PTR_ERR(pid); 2278 goto bad_fork_cleanup_thread; 2279 } 2280 } 2281 2282 /* 2283 * This has to happen after we've potentially unshared the file 2284 * descriptor table (so that the pidfd doesn't leak into the child 2285 * if the fd table isn't shared). 2286 */ 2287 if (clone_flags & CLONE_PIDFD) { 2288 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC); 2289 if (retval < 0) 2290 goto bad_fork_free_pid; 2291 2292 pidfd = retval; 2293 2294 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid, 2295 O_RDWR | O_CLOEXEC); 2296 if (IS_ERR(pidfile)) { 2297 put_unused_fd(pidfd); 2298 retval = PTR_ERR(pidfile); 2299 goto bad_fork_free_pid; 2300 } 2301 get_pid(pid); /* held by pidfile now */ 2302 2303 retval = put_user(pidfd, args->pidfd); 2304 if (retval) 2305 goto bad_fork_put_pidfd; 2306 } 2307 2308 #ifdef CONFIG_BLOCK 2309 p->plug = NULL; 2310 #endif 2311 futex_init_task(p); 2312 2313 /* 2314 * sigaltstack should be cleared when sharing the same VM 2315 */ 2316 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM) 2317 sas_ss_reset(p); 2318 2319 /* 2320 * Syscall tracing and stepping should be turned off in the 2321 * child regardless of CLONE_PTRACE. 2322 */ 2323 user_disable_single_step(p); 2324 clear_task_syscall_work(p, SYSCALL_TRACE); 2325 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU) 2326 clear_task_syscall_work(p, SYSCALL_EMU); 2327 #endif 2328 clear_tsk_latency_tracing(p); 2329 2330 /* ok, now we should be set up.. */ 2331 p->pid = pid_nr(pid); 2332 if (clone_flags & CLONE_THREAD) { 2333 p->group_leader = current->group_leader; 2334 p->tgid = current->tgid; 2335 } else { 2336 p->group_leader = p; 2337 p->tgid = p->pid; 2338 } 2339 2340 p->nr_dirtied = 0; 2341 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10); 2342 p->dirty_paused_when = 0; 2343 2344 p->pdeath_signal = 0; 2345 INIT_LIST_HEAD(&p->thread_group); 2346 p->task_works = NULL; 2347 clear_posix_cputimers_work(p); 2348 2349 #ifdef CONFIG_KRETPROBES 2350 p->kretprobe_instances.first = NULL; 2351 #endif 2352 #ifdef CONFIG_RETHOOK 2353 p->rethooks.first = NULL; 2354 #endif 2355 2356 /* 2357 * Ensure that the cgroup subsystem policies allow the new process to be 2358 * forked. It should be noted that the new process's css_set can be changed 2359 * between here and cgroup_post_fork() if an organisation operation is in 2360 * progress. 2361 */ 2362 retval = cgroup_can_fork(p, args); 2363 if (retval) 2364 goto bad_fork_put_pidfd; 2365 2366 /* 2367 * Now that the cgroups are pinned, re-clone the parent cgroup and put 2368 * the new task on the correct runqueue. All this *before* the task 2369 * becomes visible. 2370 * 2371 * This isn't part of ->can_fork() because while the re-cloning is 2372 * cgroup specific, it unconditionally needs to place the task on a 2373 * runqueue. 2374 */ 2375 sched_cgroup_fork(p, args); 2376 2377 /* 2378 * From this point on we must avoid any synchronous user-space 2379 * communication until we take the tasklist-lock. In particular, we do 2380 * not want user-space to be able to predict the process start-time by 2381 * stalling fork(2) after we recorded the start_time but before it is 2382 * visible to the system. 2383 */ 2384 2385 p->start_time = ktime_get_ns(); 2386 p->start_boottime = ktime_get_boottime_ns(); 2387 2388 /* 2389 * Make it visible to the rest of the system, but dont wake it up yet. 2390 * Need tasklist lock for parent etc handling! 2391 */ 2392 write_lock_irq(&tasklist_lock); 2393 2394 /* CLONE_PARENT re-uses the old parent */ 2395 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { 2396 p->real_parent = current->real_parent; 2397 p->parent_exec_id = current->parent_exec_id; 2398 if (clone_flags & CLONE_THREAD) 2399 p->exit_signal = -1; 2400 else 2401 p->exit_signal = current->group_leader->exit_signal; 2402 } else { 2403 p->real_parent = current; 2404 p->parent_exec_id = current->self_exec_id; 2405 p->exit_signal = args->exit_signal; 2406 } 2407 2408 klp_copy_process(p); 2409 2410 sched_core_fork(p); 2411 2412 spin_lock(¤t->sighand->siglock); 2413 2414 /* 2415 * Copy seccomp details explicitly here, in case they were changed 2416 * before holding sighand lock. 2417 */ 2418 copy_seccomp(p); 2419 2420 rv_task_fork(p); 2421 2422 rseq_fork(p, clone_flags); 2423 2424 /* Don't start children in a dying pid namespace */ 2425 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) { 2426 retval = -ENOMEM; 2427 goto bad_fork_cancel_cgroup; 2428 } 2429 2430 /* Let kill terminate clone/fork in the middle */ 2431 if (fatal_signal_pending(current)) { 2432 retval = -EINTR; 2433 goto bad_fork_cancel_cgroup; 2434 } 2435 2436 init_task_pid_links(p); 2437 if (likely(p->pid)) { 2438 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace); 2439 2440 init_task_pid(p, PIDTYPE_PID, pid); 2441 if (thread_group_leader(p)) { 2442 init_task_pid(p, PIDTYPE_TGID, pid); 2443 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current)); 2444 init_task_pid(p, PIDTYPE_SID, task_session(current)); 2445 2446 if (is_child_reaper(pid)) { 2447 ns_of_pid(pid)->child_reaper = p; 2448 p->signal->flags |= SIGNAL_UNKILLABLE; 2449 } 2450 p->signal->shared_pending.signal = delayed.signal; 2451 p->signal->tty = tty_kref_get(current->signal->tty); 2452 /* 2453 * Inherit has_child_subreaper flag under the same 2454 * tasklist_lock with adding child to the process tree 2455 * for propagate_has_child_subreaper optimization. 2456 */ 2457 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper || 2458 p->real_parent->signal->is_child_subreaper; 2459 list_add_tail(&p->sibling, &p->real_parent->children); 2460 list_add_tail_rcu(&p->tasks, &init_task.tasks); 2461 attach_pid(p, PIDTYPE_TGID); 2462 attach_pid(p, PIDTYPE_PGID); 2463 attach_pid(p, PIDTYPE_SID); 2464 __this_cpu_inc(process_counts); 2465 } else { 2466 current->signal->nr_threads++; 2467 atomic_inc(¤t->signal->live); 2468 refcount_inc(¤t->signal->sigcnt); 2469 task_join_group_stop(p); 2470 list_add_tail_rcu(&p->thread_group, 2471 &p->group_leader->thread_group); 2472 list_add_tail_rcu(&p->thread_node, 2473 &p->signal->thread_head); 2474 } 2475 attach_pid(p, PIDTYPE_PID); 2476 nr_threads++; 2477 } 2478 total_forks++; 2479 hlist_del_init(&delayed.node); 2480 spin_unlock(¤t->sighand->siglock); 2481 syscall_tracepoint_update(p); 2482 write_unlock_irq(&tasklist_lock); 2483 2484 if (pidfile) 2485 fd_install(pidfd, pidfile); 2486 2487 proc_fork_connector(p); 2488 sched_post_fork(p); 2489 cgroup_post_fork(p, args); 2490 perf_event_fork(p); 2491 2492 trace_task_newtask(p, clone_flags); 2493 uprobe_copy_process(p, clone_flags); 2494 2495 copy_oom_score_adj(clone_flags, p); 2496 2497 return p; 2498 2499 bad_fork_cancel_cgroup: 2500 sched_core_free(p); 2501 spin_unlock(¤t->sighand->siglock); 2502 write_unlock_irq(&tasklist_lock); 2503 cgroup_cancel_fork(p, args); 2504 bad_fork_put_pidfd: 2505 if (clone_flags & CLONE_PIDFD) { 2506 fput(pidfile); 2507 put_unused_fd(pidfd); 2508 } 2509 bad_fork_free_pid: 2510 if (pid != &init_struct_pid) 2511 free_pid(pid); 2512 bad_fork_cleanup_thread: 2513 exit_thread(p); 2514 bad_fork_cleanup_io: 2515 if (p->io_context) 2516 exit_io_context(p); 2517 bad_fork_cleanup_namespaces: 2518 exit_task_namespaces(p); 2519 bad_fork_cleanup_mm: 2520 if (p->mm) { 2521 mm_clear_owner(p->mm, p); 2522 mmput(p->mm); 2523 } 2524 bad_fork_cleanup_signal: 2525 if (!(clone_flags & CLONE_THREAD)) 2526 free_signal_struct(p->signal); 2527 bad_fork_cleanup_sighand: 2528 __cleanup_sighand(p->sighand); 2529 bad_fork_cleanup_fs: 2530 exit_fs(p); /* blocking */ 2531 bad_fork_cleanup_files: 2532 exit_files(p); /* blocking */ 2533 bad_fork_cleanup_semundo: 2534 exit_sem(p); 2535 bad_fork_cleanup_security: 2536 security_task_free(p); 2537 bad_fork_cleanup_audit: 2538 audit_free(p); 2539 bad_fork_cleanup_perf: 2540 perf_event_free_task(p); 2541 bad_fork_cleanup_policy: 2542 lockdep_free_task(p); 2543 #ifdef CONFIG_NUMA 2544 mpol_put(p->mempolicy); 2545 #endif 2546 bad_fork_cleanup_delayacct: 2547 delayacct_tsk_free(p); 2548 bad_fork_cleanup_count: 2549 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1); 2550 exit_creds(p); 2551 bad_fork_free: 2552 WRITE_ONCE(p->__state, TASK_DEAD); 2553 exit_task_stack_account(p); 2554 put_task_stack(p); 2555 delayed_free_task(p); 2556 fork_out: 2557 spin_lock_irq(¤t->sighand->siglock); 2558 hlist_del_init(&delayed.node); 2559 spin_unlock_irq(¤t->sighand->siglock); 2560 return ERR_PTR(retval); 2561 } 2562 2563 static inline void init_idle_pids(struct task_struct *idle) 2564 { 2565 enum pid_type type; 2566 2567 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 2568 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */ 2569 init_task_pid(idle, type, &init_struct_pid); 2570 } 2571 } 2572 2573 static int idle_dummy(void *dummy) 2574 { 2575 /* This function is never called */ 2576 return 0; 2577 } 2578 2579 struct task_struct * __init fork_idle(int cpu) 2580 { 2581 struct task_struct *task; 2582 struct kernel_clone_args args = { 2583 .flags = CLONE_VM, 2584 .fn = &idle_dummy, 2585 .fn_arg = NULL, 2586 .kthread = 1, 2587 .idle = 1, 2588 }; 2589 2590 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args); 2591 if (!IS_ERR(task)) { 2592 init_idle_pids(task); 2593 init_idle(task, cpu); 2594 } 2595 2596 return task; 2597 } 2598 2599 struct mm_struct *copy_init_mm(void) 2600 { 2601 return dup_mm(NULL, &init_mm); 2602 } 2603 2604 /* 2605 * This is like kernel_clone(), but shaved down and tailored to just 2606 * creating io_uring workers. It returns a created task, or an error pointer. 2607 * The returned task is inactive, and the caller must fire it up through 2608 * wake_up_new_task(p). All signals are blocked in the created task. 2609 */ 2610 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node) 2611 { 2612 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD| 2613 CLONE_IO; 2614 struct kernel_clone_args args = { 2615 .flags = ((lower_32_bits(flags) | CLONE_VM | 2616 CLONE_UNTRACED) & ~CSIGNAL), 2617 .exit_signal = (lower_32_bits(flags) & CSIGNAL), 2618 .fn = fn, 2619 .fn_arg = arg, 2620 .io_thread = 1, 2621 }; 2622 2623 return copy_process(NULL, 0, node, &args); 2624 } 2625 2626 /* 2627 * Ok, this is the main fork-routine. 2628 * 2629 * It copies the process, and if successful kick-starts 2630 * it and waits for it to finish using the VM if required. 2631 * 2632 * args->exit_signal is expected to be checked for sanity by the caller. 2633 */ 2634 pid_t kernel_clone(struct kernel_clone_args *args) 2635 { 2636 u64 clone_flags = args->flags; 2637 struct completion vfork; 2638 struct pid *pid; 2639 struct task_struct *p; 2640 int trace = 0; 2641 pid_t nr; 2642 2643 /* 2644 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument 2645 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are 2646 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate 2647 * field in struct clone_args and it still doesn't make sense to have 2648 * them both point at the same memory location. Performing this check 2649 * here has the advantage that we don't need to have a separate helper 2650 * to check for legacy clone(). 2651 */ 2652 if ((args->flags & CLONE_PIDFD) && 2653 (args->flags & CLONE_PARENT_SETTID) && 2654 (args->pidfd == args->parent_tid)) 2655 return -EINVAL; 2656 2657 /* 2658 * Determine whether and which event to report to ptracer. When 2659 * called from kernel_thread or CLONE_UNTRACED is explicitly 2660 * requested, no event is reported; otherwise, report if the event 2661 * for the type of forking is enabled. 2662 */ 2663 if (!(clone_flags & CLONE_UNTRACED)) { 2664 if (clone_flags & CLONE_VFORK) 2665 trace = PTRACE_EVENT_VFORK; 2666 else if (args->exit_signal != SIGCHLD) 2667 trace = PTRACE_EVENT_CLONE; 2668 else 2669 trace = PTRACE_EVENT_FORK; 2670 2671 if (likely(!ptrace_event_enabled(current, trace))) 2672 trace = 0; 2673 } 2674 2675 p = copy_process(NULL, trace, NUMA_NO_NODE, args); 2676 add_latent_entropy(); 2677 2678 if (IS_ERR(p)) 2679 return PTR_ERR(p); 2680 2681 /* 2682 * Do this prior waking up the new thread - the thread pointer 2683 * might get invalid after that point, if the thread exits quickly. 2684 */ 2685 trace_sched_process_fork(current, p); 2686 2687 pid = get_task_pid(p, PIDTYPE_PID); 2688 nr = pid_vnr(pid); 2689 2690 if (clone_flags & CLONE_PARENT_SETTID) 2691 put_user(nr, args->parent_tid); 2692 2693 if (clone_flags & CLONE_VFORK) { 2694 p->vfork_done = &vfork; 2695 init_completion(&vfork); 2696 get_task_struct(p); 2697 } 2698 2699 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) { 2700 /* lock the task to synchronize with memcg migration */ 2701 task_lock(p); 2702 lru_gen_add_mm(p->mm); 2703 task_unlock(p); 2704 } 2705 2706 wake_up_new_task(p); 2707 2708 /* forking complete and child started to run, tell ptracer */ 2709 if (unlikely(trace)) 2710 ptrace_event_pid(trace, pid); 2711 2712 if (clone_flags & CLONE_VFORK) { 2713 if (!wait_for_vfork_done(p, &vfork)) 2714 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid); 2715 } 2716 2717 put_pid(pid); 2718 return nr; 2719 } 2720 2721 /* 2722 * Create a kernel thread. 2723 */ 2724 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags) 2725 { 2726 struct kernel_clone_args args = { 2727 .flags = ((lower_32_bits(flags) | CLONE_VM | 2728 CLONE_UNTRACED) & ~CSIGNAL), 2729 .exit_signal = (lower_32_bits(flags) & CSIGNAL), 2730 .fn = fn, 2731 .fn_arg = arg, 2732 .kthread = 1, 2733 }; 2734 2735 return kernel_clone(&args); 2736 } 2737 2738 /* 2739 * Create a user mode thread. 2740 */ 2741 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags) 2742 { 2743 struct kernel_clone_args args = { 2744 .flags = ((lower_32_bits(flags) | CLONE_VM | 2745 CLONE_UNTRACED) & ~CSIGNAL), 2746 .exit_signal = (lower_32_bits(flags) & CSIGNAL), 2747 .fn = fn, 2748 .fn_arg = arg, 2749 }; 2750 2751 return kernel_clone(&args); 2752 } 2753 2754 #ifdef __ARCH_WANT_SYS_FORK 2755 SYSCALL_DEFINE0(fork) 2756 { 2757 #ifdef CONFIG_MMU 2758 struct kernel_clone_args args = { 2759 .exit_signal = SIGCHLD, 2760 }; 2761 2762 return kernel_clone(&args); 2763 #else 2764 /* can not support in nommu mode */ 2765 return -EINVAL; 2766 #endif 2767 } 2768 #endif 2769 2770 #ifdef __ARCH_WANT_SYS_VFORK 2771 SYSCALL_DEFINE0(vfork) 2772 { 2773 struct kernel_clone_args args = { 2774 .flags = CLONE_VFORK | CLONE_VM, 2775 .exit_signal = SIGCHLD, 2776 }; 2777 2778 return kernel_clone(&args); 2779 } 2780 #endif 2781 2782 #ifdef __ARCH_WANT_SYS_CLONE 2783 #ifdef CONFIG_CLONE_BACKWARDS 2784 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2785 int __user *, parent_tidptr, 2786 unsigned long, tls, 2787 int __user *, child_tidptr) 2788 #elif defined(CONFIG_CLONE_BACKWARDS2) 2789 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags, 2790 int __user *, parent_tidptr, 2791 int __user *, child_tidptr, 2792 unsigned long, tls) 2793 #elif defined(CONFIG_CLONE_BACKWARDS3) 2794 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp, 2795 int, stack_size, 2796 int __user *, parent_tidptr, 2797 int __user *, child_tidptr, 2798 unsigned long, tls) 2799 #else 2800 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2801 int __user *, parent_tidptr, 2802 int __user *, child_tidptr, 2803 unsigned long, tls) 2804 #endif 2805 { 2806 struct kernel_clone_args args = { 2807 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL), 2808 .pidfd = parent_tidptr, 2809 .child_tid = child_tidptr, 2810 .parent_tid = parent_tidptr, 2811 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL), 2812 .stack = newsp, 2813 .tls = tls, 2814 }; 2815 2816 return kernel_clone(&args); 2817 } 2818 #endif 2819 2820 #ifdef __ARCH_WANT_SYS_CLONE3 2821 2822 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs, 2823 struct clone_args __user *uargs, 2824 size_t usize) 2825 { 2826 int err; 2827 struct clone_args args; 2828 pid_t *kset_tid = kargs->set_tid; 2829 2830 BUILD_BUG_ON(offsetofend(struct clone_args, tls) != 2831 CLONE_ARGS_SIZE_VER0); 2832 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) != 2833 CLONE_ARGS_SIZE_VER1); 2834 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) != 2835 CLONE_ARGS_SIZE_VER2); 2836 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2); 2837 2838 if (unlikely(usize > PAGE_SIZE)) 2839 return -E2BIG; 2840 if (unlikely(usize < CLONE_ARGS_SIZE_VER0)) 2841 return -EINVAL; 2842 2843 err = copy_struct_from_user(&args, sizeof(args), uargs, usize); 2844 if (err) 2845 return err; 2846 2847 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL)) 2848 return -EINVAL; 2849 2850 if (unlikely(!args.set_tid && args.set_tid_size > 0)) 2851 return -EINVAL; 2852 2853 if (unlikely(args.set_tid && args.set_tid_size == 0)) 2854 return -EINVAL; 2855 2856 /* 2857 * Verify that higher 32bits of exit_signal are unset and that 2858 * it is a valid signal 2859 */ 2860 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) || 2861 !valid_signal(args.exit_signal))) 2862 return -EINVAL; 2863 2864 if ((args.flags & CLONE_INTO_CGROUP) && 2865 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2)) 2866 return -EINVAL; 2867 2868 *kargs = (struct kernel_clone_args){ 2869 .flags = args.flags, 2870 .pidfd = u64_to_user_ptr(args.pidfd), 2871 .child_tid = u64_to_user_ptr(args.child_tid), 2872 .parent_tid = u64_to_user_ptr(args.parent_tid), 2873 .exit_signal = args.exit_signal, 2874 .stack = args.stack, 2875 .stack_size = args.stack_size, 2876 .tls = args.tls, 2877 .set_tid_size = args.set_tid_size, 2878 .cgroup = args.cgroup, 2879 }; 2880 2881 if (args.set_tid && 2882 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid), 2883 (kargs->set_tid_size * sizeof(pid_t)))) 2884 return -EFAULT; 2885 2886 kargs->set_tid = kset_tid; 2887 2888 return 0; 2889 } 2890 2891 /** 2892 * clone3_stack_valid - check and prepare stack 2893 * @kargs: kernel clone args 2894 * 2895 * Verify that the stack arguments userspace gave us are sane. 2896 * In addition, set the stack direction for userspace since it's easy for us to 2897 * determine. 2898 */ 2899 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs) 2900 { 2901 if (kargs->stack == 0) { 2902 if (kargs->stack_size > 0) 2903 return false; 2904 } else { 2905 if (kargs->stack_size == 0) 2906 return false; 2907 2908 if (!access_ok((void __user *)kargs->stack, kargs->stack_size)) 2909 return false; 2910 2911 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64) 2912 kargs->stack += kargs->stack_size; 2913 #endif 2914 } 2915 2916 return true; 2917 } 2918 2919 static bool clone3_args_valid(struct kernel_clone_args *kargs) 2920 { 2921 /* Verify that no unknown flags are passed along. */ 2922 if (kargs->flags & 2923 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP)) 2924 return false; 2925 2926 /* 2927 * - make the CLONE_DETACHED bit reusable for clone3 2928 * - make the CSIGNAL bits reusable for clone3 2929 */ 2930 if (kargs->flags & (CLONE_DETACHED | CSIGNAL)) 2931 return false; 2932 2933 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) == 2934 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) 2935 return false; 2936 2937 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) && 2938 kargs->exit_signal) 2939 return false; 2940 2941 if (!clone3_stack_valid(kargs)) 2942 return false; 2943 2944 return true; 2945 } 2946 2947 /** 2948 * clone3 - create a new process with specific properties 2949 * @uargs: argument structure 2950 * @size: size of @uargs 2951 * 2952 * clone3() is the extensible successor to clone()/clone2(). 2953 * It takes a struct as argument that is versioned by its size. 2954 * 2955 * Return: On success, a positive PID for the child process. 2956 * On error, a negative errno number. 2957 */ 2958 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size) 2959 { 2960 int err; 2961 2962 struct kernel_clone_args kargs; 2963 pid_t set_tid[MAX_PID_NS_LEVEL]; 2964 2965 kargs.set_tid = set_tid; 2966 2967 err = copy_clone_args_from_user(&kargs, uargs, size); 2968 if (err) 2969 return err; 2970 2971 if (!clone3_args_valid(&kargs)) 2972 return -EINVAL; 2973 2974 return kernel_clone(&kargs); 2975 } 2976 #endif 2977 2978 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data) 2979 { 2980 struct task_struct *leader, *parent, *child; 2981 int res; 2982 2983 read_lock(&tasklist_lock); 2984 leader = top = top->group_leader; 2985 down: 2986 for_each_thread(leader, parent) { 2987 list_for_each_entry(child, &parent->children, sibling) { 2988 res = visitor(child, data); 2989 if (res) { 2990 if (res < 0) 2991 goto out; 2992 leader = child; 2993 goto down; 2994 } 2995 up: 2996 ; 2997 } 2998 } 2999 3000 if (leader != top) { 3001 child = leader; 3002 parent = child->real_parent; 3003 leader = parent->group_leader; 3004 goto up; 3005 } 3006 out: 3007 read_unlock(&tasklist_lock); 3008 } 3009 3010 #ifndef ARCH_MIN_MMSTRUCT_ALIGN 3011 #define ARCH_MIN_MMSTRUCT_ALIGN 0 3012 #endif 3013 3014 static void sighand_ctor(void *data) 3015 { 3016 struct sighand_struct *sighand = data; 3017 3018 spin_lock_init(&sighand->siglock); 3019 init_waitqueue_head(&sighand->signalfd_wqh); 3020 } 3021 3022 void __init proc_caches_init(void) 3023 { 3024 unsigned int mm_size; 3025 3026 sighand_cachep = kmem_cache_create("sighand_cache", 3027 sizeof(struct sighand_struct), 0, 3028 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU| 3029 SLAB_ACCOUNT, sighand_ctor); 3030 signal_cachep = kmem_cache_create("signal_cache", 3031 sizeof(struct signal_struct), 0, 3032 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 3033 NULL); 3034 files_cachep = kmem_cache_create("files_cache", 3035 sizeof(struct files_struct), 0, 3036 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 3037 NULL); 3038 fs_cachep = kmem_cache_create("fs_cache", 3039 sizeof(struct fs_struct), 0, 3040 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 3041 NULL); 3042 3043 /* 3044 * The mm_cpumask is located at the end of mm_struct, and is 3045 * dynamically sized based on the maximum CPU number this system 3046 * can have, taking hotplug into account (nr_cpu_ids). 3047 */ 3048 mm_size = sizeof(struct mm_struct) + cpumask_size(); 3049 3050 mm_cachep = kmem_cache_create_usercopy("mm_struct", 3051 mm_size, ARCH_MIN_MMSTRUCT_ALIGN, 3052 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 3053 offsetof(struct mm_struct, saved_auxv), 3054 sizeof_field(struct mm_struct, saved_auxv), 3055 NULL); 3056 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT); 3057 mmap_init(); 3058 nsproxy_cache_init(); 3059 } 3060 3061 /* 3062 * Check constraints on flags passed to the unshare system call. 3063 */ 3064 static int check_unshare_flags(unsigned long unshare_flags) 3065 { 3066 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND| 3067 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM| 3068 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET| 3069 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP| 3070 CLONE_NEWTIME)) 3071 return -EINVAL; 3072 /* 3073 * Not implemented, but pretend it works if there is nothing 3074 * to unshare. Note that unsharing the address space or the 3075 * signal handlers also need to unshare the signal queues (aka 3076 * CLONE_THREAD). 3077 */ 3078 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) { 3079 if (!thread_group_empty(current)) 3080 return -EINVAL; 3081 } 3082 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) { 3083 if (refcount_read(¤t->sighand->count) > 1) 3084 return -EINVAL; 3085 } 3086 if (unshare_flags & CLONE_VM) { 3087 if (!current_is_single_threaded()) 3088 return -EINVAL; 3089 } 3090 3091 return 0; 3092 } 3093 3094 /* 3095 * Unshare the filesystem structure if it is being shared 3096 */ 3097 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) 3098 { 3099 struct fs_struct *fs = current->fs; 3100 3101 if (!(unshare_flags & CLONE_FS) || !fs) 3102 return 0; 3103 3104 /* don't need lock here; in the worst case we'll do useless copy */ 3105 if (fs->users == 1) 3106 return 0; 3107 3108 *new_fsp = copy_fs_struct(fs); 3109 if (!*new_fsp) 3110 return -ENOMEM; 3111 3112 return 0; 3113 } 3114 3115 /* 3116 * Unshare file descriptor table if it is being shared 3117 */ 3118 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds, 3119 struct files_struct **new_fdp) 3120 { 3121 struct files_struct *fd = current->files; 3122 int error = 0; 3123 3124 if ((unshare_flags & CLONE_FILES) && 3125 (fd && atomic_read(&fd->count) > 1)) { 3126 *new_fdp = dup_fd(fd, max_fds, &error); 3127 if (!*new_fdp) 3128 return error; 3129 } 3130 3131 return 0; 3132 } 3133 3134 /* 3135 * unshare allows a process to 'unshare' part of the process 3136 * context which was originally shared using clone. copy_* 3137 * functions used by kernel_clone() cannot be used here directly 3138 * because they modify an inactive task_struct that is being 3139 * constructed. Here we are modifying the current, active, 3140 * task_struct. 3141 */ 3142 int ksys_unshare(unsigned long unshare_flags) 3143 { 3144 struct fs_struct *fs, *new_fs = NULL; 3145 struct files_struct *new_fd = NULL; 3146 struct cred *new_cred = NULL; 3147 struct nsproxy *new_nsproxy = NULL; 3148 int do_sysvsem = 0; 3149 int err; 3150 3151 /* 3152 * If unsharing a user namespace must also unshare the thread group 3153 * and unshare the filesystem root and working directories. 3154 */ 3155 if (unshare_flags & CLONE_NEWUSER) 3156 unshare_flags |= CLONE_THREAD | CLONE_FS; 3157 /* 3158 * If unsharing vm, must also unshare signal handlers. 3159 */ 3160 if (unshare_flags & CLONE_VM) 3161 unshare_flags |= CLONE_SIGHAND; 3162 /* 3163 * If unsharing a signal handlers, must also unshare the signal queues. 3164 */ 3165 if (unshare_flags & CLONE_SIGHAND) 3166 unshare_flags |= CLONE_THREAD; 3167 /* 3168 * If unsharing namespace, must also unshare filesystem information. 3169 */ 3170 if (unshare_flags & CLONE_NEWNS) 3171 unshare_flags |= CLONE_FS; 3172 3173 err = check_unshare_flags(unshare_flags); 3174 if (err) 3175 goto bad_unshare_out; 3176 /* 3177 * CLONE_NEWIPC must also detach from the undolist: after switching 3178 * to a new ipc namespace, the semaphore arrays from the old 3179 * namespace are unreachable. 3180 */ 3181 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) 3182 do_sysvsem = 1; 3183 err = unshare_fs(unshare_flags, &new_fs); 3184 if (err) 3185 goto bad_unshare_out; 3186 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd); 3187 if (err) 3188 goto bad_unshare_cleanup_fs; 3189 err = unshare_userns(unshare_flags, &new_cred); 3190 if (err) 3191 goto bad_unshare_cleanup_fd; 3192 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, 3193 new_cred, new_fs); 3194 if (err) 3195 goto bad_unshare_cleanup_cred; 3196 3197 if (new_cred) { 3198 err = set_cred_ucounts(new_cred); 3199 if (err) 3200 goto bad_unshare_cleanup_cred; 3201 } 3202 3203 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) { 3204 if (do_sysvsem) { 3205 /* 3206 * CLONE_SYSVSEM is equivalent to sys_exit(). 3207 */ 3208 exit_sem(current); 3209 } 3210 if (unshare_flags & CLONE_NEWIPC) { 3211 /* Orphan segments in old ns (see sem above). */ 3212 exit_shm(current); 3213 shm_init_task(current); 3214 } 3215 3216 if (new_nsproxy) 3217 switch_task_namespaces(current, new_nsproxy); 3218 3219 task_lock(current); 3220 3221 if (new_fs) { 3222 fs = current->fs; 3223 spin_lock(&fs->lock); 3224 current->fs = new_fs; 3225 if (--fs->users) 3226 new_fs = NULL; 3227 else 3228 new_fs = fs; 3229 spin_unlock(&fs->lock); 3230 } 3231 3232 if (new_fd) 3233 swap(current->files, new_fd); 3234 3235 task_unlock(current); 3236 3237 if (new_cred) { 3238 /* Install the new user namespace */ 3239 commit_creds(new_cred); 3240 new_cred = NULL; 3241 } 3242 } 3243 3244 perf_event_namespaces(current); 3245 3246 bad_unshare_cleanup_cred: 3247 if (new_cred) 3248 put_cred(new_cred); 3249 bad_unshare_cleanup_fd: 3250 if (new_fd) 3251 put_files_struct(new_fd); 3252 3253 bad_unshare_cleanup_fs: 3254 if (new_fs) 3255 free_fs_struct(new_fs); 3256 3257 bad_unshare_out: 3258 return err; 3259 } 3260 3261 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags) 3262 { 3263 return ksys_unshare(unshare_flags); 3264 } 3265 3266 /* 3267 * Helper to unshare the files of the current task. 3268 * We don't want to expose copy_files internals to 3269 * the exec layer of the kernel. 3270 */ 3271 3272 int unshare_files(void) 3273 { 3274 struct task_struct *task = current; 3275 struct files_struct *old, *copy = NULL; 3276 int error; 3277 3278 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©); 3279 if (error || !copy) 3280 return error; 3281 3282 old = task->files; 3283 task_lock(task); 3284 task->files = copy; 3285 task_unlock(task); 3286 put_files_struct(old); 3287 return 0; 3288 } 3289 3290 int sysctl_max_threads(struct ctl_table *table, int write, 3291 void *buffer, size_t *lenp, loff_t *ppos) 3292 { 3293 struct ctl_table t; 3294 int ret; 3295 int threads = max_threads; 3296 int min = 1; 3297 int max = MAX_THREADS; 3298 3299 t = *table; 3300 t.data = &threads; 3301 t.extra1 = &min; 3302 t.extra2 = &max; 3303 3304 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); 3305 if (ret || !write) 3306 return ret; 3307 3308 max_threads = threads; 3309 3310 return 0; 3311 } 3312