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/user.h> 20 #include <linux/sched/numa_balancing.h> 21 #include <linux/sched/stat.h> 22 #include <linux/sched/task.h> 23 #include <linux/sched/task_stack.h> 24 #include <linux/sched/cputime.h> 25 #include <linux/sched/ext.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/kmsan.h> 41 #include <linux/binfmts.h> 42 #include <linux/mman.h> 43 #include <linux/mmu_notifier.h> 44 #include <linux/fs.h> 45 #include <linux/mm.h> 46 #include <linux/mm_inline.h> 47 #include <linux/memblock.h> 48 #include <linux/nsproxy.h> 49 #include <linux/capability.h> 50 #include <linux/cpu.h> 51 #include <linux/cgroup.h> 52 #include <linux/security.h> 53 #include <linux/hugetlb.h> 54 #include <linux/seccomp.h> 55 #include <linux/swap.h> 56 #include <linux/syscalls.h> 57 #include <linux/syscall_user_dispatch.h> 58 #include <linux/jiffies.h> 59 #include <linux/futex.h> 60 #include <linux/compat.h> 61 #include <linux/kthread.h> 62 #include <linux/task_io_accounting_ops.h> 63 #include <linux/rcupdate.h> 64 #include <linux/ptrace.h> 65 #include <linux/mount.h> 66 #include <linux/audit.h> 67 #include <linux/memcontrol.h> 68 #include <linux/ftrace.h> 69 #include <linux/proc_fs.h> 70 #include <linux/profile.h> 71 #include <linux/rmap.h> 72 #include <linux/ksm.h> 73 #include <linux/acct.h> 74 #include <linux/userfaultfd_k.h> 75 #include <linux/tsacct_kern.h> 76 #include <linux/cn_proc.h> 77 #include <linux/freezer.h> 78 #include <linux/delayacct.h> 79 #include <linux/taskstats_kern.h> 80 #include <linux/tty.h> 81 #include <linux/fs_struct.h> 82 #include <linux/magic.h> 83 #include <linux/perf_event.h> 84 #include <linux/posix-timers.h> 85 #include <linux/user-return-notifier.h> 86 #include <linux/oom.h> 87 #include <linux/khugepaged.h> 88 #include <linux/signalfd.h> 89 #include <linux/uprobes.h> 90 #include <linux/aio.h> 91 #include <linux/compiler.h> 92 #include <linux/sysctl.h> 93 #include <linux/kcov.h> 94 #include <linux/livepatch.h> 95 #include <linux/thread_info.h> 96 #include <linux/stackleak.h> 97 #include <linux/kasan.h> 98 #include <linux/scs.h> 99 #include <linux/io_uring.h> 100 #include <linux/bpf.h> 101 #include <linux/stackprotector.h> 102 #include <linux/user_events.h> 103 #include <linux/iommu.h> 104 #include <linux/rseq.h> 105 #include <uapi/linux/pidfd.h> 106 #include <linux/pidfs.h> 107 #include <linux/tick.h> 108 109 #include <asm/pgalloc.h> 110 #include <linux/uaccess.h> 111 #include <asm/mmu_context.h> 112 #include <asm/cacheflush.h> 113 #include <asm/tlbflush.h> 114 115 #include <trace/events/sched.h> 116 117 #define CREATE_TRACE_POINTS 118 #include <trace/events/task.h> 119 120 #include <kunit/visibility.h> 121 122 /* 123 * Minimum number of threads to boot the kernel 124 */ 125 #define MIN_THREADS 20 126 127 /* 128 * Maximum number of threads 129 */ 130 #define MAX_THREADS FUTEX_TID_MASK 131 132 /* 133 * Protected counters by write_lock_irq(&tasklist_lock) 134 */ 135 unsigned long total_forks; /* Handle normal Linux uptimes. */ 136 int nr_threads; /* The idle threads do not count.. */ 137 138 static int max_threads; /* tunable limit on nr_threads */ 139 140 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x) 141 142 static const char * const resident_page_types[] = { 143 NAMED_ARRAY_INDEX(MM_FILEPAGES), 144 NAMED_ARRAY_INDEX(MM_ANONPAGES), 145 NAMED_ARRAY_INDEX(MM_SWAPENTS), 146 NAMED_ARRAY_INDEX(MM_SHMEMPAGES), 147 }; 148 149 DEFINE_PER_CPU(unsigned long, process_counts) = 0; 150 151 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */ 152 153 #ifdef CONFIG_PROVE_RCU 154 int lockdep_tasklist_lock_is_held(void) 155 { 156 return lockdep_is_held(&tasklist_lock); 157 } 158 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held); 159 #endif /* #ifdef CONFIG_PROVE_RCU */ 160 161 int nr_processes(void) 162 { 163 int cpu; 164 int total = 0; 165 166 for_each_possible_cpu(cpu) 167 total += per_cpu(process_counts, cpu); 168 169 return total; 170 } 171 172 void __weak arch_release_task_struct(struct task_struct *tsk) 173 { 174 } 175 176 static struct kmem_cache *task_struct_cachep; 177 178 static inline struct task_struct *alloc_task_struct_node(int node) 179 { 180 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node); 181 } 182 183 static inline void free_task_struct(struct task_struct *tsk) 184 { 185 kmem_cache_free(task_struct_cachep, tsk); 186 } 187 188 /* 189 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a 190 * kmemcache based allocator. 191 */ 192 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) 193 194 # ifdef CONFIG_VMAP_STACK 195 /* 196 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB 197 * flush. Try to minimize the number of calls by caching stacks. 198 */ 199 #define NR_CACHED_STACKS 2 200 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]); 201 202 struct vm_stack { 203 struct rcu_head rcu; 204 struct vm_struct *stack_vm_area; 205 }; 206 207 static bool try_release_thread_stack_to_cache(struct vm_struct *vm) 208 { 209 unsigned int i; 210 211 for (i = 0; i < NR_CACHED_STACKS; i++) { 212 struct vm_struct *tmp = NULL; 213 214 if (this_cpu_try_cmpxchg(cached_stacks[i], &tmp, vm)) 215 return true; 216 } 217 return false; 218 } 219 220 static void thread_stack_free_rcu(struct rcu_head *rh) 221 { 222 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu); 223 224 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area)) 225 return; 226 227 vfree(vm_stack); 228 } 229 230 static void thread_stack_delayed_free(struct task_struct *tsk) 231 { 232 struct vm_stack *vm_stack = tsk->stack; 233 234 vm_stack->stack_vm_area = tsk->stack_vm_area; 235 call_rcu(&vm_stack->rcu, thread_stack_free_rcu); 236 } 237 238 static int free_vm_stack_cache(unsigned int cpu) 239 { 240 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu); 241 int i; 242 243 for (i = 0; i < NR_CACHED_STACKS; i++) { 244 struct vm_struct *vm_stack = cached_vm_stacks[i]; 245 246 if (!vm_stack) 247 continue; 248 249 vfree(vm_stack->addr); 250 cached_vm_stacks[i] = NULL; 251 } 252 253 return 0; 254 } 255 256 static int memcg_charge_kernel_stack(struct vm_struct *vm) 257 { 258 int i; 259 int ret; 260 int nr_charged = 0; 261 262 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE); 263 264 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) { 265 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0); 266 if (ret) 267 goto err; 268 nr_charged++; 269 } 270 return 0; 271 err: 272 for (i = 0; i < nr_charged; i++) 273 memcg_kmem_uncharge_page(vm->pages[i], 0); 274 return ret; 275 } 276 277 static int alloc_thread_stack_node(struct task_struct *tsk, int node) 278 { 279 struct vm_struct *vm; 280 void *stack; 281 int i; 282 283 for (i = 0; i < NR_CACHED_STACKS; i++) { 284 struct vm_struct *s; 285 286 s = this_cpu_xchg(cached_stacks[i], NULL); 287 288 if (!s) 289 continue; 290 291 /* Reset stack metadata. */ 292 kasan_unpoison_range(s->addr, THREAD_SIZE); 293 294 stack = kasan_reset_tag(s->addr); 295 296 /* Clear stale pointers from reused stack. */ 297 memset(stack, 0, THREAD_SIZE); 298 299 if (memcg_charge_kernel_stack(s)) { 300 vfree(s->addr); 301 return -ENOMEM; 302 } 303 304 tsk->stack_vm_area = s; 305 tsk->stack = stack; 306 return 0; 307 } 308 309 /* 310 * Allocated stacks are cached and later reused by new threads, 311 * so memcg accounting is performed manually on assigning/releasing 312 * stacks to tasks. Drop __GFP_ACCOUNT. 313 */ 314 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN, 315 VMALLOC_START, VMALLOC_END, 316 THREADINFO_GFP & ~__GFP_ACCOUNT, 317 PAGE_KERNEL, 318 0, node, __builtin_return_address(0)); 319 if (!stack) 320 return -ENOMEM; 321 322 vm = find_vm_area(stack); 323 if (memcg_charge_kernel_stack(vm)) { 324 vfree(stack); 325 return -ENOMEM; 326 } 327 /* 328 * We can't call find_vm_area() in interrupt context, and 329 * free_thread_stack() can be called in interrupt context, 330 * so cache the vm_struct. 331 */ 332 tsk->stack_vm_area = vm; 333 stack = kasan_reset_tag(stack); 334 tsk->stack = stack; 335 return 0; 336 } 337 338 static void free_thread_stack(struct task_struct *tsk) 339 { 340 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area)) 341 thread_stack_delayed_free(tsk); 342 343 tsk->stack = NULL; 344 tsk->stack_vm_area = NULL; 345 } 346 347 # else /* !CONFIG_VMAP_STACK */ 348 349 static void thread_stack_free_rcu(struct rcu_head *rh) 350 { 351 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER); 352 } 353 354 static void thread_stack_delayed_free(struct task_struct *tsk) 355 { 356 struct rcu_head *rh = tsk->stack; 357 358 call_rcu(rh, thread_stack_free_rcu); 359 } 360 361 static int alloc_thread_stack_node(struct task_struct *tsk, int node) 362 { 363 struct page *page = alloc_pages_node(node, THREADINFO_GFP, 364 THREAD_SIZE_ORDER); 365 366 if (likely(page)) { 367 tsk->stack = kasan_reset_tag(page_address(page)); 368 return 0; 369 } 370 return -ENOMEM; 371 } 372 373 static void free_thread_stack(struct task_struct *tsk) 374 { 375 thread_stack_delayed_free(tsk); 376 tsk->stack = NULL; 377 } 378 379 # endif /* CONFIG_VMAP_STACK */ 380 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */ 381 382 static struct kmem_cache *thread_stack_cache; 383 384 static void thread_stack_free_rcu(struct rcu_head *rh) 385 { 386 kmem_cache_free(thread_stack_cache, rh); 387 } 388 389 static void thread_stack_delayed_free(struct task_struct *tsk) 390 { 391 struct rcu_head *rh = tsk->stack; 392 393 call_rcu(rh, thread_stack_free_rcu); 394 } 395 396 static int alloc_thread_stack_node(struct task_struct *tsk, int node) 397 { 398 unsigned long *stack; 399 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node); 400 stack = kasan_reset_tag(stack); 401 tsk->stack = stack; 402 return stack ? 0 : -ENOMEM; 403 } 404 405 static void free_thread_stack(struct task_struct *tsk) 406 { 407 thread_stack_delayed_free(tsk); 408 tsk->stack = NULL; 409 } 410 411 void thread_stack_cache_init(void) 412 { 413 thread_stack_cache = kmem_cache_create_usercopy("thread_stack", 414 THREAD_SIZE, THREAD_SIZE, 0, 0, 415 THREAD_SIZE, NULL); 416 BUG_ON(thread_stack_cache == NULL); 417 } 418 419 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */ 420 421 /* SLAB cache for signal_struct structures (tsk->signal) */ 422 static struct kmem_cache *signal_cachep; 423 424 /* SLAB cache for sighand_struct structures (tsk->sighand) */ 425 struct kmem_cache *sighand_cachep; 426 427 /* SLAB cache for files_struct structures (tsk->files) */ 428 struct kmem_cache *files_cachep; 429 430 /* SLAB cache for fs_struct structures (tsk->fs) */ 431 struct kmem_cache *fs_cachep; 432 433 /* SLAB cache for vm_area_struct structures */ 434 static struct kmem_cache *vm_area_cachep; 435 436 /* SLAB cache for mm_struct structures (tsk->mm) */ 437 static struct kmem_cache *mm_cachep; 438 439 #ifdef CONFIG_PER_VMA_LOCK 440 441 /* SLAB cache for vm_area_struct.lock */ 442 static struct kmem_cache *vma_lock_cachep; 443 444 static bool vma_lock_alloc(struct vm_area_struct *vma) 445 { 446 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL); 447 if (!vma->vm_lock) 448 return false; 449 450 init_rwsem(&vma->vm_lock->lock); 451 vma->vm_lock_seq = -1; 452 453 return true; 454 } 455 456 static inline void vma_lock_free(struct vm_area_struct *vma) 457 { 458 kmem_cache_free(vma_lock_cachep, vma->vm_lock); 459 } 460 461 #else /* CONFIG_PER_VMA_LOCK */ 462 463 static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; } 464 static inline void vma_lock_free(struct vm_area_struct *vma) {} 465 466 #endif /* CONFIG_PER_VMA_LOCK */ 467 468 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm) 469 { 470 struct vm_area_struct *vma; 471 472 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 473 if (!vma) 474 return NULL; 475 476 vma_init(vma, mm); 477 if (!vma_lock_alloc(vma)) { 478 kmem_cache_free(vm_area_cachep, vma); 479 return NULL; 480 } 481 482 return vma; 483 } 484 485 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig) 486 { 487 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 488 489 if (!new) 490 return NULL; 491 492 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags); 493 ASSERT_EXCLUSIVE_WRITER(orig->vm_file); 494 /* 495 * orig->shared.rb may be modified concurrently, but the clone 496 * will be reinitialized. 497 */ 498 data_race(memcpy(new, orig, sizeof(*new))); 499 if (!vma_lock_alloc(new)) { 500 kmem_cache_free(vm_area_cachep, new); 501 return NULL; 502 } 503 INIT_LIST_HEAD(&new->anon_vma_chain); 504 vma_numab_state_init(new); 505 dup_anon_vma_name(orig, new); 506 507 return new; 508 } 509 510 void __vm_area_free(struct vm_area_struct *vma) 511 { 512 vma_numab_state_free(vma); 513 free_anon_vma_name(vma); 514 vma_lock_free(vma); 515 kmem_cache_free(vm_area_cachep, vma); 516 } 517 518 #ifdef CONFIG_PER_VMA_LOCK 519 static void vm_area_free_rcu_cb(struct rcu_head *head) 520 { 521 struct vm_area_struct *vma = container_of(head, struct vm_area_struct, 522 vm_rcu); 523 524 /* The vma should not be locked while being destroyed. */ 525 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma); 526 __vm_area_free(vma); 527 } 528 #endif 529 530 void vm_area_free(struct vm_area_struct *vma) 531 { 532 #ifdef CONFIG_PER_VMA_LOCK 533 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb); 534 #else 535 __vm_area_free(vma); 536 #endif 537 } 538 539 static void account_kernel_stack(struct task_struct *tsk, int account) 540 { 541 if (IS_ENABLED(CONFIG_VMAP_STACK)) { 542 struct vm_struct *vm = task_stack_vm_area(tsk); 543 int i; 544 545 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) 546 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB, 547 account * (PAGE_SIZE / 1024)); 548 } else { 549 void *stack = task_stack_page(tsk); 550 551 /* All stack pages are in the same node. */ 552 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB, 553 account * (THREAD_SIZE / 1024)); 554 } 555 } 556 557 void exit_task_stack_account(struct task_struct *tsk) 558 { 559 account_kernel_stack(tsk, -1); 560 561 if (IS_ENABLED(CONFIG_VMAP_STACK)) { 562 struct vm_struct *vm; 563 int i; 564 565 vm = task_stack_vm_area(tsk); 566 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) 567 memcg_kmem_uncharge_page(vm->pages[i], 0); 568 } 569 } 570 571 static void release_task_stack(struct task_struct *tsk) 572 { 573 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD)) 574 return; /* Better to leak the stack than to free prematurely */ 575 576 free_thread_stack(tsk); 577 } 578 579 #ifdef CONFIG_THREAD_INFO_IN_TASK 580 void put_task_stack(struct task_struct *tsk) 581 { 582 if (refcount_dec_and_test(&tsk->stack_refcount)) 583 release_task_stack(tsk); 584 } 585 #endif 586 587 void free_task(struct task_struct *tsk) 588 { 589 #ifdef CONFIG_SECCOMP 590 WARN_ON_ONCE(tsk->seccomp.filter); 591 #endif 592 release_user_cpus_ptr(tsk); 593 scs_release(tsk); 594 595 #ifndef CONFIG_THREAD_INFO_IN_TASK 596 /* 597 * The task is finally done with both the stack and thread_info, 598 * so free both. 599 */ 600 release_task_stack(tsk); 601 #else 602 /* 603 * If the task had a separate stack allocation, it should be gone 604 * by now. 605 */ 606 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0); 607 #endif 608 rt_mutex_debug_task_free(tsk); 609 ftrace_graph_exit_task(tsk); 610 arch_release_task_struct(tsk); 611 if (tsk->flags & PF_KTHREAD) 612 free_kthread_struct(tsk); 613 bpf_task_storage_free(tsk); 614 free_task_struct(tsk); 615 } 616 EXPORT_SYMBOL(free_task); 617 618 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm) 619 { 620 struct file *exe_file; 621 622 exe_file = get_mm_exe_file(oldmm); 623 RCU_INIT_POINTER(mm->exe_file, exe_file); 624 /* 625 * We depend on the oldmm having properly denied write access to the 626 * exe_file already. 627 */ 628 if (exe_file && deny_write_access(exe_file)) 629 pr_warn_once("deny_write_access() failed in %s\n", __func__); 630 } 631 632 #ifdef CONFIG_MMU 633 static __latent_entropy int dup_mmap(struct mm_struct *mm, 634 struct mm_struct *oldmm) 635 { 636 struct vm_area_struct *mpnt, *tmp; 637 int retval; 638 unsigned long charge = 0; 639 LIST_HEAD(uf); 640 VMA_ITERATOR(vmi, mm, 0); 641 642 if (mmap_write_lock_killable(oldmm)) 643 return -EINTR; 644 flush_cache_dup_mm(oldmm); 645 uprobe_dup_mmap(oldmm, mm); 646 /* 647 * Not linked in yet - no deadlock potential: 648 */ 649 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING); 650 651 /* No ordering required: file already has been exposed. */ 652 dup_mm_exe_file(mm, oldmm); 653 654 mm->total_vm = oldmm->total_vm; 655 mm->data_vm = oldmm->data_vm; 656 mm->exec_vm = oldmm->exec_vm; 657 mm->stack_vm = oldmm->stack_vm; 658 659 /* Use __mt_dup() to efficiently build an identical maple tree. */ 660 retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL); 661 if (unlikely(retval)) 662 goto out; 663 664 mt_clear_in_rcu(vmi.mas.tree); 665 for_each_vma(vmi, mpnt) { 666 struct file *file; 667 668 vma_start_write(mpnt); 669 if (mpnt->vm_flags & VM_DONTCOPY) { 670 retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start, 671 mpnt->vm_end, GFP_KERNEL); 672 if (retval) 673 goto loop_out; 674 675 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt)); 676 continue; 677 } 678 charge = 0; 679 /* 680 * Don't duplicate many vmas if we've been oom-killed (for 681 * example) 682 */ 683 if (fatal_signal_pending(current)) { 684 retval = -EINTR; 685 goto loop_out; 686 } 687 if (mpnt->vm_flags & VM_ACCOUNT) { 688 unsigned long len = vma_pages(mpnt); 689 690 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */ 691 goto fail_nomem; 692 charge = len; 693 } 694 tmp = vm_area_dup(mpnt); 695 if (!tmp) 696 goto fail_nomem; 697 retval = vma_dup_policy(mpnt, tmp); 698 if (retval) 699 goto fail_nomem_policy; 700 tmp->vm_mm = mm; 701 retval = dup_userfaultfd(tmp, &uf); 702 if (retval) 703 goto fail_nomem_anon_vma_fork; 704 if (tmp->vm_flags & VM_WIPEONFORK) { 705 /* 706 * VM_WIPEONFORK gets a clean slate in the child. 707 * Don't prepare anon_vma until fault since we don't 708 * copy page for current vma. 709 */ 710 tmp->anon_vma = NULL; 711 } else if (anon_vma_fork(tmp, mpnt)) 712 goto fail_nomem_anon_vma_fork; 713 vm_flags_clear(tmp, VM_LOCKED_MASK); 714 /* 715 * Copy/update hugetlb private vma information. 716 */ 717 if (is_vm_hugetlb_page(tmp)) 718 hugetlb_dup_vma_private(tmp); 719 720 /* 721 * Link the vma into the MT. After using __mt_dup(), memory 722 * allocation is not necessary here, so it cannot fail. 723 */ 724 vma_iter_bulk_store(&vmi, tmp); 725 726 mm->map_count++; 727 728 if (tmp->vm_ops && tmp->vm_ops->open) 729 tmp->vm_ops->open(tmp); 730 731 file = tmp->vm_file; 732 if (file) { 733 struct address_space *mapping = file->f_mapping; 734 735 get_file(file); 736 i_mmap_lock_write(mapping); 737 if (vma_is_shared_maywrite(tmp)) 738 mapping_allow_writable(mapping); 739 flush_dcache_mmap_lock(mapping); 740 /* insert tmp into the share list, just after mpnt */ 741 vma_interval_tree_insert_after(tmp, mpnt, 742 &mapping->i_mmap); 743 flush_dcache_mmap_unlock(mapping); 744 i_mmap_unlock_write(mapping); 745 } 746 747 if (!(tmp->vm_flags & VM_WIPEONFORK)) 748 retval = copy_page_range(tmp, mpnt); 749 750 if (retval) { 751 mpnt = vma_next(&vmi); 752 goto loop_out; 753 } 754 } 755 /* a new mm has just been created */ 756 retval = arch_dup_mmap(oldmm, mm); 757 loop_out: 758 vma_iter_free(&vmi); 759 if (!retval) { 760 mt_set_in_rcu(vmi.mas.tree); 761 ksm_fork(mm, oldmm); 762 khugepaged_fork(mm, oldmm); 763 } else if (mpnt) { 764 /* 765 * The entire maple tree has already been duplicated. If the 766 * mmap duplication fails, mark the failure point with 767 * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered, 768 * stop releasing VMAs that have not been duplicated after this 769 * point. 770 */ 771 mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1); 772 mas_store(&vmi.mas, XA_ZERO_ENTRY); 773 } 774 out: 775 mmap_write_unlock(mm); 776 flush_tlb_mm(oldmm); 777 mmap_write_unlock(oldmm); 778 if (!retval) 779 dup_userfaultfd_complete(&uf); 780 else 781 dup_userfaultfd_fail(&uf); 782 return retval; 783 784 fail_nomem_anon_vma_fork: 785 mpol_put(vma_policy(tmp)); 786 fail_nomem_policy: 787 vm_area_free(tmp); 788 fail_nomem: 789 retval = -ENOMEM; 790 vm_unacct_memory(charge); 791 goto loop_out; 792 } 793 794 static inline int mm_alloc_pgd(struct mm_struct *mm) 795 { 796 mm->pgd = pgd_alloc(mm); 797 if (unlikely(!mm->pgd)) 798 return -ENOMEM; 799 return 0; 800 } 801 802 static inline void mm_free_pgd(struct mm_struct *mm) 803 { 804 pgd_free(mm, mm->pgd); 805 } 806 #else 807 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) 808 { 809 mmap_write_lock(oldmm); 810 dup_mm_exe_file(mm, oldmm); 811 mmap_write_unlock(oldmm); 812 return 0; 813 } 814 #define mm_alloc_pgd(mm) (0) 815 #define mm_free_pgd(mm) 816 #endif /* CONFIG_MMU */ 817 818 static void check_mm(struct mm_struct *mm) 819 { 820 int i; 821 822 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS, 823 "Please make sure 'struct resident_page_types[]' is updated as well"); 824 825 for (i = 0; i < NR_MM_COUNTERS; i++) { 826 long x = percpu_counter_sum(&mm->rss_stat[i]); 827 828 if (unlikely(x)) 829 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n", 830 mm, resident_page_types[i], x); 831 } 832 833 if (mm_pgtables_bytes(mm)) 834 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n", 835 mm_pgtables_bytes(mm)); 836 837 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS) 838 VM_BUG_ON_MM(mm->pmd_huge_pte, mm); 839 #endif 840 } 841 842 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL)) 843 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) 844 845 static void do_check_lazy_tlb(void *arg) 846 { 847 struct mm_struct *mm = arg; 848 849 WARN_ON_ONCE(current->active_mm == mm); 850 } 851 852 static void do_shoot_lazy_tlb(void *arg) 853 { 854 struct mm_struct *mm = arg; 855 856 if (current->active_mm == mm) { 857 WARN_ON_ONCE(current->mm); 858 current->active_mm = &init_mm; 859 switch_mm(mm, &init_mm, current); 860 } 861 } 862 863 static void cleanup_lazy_tlbs(struct mm_struct *mm) 864 { 865 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) { 866 /* 867 * In this case, lazy tlb mms are refounted and would not reach 868 * __mmdrop until all CPUs have switched away and mmdrop()ed. 869 */ 870 return; 871 } 872 873 /* 874 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it 875 * requires lazy mm users to switch to another mm when the refcount 876 * drops to zero, before the mm is freed. This requires IPIs here to 877 * switch kernel threads to init_mm. 878 * 879 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm 880 * switch with the final userspace teardown TLB flush which leaves the 881 * mm lazy on this CPU but no others, reducing the need for additional 882 * IPIs here. There are cases where a final IPI is still required here, 883 * such as the final mmdrop being performed on a different CPU than the 884 * one exiting, or kernel threads using the mm when userspace exits. 885 * 886 * IPI overheads have not found to be expensive, but they could be 887 * reduced in a number of possible ways, for example (roughly 888 * increasing order of complexity): 889 * - The last lazy reference created by exit_mm() could instead switch 890 * to init_mm, however it's probable this will run on the same CPU 891 * immediately afterwards, so this may not reduce IPIs much. 892 * - A batch of mms requiring IPIs could be gathered and freed at once. 893 * - CPUs store active_mm where it can be remotely checked without a 894 * lock, to filter out false-positives in the cpumask. 895 * - After mm_users or mm_count reaches zero, switching away from the 896 * mm could clear mm_cpumask to reduce some IPIs, perhaps together 897 * with some batching or delaying of the final IPIs. 898 * - A delayed freeing and RCU-like quiescing sequence based on mm 899 * switching to avoid IPIs completely. 900 */ 901 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1); 902 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES)) 903 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1); 904 } 905 906 /* 907 * Called when the last reference to the mm 908 * is dropped: either by a lazy thread or by 909 * mmput. Free the page directory and the mm. 910 */ 911 void __mmdrop(struct mm_struct *mm) 912 { 913 BUG_ON(mm == &init_mm); 914 WARN_ON_ONCE(mm == current->mm); 915 916 /* Ensure no CPUs are using this as their lazy tlb mm */ 917 cleanup_lazy_tlbs(mm); 918 919 WARN_ON_ONCE(mm == current->active_mm); 920 mm_free_pgd(mm); 921 destroy_context(mm); 922 mmu_notifier_subscriptions_destroy(mm); 923 check_mm(mm); 924 put_user_ns(mm->user_ns); 925 mm_pasid_drop(mm); 926 mm_destroy_cid(mm); 927 percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS); 928 929 free_mm(mm); 930 } 931 EXPORT_SYMBOL_GPL(__mmdrop); 932 933 static void mmdrop_async_fn(struct work_struct *work) 934 { 935 struct mm_struct *mm; 936 937 mm = container_of(work, struct mm_struct, async_put_work); 938 __mmdrop(mm); 939 } 940 941 static void mmdrop_async(struct mm_struct *mm) 942 { 943 if (unlikely(atomic_dec_and_test(&mm->mm_count))) { 944 INIT_WORK(&mm->async_put_work, mmdrop_async_fn); 945 schedule_work(&mm->async_put_work); 946 } 947 } 948 949 static inline void free_signal_struct(struct signal_struct *sig) 950 { 951 taskstats_tgid_free(sig); 952 sched_autogroup_exit(sig); 953 /* 954 * __mmdrop is not safe to call from softirq context on x86 due to 955 * pgd_dtor so postpone it to the async context 956 */ 957 if (sig->oom_mm) 958 mmdrop_async(sig->oom_mm); 959 kmem_cache_free(signal_cachep, sig); 960 } 961 962 static inline void put_signal_struct(struct signal_struct *sig) 963 { 964 if (refcount_dec_and_test(&sig->sigcnt)) 965 free_signal_struct(sig); 966 } 967 968 void __put_task_struct(struct task_struct *tsk) 969 { 970 WARN_ON(!tsk->exit_state); 971 WARN_ON(refcount_read(&tsk->usage)); 972 WARN_ON(tsk == current); 973 974 sched_ext_free(tsk); 975 io_uring_free(tsk); 976 cgroup_free(tsk); 977 task_numa_free(tsk, true); 978 security_task_free(tsk); 979 exit_creds(tsk); 980 delayacct_tsk_free(tsk); 981 put_signal_struct(tsk->signal); 982 sched_core_free(tsk); 983 free_task(tsk); 984 } 985 EXPORT_SYMBOL_GPL(__put_task_struct); 986 987 void __put_task_struct_rcu_cb(struct rcu_head *rhp) 988 { 989 struct task_struct *task = container_of(rhp, struct task_struct, rcu); 990 991 __put_task_struct(task); 992 } 993 EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb); 994 995 void __init __weak arch_task_cache_init(void) { } 996 997 /* 998 * set_max_threads 999 */ 1000 static void __init set_max_threads(unsigned int max_threads_suggested) 1001 { 1002 u64 threads; 1003 unsigned long nr_pages = memblock_estimated_nr_free_pages(); 1004 1005 /* 1006 * The number of threads shall be limited such that the thread 1007 * structures may only consume a small part of the available memory. 1008 */ 1009 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64) 1010 threads = MAX_THREADS; 1011 else 1012 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE, 1013 (u64) THREAD_SIZE * 8UL); 1014 1015 if (threads > max_threads_suggested) 1016 threads = max_threads_suggested; 1017 1018 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS); 1019 } 1020 1021 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT 1022 /* Initialized by the architecture: */ 1023 int arch_task_struct_size __read_mostly; 1024 #endif 1025 1026 static void __init task_struct_whitelist(unsigned long *offset, unsigned long *size) 1027 { 1028 /* Fetch thread_struct whitelist for the architecture. */ 1029 arch_thread_struct_whitelist(offset, size); 1030 1031 /* 1032 * Handle zero-sized whitelist or empty thread_struct, otherwise 1033 * adjust offset to position of thread_struct in task_struct. 1034 */ 1035 if (unlikely(*size == 0)) 1036 *offset = 0; 1037 else 1038 *offset += offsetof(struct task_struct, thread); 1039 } 1040 1041 void __init fork_init(void) 1042 { 1043 int i; 1044 #ifndef ARCH_MIN_TASKALIGN 1045 #define ARCH_MIN_TASKALIGN 0 1046 #endif 1047 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN); 1048 unsigned long useroffset, usersize; 1049 1050 /* create a slab on which task_structs can be allocated */ 1051 task_struct_whitelist(&useroffset, &usersize); 1052 task_struct_cachep = kmem_cache_create_usercopy("task_struct", 1053 arch_task_struct_size, align, 1054 SLAB_PANIC|SLAB_ACCOUNT, 1055 useroffset, usersize, NULL); 1056 1057 /* do the arch specific task caches init */ 1058 arch_task_cache_init(); 1059 1060 set_max_threads(MAX_THREADS); 1061 1062 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; 1063 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; 1064 init_task.signal->rlim[RLIMIT_SIGPENDING] = 1065 init_task.signal->rlim[RLIMIT_NPROC]; 1066 1067 for (i = 0; i < UCOUNT_COUNTS; i++) 1068 init_user_ns.ucount_max[i] = max_threads/2; 1069 1070 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY); 1071 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY); 1072 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY); 1073 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY); 1074 1075 #ifdef CONFIG_VMAP_STACK 1076 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache", 1077 NULL, free_vm_stack_cache); 1078 #endif 1079 1080 scs_init(); 1081 1082 lockdep_init_task(&init_task); 1083 uprobes_init(); 1084 } 1085 1086 int __weak arch_dup_task_struct(struct task_struct *dst, 1087 struct task_struct *src) 1088 { 1089 *dst = *src; 1090 return 0; 1091 } 1092 1093 void set_task_stack_end_magic(struct task_struct *tsk) 1094 { 1095 unsigned long *stackend; 1096 1097 stackend = end_of_stack(tsk); 1098 *stackend = STACK_END_MAGIC; /* for overflow detection */ 1099 } 1100 1101 static struct task_struct *dup_task_struct(struct task_struct *orig, int node) 1102 { 1103 struct task_struct *tsk; 1104 int err; 1105 1106 if (node == NUMA_NO_NODE) 1107 node = tsk_fork_get_node(orig); 1108 tsk = alloc_task_struct_node(node); 1109 if (!tsk) 1110 return NULL; 1111 1112 err = arch_dup_task_struct(tsk, orig); 1113 if (err) 1114 goto free_tsk; 1115 1116 err = alloc_thread_stack_node(tsk, node); 1117 if (err) 1118 goto free_tsk; 1119 1120 #ifdef CONFIG_THREAD_INFO_IN_TASK 1121 refcount_set(&tsk->stack_refcount, 1); 1122 #endif 1123 account_kernel_stack(tsk, 1); 1124 1125 err = scs_prepare(tsk, node); 1126 if (err) 1127 goto free_stack; 1128 1129 #ifdef CONFIG_SECCOMP 1130 /* 1131 * We must handle setting up seccomp filters once we're under 1132 * the sighand lock in case orig has changed between now and 1133 * then. Until then, filter must be NULL to avoid messing up 1134 * the usage counts on the error path calling free_task. 1135 */ 1136 tsk->seccomp.filter = NULL; 1137 #endif 1138 1139 setup_thread_stack(tsk, orig); 1140 clear_user_return_notifier(tsk); 1141 clear_tsk_need_resched(tsk); 1142 set_task_stack_end_magic(tsk); 1143 clear_syscall_work_syscall_user_dispatch(tsk); 1144 1145 #ifdef CONFIG_STACKPROTECTOR 1146 tsk->stack_canary = get_random_canary(); 1147 #endif 1148 if (orig->cpus_ptr == &orig->cpus_mask) 1149 tsk->cpus_ptr = &tsk->cpus_mask; 1150 dup_user_cpus_ptr(tsk, orig, node); 1151 1152 /* 1153 * One for the user space visible state that goes away when reaped. 1154 * One for the scheduler. 1155 */ 1156 refcount_set(&tsk->rcu_users, 2); 1157 /* One for the rcu users */ 1158 refcount_set(&tsk->usage, 1); 1159 #ifdef CONFIG_BLK_DEV_IO_TRACE 1160 tsk->btrace_seq = 0; 1161 #endif 1162 tsk->splice_pipe = NULL; 1163 tsk->task_frag.page = NULL; 1164 tsk->wake_q.next = NULL; 1165 tsk->worker_private = NULL; 1166 1167 kcov_task_init(tsk); 1168 kmsan_task_create(tsk); 1169 kmap_local_fork(tsk); 1170 1171 #ifdef CONFIG_FAULT_INJECTION 1172 tsk->fail_nth = 0; 1173 #endif 1174 1175 #ifdef CONFIG_BLK_CGROUP 1176 tsk->throttle_disk = NULL; 1177 tsk->use_memdelay = 0; 1178 #endif 1179 1180 #ifdef CONFIG_ARCH_HAS_CPU_PASID 1181 tsk->pasid_activated = 0; 1182 #endif 1183 1184 #ifdef CONFIG_MEMCG 1185 tsk->active_memcg = NULL; 1186 #endif 1187 1188 #ifdef CONFIG_X86_BUS_LOCK_DETECT 1189 tsk->reported_split_lock = 0; 1190 #endif 1191 1192 #ifdef CONFIG_SCHED_MM_CID 1193 tsk->mm_cid = -1; 1194 tsk->last_mm_cid = -1; 1195 tsk->mm_cid_active = 0; 1196 tsk->migrate_from_cpu = -1; 1197 #endif 1198 return tsk; 1199 1200 free_stack: 1201 exit_task_stack_account(tsk); 1202 free_thread_stack(tsk); 1203 free_tsk: 1204 free_task_struct(tsk); 1205 return NULL; 1206 } 1207 1208 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); 1209 1210 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT; 1211 1212 static int __init coredump_filter_setup(char *s) 1213 { 1214 default_dump_filter = 1215 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) & 1216 MMF_DUMP_FILTER_MASK; 1217 return 1; 1218 } 1219 1220 __setup("coredump_filter=", coredump_filter_setup); 1221 1222 #include <linux/init_task.h> 1223 1224 static void mm_init_aio(struct mm_struct *mm) 1225 { 1226 #ifdef CONFIG_AIO 1227 spin_lock_init(&mm->ioctx_lock); 1228 mm->ioctx_table = NULL; 1229 #endif 1230 } 1231 1232 static __always_inline void mm_clear_owner(struct mm_struct *mm, 1233 struct task_struct *p) 1234 { 1235 #ifdef CONFIG_MEMCG 1236 if (mm->owner == p) 1237 WRITE_ONCE(mm->owner, NULL); 1238 #endif 1239 } 1240 1241 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p) 1242 { 1243 #ifdef CONFIG_MEMCG 1244 mm->owner = p; 1245 #endif 1246 } 1247 1248 static void mm_init_uprobes_state(struct mm_struct *mm) 1249 { 1250 #ifdef CONFIG_UPROBES 1251 mm->uprobes_state.xol_area = NULL; 1252 #endif 1253 } 1254 1255 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p, 1256 struct user_namespace *user_ns) 1257 { 1258 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS); 1259 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock); 1260 atomic_set(&mm->mm_users, 1); 1261 atomic_set(&mm->mm_count, 1); 1262 seqcount_init(&mm->write_protect_seq); 1263 mmap_init_lock(mm); 1264 INIT_LIST_HEAD(&mm->mmlist); 1265 #ifdef CONFIG_PER_VMA_LOCK 1266 mm->mm_lock_seq = 0; 1267 #endif 1268 mm_pgtables_bytes_init(mm); 1269 mm->map_count = 0; 1270 mm->locked_vm = 0; 1271 atomic64_set(&mm->pinned_vm, 0); 1272 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat)); 1273 spin_lock_init(&mm->page_table_lock); 1274 spin_lock_init(&mm->arg_lock); 1275 mm_init_cpumask(mm); 1276 mm_init_aio(mm); 1277 mm_init_owner(mm, p); 1278 mm_pasid_init(mm); 1279 RCU_INIT_POINTER(mm->exe_file, NULL); 1280 mmu_notifier_subscriptions_init(mm); 1281 init_tlb_flush_pending(mm); 1282 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS) 1283 mm->pmd_huge_pte = NULL; 1284 #endif 1285 mm_init_uprobes_state(mm); 1286 hugetlb_count_init(mm); 1287 1288 if (current->mm) { 1289 mm->flags = mmf_init_flags(current->mm->flags); 1290 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK; 1291 } else { 1292 mm->flags = default_dump_filter; 1293 mm->def_flags = 0; 1294 } 1295 1296 if (mm_alloc_pgd(mm)) 1297 goto fail_nopgd; 1298 1299 if (init_new_context(p, mm)) 1300 goto fail_nocontext; 1301 1302 if (mm_alloc_cid(mm, p)) 1303 goto fail_cid; 1304 1305 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT, 1306 NR_MM_COUNTERS)) 1307 goto fail_pcpu; 1308 1309 mm->user_ns = get_user_ns(user_ns); 1310 lru_gen_init_mm(mm); 1311 return mm; 1312 1313 fail_pcpu: 1314 mm_destroy_cid(mm); 1315 fail_cid: 1316 destroy_context(mm); 1317 fail_nocontext: 1318 mm_free_pgd(mm); 1319 fail_nopgd: 1320 free_mm(mm); 1321 return NULL; 1322 } 1323 1324 /* 1325 * Allocate and initialize an mm_struct. 1326 */ 1327 struct mm_struct *mm_alloc(void) 1328 { 1329 struct mm_struct *mm; 1330 1331 mm = allocate_mm(); 1332 if (!mm) 1333 return NULL; 1334 1335 memset(mm, 0, sizeof(*mm)); 1336 return mm_init(mm, current, current_user_ns()); 1337 } 1338 EXPORT_SYMBOL_IF_KUNIT(mm_alloc); 1339 1340 static inline void __mmput(struct mm_struct *mm) 1341 { 1342 VM_BUG_ON(atomic_read(&mm->mm_users)); 1343 1344 uprobe_clear_state(mm); 1345 exit_aio(mm); 1346 ksm_exit(mm); 1347 khugepaged_exit(mm); /* must run before exit_mmap */ 1348 exit_mmap(mm); 1349 mm_put_huge_zero_folio(mm); 1350 set_mm_exe_file(mm, NULL); 1351 if (!list_empty(&mm->mmlist)) { 1352 spin_lock(&mmlist_lock); 1353 list_del(&mm->mmlist); 1354 spin_unlock(&mmlist_lock); 1355 } 1356 if (mm->binfmt) 1357 module_put(mm->binfmt->module); 1358 lru_gen_del_mm(mm); 1359 mmdrop(mm); 1360 } 1361 1362 /* 1363 * Decrement the use count and release all resources for an mm. 1364 */ 1365 void mmput(struct mm_struct *mm) 1366 { 1367 might_sleep(); 1368 1369 if (atomic_dec_and_test(&mm->mm_users)) 1370 __mmput(mm); 1371 } 1372 EXPORT_SYMBOL_GPL(mmput); 1373 1374 #ifdef CONFIG_MMU 1375 static void mmput_async_fn(struct work_struct *work) 1376 { 1377 struct mm_struct *mm = container_of(work, struct mm_struct, 1378 async_put_work); 1379 1380 __mmput(mm); 1381 } 1382 1383 void mmput_async(struct mm_struct *mm) 1384 { 1385 if (atomic_dec_and_test(&mm->mm_users)) { 1386 INIT_WORK(&mm->async_put_work, mmput_async_fn); 1387 schedule_work(&mm->async_put_work); 1388 } 1389 } 1390 EXPORT_SYMBOL_GPL(mmput_async); 1391 #endif 1392 1393 /** 1394 * set_mm_exe_file - change a reference to the mm's executable file 1395 * @mm: The mm to change. 1396 * @new_exe_file: The new file to use. 1397 * 1398 * This changes mm's executable file (shown as symlink /proc/[pid]/exe). 1399 * 1400 * Main users are mmput() and sys_execve(). Callers prevent concurrent 1401 * invocations: in mmput() nobody alive left, in execve it happens before 1402 * the new mm is made visible to anyone. 1403 * 1404 * Can only fail if new_exe_file != NULL. 1405 */ 1406 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 1407 { 1408 struct file *old_exe_file; 1409 1410 /* 1411 * It is safe to dereference the exe_file without RCU as 1412 * this function is only called if nobody else can access 1413 * this mm -- see comment above for justification. 1414 */ 1415 old_exe_file = rcu_dereference_raw(mm->exe_file); 1416 1417 if (new_exe_file) { 1418 /* 1419 * We expect the caller (i.e., sys_execve) to already denied 1420 * write access, so this is unlikely to fail. 1421 */ 1422 if (unlikely(deny_write_access(new_exe_file))) 1423 return -EACCES; 1424 get_file(new_exe_file); 1425 } 1426 rcu_assign_pointer(mm->exe_file, new_exe_file); 1427 if (old_exe_file) { 1428 allow_write_access(old_exe_file); 1429 fput(old_exe_file); 1430 } 1431 return 0; 1432 } 1433 1434 /** 1435 * replace_mm_exe_file - replace a reference to the mm's executable file 1436 * @mm: The mm to change. 1437 * @new_exe_file: The new file to use. 1438 * 1439 * This changes mm's executable file (shown as symlink /proc/[pid]/exe). 1440 * 1441 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE). 1442 */ 1443 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 1444 { 1445 struct vm_area_struct *vma; 1446 struct file *old_exe_file; 1447 int ret = 0; 1448 1449 /* Forbid mm->exe_file change if old file still mapped. */ 1450 old_exe_file = get_mm_exe_file(mm); 1451 if (old_exe_file) { 1452 VMA_ITERATOR(vmi, mm, 0); 1453 mmap_read_lock(mm); 1454 for_each_vma(vmi, vma) { 1455 if (!vma->vm_file) 1456 continue; 1457 if (path_equal(&vma->vm_file->f_path, 1458 &old_exe_file->f_path)) { 1459 ret = -EBUSY; 1460 break; 1461 } 1462 } 1463 mmap_read_unlock(mm); 1464 fput(old_exe_file); 1465 if (ret) 1466 return ret; 1467 } 1468 1469 ret = deny_write_access(new_exe_file); 1470 if (ret) 1471 return -EACCES; 1472 get_file(new_exe_file); 1473 1474 /* set the new file */ 1475 mmap_write_lock(mm); 1476 old_exe_file = rcu_dereference_raw(mm->exe_file); 1477 rcu_assign_pointer(mm->exe_file, new_exe_file); 1478 mmap_write_unlock(mm); 1479 1480 if (old_exe_file) { 1481 allow_write_access(old_exe_file); 1482 fput(old_exe_file); 1483 } 1484 return 0; 1485 } 1486 1487 /** 1488 * get_mm_exe_file - acquire a reference to the mm's executable file 1489 * @mm: The mm of interest. 1490 * 1491 * Returns %NULL if mm has no associated executable file. 1492 * User must release file via fput(). 1493 */ 1494 struct file *get_mm_exe_file(struct mm_struct *mm) 1495 { 1496 struct file *exe_file; 1497 1498 rcu_read_lock(); 1499 exe_file = get_file_rcu(&mm->exe_file); 1500 rcu_read_unlock(); 1501 return exe_file; 1502 } 1503 1504 /** 1505 * get_task_exe_file - acquire a reference to the task's executable file 1506 * @task: The task. 1507 * 1508 * Returns %NULL if task's mm (if any) has no associated executable file or 1509 * this is a kernel thread with borrowed mm (see the comment above get_task_mm). 1510 * User must release file via fput(). 1511 */ 1512 struct file *get_task_exe_file(struct task_struct *task) 1513 { 1514 struct file *exe_file = NULL; 1515 struct mm_struct *mm; 1516 1517 task_lock(task); 1518 mm = task->mm; 1519 if (mm) { 1520 if (!(task->flags & PF_KTHREAD)) 1521 exe_file = get_mm_exe_file(mm); 1522 } 1523 task_unlock(task); 1524 return exe_file; 1525 } 1526 1527 /** 1528 * get_task_mm - acquire a reference to the task's mm 1529 * @task: The task. 1530 * 1531 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning 1532 * this kernel workthread has transiently adopted a user mm with use_mm, 1533 * to do its AIO) is not set and if so returns a reference to it, after 1534 * bumping up the use count. User must release the mm via mmput() 1535 * after use. Typically used by /proc and ptrace. 1536 */ 1537 struct mm_struct *get_task_mm(struct task_struct *task) 1538 { 1539 struct mm_struct *mm; 1540 1541 if (task->flags & PF_KTHREAD) 1542 return NULL; 1543 1544 task_lock(task); 1545 mm = task->mm; 1546 if (mm) 1547 mmget(mm); 1548 task_unlock(task); 1549 return mm; 1550 } 1551 EXPORT_SYMBOL_GPL(get_task_mm); 1552 1553 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode) 1554 { 1555 struct mm_struct *mm; 1556 int err; 1557 1558 err = down_read_killable(&task->signal->exec_update_lock); 1559 if (err) 1560 return ERR_PTR(err); 1561 1562 mm = get_task_mm(task); 1563 if (!mm) { 1564 mm = ERR_PTR(-ESRCH); 1565 } else if (mm != current->mm && !ptrace_may_access(task, mode)) { 1566 mmput(mm); 1567 mm = ERR_PTR(-EACCES); 1568 } 1569 up_read(&task->signal->exec_update_lock); 1570 1571 return mm; 1572 } 1573 1574 static void complete_vfork_done(struct task_struct *tsk) 1575 { 1576 struct completion *vfork; 1577 1578 task_lock(tsk); 1579 vfork = tsk->vfork_done; 1580 if (likely(vfork)) { 1581 tsk->vfork_done = NULL; 1582 complete(vfork); 1583 } 1584 task_unlock(tsk); 1585 } 1586 1587 static int wait_for_vfork_done(struct task_struct *child, 1588 struct completion *vfork) 1589 { 1590 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE; 1591 int killed; 1592 1593 cgroup_enter_frozen(); 1594 killed = wait_for_completion_state(vfork, state); 1595 cgroup_leave_frozen(false); 1596 1597 if (killed) { 1598 task_lock(child); 1599 child->vfork_done = NULL; 1600 task_unlock(child); 1601 } 1602 1603 put_task_struct(child); 1604 return killed; 1605 } 1606 1607 /* Please note the differences between mmput and mm_release. 1608 * mmput is called whenever we stop holding onto a mm_struct, 1609 * error success whatever. 1610 * 1611 * mm_release is called after a mm_struct has been removed 1612 * from the current process. 1613 * 1614 * This difference is important for error handling, when we 1615 * only half set up a mm_struct for a new process and need to restore 1616 * the old one. Because we mmput the new mm_struct before 1617 * restoring the old one. . . 1618 * Eric Biederman 10 January 1998 1619 */ 1620 static void mm_release(struct task_struct *tsk, struct mm_struct *mm) 1621 { 1622 uprobe_free_utask(tsk); 1623 1624 /* Get rid of any cached register state */ 1625 deactivate_mm(tsk, mm); 1626 1627 /* 1628 * Signal userspace if we're not exiting with a core dump 1629 * because we want to leave the value intact for debugging 1630 * purposes. 1631 */ 1632 if (tsk->clear_child_tid) { 1633 if (atomic_read(&mm->mm_users) > 1) { 1634 /* 1635 * We don't check the error code - if userspace has 1636 * not set up a proper pointer then tough luck. 1637 */ 1638 put_user(0, tsk->clear_child_tid); 1639 do_futex(tsk->clear_child_tid, FUTEX_WAKE, 1640 1, NULL, NULL, 0, 0); 1641 } 1642 tsk->clear_child_tid = NULL; 1643 } 1644 1645 /* 1646 * All done, finally we can wake up parent and return this mm to him. 1647 * Also kthread_stop() uses this completion for synchronization. 1648 */ 1649 if (tsk->vfork_done) 1650 complete_vfork_done(tsk); 1651 } 1652 1653 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm) 1654 { 1655 futex_exit_release(tsk); 1656 mm_release(tsk, mm); 1657 } 1658 1659 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm) 1660 { 1661 futex_exec_release(tsk); 1662 mm_release(tsk, mm); 1663 } 1664 1665 /** 1666 * dup_mm() - duplicates an existing mm structure 1667 * @tsk: the task_struct with which the new mm will be associated. 1668 * @oldmm: the mm to duplicate. 1669 * 1670 * Allocates a new mm structure and duplicates the provided @oldmm structure 1671 * content into it. 1672 * 1673 * Return: the duplicated mm or NULL on failure. 1674 */ 1675 static struct mm_struct *dup_mm(struct task_struct *tsk, 1676 struct mm_struct *oldmm) 1677 { 1678 struct mm_struct *mm; 1679 int err; 1680 1681 mm = allocate_mm(); 1682 if (!mm) 1683 goto fail_nomem; 1684 1685 memcpy(mm, oldmm, sizeof(*mm)); 1686 1687 if (!mm_init(mm, tsk, mm->user_ns)) 1688 goto fail_nomem; 1689 1690 uprobe_start_dup_mmap(); 1691 err = dup_mmap(mm, oldmm); 1692 if (err) 1693 goto free_pt; 1694 uprobe_end_dup_mmap(); 1695 1696 mm->hiwater_rss = get_mm_rss(mm); 1697 mm->hiwater_vm = mm->total_vm; 1698 1699 if (mm->binfmt && !try_module_get(mm->binfmt->module)) 1700 goto free_pt; 1701 1702 return mm; 1703 1704 free_pt: 1705 /* don't put binfmt in mmput, we haven't got module yet */ 1706 mm->binfmt = NULL; 1707 mm_init_owner(mm, NULL); 1708 mmput(mm); 1709 if (err) 1710 uprobe_end_dup_mmap(); 1711 1712 fail_nomem: 1713 return NULL; 1714 } 1715 1716 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk) 1717 { 1718 struct mm_struct *mm, *oldmm; 1719 1720 tsk->min_flt = tsk->maj_flt = 0; 1721 tsk->nvcsw = tsk->nivcsw = 0; 1722 #ifdef CONFIG_DETECT_HUNG_TASK 1723 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw; 1724 tsk->last_switch_time = 0; 1725 #endif 1726 1727 tsk->mm = NULL; 1728 tsk->active_mm = NULL; 1729 1730 /* 1731 * Are we cloning a kernel thread? 1732 * 1733 * We need to steal a active VM for that.. 1734 */ 1735 oldmm = current->mm; 1736 if (!oldmm) 1737 return 0; 1738 1739 if (clone_flags & CLONE_VM) { 1740 mmget(oldmm); 1741 mm = oldmm; 1742 } else { 1743 mm = dup_mm(tsk, current->mm); 1744 if (!mm) 1745 return -ENOMEM; 1746 } 1747 1748 tsk->mm = mm; 1749 tsk->active_mm = mm; 1750 sched_mm_cid_fork(tsk); 1751 return 0; 1752 } 1753 1754 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 1755 { 1756 struct fs_struct *fs = current->fs; 1757 if (clone_flags & CLONE_FS) { 1758 /* tsk->fs is already what we want */ 1759 spin_lock(&fs->lock); 1760 /* "users" and "in_exec" locked for check_unsafe_exec() */ 1761 if (fs->in_exec) { 1762 spin_unlock(&fs->lock); 1763 return -EAGAIN; 1764 } 1765 fs->users++; 1766 spin_unlock(&fs->lock); 1767 return 0; 1768 } 1769 tsk->fs = copy_fs_struct(fs); 1770 if (!tsk->fs) 1771 return -ENOMEM; 1772 return 0; 1773 } 1774 1775 static int copy_files(unsigned long clone_flags, struct task_struct *tsk, 1776 int no_files) 1777 { 1778 struct files_struct *oldf, *newf; 1779 1780 /* 1781 * A background process may not have any files ... 1782 */ 1783 oldf = current->files; 1784 if (!oldf) 1785 return 0; 1786 1787 if (no_files) { 1788 tsk->files = NULL; 1789 return 0; 1790 } 1791 1792 if (clone_flags & CLONE_FILES) { 1793 atomic_inc(&oldf->count); 1794 return 0; 1795 } 1796 1797 newf = dup_fd(oldf, NULL); 1798 if (IS_ERR(newf)) 1799 return PTR_ERR(newf); 1800 1801 tsk->files = newf; 1802 return 0; 1803 } 1804 1805 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk) 1806 { 1807 struct sighand_struct *sig; 1808 1809 if (clone_flags & CLONE_SIGHAND) { 1810 refcount_inc(¤t->sighand->count); 1811 return 0; 1812 } 1813 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 1814 RCU_INIT_POINTER(tsk->sighand, sig); 1815 if (!sig) 1816 return -ENOMEM; 1817 1818 refcount_set(&sig->count, 1); 1819 spin_lock_irq(¤t->sighand->siglock); 1820 memcpy(sig->action, current->sighand->action, sizeof(sig->action)); 1821 spin_unlock_irq(¤t->sighand->siglock); 1822 1823 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */ 1824 if (clone_flags & CLONE_CLEAR_SIGHAND) 1825 flush_signal_handlers(tsk, 0); 1826 1827 return 0; 1828 } 1829 1830 void __cleanup_sighand(struct sighand_struct *sighand) 1831 { 1832 if (refcount_dec_and_test(&sighand->count)) { 1833 signalfd_cleanup(sighand); 1834 /* 1835 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it 1836 * without an RCU grace period, see __lock_task_sighand(). 1837 */ 1838 kmem_cache_free(sighand_cachep, sighand); 1839 } 1840 } 1841 1842 /* 1843 * Initialize POSIX timer handling for a thread group. 1844 */ 1845 static void posix_cpu_timers_init_group(struct signal_struct *sig) 1846 { 1847 struct posix_cputimers *pct = &sig->posix_cputimers; 1848 unsigned long cpu_limit; 1849 1850 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); 1851 posix_cputimers_group_init(pct, cpu_limit); 1852 } 1853 1854 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk) 1855 { 1856 struct signal_struct *sig; 1857 1858 if (clone_flags & CLONE_THREAD) 1859 return 0; 1860 1861 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL); 1862 tsk->signal = sig; 1863 if (!sig) 1864 return -ENOMEM; 1865 1866 sig->nr_threads = 1; 1867 sig->quick_threads = 1; 1868 atomic_set(&sig->live, 1); 1869 refcount_set(&sig->sigcnt, 1); 1870 1871 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */ 1872 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node); 1873 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head); 1874 1875 init_waitqueue_head(&sig->wait_chldexit); 1876 sig->curr_target = tsk; 1877 init_sigpending(&sig->shared_pending); 1878 INIT_HLIST_HEAD(&sig->multiprocess); 1879 seqlock_init(&sig->stats_lock); 1880 prev_cputime_init(&sig->prev_cputime); 1881 1882 #ifdef CONFIG_POSIX_TIMERS 1883 INIT_HLIST_HEAD(&sig->posix_timers); 1884 INIT_HLIST_HEAD(&sig->ignored_posix_timers); 1885 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1886 sig->real_timer.function = it_real_fn; 1887 #endif 1888 1889 task_lock(current->group_leader); 1890 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); 1891 task_unlock(current->group_leader); 1892 1893 posix_cpu_timers_init_group(sig); 1894 1895 tty_audit_fork(sig); 1896 sched_autogroup_fork(sig); 1897 1898 sig->oom_score_adj = current->signal->oom_score_adj; 1899 sig->oom_score_adj_min = current->signal->oom_score_adj_min; 1900 1901 mutex_init(&sig->cred_guard_mutex); 1902 init_rwsem(&sig->exec_update_lock); 1903 1904 return 0; 1905 } 1906 1907 static void copy_seccomp(struct task_struct *p) 1908 { 1909 #ifdef CONFIG_SECCOMP 1910 /* 1911 * Must be called with sighand->lock held, which is common to 1912 * all threads in the group. Holding cred_guard_mutex is not 1913 * needed because this new task is not yet running and cannot 1914 * be racing exec. 1915 */ 1916 assert_spin_locked(¤t->sighand->siglock); 1917 1918 /* Ref-count the new filter user, and assign it. */ 1919 get_seccomp_filter(current); 1920 p->seccomp = current->seccomp; 1921 1922 /* 1923 * Explicitly enable no_new_privs here in case it got set 1924 * between the task_struct being duplicated and holding the 1925 * sighand lock. The seccomp state and nnp must be in sync. 1926 */ 1927 if (task_no_new_privs(current)) 1928 task_set_no_new_privs(p); 1929 1930 /* 1931 * If the parent gained a seccomp mode after copying thread 1932 * flags and between before we held the sighand lock, we have 1933 * to manually enable the seccomp thread flag here. 1934 */ 1935 if (p->seccomp.mode != SECCOMP_MODE_DISABLED) 1936 set_task_syscall_work(p, SECCOMP); 1937 #endif 1938 } 1939 1940 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr) 1941 { 1942 current->clear_child_tid = tidptr; 1943 1944 return task_pid_vnr(current); 1945 } 1946 1947 static void rt_mutex_init_task(struct task_struct *p) 1948 { 1949 raw_spin_lock_init(&p->pi_lock); 1950 #ifdef CONFIG_RT_MUTEXES 1951 p->pi_waiters = RB_ROOT_CACHED; 1952 p->pi_top_task = NULL; 1953 p->pi_blocked_on = NULL; 1954 #endif 1955 } 1956 1957 static inline void init_task_pid_links(struct task_struct *task) 1958 { 1959 enum pid_type type; 1960 1961 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) 1962 INIT_HLIST_NODE(&task->pid_links[type]); 1963 } 1964 1965 static inline void 1966 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid) 1967 { 1968 if (type == PIDTYPE_PID) 1969 task->thread_pid = pid; 1970 else 1971 task->signal->pids[type] = pid; 1972 } 1973 1974 static inline void rcu_copy_process(struct task_struct *p) 1975 { 1976 #ifdef CONFIG_PREEMPT_RCU 1977 p->rcu_read_lock_nesting = 0; 1978 p->rcu_read_unlock_special.s = 0; 1979 p->rcu_blocked_node = NULL; 1980 INIT_LIST_HEAD(&p->rcu_node_entry); 1981 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 1982 #ifdef CONFIG_TASKS_RCU 1983 p->rcu_tasks_holdout = false; 1984 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list); 1985 p->rcu_tasks_idle_cpu = -1; 1986 INIT_LIST_HEAD(&p->rcu_tasks_exit_list); 1987 #endif /* #ifdef CONFIG_TASKS_RCU */ 1988 #ifdef CONFIG_TASKS_TRACE_RCU 1989 p->trc_reader_nesting = 0; 1990 p->trc_reader_special.s = 0; 1991 INIT_LIST_HEAD(&p->trc_holdout_list); 1992 INIT_LIST_HEAD(&p->trc_blkd_node); 1993 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */ 1994 } 1995 1996 /** 1997 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd 1998 * @pid: the struct pid for which to create a pidfd 1999 * @flags: flags of the new @pidfd 2000 * @ret: Where to return the file for the pidfd. 2001 * 2002 * Allocate a new file that stashes @pid and reserve a new pidfd number in the 2003 * caller's file descriptor table. The pidfd is reserved but not installed yet. 2004 * 2005 * The helper doesn't perform checks on @pid which makes it useful for pidfds 2006 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and 2007 * pidfd file are prepared. 2008 * 2009 * If this function returns successfully the caller is responsible to either 2010 * call fd_install() passing the returned pidfd and pidfd file as arguments in 2011 * order to install the pidfd into its file descriptor table or they must use 2012 * put_unused_fd() and fput() on the returned pidfd and pidfd file 2013 * respectively. 2014 * 2015 * This function is useful when a pidfd must already be reserved but there 2016 * might still be points of failure afterwards and the caller wants to ensure 2017 * that no pidfd is leaked into its file descriptor table. 2018 * 2019 * Return: On success, a reserved pidfd is returned from the function and a new 2020 * pidfd file is returned in the last argument to the function. On 2021 * error, a negative error code is returned from the function and the 2022 * last argument remains unchanged. 2023 */ 2024 static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret) 2025 { 2026 int pidfd; 2027 struct file *pidfd_file; 2028 2029 pidfd = get_unused_fd_flags(O_CLOEXEC); 2030 if (pidfd < 0) 2031 return pidfd; 2032 2033 pidfd_file = pidfs_alloc_file(pid, flags | O_RDWR); 2034 if (IS_ERR(pidfd_file)) { 2035 put_unused_fd(pidfd); 2036 return PTR_ERR(pidfd_file); 2037 } 2038 /* 2039 * anon_inode_getfile() ignores everything outside of the 2040 * O_ACCMODE | O_NONBLOCK mask, set PIDFD_THREAD manually. 2041 */ 2042 pidfd_file->f_flags |= (flags & PIDFD_THREAD); 2043 *ret = pidfd_file; 2044 return pidfd; 2045 } 2046 2047 /** 2048 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd 2049 * @pid: the struct pid for which to create a pidfd 2050 * @flags: flags of the new @pidfd 2051 * @ret: Where to return the pidfd. 2052 * 2053 * Allocate a new file that stashes @pid and reserve a new pidfd number in the 2054 * caller's file descriptor table. The pidfd is reserved but not installed yet. 2055 * 2056 * The helper verifies that @pid is still in use, without PIDFD_THREAD the 2057 * task identified by @pid must be a thread-group leader. 2058 * 2059 * If this function returns successfully the caller is responsible to either 2060 * call fd_install() passing the returned pidfd and pidfd file as arguments in 2061 * order to install the pidfd into its file descriptor table or they must use 2062 * put_unused_fd() and fput() on the returned pidfd and pidfd file 2063 * respectively. 2064 * 2065 * This function is useful when a pidfd must already be reserved but there 2066 * might still be points of failure afterwards and the caller wants to ensure 2067 * that no pidfd is leaked into its file descriptor table. 2068 * 2069 * Return: On success, a reserved pidfd is returned from the function and a new 2070 * pidfd file is returned in the last argument to the function. On 2071 * error, a negative error code is returned from the function and the 2072 * last argument remains unchanged. 2073 */ 2074 int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret) 2075 { 2076 bool thread = flags & PIDFD_THREAD; 2077 2078 if (!pid || !pid_has_task(pid, thread ? PIDTYPE_PID : PIDTYPE_TGID)) 2079 return -EINVAL; 2080 2081 return __pidfd_prepare(pid, flags, ret); 2082 } 2083 2084 static void __delayed_free_task(struct rcu_head *rhp) 2085 { 2086 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); 2087 2088 free_task(tsk); 2089 } 2090 2091 static __always_inline void delayed_free_task(struct task_struct *tsk) 2092 { 2093 if (IS_ENABLED(CONFIG_MEMCG)) 2094 call_rcu(&tsk->rcu, __delayed_free_task); 2095 else 2096 free_task(tsk); 2097 } 2098 2099 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk) 2100 { 2101 /* Skip if kernel thread */ 2102 if (!tsk->mm) 2103 return; 2104 2105 /* Skip if spawning a thread or using vfork */ 2106 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM) 2107 return; 2108 2109 /* We need to synchronize with __set_oom_adj */ 2110 mutex_lock(&oom_adj_mutex); 2111 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags); 2112 /* Update the values in case they were changed after copy_signal */ 2113 tsk->signal->oom_score_adj = current->signal->oom_score_adj; 2114 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min; 2115 mutex_unlock(&oom_adj_mutex); 2116 } 2117 2118 #ifdef CONFIG_RV 2119 static void rv_task_fork(struct task_struct *p) 2120 { 2121 int i; 2122 2123 for (i = 0; i < RV_PER_TASK_MONITORS; i++) 2124 p->rv[i].da_mon.monitoring = false; 2125 } 2126 #else 2127 #define rv_task_fork(p) do {} while (0) 2128 #endif 2129 2130 /* 2131 * This creates a new process as a copy of the old one, 2132 * but does not actually start it yet. 2133 * 2134 * It copies the registers, and all the appropriate 2135 * parts of the process environment (as per the clone 2136 * flags). The actual kick-off is left to the caller. 2137 */ 2138 __latent_entropy struct task_struct *copy_process( 2139 struct pid *pid, 2140 int trace, 2141 int node, 2142 struct kernel_clone_args *args) 2143 { 2144 int pidfd = -1, retval; 2145 struct task_struct *p; 2146 struct multiprocess_signals delayed; 2147 struct file *pidfile = NULL; 2148 const u64 clone_flags = args->flags; 2149 struct nsproxy *nsp = current->nsproxy; 2150 2151 /* 2152 * Don't allow sharing the root directory with processes in a different 2153 * namespace 2154 */ 2155 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) 2156 return ERR_PTR(-EINVAL); 2157 2158 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS)) 2159 return ERR_PTR(-EINVAL); 2160 2161 /* 2162 * Thread groups must share signals as well, and detached threads 2163 * can only be started up within the thread group. 2164 */ 2165 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) 2166 return ERR_PTR(-EINVAL); 2167 2168 /* 2169 * Shared signal handlers imply shared VM. By way of the above, 2170 * thread groups also imply shared VM. Blocking this case allows 2171 * for various simplifications in other code. 2172 */ 2173 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) 2174 return ERR_PTR(-EINVAL); 2175 2176 /* 2177 * Siblings of global init remain as zombies on exit since they are 2178 * not reaped by their parent (swapper). To solve this and to avoid 2179 * multi-rooted process trees, prevent global and container-inits 2180 * from creating siblings. 2181 */ 2182 if ((clone_flags & CLONE_PARENT) && 2183 current->signal->flags & SIGNAL_UNKILLABLE) 2184 return ERR_PTR(-EINVAL); 2185 2186 /* 2187 * If the new process will be in a different pid or user namespace 2188 * do not allow it to share a thread group with the forking task. 2189 */ 2190 if (clone_flags & CLONE_THREAD) { 2191 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) || 2192 (task_active_pid_ns(current) != nsp->pid_ns_for_children)) 2193 return ERR_PTR(-EINVAL); 2194 } 2195 2196 if (clone_flags & CLONE_PIDFD) { 2197 /* 2198 * - CLONE_DETACHED is blocked so that we can potentially 2199 * reuse it later for CLONE_PIDFD. 2200 */ 2201 if (clone_flags & CLONE_DETACHED) 2202 return ERR_PTR(-EINVAL); 2203 } 2204 2205 /* 2206 * Force any signals received before this point to be delivered 2207 * before the fork happens. Collect up signals sent to multiple 2208 * processes that happen during the fork and delay them so that 2209 * they appear to happen after the fork. 2210 */ 2211 sigemptyset(&delayed.signal); 2212 INIT_HLIST_NODE(&delayed.node); 2213 2214 spin_lock_irq(¤t->sighand->siglock); 2215 if (!(clone_flags & CLONE_THREAD)) 2216 hlist_add_head(&delayed.node, ¤t->signal->multiprocess); 2217 recalc_sigpending(); 2218 spin_unlock_irq(¤t->sighand->siglock); 2219 retval = -ERESTARTNOINTR; 2220 if (task_sigpending(current)) 2221 goto fork_out; 2222 2223 retval = -ENOMEM; 2224 p = dup_task_struct(current, node); 2225 if (!p) 2226 goto fork_out; 2227 p->flags &= ~PF_KTHREAD; 2228 if (args->kthread) 2229 p->flags |= PF_KTHREAD; 2230 if (args->user_worker) { 2231 /* 2232 * Mark us a user worker, and block any signal that isn't 2233 * fatal or STOP 2234 */ 2235 p->flags |= PF_USER_WORKER; 2236 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP)); 2237 } 2238 if (args->io_thread) 2239 p->flags |= PF_IO_WORKER; 2240 2241 if (args->name) 2242 strscpy_pad(p->comm, args->name, sizeof(p->comm)); 2243 2244 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL; 2245 /* 2246 * Clear TID on mm_release()? 2247 */ 2248 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL; 2249 2250 ftrace_graph_init_task(p); 2251 2252 rt_mutex_init_task(p); 2253 2254 lockdep_assert_irqs_enabled(); 2255 #ifdef CONFIG_PROVE_LOCKING 2256 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 2257 #endif 2258 retval = copy_creds(p, clone_flags); 2259 if (retval < 0) 2260 goto bad_fork_free; 2261 2262 retval = -EAGAIN; 2263 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) { 2264 if (p->real_cred->user != INIT_USER && 2265 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) 2266 goto bad_fork_cleanup_count; 2267 } 2268 current->flags &= ~PF_NPROC_EXCEEDED; 2269 2270 /* 2271 * If multiple threads are within copy_process(), then this check 2272 * triggers too late. This doesn't hurt, the check is only there 2273 * to stop root fork bombs. 2274 */ 2275 retval = -EAGAIN; 2276 if (data_race(nr_threads >= max_threads)) 2277 goto bad_fork_cleanup_count; 2278 2279 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */ 2280 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY); 2281 p->flags |= PF_FORKNOEXEC; 2282 INIT_LIST_HEAD(&p->children); 2283 INIT_LIST_HEAD(&p->sibling); 2284 rcu_copy_process(p); 2285 p->vfork_done = NULL; 2286 spin_lock_init(&p->alloc_lock); 2287 2288 init_sigpending(&p->pending); 2289 2290 p->utime = p->stime = p->gtime = 0; 2291 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 2292 p->utimescaled = p->stimescaled = 0; 2293 #endif 2294 prev_cputime_init(&p->prev_cputime); 2295 2296 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN 2297 seqcount_init(&p->vtime.seqcount); 2298 p->vtime.starttime = 0; 2299 p->vtime.state = VTIME_INACTIVE; 2300 #endif 2301 2302 #ifdef CONFIG_IO_URING 2303 p->io_uring = NULL; 2304 #endif 2305 2306 p->default_timer_slack_ns = current->timer_slack_ns; 2307 2308 #ifdef CONFIG_PSI 2309 p->psi_flags = 0; 2310 #endif 2311 2312 task_io_accounting_init(&p->ioac); 2313 acct_clear_integrals(p); 2314 2315 posix_cputimers_init(&p->posix_cputimers); 2316 tick_dep_init_task(p); 2317 2318 p->io_context = NULL; 2319 audit_set_context(p, NULL); 2320 cgroup_fork(p); 2321 if (args->kthread) { 2322 if (!set_kthread_struct(p)) 2323 goto bad_fork_cleanup_delayacct; 2324 } 2325 #ifdef CONFIG_NUMA 2326 p->mempolicy = mpol_dup(p->mempolicy); 2327 if (IS_ERR(p->mempolicy)) { 2328 retval = PTR_ERR(p->mempolicy); 2329 p->mempolicy = NULL; 2330 goto bad_fork_cleanup_delayacct; 2331 } 2332 #endif 2333 #ifdef CONFIG_CPUSETS 2334 p->cpuset_mem_spread_rotor = NUMA_NO_NODE; 2335 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock); 2336 #endif 2337 #ifdef CONFIG_TRACE_IRQFLAGS 2338 memset(&p->irqtrace, 0, sizeof(p->irqtrace)); 2339 p->irqtrace.hardirq_disable_ip = _THIS_IP_; 2340 p->irqtrace.softirq_enable_ip = _THIS_IP_; 2341 p->softirqs_enabled = 1; 2342 p->softirq_context = 0; 2343 #endif 2344 2345 p->pagefault_disabled = 0; 2346 2347 #ifdef CONFIG_LOCKDEP 2348 lockdep_init_task(p); 2349 #endif 2350 2351 #ifdef CONFIG_DEBUG_MUTEXES 2352 p->blocked_on = NULL; /* not blocked yet */ 2353 #endif 2354 #ifdef CONFIG_BCACHE 2355 p->sequential_io = 0; 2356 p->sequential_io_avg = 0; 2357 #endif 2358 #ifdef CONFIG_BPF_SYSCALL 2359 RCU_INIT_POINTER(p->bpf_storage, NULL); 2360 p->bpf_ctx = NULL; 2361 #endif 2362 2363 /* Perform scheduler related setup. Assign this task to a CPU. */ 2364 retval = sched_fork(clone_flags, p); 2365 if (retval) 2366 goto bad_fork_cleanup_policy; 2367 2368 retval = perf_event_init_task(p, clone_flags); 2369 if (retval) 2370 goto bad_fork_sched_cancel_fork; 2371 retval = audit_alloc(p); 2372 if (retval) 2373 goto bad_fork_cleanup_perf; 2374 /* copy all the process information */ 2375 shm_init_task(p); 2376 retval = security_task_alloc(p, clone_flags); 2377 if (retval) 2378 goto bad_fork_cleanup_audit; 2379 retval = copy_semundo(clone_flags, p); 2380 if (retval) 2381 goto bad_fork_cleanup_security; 2382 retval = copy_files(clone_flags, p, args->no_files); 2383 if (retval) 2384 goto bad_fork_cleanup_semundo; 2385 retval = copy_fs(clone_flags, p); 2386 if (retval) 2387 goto bad_fork_cleanup_files; 2388 retval = copy_sighand(clone_flags, p); 2389 if (retval) 2390 goto bad_fork_cleanup_fs; 2391 retval = copy_signal(clone_flags, p); 2392 if (retval) 2393 goto bad_fork_cleanup_sighand; 2394 retval = copy_mm(clone_flags, p); 2395 if (retval) 2396 goto bad_fork_cleanup_signal; 2397 retval = copy_namespaces(clone_flags, p); 2398 if (retval) 2399 goto bad_fork_cleanup_mm; 2400 retval = copy_io(clone_flags, p); 2401 if (retval) 2402 goto bad_fork_cleanup_namespaces; 2403 retval = copy_thread(p, args); 2404 if (retval) 2405 goto bad_fork_cleanup_io; 2406 2407 stackleak_task_init(p); 2408 2409 if (pid != &init_struct_pid) { 2410 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid, 2411 args->set_tid_size); 2412 if (IS_ERR(pid)) { 2413 retval = PTR_ERR(pid); 2414 goto bad_fork_cleanup_thread; 2415 } 2416 } 2417 2418 /* 2419 * This has to happen after we've potentially unshared the file 2420 * descriptor table (so that the pidfd doesn't leak into the child 2421 * if the fd table isn't shared). 2422 */ 2423 if (clone_flags & CLONE_PIDFD) { 2424 int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0; 2425 2426 /* Note that no task has been attached to @pid yet. */ 2427 retval = __pidfd_prepare(pid, flags, &pidfile); 2428 if (retval < 0) 2429 goto bad_fork_free_pid; 2430 pidfd = retval; 2431 2432 retval = put_user(pidfd, args->pidfd); 2433 if (retval) 2434 goto bad_fork_put_pidfd; 2435 } 2436 2437 #ifdef CONFIG_BLOCK 2438 p->plug = NULL; 2439 #endif 2440 futex_init_task(p); 2441 2442 /* 2443 * sigaltstack should be cleared when sharing the same VM 2444 */ 2445 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM) 2446 sas_ss_reset(p); 2447 2448 /* 2449 * Syscall tracing and stepping should be turned off in the 2450 * child regardless of CLONE_PTRACE. 2451 */ 2452 user_disable_single_step(p); 2453 clear_task_syscall_work(p, SYSCALL_TRACE); 2454 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU) 2455 clear_task_syscall_work(p, SYSCALL_EMU); 2456 #endif 2457 clear_tsk_latency_tracing(p); 2458 2459 /* ok, now we should be set up.. */ 2460 p->pid = pid_nr(pid); 2461 if (clone_flags & CLONE_THREAD) { 2462 p->group_leader = current->group_leader; 2463 p->tgid = current->tgid; 2464 } else { 2465 p->group_leader = p; 2466 p->tgid = p->pid; 2467 } 2468 2469 p->nr_dirtied = 0; 2470 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10); 2471 p->dirty_paused_when = 0; 2472 2473 p->pdeath_signal = 0; 2474 p->task_works = NULL; 2475 clear_posix_cputimers_work(p); 2476 2477 #ifdef CONFIG_KRETPROBES 2478 p->kretprobe_instances.first = NULL; 2479 #endif 2480 #ifdef CONFIG_RETHOOK 2481 p->rethooks.first = NULL; 2482 #endif 2483 2484 /* 2485 * Ensure that the cgroup subsystem policies allow the new process to be 2486 * forked. It should be noted that the new process's css_set can be changed 2487 * between here and cgroup_post_fork() if an organisation operation is in 2488 * progress. 2489 */ 2490 retval = cgroup_can_fork(p, args); 2491 if (retval) 2492 goto bad_fork_put_pidfd; 2493 2494 /* 2495 * Now that the cgroups are pinned, re-clone the parent cgroup and put 2496 * the new task on the correct runqueue. All this *before* the task 2497 * becomes visible. 2498 * 2499 * This isn't part of ->can_fork() because while the re-cloning is 2500 * cgroup specific, it unconditionally needs to place the task on a 2501 * runqueue. 2502 */ 2503 retval = sched_cgroup_fork(p, args); 2504 if (retval) 2505 goto bad_fork_cancel_cgroup; 2506 2507 /* 2508 * From this point on we must avoid any synchronous user-space 2509 * communication until we take the tasklist-lock. In particular, we do 2510 * not want user-space to be able to predict the process start-time by 2511 * stalling fork(2) after we recorded the start_time but before it is 2512 * visible to the system. 2513 */ 2514 2515 p->start_time = ktime_get_ns(); 2516 p->start_boottime = ktime_get_boottime_ns(); 2517 2518 /* 2519 * Make it visible to the rest of the system, but dont wake it up yet. 2520 * Need tasklist lock for parent etc handling! 2521 */ 2522 write_lock_irq(&tasklist_lock); 2523 2524 /* CLONE_PARENT re-uses the old parent */ 2525 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { 2526 p->real_parent = current->real_parent; 2527 p->parent_exec_id = current->parent_exec_id; 2528 if (clone_flags & CLONE_THREAD) 2529 p->exit_signal = -1; 2530 else 2531 p->exit_signal = current->group_leader->exit_signal; 2532 } else { 2533 p->real_parent = current; 2534 p->parent_exec_id = current->self_exec_id; 2535 p->exit_signal = args->exit_signal; 2536 } 2537 2538 klp_copy_process(p); 2539 2540 sched_core_fork(p); 2541 2542 spin_lock(¤t->sighand->siglock); 2543 2544 rv_task_fork(p); 2545 2546 rseq_fork(p, clone_flags); 2547 2548 /* Don't start children in a dying pid namespace */ 2549 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) { 2550 retval = -ENOMEM; 2551 goto bad_fork_core_free; 2552 } 2553 2554 /* Let kill terminate clone/fork in the middle */ 2555 if (fatal_signal_pending(current)) { 2556 retval = -EINTR; 2557 goto bad_fork_core_free; 2558 } 2559 2560 /* No more failure paths after this point. */ 2561 2562 /* 2563 * Copy seccomp details explicitly here, in case they were changed 2564 * before holding sighand lock. 2565 */ 2566 copy_seccomp(p); 2567 2568 init_task_pid_links(p); 2569 if (likely(p->pid)) { 2570 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace); 2571 2572 init_task_pid(p, PIDTYPE_PID, pid); 2573 if (thread_group_leader(p)) { 2574 init_task_pid(p, PIDTYPE_TGID, pid); 2575 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current)); 2576 init_task_pid(p, PIDTYPE_SID, task_session(current)); 2577 2578 if (is_child_reaper(pid)) { 2579 ns_of_pid(pid)->child_reaper = p; 2580 p->signal->flags |= SIGNAL_UNKILLABLE; 2581 } 2582 p->signal->shared_pending.signal = delayed.signal; 2583 p->signal->tty = tty_kref_get(current->signal->tty); 2584 /* 2585 * Inherit has_child_subreaper flag under the same 2586 * tasklist_lock with adding child to the process tree 2587 * for propagate_has_child_subreaper optimization. 2588 */ 2589 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper || 2590 p->real_parent->signal->is_child_subreaper; 2591 list_add_tail(&p->sibling, &p->real_parent->children); 2592 list_add_tail_rcu(&p->tasks, &init_task.tasks); 2593 attach_pid(p, PIDTYPE_TGID); 2594 attach_pid(p, PIDTYPE_PGID); 2595 attach_pid(p, PIDTYPE_SID); 2596 __this_cpu_inc(process_counts); 2597 } else { 2598 current->signal->nr_threads++; 2599 current->signal->quick_threads++; 2600 atomic_inc(¤t->signal->live); 2601 refcount_inc(¤t->signal->sigcnt); 2602 task_join_group_stop(p); 2603 list_add_tail_rcu(&p->thread_node, 2604 &p->signal->thread_head); 2605 } 2606 attach_pid(p, PIDTYPE_PID); 2607 nr_threads++; 2608 } 2609 total_forks++; 2610 hlist_del_init(&delayed.node); 2611 spin_unlock(¤t->sighand->siglock); 2612 syscall_tracepoint_update(p); 2613 write_unlock_irq(&tasklist_lock); 2614 2615 if (pidfile) 2616 fd_install(pidfd, pidfile); 2617 2618 proc_fork_connector(p); 2619 sched_post_fork(p); 2620 cgroup_post_fork(p, args); 2621 perf_event_fork(p); 2622 2623 trace_task_newtask(p, clone_flags); 2624 uprobe_copy_process(p, clone_flags); 2625 user_events_fork(p, clone_flags); 2626 2627 copy_oom_score_adj(clone_flags, p); 2628 2629 return p; 2630 2631 bad_fork_core_free: 2632 sched_core_free(p); 2633 spin_unlock(¤t->sighand->siglock); 2634 write_unlock_irq(&tasklist_lock); 2635 bad_fork_cancel_cgroup: 2636 cgroup_cancel_fork(p, args); 2637 bad_fork_put_pidfd: 2638 if (clone_flags & CLONE_PIDFD) { 2639 fput(pidfile); 2640 put_unused_fd(pidfd); 2641 } 2642 bad_fork_free_pid: 2643 if (pid != &init_struct_pid) 2644 free_pid(pid); 2645 bad_fork_cleanup_thread: 2646 exit_thread(p); 2647 bad_fork_cleanup_io: 2648 if (p->io_context) 2649 exit_io_context(p); 2650 bad_fork_cleanup_namespaces: 2651 exit_task_namespaces(p); 2652 bad_fork_cleanup_mm: 2653 if (p->mm) { 2654 mm_clear_owner(p->mm, p); 2655 mmput(p->mm); 2656 } 2657 bad_fork_cleanup_signal: 2658 if (!(clone_flags & CLONE_THREAD)) 2659 free_signal_struct(p->signal); 2660 bad_fork_cleanup_sighand: 2661 __cleanup_sighand(p->sighand); 2662 bad_fork_cleanup_fs: 2663 exit_fs(p); /* blocking */ 2664 bad_fork_cleanup_files: 2665 exit_files(p); /* blocking */ 2666 bad_fork_cleanup_semundo: 2667 exit_sem(p); 2668 bad_fork_cleanup_security: 2669 security_task_free(p); 2670 bad_fork_cleanup_audit: 2671 audit_free(p); 2672 bad_fork_cleanup_perf: 2673 perf_event_free_task(p); 2674 bad_fork_sched_cancel_fork: 2675 sched_cancel_fork(p); 2676 bad_fork_cleanup_policy: 2677 lockdep_free_task(p); 2678 #ifdef CONFIG_NUMA 2679 mpol_put(p->mempolicy); 2680 #endif 2681 bad_fork_cleanup_delayacct: 2682 delayacct_tsk_free(p); 2683 bad_fork_cleanup_count: 2684 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1); 2685 exit_creds(p); 2686 bad_fork_free: 2687 WRITE_ONCE(p->__state, TASK_DEAD); 2688 exit_task_stack_account(p); 2689 put_task_stack(p); 2690 delayed_free_task(p); 2691 fork_out: 2692 spin_lock_irq(¤t->sighand->siglock); 2693 hlist_del_init(&delayed.node); 2694 spin_unlock_irq(¤t->sighand->siglock); 2695 return ERR_PTR(retval); 2696 } 2697 2698 static inline void init_idle_pids(struct task_struct *idle) 2699 { 2700 enum pid_type type; 2701 2702 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) { 2703 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */ 2704 init_task_pid(idle, type, &init_struct_pid); 2705 } 2706 } 2707 2708 static int idle_dummy(void *dummy) 2709 { 2710 /* This function is never called */ 2711 return 0; 2712 } 2713 2714 struct task_struct * __init fork_idle(int cpu) 2715 { 2716 struct task_struct *task; 2717 struct kernel_clone_args args = { 2718 .flags = CLONE_VM, 2719 .fn = &idle_dummy, 2720 .fn_arg = NULL, 2721 .kthread = 1, 2722 .idle = 1, 2723 }; 2724 2725 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args); 2726 if (!IS_ERR(task)) { 2727 init_idle_pids(task); 2728 init_idle(task, cpu); 2729 } 2730 2731 return task; 2732 } 2733 2734 /* 2735 * This is like kernel_clone(), but shaved down and tailored to just 2736 * creating io_uring workers. It returns a created task, or an error pointer. 2737 * The returned task is inactive, and the caller must fire it up through 2738 * wake_up_new_task(p). All signals are blocked in the created task. 2739 */ 2740 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node) 2741 { 2742 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD| 2743 CLONE_IO; 2744 struct kernel_clone_args args = { 2745 .flags = ((lower_32_bits(flags) | CLONE_VM | 2746 CLONE_UNTRACED) & ~CSIGNAL), 2747 .exit_signal = (lower_32_bits(flags) & CSIGNAL), 2748 .fn = fn, 2749 .fn_arg = arg, 2750 .io_thread = 1, 2751 .user_worker = 1, 2752 }; 2753 2754 return copy_process(NULL, 0, node, &args); 2755 } 2756 2757 /* 2758 * Ok, this is the main fork-routine. 2759 * 2760 * It copies the process, and if successful kick-starts 2761 * it and waits for it to finish using the VM if required. 2762 * 2763 * args->exit_signal is expected to be checked for sanity by the caller. 2764 */ 2765 pid_t kernel_clone(struct kernel_clone_args *args) 2766 { 2767 u64 clone_flags = args->flags; 2768 struct completion vfork; 2769 struct pid *pid; 2770 struct task_struct *p; 2771 int trace = 0; 2772 pid_t nr; 2773 2774 /* 2775 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument 2776 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are 2777 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate 2778 * field in struct clone_args and it still doesn't make sense to have 2779 * them both point at the same memory location. Performing this check 2780 * here has the advantage that we don't need to have a separate helper 2781 * to check for legacy clone(). 2782 */ 2783 if ((clone_flags & CLONE_PIDFD) && 2784 (clone_flags & CLONE_PARENT_SETTID) && 2785 (args->pidfd == args->parent_tid)) 2786 return -EINVAL; 2787 2788 /* 2789 * Determine whether and which event to report to ptracer. When 2790 * called from kernel_thread or CLONE_UNTRACED is explicitly 2791 * requested, no event is reported; otherwise, report if the event 2792 * for the type of forking is enabled. 2793 */ 2794 if (!(clone_flags & CLONE_UNTRACED)) { 2795 if (clone_flags & CLONE_VFORK) 2796 trace = PTRACE_EVENT_VFORK; 2797 else if (args->exit_signal != SIGCHLD) 2798 trace = PTRACE_EVENT_CLONE; 2799 else 2800 trace = PTRACE_EVENT_FORK; 2801 2802 if (likely(!ptrace_event_enabled(current, trace))) 2803 trace = 0; 2804 } 2805 2806 p = copy_process(NULL, trace, NUMA_NO_NODE, args); 2807 add_latent_entropy(); 2808 2809 if (IS_ERR(p)) 2810 return PTR_ERR(p); 2811 2812 /* 2813 * Do this prior waking up the new thread - the thread pointer 2814 * might get invalid after that point, if the thread exits quickly. 2815 */ 2816 trace_sched_process_fork(current, p); 2817 2818 pid = get_task_pid(p, PIDTYPE_PID); 2819 nr = pid_vnr(pid); 2820 2821 if (clone_flags & CLONE_PARENT_SETTID) 2822 put_user(nr, args->parent_tid); 2823 2824 if (clone_flags & CLONE_VFORK) { 2825 p->vfork_done = &vfork; 2826 init_completion(&vfork); 2827 get_task_struct(p); 2828 } 2829 2830 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) { 2831 /* lock the task to synchronize with memcg migration */ 2832 task_lock(p); 2833 lru_gen_add_mm(p->mm); 2834 task_unlock(p); 2835 } 2836 2837 wake_up_new_task(p); 2838 2839 /* forking complete and child started to run, tell ptracer */ 2840 if (unlikely(trace)) 2841 ptrace_event_pid(trace, pid); 2842 2843 if (clone_flags & CLONE_VFORK) { 2844 if (!wait_for_vfork_done(p, &vfork)) 2845 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid); 2846 } 2847 2848 put_pid(pid); 2849 return nr; 2850 } 2851 2852 /* 2853 * Create a kernel thread. 2854 */ 2855 pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name, 2856 unsigned long flags) 2857 { 2858 struct kernel_clone_args args = { 2859 .flags = ((lower_32_bits(flags) | CLONE_VM | 2860 CLONE_UNTRACED) & ~CSIGNAL), 2861 .exit_signal = (lower_32_bits(flags) & CSIGNAL), 2862 .fn = fn, 2863 .fn_arg = arg, 2864 .name = name, 2865 .kthread = 1, 2866 }; 2867 2868 return kernel_clone(&args); 2869 } 2870 2871 /* 2872 * Create a user mode thread. 2873 */ 2874 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags) 2875 { 2876 struct kernel_clone_args args = { 2877 .flags = ((lower_32_bits(flags) | CLONE_VM | 2878 CLONE_UNTRACED) & ~CSIGNAL), 2879 .exit_signal = (lower_32_bits(flags) & CSIGNAL), 2880 .fn = fn, 2881 .fn_arg = arg, 2882 }; 2883 2884 return kernel_clone(&args); 2885 } 2886 2887 #ifdef __ARCH_WANT_SYS_FORK 2888 SYSCALL_DEFINE0(fork) 2889 { 2890 #ifdef CONFIG_MMU 2891 struct kernel_clone_args args = { 2892 .exit_signal = SIGCHLD, 2893 }; 2894 2895 return kernel_clone(&args); 2896 #else 2897 /* can not support in nommu mode */ 2898 return -EINVAL; 2899 #endif 2900 } 2901 #endif 2902 2903 #ifdef __ARCH_WANT_SYS_VFORK 2904 SYSCALL_DEFINE0(vfork) 2905 { 2906 struct kernel_clone_args args = { 2907 .flags = CLONE_VFORK | CLONE_VM, 2908 .exit_signal = SIGCHLD, 2909 }; 2910 2911 return kernel_clone(&args); 2912 } 2913 #endif 2914 2915 #ifdef __ARCH_WANT_SYS_CLONE 2916 #ifdef CONFIG_CLONE_BACKWARDS 2917 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2918 int __user *, parent_tidptr, 2919 unsigned long, tls, 2920 int __user *, child_tidptr) 2921 #elif defined(CONFIG_CLONE_BACKWARDS2) 2922 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags, 2923 int __user *, parent_tidptr, 2924 int __user *, child_tidptr, 2925 unsigned long, tls) 2926 #elif defined(CONFIG_CLONE_BACKWARDS3) 2927 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp, 2928 int, stack_size, 2929 int __user *, parent_tidptr, 2930 int __user *, child_tidptr, 2931 unsigned long, tls) 2932 #else 2933 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp, 2934 int __user *, parent_tidptr, 2935 int __user *, child_tidptr, 2936 unsigned long, tls) 2937 #endif 2938 { 2939 struct kernel_clone_args args = { 2940 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL), 2941 .pidfd = parent_tidptr, 2942 .child_tid = child_tidptr, 2943 .parent_tid = parent_tidptr, 2944 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL), 2945 .stack = newsp, 2946 .tls = tls, 2947 }; 2948 2949 return kernel_clone(&args); 2950 } 2951 #endif 2952 2953 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs, 2954 struct clone_args __user *uargs, 2955 size_t usize) 2956 { 2957 int err; 2958 struct clone_args args; 2959 pid_t *kset_tid = kargs->set_tid; 2960 2961 BUILD_BUG_ON(offsetofend(struct clone_args, tls) != 2962 CLONE_ARGS_SIZE_VER0); 2963 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) != 2964 CLONE_ARGS_SIZE_VER1); 2965 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) != 2966 CLONE_ARGS_SIZE_VER2); 2967 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2); 2968 2969 if (unlikely(usize > PAGE_SIZE)) 2970 return -E2BIG; 2971 if (unlikely(usize < CLONE_ARGS_SIZE_VER0)) 2972 return -EINVAL; 2973 2974 err = copy_struct_from_user(&args, sizeof(args), uargs, usize); 2975 if (err) 2976 return err; 2977 2978 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL)) 2979 return -EINVAL; 2980 2981 if (unlikely(!args.set_tid && args.set_tid_size > 0)) 2982 return -EINVAL; 2983 2984 if (unlikely(args.set_tid && args.set_tid_size == 0)) 2985 return -EINVAL; 2986 2987 /* 2988 * Verify that higher 32bits of exit_signal are unset and that 2989 * it is a valid signal 2990 */ 2991 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) || 2992 !valid_signal(args.exit_signal))) 2993 return -EINVAL; 2994 2995 if ((args.flags & CLONE_INTO_CGROUP) && 2996 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2)) 2997 return -EINVAL; 2998 2999 *kargs = (struct kernel_clone_args){ 3000 .flags = args.flags, 3001 .pidfd = u64_to_user_ptr(args.pidfd), 3002 .child_tid = u64_to_user_ptr(args.child_tid), 3003 .parent_tid = u64_to_user_ptr(args.parent_tid), 3004 .exit_signal = args.exit_signal, 3005 .stack = args.stack, 3006 .stack_size = args.stack_size, 3007 .tls = args.tls, 3008 .set_tid_size = args.set_tid_size, 3009 .cgroup = args.cgroup, 3010 }; 3011 3012 if (args.set_tid && 3013 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid), 3014 (kargs->set_tid_size * sizeof(pid_t)))) 3015 return -EFAULT; 3016 3017 kargs->set_tid = kset_tid; 3018 3019 return 0; 3020 } 3021 3022 /** 3023 * clone3_stack_valid - check and prepare stack 3024 * @kargs: kernel clone args 3025 * 3026 * Verify that the stack arguments userspace gave us are sane. 3027 * In addition, set the stack direction for userspace since it's easy for us to 3028 * determine. 3029 */ 3030 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs) 3031 { 3032 if (kargs->stack == 0) { 3033 if (kargs->stack_size > 0) 3034 return false; 3035 } else { 3036 if (kargs->stack_size == 0) 3037 return false; 3038 3039 if (!access_ok((void __user *)kargs->stack, kargs->stack_size)) 3040 return false; 3041 3042 #if !defined(CONFIG_STACK_GROWSUP) 3043 kargs->stack += kargs->stack_size; 3044 #endif 3045 } 3046 3047 return true; 3048 } 3049 3050 static bool clone3_args_valid(struct kernel_clone_args *kargs) 3051 { 3052 /* Verify that no unknown flags are passed along. */ 3053 if (kargs->flags & 3054 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP)) 3055 return false; 3056 3057 /* 3058 * - make the CLONE_DETACHED bit reusable for clone3 3059 * - make the CSIGNAL bits reusable for clone3 3060 */ 3061 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME)))) 3062 return false; 3063 3064 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) == 3065 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) 3066 return false; 3067 3068 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) && 3069 kargs->exit_signal) 3070 return false; 3071 3072 if (!clone3_stack_valid(kargs)) 3073 return false; 3074 3075 return true; 3076 } 3077 3078 /** 3079 * sys_clone3 - create a new process with specific properties 3080 * @uargs: argument structure 3081 * @size: size of @uargs 3082 * 3083 * clone3() is the extensible successor to clone()/clone2(). 3084 * It takes a struct as argument that is versioned by its size. 3085 * 3086 * Return: On success, a positive PID for the child process. 3087 * On error, a negative errno number. 3088 */ 3089 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size) 3090 { 3091 int err; 3092 3093 struct kernel_clone_args kargs; 3094 pid_t set_tid[MAX_PID_NS_LEVEL]; 3095 3096 #ifdef __ARCH_BROKEN_SYS_CLONE3 3097 #warning clone3() entry point is missing, please fix 3098 return -ENOSYS; 3099 #endif 3100 3101 kargs.set_tid = set_tid; 3102 3103 err = copy_clone_args_from_user(&kargs, uargs, size); 3104 if (err) 3105 return err; 3106 3107 if (!clone3_args_valid(&kargs)) 3108 return -EINVAL; 3109 3110 return kernel_clone(&kargs); 3111 } 3112 3113 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data) 3114 { 3115 struct task_struct *leader, *parent, *child; 3116 int res; 3117 3118 read_lock(&tasklist_lock); 3119 leader = top = top->group_leader; 3120 down: 3121 for_each_thread(leader, parent) { 3122 list_for_each_entry(child, &parent->children, sibling) { 3123 res = visitor(child, data); 3124 if (res) { 3125 if (res < 0) 3126 goto out; 3127 leader = child; 3128 goto down; 3129 } 3130 up: 3131 ; 3132 } 3133 } 3134 3135 if (leader != top) { 3136 child = leader; 3137 parent = child->real_parent; 3138 leader = parent->group_leader; 3139 goto up; 3140 } 3141 out: 3142 read_unlock(&tasklist_lock); 3143 } 3144 3145 #ifndef ARCH_MIN_MMSTRUCT_ALIGN 3146 #define ARCH_MIN_MMSTRUCT_ALIGN 0 3147 #endif 3148 3149 static void sighand_ctor(void *data) 3150 { 3151 struct sighand_struct *sighand = data; 3152 3153 spin_lock_init(&sighand->siglock); 3154 init_waitqueue_head(&sighand->signalfd_wqh); 3155 } 3156 3157 void __init mm_cache_init(void) 3158 { 3159 unsigned int mm_size; 3160 3161 /* 3162 * The mm_cpumask is located at the end of mm_struct, and is 3163 * dynamically sized based on the maximum CPU number this system 3164 * can have, taking hotplug into account (nr_cpu_ids). 3165 */ 3166 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size(); 3167 3168 mm_cachep = kmem_cache_create_usercopy("mm_struct", 3169 mm_size, ARCH_MIN_MMSTRUCT_ALIGN, 3170 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 3171 offsetof(struct mm_struct, saved_auxv), 3172 sizeof_field(struct mm_struct, saved_auxv), 3173 NULL); 3174 } 3175 3176 void __init proc_caches_init(void) 3177 { 3178 sighand_cachep = kmem_cache_create("sighand_cache", 3179 sizeof(struct sighand_struct), 0, 3180 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU| 3181 SLAB_ACCOUNT, sighand_ctor); 3182 signal_cachep = kmem_cache_create("signal_cache", 3183 sizeof(struct signal_struct), 0, 3184 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 3185 NULL); 3186 files_cachep = kmem_cache_create("files_cache", 3187 sizeof(struct files_struct), 0, 3188 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 3189 NULL); 3190 fs_cachep = kmem_cache_create("fs_cache", 3191 sizeof(struct fs_struct), 0, 3192 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, 3193 NULL); 3194 3195 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT); 3196 #ifdef CONFIG_PER_VMA_LOCK 3197 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT); 3198 #endif 3199 mmap_init(); 3200 nsproxy_cache_init(); 3201 } 3202 3203 /* 3204 * Check constraints on flags passed to the unshare system call. 3205 */ 3206 static int check_unshare_flags(unsigned long unshare_flags) 3207 { 3208 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND| 3209 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM| 3210 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET| 3211 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP| 3212 CLONE_NEWTIME)) 3213 return -EINVAL; 3214 /* 3215 * Not implemented, but pretend it works if there is nothing 3216 * to unshare. Note that unsharing the address space or the 3217 * signal handlers also need to unshare the signal queues (aka 3218 * CLONE_THREAD). 3219 */ 3220 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) { 3221 if (!thread_group_empty(current)) 3222 return -EINVAL; 3223 } 3224 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) { 3225 if (refcount_read(¤t->sighand->count) > 1) 3226 return -EINVAL; 3227 } 3228 if (unshare_flags & CLONE_VM) { 3229 if (!current_is_single_threaded()) 3230 return -EINVAL; 3231 } 3232 3233 return 0; 3234 } 3235 3236 /* 3237 * Unshare the filesystem structure if it is being shared 3238 */ 3239 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) 3240 { 3241 struct fs_struct *fs = current->fs; 3242 3243 if (!(unshare_flags & CLONE_FS) || !fs) 3244 return 0; 3245 3246 /* don't need lock here; in the worst case we'll do useless copy */ 3247 if (fs->users == 1) 3248 return 0; 3249 3250 *new_fsp = copy_fs_struct(fs); 3251 if (!*new_fsp) 3252 return -ENOMEM; 3253 3254 return 0; 3255 } 3256 3257 /* 3258 * Unshare file descriptor table if it is being shared 3259 */ 3260 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp) 3261 { 3262 struct files_struct *fd = current->files; 3263 3264 if ((unshare_flags & CLONE_FILES) && 3265 (fd && atomic_read(&fd->count) > 1)) { 3266 fd = dup_fd(fd, NULL); 3267 if (IS_ERR(fd)) 3268 return PTR_ERR(fd); 3269 *new_fdp = fd; 3270 } 3271 3272 return 0; 3273 } 3274 3275 /* 3276 * unshare allows a process to 'unshare' part of the process 3277 * context which was originally shared using clone. copy_* 3278 * functions used by kernel_clone() cannot be used here directly 3279 * because they modify an inactive task_struct that is being 3280 * constructed. Here we are modifying the current, active, 3281 * task_struct. 3282 */ 3283 int ksys_unshare(unsigned long unshare_flags) 3284 { 3285 struct fs_struct *fs, *new_fs = NULL; 3286 struct files_struct *new_fd = NULL; 3287 struct cred *new_cred = NULL; 3288 struct nsproxy *new_nsproxy = NULL; 3289 int do_sysvsem = 0; 3290 int err; 3291 3292 /* 3293 * If unsharing a user namespace must also unshare the thread group 3294 * and unshare the filesystem root and working directories. 3295 */ 3296 if (unshare_flags & CLONE_NEWUSER) 3297 unshare_flags |= CLONE_THREAD | CLONE_FS; 3298 /* 3299 * If unsharing vm, must also unshare signal handlers. 3300 */ 3301 if (unshare_flags & CLONE_VM) 3302 unshare_flags |= CLONE_SIGHAND; 3303 /* 3304 * If unsharing a signal handlers, must also unshare the signal queues. 3305 */ 3306 if (unshare_flags & CLONE_SIGHAND) 3307 unshare_flags |= CLONE_THREAD; 3308 /* 3309 * If unsharing namespace, must also unshare filesystem information. 3310 */ 3311 if (unshare_flags & CLONE_NEWNS) 3312 unshare_flags |= CLONE_FS; 3313 3314 err = check_unshare_flags(unshare_flags); 3315 if (err) 3316 goto bad_unshare_out; 3317 /* 3318 * CLONE_NEWIPC must also detach from the undolist: after switching 3319 * to a new ipc namespace, the semaphore arrays from the old 3320 * namespace are unreachable. 3321 */ 3322 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM)) 3323 do_sysvsem = 1; 3324 err = unshare_fs(unshare_flags, &new_fs); 3325 if (err) 3326 goto bad_unshare_out; 3327 err = unshare_fd(unshare_flags, &new_fd); 3328 if (err) 3329 goto bad_unshare_cleanup_fs; 3330 err = unshare_userns(unshare_flags, &new_cred); 3331 if (err) 3332 goto bad_unshare_cleanup_fd; 3333 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, 3334 new_cred, new_fs); 3335 if (err) 3336 goto bad_unshare_cleanup_cred; 3337 3338 if (new_cred) { 3339 err = set_cred_ucounts(new_cred); 3340 if (err) 3341 goto bad_unshare_cleanup_cred; 3342 } 3343 3344 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) { 3345 if (do_sysvsem) { 3346 /* 3347 * CLONE_SYSVSEM is equivalent to sys_exit(). 3348 */ 3349 exit_sem(current); 3350 } 3351 if (unshare_flags & CLONE_NEWIPC) { 3352 /* Orphan segments in old ns (see sem above). */ 3353 exit_shm(current); 3354 shm_init_task(current); 3355 } 3356 3357 if (new_nsproxy) 3358 switch_task_namespaces(current, new_nsproxy); 3359 3360 task_lock(current); 3361 3362 if (new_fs) { 3363 fs = current->fs; 3364 spin_lock(&fs->lock); 3365 current->fs = new_fs; 3366 if (--fs->users) 3367 new_fs = NULL; 3368 else 3369 new_fs = fs; 3370 spin_unlock(&fs->lock); 3371 } 3372 3373 if (new_fd) 3374 swap(current->files, new_fd); 3375 3376 task_unlock(current); 3377 3378 if (new_cred) { 3379 /* Install the new user namespace */ 3380 commit_creds(new_cred); 3381 new_cred = NULL; 3382 } 3383 } 3384 3385 perf_event_namespaces(current); 3386 3387 bad_unshare_cleanup_cred: 3388 if (new_cred) 3389 put_cred(new_cred); 3390 bad_unshare_cleanup_fd: 3391 if (new_fd) 3392 put_files_struct(new_fd); 3393 3394 bad_unshare_cleanup_fs: 3395 if (new_fs) 3396 free_fs_struct(new_fs); 3397 3398 bad_unshare_out: 3399 return err; 3400 } 3401 3402 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags) 3403 { 3404 return ksys_unshare(unshare_flags); 3405 } 3406 3407 /* 3408 * Helper to unshare the files of the current task. 3409 * We don't want to expose copy_files internals to 3410 * the exec layer of the kernel. 3411 */ 3412 3413 int unshare_files(void) 3414 { 3415 struct task_struct *task = current; 3416 struct files_struct *old, *copy = NULL; 3417 int error; 3418 3419 error = unshare_fd(CLONE_FILES, ©); 3420 if (error || !copy) 3421 return error; 3422 3423 old = task->files; 3424 task_lock(task); 3425 task->files = copy; 3426 task_unlock(task); 3427 put_files_struct(old); 3428 return 0; 3429 } 3430 3431 int sysctl_max_threads(const struct ctl_table *table, int write, 3432 void *buffer, size_t *lenp, loff_t *ppos) 3433 { 3434 struct ctl_table t; 3435 int ret; 3436 int threads = max_threads; 3437 int min = 1; 3438 int max = MAX_THREADS; 3439 3440 t = *table; 3441 t.data = &threads; 3442 t.extra1 = &min; 3443 t.extra2 = &max; 3444 3445 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); 3446 if (ret || !write) 3447 return ret; 3448 3449 max_threads = threads; 3450 3451 return 0; 3452 } 3453