1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
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()'
15 #include <linux/anon_inodes.h>
16 #include <linux/slab.h>
17 #include <linux/sched/autogroup.h>
18 #include <linux/sched/mm.h>
19 #include <linux/sched/coredump.h>
20 #include <linux/sched/user.h>
21 #include <linux/sched/numa_balancing.h>
22 #include <linux/sched/stat.h>
23 #include <linux/sched/task.h>
24 #include <linux/sched/task_stack.h>
25 #include <linux/sched/cputime.h>
26 #include <linux/seq_file.h>
27 #include <linux/rtmutex.h>
28 #include <linux/init.h>
29 #include <linux/unistd.h>
30 #include <linux/module.h>
31 #include <linux/vmalloc.h>
32 #include <linux/completion.h>
33 #include <linux/personality.h>
34 #include <linux/mempolicy.h>
35 #include <linux/sem.h>
36 #include <linux/file.h>
37 #include <linux/fdtable.h>
38 #include <linux/iocontext.h>
39 #include <linux/key.h>
40 #include <linux/binfmts.h>
41 #include <linux/mman.h>
42 #include <linux/mmu_notifier.h>
45 #include <linux/mm_inline.h>
46 #include <linux/nsproxy.h>
47 #include <linux/capability.h>
48 #include <linux/cpu.h>
49 #include <linux/cgroup.h>
50 #include <linux/security.h>
51 #include <linux/hugetlb.h>
52 #include <linux/seccomp.h>
53 #include <linux/swap.h>
54 #include <linux/syscalls.h>
55 #include <linux/jiffies.h>
56 #include <linux/futex.h>
57 #include <linux/compat.h>
58 #include <linux/kthread.h>
59 #include <linux/task_io_accounting_ops.h>
60 #include <linux/rcupdate.h>
61 #include <linux/ptrace.h>
62 #include <linux/mount.h>
63 #include <linux/audit.h>
64 #include <linux/memcontrol.h>
65 #include <linux/ftrace.h>
66 #include <linux/proc_fs.h>
67 #include <linux/profile.h>
68 #include <linux/rmap.h>
69 #include <linux/ksm.h>
70 #include <linux/acct.h>
71 #include <linux/userfaultfd_k.h>
72 #include <linux/tsacct_kern.h>
73 #include <linux/cn_proc.h>
74 #include <linux/freezer.h>
75 #include <linux/delayacct.h>
76 #include <linux/taskstats_kern.h>
77 #include <linux/random.h>
78 #include <linux/tty.h>
79 #include <linux/fs_struct.h>
80 #include <linux/magic.h>
81 #include <linux/perf_event.h>
82 #include <linux/posix-timers.h>
83 #include <linux/user-return-notifier.h>
84 #include <linux/oom.h>
85 #include <linux/khugepaged.h>
86 #include <linux/signalfd.h>
87 #include <linux/uprobes.h>
88 #include <linux/aio.h>
89 #include <linux/compiler.h>
90 #include <linux/sysctl.h>
91 #include <linux/kcov.h>
92 #include <linux/livepatch.h>
93 #include <linux/thread_info.h>
94 #include <linux/stackleak.h>
95 #include <linux/kasan.h>
96 #include <linux/scs.h>
97 #include <linux/io_uring.h>
98 #include <linux/bpf.h>
99 #include <linux/sched/mm.h>
101 #include <asm/pgalloc.h>
102 #include <linux/uaccess.h>
103 #include <asm/mmu_context.h>
104 #include <asm/cacheflush.h>
105 #include <asm/tlbflush.h>
107 #include <trace/events/sched.h>
109 #define CREATE_TRACE_POINTS
110 #include <trace/events/task.h>
113 * Minimum number of threads to boot the kernel
115 #define MIN_THREADS 20
118 * Maximum number of threads
120 #define MAX_THREADS FUTEX_TID_MASK
123 * Protected counters by write_lock_irq(&tasklist_lock)
125 unsigned long total_forks; /* Handle normal Linux uptimes. */
126 int nr_threads; /* The idle threads do not count.. */
128 static int max_threads; /* tunable limit on nr_threads */
130 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
132 static const char * const resident_page_types[] = {
133 NAMED_ARRAY_INDEX(MM_FILEPAGES),
134 NAMED_ARRAY_INDEX(MM_ANONPAGES),
135 NAMED_ARRAY_INDEX(MM_SWAPENTS),
136 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
139 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
141 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
143 #ifdef CONFIG_PROVE_RCU
144 int lockdep_tasklist_lock_is_held(void)
146 return lockdep_is_held(&tasklist_lock);
148 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
149 #endif /* #ifdef CONFIG_PROVE_RCU */
151 int nr_processes(void)
156 for_each_possible_cpu(cpu)
157 total += per_cpu(process_counts, cpu);
162 void __weak arch_release_task_struct(struct task_struct *tsk)
166 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
167 static struct kmem_cache *task_struct_cachep;
169 static inline struct task_struct *alloc_task_struct_node(int node)
171 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
174 static inline void free_task_struct(struct task_struct *tsk)
176 kmem_cache_free(task_struct_cachep, tsk);
180 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
183 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
184 * kmemcache based allocator.
186 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
188 # ifdef CONFIG_VMAP_STACK
190 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
191 * flush. Try to minimize the number of calls by caching stacks.
193 #define NR_CACHED_STACKS 2
194 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
198 struct vm_struct *stack_vm_area;
201 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
205 for (i = 0; i < NR_CACHED_STACKS; i++) {
206 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
213 static void thread_stack_free_rcu(struct rcu_head *rh)
215 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
217 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
223 static void thread_stack_delayed_free(struct task_struct *tsk)
225 struct vm_stack *vm_stack = tsk->stack;
227 vm_stack->stack_vm_area = tsk->stack_vm_area;
228 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
231 static int free_vm_stack_cache(unsigned int cpu)
233 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
236 for (i = 0; i < NR_CACHED_STACKS; i++) {
237 struct vm_struct *vm_stack = cached_vm_stacks[i];
242 vfree(vm_stack->addr);
243 cached_vm_stacks[i] = NULL;
249 static int memcg_charge_kernel_stack(struct vm_struct *vm)
254 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
255 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
257 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
258 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
265 * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
266 * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
269 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
270 memcg_kmem_uncharge_page(vm->pages[i], 0);
274 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
276 struct vm_struct *vm;
280 for (i = 0; i < NR_CACHED_STACKS; i++) {
283 s = this_cpu_xchg(cached_stacks[i], NULL);
288 /* Reset stack metadata. */
289 kasan_unpoison_range(s->addr, THREAD_SIZE);
291 stack = kasan_reset_tag(s->addr);
293 /* Clear stale pointers from reused stack. */
294 memset(stack, 0, THREAD_SIZE);
296 if (memcg_charge_kernel_stack(s)) {
301 tsk->stack_vm_area = s;
307 * Allocated stacks are cached and later reused by new threads,
308 * so memcg accounting is performed manually on assigning/releasing
309 * stacks to tasks. Drop __GFP_ACCOUNT.
311 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
312 VMALLOC_START, VMALLOC_END,
313 THREADINFO_GFP & ~__GFP_ACCOUNT,
315 0, node, __builtin_return_address(0));
319 vm = find_vm_area(stack);
320 if (memcg_charge_kernel_stack(vm)) {
325 * We can't call find_vm_area() in interrupt context, and
326 * free_thread_stack() can be called in interrupt context,
327 * so cache the vm_struct.
329 tsk->stack_vm_area = vm;
330 stack = kasan_reset_tag(stack);
335 static void free_thread_stack(struct task_struct *tsk)
337 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
338 thread_stack_delayed_free(tsk);
341 tsk->stack_vm_area = NULL;
344 # else /* !CONFIG_VMAP_STACK */
346 static void thread_stack_free_rcu(struct rcu_head *rh)
348 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
351 static void thread_stack_delayed_free(struct task_struct *tsk)
353 struct rcu_head *rh = tsk->stack;
355 call_rcu(rh, thread_stack_free_rcu);
358 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
360 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
364 tsk->stack = kasan_reset_tag(page_address(page));
370 static void free_thread_stack(struct task_struct *tsk)
372 thread_stack_delayed_free(tsk);
376 # endif /* CONFIG_VMAP_STACK */
377 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
379 static struct kmem_cache *thread_stack_cache;
381 static void thread_stack_free_rcu(struct rcu_head *rh)
383 kmem_cache_free(thread_stack_cache, rh);
386 static void thread_stack_delayed_free(struct task_struct *tsk)
388 struct rcu_head *rh = tsk->stack;
390 call_rcu(rh, thread_stack_free_rcu);
393 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
395 unsigned long *stack;
396 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
397 stack = kasan_reset_tag(stack);
399 return stack ? 0 : -ENOMEM;
402 static void free_thread_stack(struct task_struct *tsk)
404 thread_stack_delayed_free(tsk);
408 void thread_stack_cache_init(void)
410 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
411 THREAD_SIZE, THREAD_SIZE, 0, 0,
413 BUG_ON(thread_stack_cache == NULL);
416 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
417 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
419 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
421 unsigned long *stack;
423 stack = arch_alloc_thread_stack_node(tsk, node);
425 return stack ? 0 : -ENOMEM;
428 static void free_thread_stack(struct task_struct *tsk)
430 arch_free_thread_stack(tsk);
434 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
436 /* SLAB cache for signal_struct structures (tsk->signal) */
437 static struct kmem_cache *signal_cachep;
439 /* SLAB cache for sighand_struct structures (tsk->sighand) */
440 struct kmem_cache *sighand_cachep;
442 /* SLAB cache for files_struct structures (tsk->files) */
443 struct kmem_cache *files_cachep;
445 /* SLAB cache for fs_struct structures (tsk->fs) */
446 struct kmem_cache *fs_cachep;
448 /* SLAB cache for vm_area_struct structures */
449 static struct kmem_cache *vm_area_cachep;
451 /* SLAB cache for mm_struct structures (tsk->mm) */
452 static struct kmem_cache *mm_cachep;
454 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
456 struct vm_area_struct *vma;
458 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
464 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
466 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
469 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
470 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
472 * orig->shared.rb may be modified concurrently, but the clone
473 * will be reinitialized.
475 *new = data_race(*orig);
476 INIT_LIST_HEAD(&new->anon_vma_chain);
477 dup_anon_vma_name(orig, new);
482 void vm_area_free(struct vm_area_struct *vma)
484 free_anon_vma_name(vma);
485 kmem_cache_free(vm_area_cachep, vma);
488 static void account_kernel_stack(struct task_struct *tsk, int account)
490 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
491 struct vm_struct *vm = task_stack_vm_area(tsk);
494 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
495 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
496 account * (PAGE_SIZE / 1024));
498 void *stack = task_stack_page(tsk);
500 /* All stack pages are in the same node. */
501 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
502 account * (THREAD_SIZE / 1024));
506 void exit_task_stack_account(struct task_struct *tsk)
508 account_kernel_stack(tsk, -1);
510 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
511 struct vm_struct *vm;
514 vm = task_stack_vm_area(tsk);
515 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
516 memcg_kmem_uncharge_page(vm->pages[i], 0);
520 static void release_task_stack(struct task_struct *tsk)
522 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
523 return; /* Better to leak the stack than to free prematurely */
525 free_thread_stack(tsk);
528 #ifdef CONFIG_THREAD_INFO_IN_TASK
529 void put_task_stack(struct task_struct *tsk)
531 if (refcount_dec_and_test(&tsk->stack_refcount))
532 release_task_stack(tsk);
536 void free_task(struct task_struct *tsk)
538 release_user_cpus_ptr(tsk);
541 #ifndef CONFIG_THREAD_INFO_IN_TASK
543 * The task is finally done with both the stack and thread_info,
546 release_task_stack(tsk);
549 * If the task had a separate stack allocation, it should be gone
552 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
554 rt_mutex_debug_task_free(tsk);
555 ftrace_graph_exit_task(tsk);
556 arch_release_task_struct(tsk);
557 if (tsk->flags & PF_KTHREAD)
558 free_kthread_struct(tsk);
559 free_task_struct(tsk);
561 EXPORT_SYMBOL(free_task);
563 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
565 struct file *exe_file;
567 exe_file = get_mm_exe_file(oldmm);
568 RCU_INIT_POINTER(mm->exe_file, exe_file);
570 * We depend on the oldmm having properly denied write access to the
573 if (exe_file && deny_write_access(exe_file))
574 pr_warn_once("deny_write_access() failed in %s\n", __func__);
578 static __latent_entropy int dup_mmap(struct mm_struct *mm,
579 struct mm_struct *oldmm)
581 struct vm_area_struct *mpnt, *tmp;
583 unsigned long charge = 0;
585 MA_STATE(old_mas, &oldmm->mm_mt, 0, 0);
586 MA_STATE(mas, &mm->mm_mt, 0, 0);
588 uprobe_start_dup_mmap();
589 if (mmap_write_lock_killable(oldmm)) {
591 goto fail_uprobe_end;
593 flush_cache_dup_mm(oldmm);
594 uprobe_dup_mmap(oldmm, mm);
596 * Not linked in yet - no deadlock potential:
598 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
600 /* No ordering required: file already has been exposed. */
601 dup_mm_exe_file(mm, oldmm);
603 mm->total_vm = oldmm->total_vm;
604 mm->data_vm = oldmm->data_vm;
605 mm->exec_vm = oldmm->exec_vm;
606 mm->stack_vm = oldmm->stack_vm;
608 retval = ksm_fork(mm, oldmm);
611 khugepaged_fork(mm, oldmm);
613 retval = mas_expected_entries(&mas, oldmm->map_count);
617 mas_for_each(&old_mas, mpnt, ULONG_MAX) {
620 if (mpnt->vm_flags & VM_DONTCOPY) {
621 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
626 * Don't duplicate many vmas if we've been oom-killed (for
629 if (fatal_signal_pending(current)) {
633 if (mpnt->vm_flags & VM_ACCOUNT) {
634 unsigned long len = vma_pages(mpnt);
636 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
640 tmp = vm_area_dup(mpnt);
643 retval = vma_dup_policy(mpnt, tmp);
645 goto fail_nomem_policy;
647 retval = dup_userfaultfd(tmp, &uf);
649 goto fail_nomem_anon_vma_fork;
650 if (tmp->vm_flags & VM_WIPEONFORK) {
652 * VM_WIPEONFORK gets a clean slate in the child.
653 * Don't prepare anon_vma until fault since we don't
654 * copy page for current vma.
656 tmp->anon_vma = NULL;
657 } else if (anon_vma_fork(tmp, mpnt))
658 goto fail_nomem_anon_vma_fork;
659 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
662 struct address_space *mapping = file->f_mapping;
665 i_mmap_lock_write(mapping);
666 if (tmp->vm_flags & VM_SHARED)
667 mapping_allow_writable(mapping);
668 flush_dcache_mmap_lock(mapping);
669 /* insert tmp into the share list, just after mpnt */
670 vma_interval_tree_insert_after(tmp, mpnt,
672 flush_dcache_mmap_unlock(mapping);
673 i_mmap_unlock_write(mapping);
677 * Copy/update hugetlb private vma information.
679 if (is_vm_hugetlb_page(tmp))
680 hugetlb_dup_vma_private(tmp);
682 /* Link the vma into the MT */
683 mas.index = tmp->vm_start;
684 mas.last = tmp->vm_end - 1;
685 mas_store(&mas, tmp);
686 if (mas_is_err(&mas))
687 goto fail_nomem_mas_store;
690 if (!(tmp->vm_flags & VM_WIPEONFORK))
691 retval = copy_page_range(tmp, mpnt);
693 if (tmp->vm_ops && tmp->vm_ops->open)
694 tmp->vm_ops->open(tmp);
699 /* a new mm has just been created */
700 retval = arch_dup_mmap(oldmm, mm);
704 mmap_write_unlock(mm);
706 mmap_write_unlock(oldmm);
707 dup_userfaultfd_complete(&uf);
709 uprobe_end_dup_mmap();
712 fail_nomem_mas_store:
713 unlink_anon_vmas(tmp);
714 fail_nomem_anon_vma_fork:
715 mpol_put(vma_policy(tmp));
720 vm_unacct_memory(charge);
724 static inline int mm_alloc_pgd(struct mm_struct *mm)
726 mm->pgd = pgd_alloc(mm);
727 if (unlikely(!mm->pgd))
732 static inline void mm_free_pgd(struct mm_struct *mm)
734 pgd_free(mm, mm->pgd);
737 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
739 mmap_write_lock(oldmm);
740 dup_mm_exe_file(mm, oldmm);
741 mmap_write_unlock(oldmm);
744 #define mm_alloc_pgd(mm) (0)
745 #define mm_free_pgd(mm)
746 #endif /* CONFIG_MMU */
748 static void check_mm(struct mm_struct *mm)
752 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
753 "Please make sure 'struct resident_page_types[]' is updated as well");
755 for (i = 0; i < NR_MM_COUNTERS; i++) {
756 long x = atomic_long_read(&mm->rss_stat.count[i]);
759 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
760 mm, resident_page_types[i], x);
763 if (mm_pgtables_bytes(mm))
764 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
765 mm_pgtables_bytes(mm));
767 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
768 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
772 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
773 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
776 * Called when the last reference to the mm
777 * is dropped: either by a lazy thread or by
778 * mmput. Free the page directory and the mm.
780 void __mmdrop(struct mm_struct *mm)
782 BUG_ON(mm == &init_mm);
783 WARN_ON_ONCE(mm == current->mm);
784 WARN_ON_ONCE(mm == current->active_mm);
787 mmu_notifier_subscriptions_destroy(mm);
789 put_user_ns(mm->user_ns);
793 EXPORT_SYMBOL_GPL(__mmdrop);
795 static void mmdrop_async_fn(struct work_struct *work)
797 struct mm_struct *mm;
799 mm = container_of(work, struct mm_struct, async_put_work);
803 static void mmdrop_async(struct mm_struct *mm)
805 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
806 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
807 schedule_work(&mm->async_put_work);
811 static inline void free_signal_struct(struct signal_struct *sig)
813 taskstats_tgid_free(sig);
814 sched_autogroup_exit(sig);
816 * __mmdrop is not safe to call from softirq context on x86 due to
817 * pgd_dtor so postpone it to the async context
820 mmdrop_async(sig->oom_mm);
821 kmem_cache_free(signal_cachep, sig);
824 static inline void put_signal_struct(struct signal_struct *sig)
826 if (refcount_dec_and_test(&sig->sigcnt))
827 free_signal_struct(sig);
830 void __put_task_struct(struct task_struct *tsk)
832 WARN_ON(!tsk->exit_state);
833 WARN_ON(refcount_read(&tsk->usage));
834 WARN_ON(tsk == current);
838 task_numa_free(tsk, true);
839 security_task_free(tsk);
840 bpf_task_storage_free(tsk);
842 delayacct_tsk_free(tsk);
843 put_signal_struct(tsk->signal);
844 sched_core_free(tsk);
847 EXPORT_SYMBOL_GPL(__put_task_struct);
849 void __init __weak arch_task_cache_init(void) { }
854 static void set_max_threads(unsigned int max_threads_suggested)
857 unsigned long nr_pages = totalram_pages();
860 * The number of threads shall be limited such that the thread
861 * structures may only consume a small part of the available memory.
863 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
864 threads = MAX_THREADS;
866 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
867 (u64) THREAD_SIZE * 8UL);
869 if (threads > max_threads_suggested)
870 threads = max_threads_suggested;
872 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
875 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
876 /* Initialized by the architecture: */
877 int arch_task_struct_size __read_mostly;
880 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
881 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
883 /* Fetch thread_struct whitelist for the architecture. */
884 arch_thread_struct_whitelist(offset, size);
887 * Handle zero-sized whitelist or empty thread_struct, otherwise
888 * adjust offset to position of thread_struct in task_struct.
890 if (unlikely(*size == 0))
893 *offset += offsetof(struct task_struct, thread);
895 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
897 void __init fork_init(void)
900 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
901 #ifndef ARCH_MIN_TASKALIGN
902 #define ARCH_MIN_TASKALIGN 0
904 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
905 unsigned long useroffset, usersize;
907 /* create a slab on which task_structs can be allocated */
908 task_struct_whitelist(&useroffset, &usersize);
909 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
910 arch_task_struct_size, align,
911 SLAB_PANIC|SLAB_ACCOUNT,
912 useroffset, usersize, NULL);
915 /* do the arch specific task caches init */
916 arch_task_cache_init();
918 set_max_threads(MAX_THREADS);
920 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
921 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
922 init_task.signal->rlim[RLIMIT_SIGPENDING] =
923 init_task.signal->rlim[RLIMIT_NPROC];
925 for (i = 0; i < MAX_PER_NAMESPACE_UCOUNTS; i++)
926 init_user_ns.ucount_max[i] = max_threads/2;
928 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
929 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
930 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
931 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
933 #ifdef CONFIG_VMAP_STACK
934 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
935 NULL, free_vm_stack_cache);
940 lockdep_init_task(&init_task);
944 int __weak arch_dup_task_struct(struct task_struct *dst,
945 struct task_struct *src)
951 void set_task_stack_end_magic(struct task_struct *tsk)
953 unsigned long *stackend;
955 stackend = end_of_stack(tsk);
956 *stackend = STACK_END_MAGIC; /* for overflow detection */
959 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
961 struct task_struct *tsk;
964 if (node == NUMA_NO_NODE)
965 node = tsk_fork_get_node(orig);
966 tsk = alloc_task_struct_node(node);
970 err = arch_dup_task_struct(tsk, orig);
974 err = alloc_thread_stack_node(tsk, node);
978 #ifdef CONFIG_THREAD_INFO_IN_TASK
979 refcount_set(&tsk->stack_refcount, 1);
981 account_kernel_stack(tsk, 1);
983 err = scs_prepare(tsk, node);
987 #ifdef CONFIG_SECCOMP
989 * We must handle setting up seccomp filters once we're under
990 * the sighand lock in case orig has changed between now and
991 * then. Until then, filter must be NULL to avoid messing up
992 * the usage counts on the error path calling free_task.
994 tsk->seccomp.filter = NULL;
997 setup_thread_stack(tsk, orig);
998 clear_user_return_notifier(tsk);
999 clear_tsk_need_resched(tsk);
1000 set_task_stack_end_magic(tsk);
1001 clear_syscall_work_syscall_user_dispatch(tsk);
1003 #ifdef CONFIG_STACKPROTECTOR
1004 tsk->stack_canary = get_random_canary();
1006 if (orig->cpus_ptr == &orig->cpus_mask)
1007 tsk->cpus_ptr = &tsk->cpus_mask;
1008 dup_user_cpus_ptr(tsk, orig, node);
1011 * One for the user space visible state that goes away when reaped.
1012 * One for the scheduler.
1014 refcount_set(&tsk->rcu_users, 2);
1015 /* One for the rcu users */
1016 refcount_set(&tsk->usage, 1);
1017 #ifdef CONFIG_BLK_DEV_IO_TRACE
1018 tsk->btrace_seq = 0;
1020 tsk->splice_pipe = NULL;
1021 tsk->task_frag.page = NULL;
1022 tsk->wake_q.next = NULL;
1023 tsk->worker_private = NULL;
1025 kcov_task_init(tsk);
1026 kmap_local_fork(tsk);
1028 #ifdef CONFIG_FAULT_INJECTION
1032 #ifdef CONFIG_BLK_CGROUP
1033 tsk->throttle_queue = NULL;
1034 tsk->use_memdelay = 0;
1037 #ifdef CONFIG_IOMMU_SVA
1038 tsk->pasid_activated = 0;
1042 tsk->active_memcg = NULL;
1045 #ifdef CONFIG_CPU_SUP_INTEL
1046 tsk->reported_split_lock = 0;
1052 exit_task_stack_account(tsk);
1053 free_thread_stack(tsk);
1055 free_task_struct(tsk);
1059 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1061 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1063 static int __init coredump_filter_setup(char *s)
1065 default_dump_filter =
1066 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1067 MMF_DUMP_FILTER_MASK;
1071 __setup("coredump_filter=", coredump_filter_setup);
1073 #include <linux/init_task.h>
1075 static void mm_init_aio(struct mm_struct *mm)
1078 spin_lock_init(&mm->ioctx_lock);
1079 mm->ioctx_table = NULL;
1083 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1084 struct task_struct *p)
1088 WRITE_ONCE(mm->owner, NULL);
1092 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1099 static void mm_init_uprobes_state(struct mm_struct *mm)
1101 #ifdef CONFIG_UPROBES
1102 mm->uprobes_state.xol_area = NULL;
1106 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1107 struct user_namespace *user_ns)
1109 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1110 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1111 atomic_set(&mm->mm_users, 1);
1112 atomic_set(&mm->mm_count, 1);
1113 seqcount_init(&mm->write_protect_seq);
1115 INIT_LIST_HEAD(&mm->mmlist);
1116 mm_pgtables_bytes_init(mm);
1119 atomic64_set(&mm->pinned_vm, 0);
1120 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1121 spin_lock_init(&mm->page_table_lock);
1122 spin_lock_init(&mm->arg_lock);
1123 mm_init_cpumask(mm);
1125 mm_init_owner(mm, p);
1127 RCU_INIT_POINTER(mm->exe_file, NULL);
1128 mmu_notifier_subscriptions_init(mm);
1129 init_tlb_flush_pending(mm);
1130 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1131 mm->pmd_huge_pte = NULL;
1133 mm_init_uprobes_state(mm);
1134 hugetlb_count_init(mm);
1137 mm->flags = current->mm->flags & MMF_INIT_MASK;
1138 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1140 mm->flags = default_dump_filter;
1144 if (mm_alloc_pgd(mm))
1147 if (init_new_context(p, mm))
1148 goto fail_nocontext;
1150 mm->user_ns = get_user_ns(user_ns);
1151 lru_gen_init_mm(mm);
1162 * Allocate and initialize an mm_struct.
1164 struct mm_struct *mm_alloc(void)
1166 struct mm_struct *mm;
1172 memset(mm, 0, sizeof(*mm));
1173 return mm_init(mm, current, current_user_ns());
1176 static inline void __mmput(struct mm_struct *mm)
1178 VM_BUG_ON(atomic_read(&mm->mm_users));
1180 uprobe_clear_state(mm);
1183 khugepaged_exit(mm); /* must run before exit_mmap */
1185 mm_put_huge_zero_page(mm);
1186 set_mm_exe_file(mm, NULL);
1187 if (!list_empty(&mm->mmlist)) {
1188 spin_lock(&mmlist_lock);
1189 list_del(&mm->mmlist);
1190 spin_unlock(&mmlist_lock);
1193 module_put(mm->binfmt->module);
1199 * Decrement the use count and release all resources for an mm.
1201 void mmput(struct mm_struct *mm)
1205 if (atomic_dec_and_test(&mm->mm_users))
1208 EXPORT_SYMBOL_GPL(mmput);
1211 static void mmput_async_fn(struct work_struct *work)
1213 struct mm_struct *mm = container_of(work, struct mm_struct,
1219 void mmput_async(struct mm_struct *mm)
1221 if (atomic_dec_and_test(&mm->mm_users)) {
1222 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1223 schedule_work(&mm->async_put_work);
1229 * set_mm_exe_file - change a reference to the mm's executable file
1231 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1233 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1234 * invocations: in mmput() nobody alive left, in execve task is single
1237 * Can only fail if new_exe_file != NULL.
1239 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1241 struct file *old_exe_file;
1244 * It is safe to dereference the exe_file without RCU as
1245 * this function is only called if nobody else can access
1246 * this mm -- see comment above for justification.
1248 old_exe_file = rcu_dereference_raw(mm->exe_file);
1252 * We expect the caller (i.e., sys_execve) to already denied
1253 * write access, so this is unlikely to fail.
1255 if (unlikely(deny_write_access(new_exe_file)))
1257 get_file(new_exe_file);
1259 rcu_assign_pointer(mm->exe_file, new_exe_file);
1261 allow_write_access(old_exe_file);
1268 * replace_mm_exe_file - replace a reference to the mm's executable file
1270 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1271 * dealing with concurrent invocation and without grabbing the mmap lock in
1274 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1276 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1278 struct vm_area_struct *vma;
1279 struct file *old_exe_file;
1282 /* Forbid mm->exe_file change if old file still mapped. */
1283 old_exe_file = get_mm_exe_file(mm);
1285 VMA_ITERATOR(vmi, mm, 0);
1287 for_each_vma(vmi, vma) {
1290 if (path_equal(&vma->vm_file->f_path,
1291 &old_exe_file->f_path)) {
1296 mmap_read_unlock(mm);
1302 /* set the new file, lockless */
1303 ret = deny_write_access(new_exe_file);
1306 get_file(new_exe_file);
1308 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1311 * Don't race with dup_mmap() getting the file and disallowing
1312 * write access while someone might open the file writable.
1315 allow_write_access(old_exe_file);
1317 mmap_read_unlock(mm);
1323 * get_mm_exe_file - acquire a reference to the mm's executable file
1325 * Returns %NULL if mm has no associated executable file.
1326 * User must release file via fput().
1328 struct file *get_mm_exe_file(struct mm_struct *mm)
1330 struct file *exe_file;
1333 exe_file = rcu_dereference(mm->exe_file);
1334 if (exe_file && !get_file_rcu(exe_file))
1341 * get_task_exe_file - acquire a reference to the task's executable file
1343 * Returns %NULL if task's mm (if any) has no associated executable file or
1344 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1345 * User must release file via fput().
1347 struct file *get_task_exe_file(struct task_struct *task)
1349 struct file *exe_file = NULL;
1350 struct mm_struct *mm;
1355 if (!(task->flags & PF_KTHREAD))
1356 exe_file = get_mm_exe_file(mm);
1363 * get_task_mm - acquire a reference to the task's mm
1365 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1366 * this kernel workthread has transiently adopted a user mm with use_mm,
1367 * to do its AIO) is not set and if so returns a reference to it, after
1368 * bumping up the use count. User must release the mm via mmput()
1369 * after use. Typically used by /proc and ptrace.
1371 struct mm_struct *get_task_mm(struct task_struct *task)
1373 struct mm_struct *mm;
1378 if (task->flags & PF_KTHREAD)
1386 EXPORT_SYMBOL_GPL(get_task_mm);
1388 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1390 struct mm_struct *mm;
1393 err = down_read_killable(&task->signal->exec_update_lock);
1395 return ERR_PTR(err);
1397 mm = get_task_mm(task);
1398 if (mm && mm != current->mm &&
1399 !ptrace_may_access(task, mode)) {
1401 mm = ERR_PTR(-EACCES);
1403 up_read(&task->signal->exec_update_lock);
1408 static void complete_vfork_done(struct task_struct *tsk)
1410 struct completion *vfork;
1413 vfork = tsk->vfork_done;
1414 if (likely(vfork)) {
1415 tsk->vfork_done = NULL;
1421 static int wait_for_vfork_done(struct task_struct *child,
1422 struct completion *vfork)
1426 freezer_do_not_count();
1427 cgroup_enter_frozen();
1428 killed = wait_for_completion_killable(vfork);
1429 cgroup_leave_frozen(false);
1434 child->vfork_done = NULL;
1438 put_task_struct(child);
1442 /* Please note the differences between mmput and mm_release.
1443 * mmput is called whenever we stop holding onto a mm_struct,
1444 * error success whatever.
1446 * mm_release is called after a mm_struct has been removed
1447 * from the current process.
1449 * This difference is important for error handling, when we
1450 * only half set up a mm_struct for a new process and need to restore
1451 * the old one. Because we mmput the new mm_struct before
1452 * restoring the old one. . .
1453 * Eric Biederman 10 January 1998
1455 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1457 uprobe_free_utask(tsk);
1459 /* Get rid of any cached register state */
1460 deactivate_mm(tsk, mm);
1463 * Signal userspace if we're not exiting with a core dump
1464 * because we want to leave the value intact for debugging
1467 if (tsk->clear_child_tid) {
1468 if (atomic_read(&mm->mm_users) > 1) {
1470 * We don't check the error code - if userspace has
1471 * not set up a proper pointer then tough luck.
1473 put_user(0, tsk->clear_child_tid);
1474 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1475 1, NULL, NULL, 0, 0);
1477 tsk->clear_child_tid = NULL;
1481 * All done, finally we can wake up parent and return this mm to him.
1482 * Also kthread_stop() uses this completion for synchronization.
1484 if (tsk->vfork_done)
1485 complete_vfork_done(tsk);
1488 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1490 futex_exit_release(tsk);
1491 mm_release(tsk, mm);
1494 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1496 futex_exec_release(tsk);
1497 mm_release(tsk, mm);
1501 * dup_mm() - duplicates an existing mm structure
1502 * @tsk: the task_struct with which the new mm will be associated.
1503 * @oldmm: the mm to duplicate.
1505 * Allocates a new mm structure and duplicates the provided @oldmm structure
1508 * Return: the duplicated mm or NULL on failure.
1510 static struct mm_struct *dup_mm(struct task_struct *tsk,
1511 struct mm_struct *oldmm)
1513 struct mm_struct *mm;
1520 memcpy(mm, oldmm, sizeof(*mm));
1522 if (!mm_init(mm, tsk, mm->user_ns))
1525 err = dup_mmap(mm, oldmm);
1529 mm->hiwater_rss = get_mm_rss(mm);
1530 mm->hiwater_vm = mm->total_vm;
1532 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1538 /* don't put binfmt in mmput, we haven't got module yet */
1540 mm_init_owner(mm, NULL);
1547 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1549 struct mm_struct *mm, *oldmm;
1551 tsk->min_flt = tsk->maj_flt = 0;
1552 tsk->nvcsw = tsk->nivcsw = 0;
1553 #ifdef CONFIG_DETECT_HUNG_TASK
1554 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1555 tsk->last_switch_time = 0;
1559 tsk->active_mm = NULL;
1562 * Are we cloning a kernel thread?
1564 * We need to steal a active VM for that..
1566 oldmm = current->mm;
1570 if (clone_flags & CLONE_VM) {
1574 mm = dup_mm(tsk, current->mm);
1580 tsk->active_mm = mm;
1584 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1586 struct fs_struct *fs = current->fs;
1587 if (clone_flags & CLONE_FS) {
1588 /* tsk->fs is already what we want */
1589 spin_lock(&fs->lock);
1591 spin_unlock(&fs->lock);
1595 spin_unlock(&fs->lock);
1598 tsk->fs = copy_fs_struct(fs);
1604 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1606 struct files_struct *oldf, *newf;
1610 * A background process may not have any files ...
1612 oldf = current->files;
1616 if (clone_flags & CLONE_FILES) {
1617 atomic_inc(&oldf->count);
1621 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1631 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1633 struct sighand_struct *sig;
1635 if (clone_flags & CLONE_SIGHAND) {
1636 refcount_inc(¤t->sighand->count);
1639 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1640 RCU_INIT_POINTER(tsk->sighand, sig);
1644 refcount_set(&sig->count, 1);
1645 spin_lock_irq(¤t->sighand->siglock);
1646 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1647 spin_unlock_irq(¤t->sighand->siglock);
1649 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1650 if (clone_flags & CLONE_CLEAR_SIGHAND)
1651 flush_signal_handlers(tsk, 0);
1656 void __cleanup_sighand(struct sighand_struct *sighand)
1658 if (refcount_dec_and_test(&sighand->count)) {
1659 signalfd_cleanup(sighand);
1661 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1662 * without an RCU grace period, see __lock_task_sighand().
1664 kmem_cache_free(sighand_cachep, sighand);
1669 * Initialize POSIX timer handling for a thread group.
1671 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1673 struct posix_cputimers *pct = &sig->posix_cputimers;
1674 unsigned long cpu_limit;
1676 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1677 posix_cputimers_group_init(pct, cpu_limit);
1680 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1682 struct signal_struct *sig;
1684 if (clone_flags & CLONE_THREAD)
1687 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1692 sig->nr_threads = 1;
1693 atomic_set(&sig->live, 1);
1694 refcount_set(&sig->sigcnt, 1);
1696 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1697 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1698 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1700 init_waitqueue_head(&sig->wait_chldexit);
1701 sig->curr_target = tsk;
1702 init_sigpending(&sig->shared_pending);
1703 INIT_HLIST_HEAD(&sig->multiprocess);
1704 seqlock_init(&sig->stats_lock);
1705 prev_cputime_init(&sig->prev_cputime);
1707 #ifdef CONFIG_POSIX_TIMERS
1708 INIT_LIST_HEAD(&sig->posix_timers);
1709 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1710 sig->real_timer.function = it_real_fn;
1713 task_lock(current->group_leader);
1714 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1715 task_unlock(current->group_leader);
1717 posix_cpu_timers_init_group(sig);
1719 tty_audit_fork(sig);
1720 sched_autogroup_fork(sig);
1722 sig->oom_score_adj = current->signal->oom_score_adj;
1723 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1725 mutex_init(&sig->cred_guard_mutex);
1726 init_rwsem(&sig->exec_update_lock);
1731 static void copy_seccomp(struct task_struct *p)
1733 #ifdef CONFIG_SECCOMP
1735 * Must be called with sighand->lock held, which is common to
1736 * all threads in the group. Holding cred_guard_mutex is not
1737 * needed because this new task is not yet running and cannot
1740 assert_spin_locked(¤t->sighand->siglock);
1742 /* Ref-count the new filter user, and assign it. */
1743 get_seccomp_filter(current);
1744 p->seccomp = current->seccomp;
1747 * Explicitly enable no_new_privs here in case it got set
1748 * between the task_struct being duplicated and holding the
1749 * sighand lock. The seccomp state and nnp must be in sync.
1751 if (task_no_new_privs(current))
1752 task_set_no_new_privs(p);
1755 * If the parent gained a seccomp mode after copying thread
1756 * flags and between before we held the sighand lock, we have
1757 * to manually enable the seccomp thread flag here.
1759 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1760 set_task_syscall_work(p, SECCOMP);
1764 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1766 current->clear_child_tid = tidptr;
1768 return task_pid_vnr(current);
1771 static void rt_mutex_init_task(struct task_struct *p)
1773 raw_spin_lock_init(&p->pi_lock);
1774 #ifdef CONFIG_RT_MUTEXES
1775 p->pi_waiters = RB_ROOT_CACHED;
1776 p->pi_top_task = NULL;
1777 p->pi_blocked_on = NULL;
1781 static inline void init_task_pid_links(struct task_struct *task)
1785 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1786 INIT_HLIST_NODE(&task->pid_links[type]);
1790 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1792 if (type == PIDTYPE_PID)
1793 task->thread_pid = pid;
1795 task->signal->pids[type] = pid;
1798 static inline void rcu_copy_process(struct task_struct *p)
1800 #ifdef CONFIG_PREEMPT_RCU
1801 p->rcu_read_lock_nesting = 0;
1802 p->rcu_read_unlock_special.s = 0;
1803 p->rcu_blocked_node = NULL;
1804 INIT_LIST_HEAD(&p->rcu_node_entry);
1805 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1806 #ifdef CONFIG_TASKS_RCU
1807 p->rcu_tasks_holdout = false;
1808 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1809 p->rcu_tasks_idle_cpu = -1;
1810 #endif /* #ifdef CONFIG_TASKS_RCU */
1811 #ifdef CONFIG_TASKS_TRACE_RCU
1812 p->trc_reader_nesting = 0;
1813 p->trc_reader_special.s = 0;
1814 INIT_LIST_HEAD(&p->trc_holdout_list);
1815 INIT_LIST_HEAD(&p->trc_blkd_node);
1816 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1819 struct pid *pidfd_pid(const struct file *file)
1821 if (file->f_op == &pidfd_fops)
1822 return file->private_data;
1824 return ERR_PTR(-EBADF);
1827 static int pidfd_release(struct inode *inode, struct file *file)
1829 struct pid *pid = file->private_data;
1831 file->private_data = NULL;
1836 #ifdef CONFIG_PROC_FS
1838 * pidfd_show_fdinfo - print information about a pidfd
1839 * @m: proc fdinfo file
1840 * @f: file referencing a pidfd
1843 * This function will print the pid that a given pidfd refers to in the
1844 * pid namespace of the procfs instance.
1845 * If the pid namespace of the process is not a descendant of the pid
1846 * namespace of the procfs instance 0 will be shown as its pid. This is
1847 * similar to calling getppid() on a process whose parent is outside of
1848 * its pid namespace.
1851 * If pid namespaces are supported then this function will also print
1852 * the pid of a given pidfd refers to for all descendant pid namespaces
1853 * starting from the current pid namespace of the instance, i.e. the
1854 * Pid field and the first entry in the NSpid field will be identical.
1855 * If the pid namespace of the process is not a descendant of the pid
1856 * namespace of the procfs instance 0 will be shown as its first NSpid
1857 * entry and no others will be shown.
1858 * Note that this differs from the Pid and NSpid fields in
1859 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1860 * the pid namespace of the procfs instance. The difference becomes
1861 * obvious when sending around a pidfd between pid namespaces from a
1862 * different branch of the tree, i.e. where no ancestral relation is
1863 * present between the pid namespaces:
1864 * - create two new pid namespaces ns1 and ns2 in the initial pid
1865 * namespace (also take care to create new mount namespaces in the
1866 * new pid namespace and mount procfs)
1867 * - create a process with a pidfd in ns1
1868 * - send pidfd from ns1 to ns2
1869 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1870 * have exactly one entry, which is 0
1872 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1874 struct pid *pid = f->private_data;
1875 struct pid_namespace *ns;
1878 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1879 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1880 nr = pid_nr_ns(pid, ns);
1883 seq_put_decimal_ll(m, "Pid:\t", nr);
1885 #ifdef CONFIG_PID_NS
1886 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1890 /* If nr is non-zero it means that 'pid' is valid and that
1891 * ns, i.e. the pid namespace associated with the procfs
1892 * instance, is in the pid namespace hierarchy of pid.
1893 * Start at one below the already printed level.
1895 for (i = ns->level + 1; i <= pid->level; i++)
1896 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1904 * Poll support for process exit notification.
1906 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1908 struct pid *pid = file->private_data;
1909 __poll_t poll_flags = 0;
1911 poll_wait(file, &pid->wait_pidfd, pts);
1914 * Inform pollers only when the whole thread group exits.
1915 * If the thread group leader exits before all other threads in the
1916 * group, then poll(2) should block, similar to the wait(2) family.
1918 if (thread_group_exited(pid))
1919 poll_flags = EPOLLIN | EPOLLRDNORM;
1924 const struct file_operations pidfd_fops = {
1925 .release = pidfd_release,
1927 #ifdef CONFIG_PROC_FS
1928 .show_fdinfo = pidfd_show_fdinfo,
1932 static void __delayed_free_task(struct rcu_head *rhp)
1934 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1939 static __always_inline void delayed_free_task(struct task_struct *tsk)
1941 if (IS_ENABLED(CONFIG_MEMCG))
1942 call_rcu(&tsk->rcu, __delayed_free_task);
1947 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1949 /* Skip if kernel thread */
1953 /* Skip if spawning a thread or using vfork */
1954 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1957 /* We need to synchronize with __set_oom_adj */
1958 mutex_lock(&oom_adj_mutex);
1959 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1960 /* Update the values in case they were changed after copy_signal */
1961 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1962 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1963 mutex_unlock(&oom_adj_mutex);
1967 static void rv_task_fork(struct task_struct *p)
1971 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
1972 p->rv[i].da_mon.monitoring = false;
1975 #define rv_task_fork(p) do {} while (0)
1979 * This creates a new process as a copy of the old one,
1980 * but does not actually start it yet.
1982 * It copies the registers, and all the appropriate
1983 * parts of the process environment (as per the clone
1984 * flags). The actual kick-off is left to the caller.
1986 static __latent_entropy struct task_struct *copy_process(
1990 struct kernel_clone_args *args)
1992 int pidfd = -1, retval;
1993 struct task_struct *p;
1994 struct multiprocess_signals delayed;
1995 struct file *pidfile = NULL;
1996 const u64 clone_flags = args->flags;
1997 struct nsproxy *nsp = current->nsproxy;
2000 * Don't allow sharing the root directory with processes in a different
2003 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2004 return ERR_PTR(-EINVAL);
2006 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2007 return ERR_PTR(-EINVAL);
2010 * Thread groups must share signals as well, and detached threads
2011 * can only be started up within the thread group.
2013 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2014 return ERR_PTR(-EINVAL);
2017 * Shared signal handlers imply shared VM. By way of the above,
2018 * thread groups also imply shared VM. Blocking this case allows
2019 * for various simplifications in other code.
2021 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2022 return ERR_PTR(-EINVAL);
2025 * Siblings of global init remain as zombies on exit since they are
2026 * not reaped by their parent (swapper). To solve this and to avoid
2027 * multi-rooted process trees, prevent global and container-inits
2028 * from creating siblings.
2030 if ((clone_flags & CLONE_PARENT) &&
2031 current->signal->flags & SIGNAL_UNKILLABLE)
2032 return ERR_PTR(-EINVAL);
2035 * If the new process will be in a different pid or user namespace
2036 * do not allow it to share a thread group with the forking task.
2038 if (clone_flags & CLONE_THREAD) {
2039 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2040 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2041 return ERR_PTR(-EINVAL);
2045 * If the new process will be in a different time namespace
2046 * do not allow it to share VM or a thread group with the forking task.
2048 * On vfork, the child process enters the target time namespace only
2051 if ((clone_flags & (CLONE_VM | CLONE_VFORK)) == CLONE_VM) {
2052 if (nsp->time_ns != nsp->time_ns_for_children)
2053 return ERR_PTR(-EINVAL);
2056 if (clone_flags & CLONE_PIDFD) {
2058 * - CLONE_DETACHED is blocked so that we can potentially
2059 * reuse it later for CLONE_PIDFD.
2060 * - CLONE_THREAD is blocked until someone really needs it.
2062 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2063 return ERR_PTR(-EINVAL);
2067 * Force any signals received before this point to be delivered
2068 * before the fork happens. Collect up signals sent to multiple
2069 * processes that happen during the fork and delay them so that
2070 * they appear to happen after the fork.
2072 sigemptyset(&delayed.signal);
2073 INIT_HLIST_NODE(&delayed.node);
2075 spin_lock_irq(¤t->sighand->siglock);
2076 if (!(clone_flags & CLONE_THREAD))
2077 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2078 recalc_sigpending();
2079 spin_unlock_irq(¤t->sighand->siglock);
2080 retval = -ERESTARTNOINTR;
2081 if (task_sigpending(current))
2085 p = dup_task_struct(current, node);
2088 p->flags &= ~PF_KTHREAD;
2090 p->flags |= PF_KTHREAD;
2091 if (args->io_thread) {
2093 * Mark us an IO worker, and block any signal that isn't
2096 p->flags |= PF_IO_WORKER;
2097 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2100 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2102 * Clear TID on mm_release()?
2104 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2106 ftrace_graph_init_task(p);
2108 rt_mutex_init_task(p);
2110 lockdep_assert_irqs_enabled();
2111 #ifdef CONFIG_PROVE_LOCKING
2112 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2114 retval = copy_creds(p, clone_flags);
2119 if (is_ucounts_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2120 if (p->real_cred->user != INIT_USER &&
2121 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2122 goto bad_fork_cleanup_count;
2124 current->flags &= ~PF_NPROC_EXCEEDED;
2127 * If multiple threads are within copy_process(), then this check
2128 * triggers too late. This doesn't hurt, the check is only there
2129 * to stop root fork bombs.
2132 if (data_race(nr_threads >= max_threads))
2133 goto bad_fork_cleanup_count;
2135 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2136 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2137 p->flags |= PF_FORKNOEXEC;
2138 INIT_LIST_HEAD(&p->children);
2139 INIT_LIST_HEAD(&p->sibling);
2140 rcu_copy_process(p);
2141 p->vfork_done = NULL;
2142 spin_lock_init(&p->alloc_lock);
2144 init_sigpending(&p->pending);
2146 p->utime = p->stime = p->gtime = 0;
2147 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2148 p->utimescaled = p->stimescaled = 0;
2150 prev_cputime_init(&p->prev_cputime);
2152 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2153 seqcount_init(&p->vtime.seqcount);
2154 p->vtime.starttime = 0;
2155 p->vtime.state = VTIME_INACTIVE;
2158 #ifdef CONFIG_IO_URING
2162 #if defined(SPLIT_RSS_COUNTING)
2163 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2166 p->default_timer_slack_ns = current->timer_slack_ns;
2172 task_io_accounting_init(&p->ioac);
2173 acct_clear_integrals(p);
2175 posix_cputimers_init(&p->posix_cputimers);
2177 p->io_context = NULL;
2178 audit_set_context(p, NULL);
2180 if (args->kthread) {
2181 if (!set_kthread_struct(p))
2182 goto bad_fork_cleanup_delayacct;
2185 p->mempolicy = mpol_dup(p->mempolicy);
2186 if (IS_ERR(p->mempolicy)) {
2187 retval = PTR_ERR(p->mempolicy);
2188 p->mempolicy = NULL;
2189 goto bad_fork_cleanup_delayacct;
2192 #ifdef CONFIG_CPUSETS
2193 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2194 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2195 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2197 #ifdef CONFIG_TRACE_IRQFLAGS
2198 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2199 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2200 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2201 p->softirqs_enabled = 1;
2202 p->softirq_context = 0;
2205 p->pagefault_disabled = 0;
2207 #ifdef CONFIG_LOCKDEP
2208 lockdep_init_task(p);
2211 #ifdef CONFIG_DEBUG_MUTEXES
2212 p->blocked_on = NULL; /* not blocked yet */
2214 #ifdef CONFIG_BCACHE
2215 p->sequential_io = 0;
2216 p->sequential_io_avg = 0;
2218 #ifdef CONFIG_BPF_SYSCALL
2219 RCU_INIT_POINTER(p->bpf_storage, NULL);
2223 /* Perform scheduler related setup. Assign this task to a CPU. */
2224 retval = sched_fork(clone_flags, p);
2226 goto bad_fork_cleanup_policy;
2228 retval = perf_event_init_task(p, clone_flags);
2230 goto bad_fork_cleanup_policy;
2231 retval = audit_alloc(p);
2233 goto bad_fork_cleanup_perf;
2234 /* copy all the process information */
2236 retval = security_task_alloc(p, clone_flags);
2238 goto bad_fork_cleanup_audit;
2239 retval = copy_semundo(clone_flags, p);
2241 goto bad_fork_cleanup_security;
2242 retval = copy_files(clone_flags, p);
2244 goto bad_fork_cleanup_semundo;
2245 retval = copy_fs(clone_flags, p);
2247 goto bad_fork_cleanup_files;
2248 retval = copy_sighand(clone_flags, p);
2250 goto bad_fork_cleanup_fs;
2251 retval = copy_signal(clone_flags, p);
2253 goto bad_fork_cleanup_sighand;
2254 retval = copy_mm(clone_flags, p);
2256 goto bad_fork_cleanup_signal;
2257 retval = copy_namespaces(clone_flags, p);
2259 goto bad_fork_cleanup_mm;
2260 retval = copy_io(clone_flags, p);
2262 goto bad_fork_cleanup_namespaces;
2263 retval = copy_thread(p, args);
2265 goto bad_fork_cleanup_io;
2267 stackleak_task_init(p);
2269 if (pid != &init_struct_pid) {
2270 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2271 args->set_tid_size);
2273 retval = PTR_ERR(pid);
2274 goto bad_fork_cleanup_thread;
2279 * This has to happen after we've potentially unshared the file
2280 * descriptor table (so that the pidfd doesn't leak into the child
2281 * if the fd table isn't shared).
2283 if (clone_flags & CLONE_PIDFD) {
2284 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2286 goto bad_fork_free_pid;
2290 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2291 O_RDWR | O_CLOEXEC);
2292 if (IS_ERR(pidfile)) {
2293 put_unused_fd(pidfd);
2294 retval = PTR_ERR(pidfile);
2295 goto bad_fork_free_pid;
2297 get_pid(pid); /* held by pidfile now */
2299 retval = put_user(pidfd, args->pidfd);
2301 goto bad_fork_put_pidfd;
2310 * sigaltstack should be cleared when sharing the same VM
2312 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2316 * Syscall tracing and stepping should be turned off in the
2317 * child regardless of CLONE_PTRACE.
2319 user_disable_single_step(p);
2320 clear_task_syscall_work(p, SYSCALL_TRACE);
2321 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2322 clear_task_syscall_work(p, SYSCALL_EMU);
2324 clear_tsk_latency_tracing(p);
2326 /* ok, now we should be set up.. */
2327 p->pid = pid_nr(pid);
2328 if (clone_flags & CLONE_THREAD) {
2329 p->group_leader = current->group_leader;
2330 p->tgid = current->tgid;
2332 p->group_leader = p;
2337 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2338 p->dirty_paused_when = 0;
2340 p->pdeath_signal = 0;
2341 INIT_LIST_HEAD(&p->thread_group);
2342 p->task_works = NULL;
2343 clear_posix_cputimers_work(p);
2345 #ifdef CONFIG_KRETPROBES
2346 p->kretprobe_instances.first = NULL;
2348 #ifdef CONFIG_RETHOOK
2349 p->rethooks.first = NULL;
2353 * Ensure that the cgroup subsystem policies allow the new process to be
2354 * forked. It should be noted that the new process's css_set can be changed
2355 * between here and cgroup_post_fork() if an organisation operation is in
2358 retval = cgroup_can_fork(p, args);
2360 goto bad_fork_put_pidfd;
2363 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2364 * the new task on the correct runqueue. All this *before* the task
2367 * This isn't part of ->can_fork() because while the re-cloning is
2368 * cgroup specific, it unconditionally needs to place the task on a
2371 sched_cgroup_fork(p, args);
2374 * From this point on we must avoid any synchronous user-space
2375 * communication until we take the tasklist-lock. In particular, we do
2376 * not want user-space to be able to predict the process start-time by
2377 * stalling fork(2) after we recorded the start_time but before it is
2378 * visible to the system.
2381 p->start_time = ktime_get_ns();
2382 p->start_boottime = ktime_get_boottime_ns();
2385 * Make it visible to the rest of the system, but dont wake it up yet.
2386 * Need tasklist lock for parent etc handling!
2388 write_lock_irq(&tasklist_lock);
2390 /* CLONE_PARENT re-uses the old parent */
2391 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2392 p->real_parent = current->real_parent;
2393 p->parent_exec_id = current->parent_exec_id;
2394 if (clone_flags & CLONE_THREAD)
2395 p->exit_signal = -1;
2397 p->exit_signal = current->group_leader->exit_signal;
2399 p->real_parent = current;
2400 p->parent_exec_id = current->self_exec_id;
2401 p->exit_signal = args->exit_signal;
2404 klp_copy_process(p);
2408 spin_lock(¤t->sighand->siglock);
2411 * Copy seccomp details explicitly here, in case they were changed
2412 * before holding sighand lock.
2418 rseq_fork(p, clone_flags);
2420 /* Don't start children in a dying pid namespace */
2421 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2423 goto bad_fork_cancel_cgroup;
2426 /* Let kill terminate clone/fork in the middle */
2427 if (fatal_signal_pending(current)) {
2429 goto bad_fork_cancel_cgroup;
2432 init_task_pid_links(p);
2433 if (likely(p->pid)) {
2434 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2436 init_task_pid(p, PIDTYPE_PID, pid);
2437 if (thread_group_leader(p)) {
2438 init_task_pid(p, PIDTYPE_TGID, pid);
2439 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2440 init_task_pid(p, PIDTYPE_SID, task_session(current));
2442 if (is_child_reaper(pid)) {
2443 ns_of_pid(pid)->child_reaper = p;
2444 p->signal->flags |= SIGNAL_UNKILLABLE;
2446 p->signal->shared_pending.signal = delayed.signal;
2447 p->signal->tty = tty_kref_get(current->signal->tty);
2449 * Inherit has_child_subreaper flag under the same
2450 * tasklist_lock with adding child to the process tree
2451 * for propagate_has_child_subreaper optimization.
2453 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2454 p->real_parent->signal->is_child_subreaper;
2455 list_add_tail(&p->sibling, &p->real_parent->children);
2456 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2457 attach_pid(p, PIDTYPE_TGID);
2458 attach_pid(p, PIDTYPE_PGID);
2459 attach_pid(p, PIDTYPE_SID);
2460 __this_cpu_inc(process_counts);
2462 current->signal->nr_threads++;
2463 atomic_inc(¤t->signal->live);
2464 refcount_inc(¤t->signal->sigcnt);
2465 task_join_group_stop(p);
2466 list_add_tail_rcu(&p->thread_group,
2467 &p->group_leader->thread_group);
2468 list_add_tail_rcu(&p->thread_node,
2469 &p->signal->thread_head);
2471 attach_pid(p, PIDTYPE_PID);
2475 hlist_del_init(&delayed.node);
2476 spin_unlock(¤t->sighand->siglock);
2477 syscall_tracepoint_update(p);
2478 write_unlock_irq(&tasklist_lock);
2481 fd_install(pidfd, pidfile);
2483 proc_fork_connector(p);
2485 cgroup_post_fork(p, args);
2488 trace_task_newtask(p, clone_flags);
2489 uprobe_copy_process(p, clone_flags);
2491 copy_oom_score_adj(clone_flags, p);
2495 bad_fork_cancel_cgroup:
2497 spin_unlock(¤t->sighand->siglock);
2498 write_unlock_irq(&tasklist_lock);
2499 cgroup_cancel_fork(p, args);
2501 if (clone_flags & CLONE_PIDFD) {
2503 put_unused_fd(pidfd);
2506 if (pid != &init_struct_pid)
2508 bad_fork_cleanup_thread:
2510 bad_fork_cleanup_io:
2513 bad_fork_cleanup_namespaces:
2514 exit_task_namespaces(p);
2515 bad_fork_cleanup_mm:
2517 mm_clear_owner(p->mm, p);
2520 bad_fork_cleanup_signal:
2521 if (!(clone_flags & CLONE_THREAD))
2522 free_signal_struct(p->signal);
2523 bad_fork_cleanup_sighand:
2524 __cleanup_sighand(p->sighand);
2525 bad_fork_cleanup_fs:
2526 exit_fs(p); /* blocking */
2527 bad_fork_cleanup_files:
2528 exit_files(p); /* blocking */
2529 bad_fork_cleanup_semundo:
2531 bad_fork_cleanup_security:
2532 security_task_free(p);
2533 bad_fork_cleanup_audit:
2535 bad_fork_cleanup_perf:
2536 perf_event_free_task(p);
2537 bad_fork_cleanup_policy:
2538 lockdep_free_task(p);
2540 mpol_put(p->mempolicy);
2542 bad_fork_cleanup_delayacct:
2543 delayacct_tsk_free(p);
2544 bad_fork_cleanup_count:
2545 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2548 WRITE_ONCE(p->__state, TASK_DEAD);
2549 exit_task_stack_account(p);
2551 delayed_free_task(p);
2553 spin_lock_irq(¤t->sighand->siglock);
2554 hlist_del_init(&delayed.node);
2555 spin_unlock_irq(¤t->sighand->siglock);
2556 return ERR_PTR(retval);
2559 static inline void init_idle_pids(struct task_struct *idle)
2563 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2564 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2565 init_task_pid(idle, type, &init_struct_pid);
2569 static int idle_dummy(void *dummy)
2571 /* This function is never called */
2575 struct task_struct * __init fork_idle(int cpu)
2577 struct task_struct *task;
2578 struct kernel_clone_args args = {
2586 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2587 if (!IS_ERR(task)) {
2588 init_idle_pids(task);
2589 init_idle(task, cpu);
2595 struct mm_struct *copy_init_mm(void)
2597 return dup_mm(NULL, &init_mm);
2601 * This is like kernel_clone(), but shaved down and tailored to just
2602 * creating io_uring workers. It returns a created task, or an error pointer.
2603 * The returned task is inactive, and the caller must fire it up through
2604 * wake_up_new_task(p). All signals are blocked in the created task.
2606 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2608 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2610 struct kernel_clone_args args = {
2611 .flags = ((lower_32_bits(flags) | CLONE_VM |
2612 CLONE_UNTRACED) & ~CSIGNAL),
2613 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2619 return copy_process(NULL, 0, node, &args);
2623 * Ok, this is the main fork-routine.
2625 * It copies the process, and if successful kick-starts
2626 * it and waits for it to finish using the VM if required.
2628 * args->exit_signal is expected to be checked for sanity by the caller.
2630 pid_t kernel_clone(struct kernel_clone_args *args)
2632 u64 clone_flags = args->flags;
2633 struct completion vfork;
2635 struct task_struct *p;
2640 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2641 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2642 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2643 * field in struct clone_args and it still doesn't make sense to have
2644 * them both point at the same memory location. Performing this check
2645 * here has the advantage that we don't need to have a separate helper
2646 * to check for legacy clone().
2648 if ((args->flags & CLONE_PIDFD) &&
2649 (args->flags & CLONE_PARENT_SETTID) &&
2650 (args->pidfd == args->parent_tid))
2654 * Determine whether and which event to report to ptracer. When
2655 * called from kernel_thread or CLONE_UNTRACED is explicitly
2656 * requested, no event is reported; otherwise, report if the event
2657 * for the type of forking is enabled.
2659 if (!(clone_flags & CLONE_UNTRACED)) {
2660 if (clone_flags & CLONE_VFORK)
2661 trace = PTRACE_EVENT_VFORK;
2662 else if (args->exit_signal != SIGCHLD)
2663 trace = PTRACE_EVENT_CLONE;
2665 trace = PTRACE_EVENT_FORK;
2667 if (likely(!ptrace_event_enabled(current, trace)))
2671 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2672 add_latent_entropy();
2678 * Do this prior waking up the new thread - the thread pointer
2679 * might get invalid after that point, if the thread exits quickly.
2681 trace_sched_process_fork(current, p);
2683 pid = get_task_pid(p, PIDTYPE_PID);
2686 if (clone_flags & CLONE_PARENT_SETTID)
2687 put_user(nr, args->parent_tid);
2689 if (clone_flags & CLONE_VFORK) {
2690 p->vfork_done = &vfork;
2691 init_completion(&vfork);
2695 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2696 /* lock the task to synchronize with memcg migration */
2698 lru_gen_add_mm(p->mm);
2702 wake_up_new_task(p);
2704 /* forking complete and child started to run, tell ptracer */
2705 if (unlikely(trace))
2706 ptrace_event_pid(trace, pid);
2708 if (clone_flags & CLONE_VFORK) {
2709 if (!wait_for_vfork_done(p, &vfork))
2710 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2718 * Create a kernel thread.
2720 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2722 struct kernel_clone_args args = {
2723 .flags = ((lower_32_bits(flags) | CLONE_VM |
2724 CLONE_UNTRACED) & ~CSIGNAL),
2725 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2731 return kernel_clone(&args);
2735 * Create a user mode thread.
2737 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2739 struct kernel_clone_args args = {
2740 .flags = ((lower_32_bits(flags) | CLONE_VM |
2741 CLONE_UNTRACED) & ~CSIGNAL),
2742 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2747 return kernel_clone(&args);
2750 #ifdef __ARCH_WANT_SYS_FORK
2751 SYSCALL_DEFINE0(fork)
2754 struct kernel_clone_args args = {
2755 .exit_signal = SIGCHLD,
2758 return kernel_clone(&args);
2760 /* can not support in nommu mode */
2766 #ifdef __ARCH_WANT_SYS_VFORK
2767 SYSCALL_DEFINE0(vfork)
2769 struct kernel_clone_args args = {
2770 .flags = CLONE_VFORK | CLONE_VM,
2771 .exit_signal = SIGCHLD,
2774 return kernel_clone(&args);
2778 #ifdef __ARCH_WANT_SYS_CLONE
2779 #ifdef CONFIG_CLONE_BACKWARDS
2780 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2781 int __user *, parent_tidptr,
2783 int __user *, child_tidptr)
2784 #elif defined(CONFIG_CLONE_BACKWARDS2)
2785 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2786 int __user *, parent_tidptr,
2787 int __user *, child_tidptr,
2789 #elif defined(CONFIG_CLONE_BACKWARDS3)
2790 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2792 int __user *, parent_tidptr,
2793 int __user *, child_tidptr,
2796 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2797 int __user *, parent_tidptr,
2798 int __user *, child_tidptr,
2802 struct kernel_clone_args args = {
2803 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2804 .pidfd = parent_tidptr,
2805 .child_tid = child_tidptr,
2806 .parent_tid = parent_tidptr,
2807 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2812 return kernel_clone(&args);
2816 #ifdef __ARCH_WANT_SYS_CLONE3
2818 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2819 struct clone_args __user *uargs,
2823 struct clone_args args;
2824 pid_t *kset_tid = kargs->set_tid;
2826 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2827 CLONE_ARGS_SIZE_VER0);
2828 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2829 CLONE_ARGS_SIZE_VER1);
2830 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2831 CLONE_ARGS_SIZE_VER2);
2832 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2834 if (unlikely(usize > PAGE_SIZE))
2836 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2839 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2843 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2846 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2849 if (unlikely(args.set_tid && args.set_tid_size == 0))
2853 * Verify that higher 32bits of exit_signal are unset and that
2854 * it is a valid signal
2856 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2857 !valid_signal(args.exit_signal)))
2860 if ((args.flags & CLONE_INTO_CGROUP) &&
2861 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2864 *kargs = (struct kernel_clone_args){
2865 .flags = args.flags,
2866 .pidfd = u64_to_user_ptr(args.pidfd),
2867 .child_tid = u64_to_user_ptr(args.child_tid),
2868 .parent_tid = u64_to_user_ptr(args.parent_tid),
2869 .exit_signal = args.exit_signal,
2870 .stack = args.stack,
2871 .stack_size = args.stack_size,
2873 .set_tid_size = args.set_tid_size,
2874 .cgroup = args.cgroup,
2878 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2879 (kargs->set_tid_size * sizeof(pid_t))))
2882 kargs->set_tid = kset_tid;
2888 * clone3_stack_valid - check and prepare stack
2889 * @kargs: kernel clone args
2891 * Verify that the stack arguments userspace gave us are sane.
2892 * In addition, set the stack direction for userspace since it's easy for us to
2895 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2897 if (kargs->stack == 0) {
2898 if (kargs->stack_size > 0)
2901 if (kargs->stack_size == 0)
2904 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2907 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2908 kargs->stack += kargs->stack_size;
2915 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2917 /* Verify that no unknown flags are passed along. */
2919 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2923 * - make the CLONE_DETACHED bit reusable for clone3
2924 * - make the CSIGNAL bits reusable for clone3
2926 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2929 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2930 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2933 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2937 if (!clone3_stack_valid(kargs))
2944 * clone3 - create a new process with specific properties
2945 * @uargs: argument structure
2946 * @size: size of @uargs
2948 * clone3() is the extensible successor to clone()/clone2().
2949 * It takes a struct as argument that is versioned by its size.
2951 * Return: On success, a positive PID for the child process.
2952 * On error, a negative errno number.
2954 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2958 struct kernel_clone_args kargs;
2959 pid_t set_tid[MAX_PID_NS_LEVEL];
2961 kargs.set_tid = set_tid;
2963 err = copy_clone_args_from_user(&kargs, uargs, size);
2967 if (!clone3_args_valid(&kargs))
2970 return kernel_clone(&kargs);
2974 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2976 struct task_struct *leader, *parent, *child;
2979 read_lock(&tasklist_lock);
2980 leader = top = top->group_leader;
2982 for_each_thread(leader, parent) {
2983 list_for_each_entry(child, &parent->children, sibling) {
2984 res = visitor(child, data);
2996 if (leader != top) {
2998 parent = child->real_parent;
2999 leader = parent->group_leader;
3003 read_unlock(&tasklist_lock);
3006 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3007 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3010 static void sighand_ctor(void *data)
3012 struct sighand_struct *sighand = data;
3014 spin_lock_init(&sighand->siglock);
3015 init_waitqueue_head(&sighand->signalfd_wqh);
3018 void __init proc_caches_init(void)
3020 unsigned int mm_size;
3022 sighand_cachep = kmem_cache_create("sighand_cache",
3023 sizeof(struct sighand_struct), 0,
3024 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3025 SLAB_ACCOUNT, sighand_ctor);
3026 signal_cachep = kmem_cache_create("signal_cache",
3027 sizeof(struct signal_struct), 0,
3028 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3030 files_cachep = kmem_cache_create("files_cache",
3031 sizeof(struct files_struct), 0,
3032 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3034 fs_cachep = kmem_cache_create("fs_cache",
3035 sizeof(struct fs_struct), 0,
3036 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3040 * The mm_cpumask is located at the end of mm_struct, and is
3041 * dynamically sized based on the maximum CPU number this system
3042 * can have, taking hotplug into account (nr_cpu_ids).
3044 mm_size = sizeof(struct mm_struct) + cpumask_size();
3046 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3047 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3048 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3049 offsetof(struct mm_struct, saved_auxv),
3050 sizeof_field(struct mm_struct, saved_auxv),
3052 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3054 nsproxy_cache_init();
3058 * Check constraints on flags passed to the unshare system call.
3060 static int check_unshare_flags(unsigned long unshare_flags)
3062 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3063 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3064 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3065 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3069 * Not implemented, but pretend it works if there is nothing
3070 * to unshare. Note that unsharing the address space or the
3071 * signal handlers also need to unshare the signal queues (aka
3074 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3075 if (!thread_group_empty(current))
3078 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3079 if (refcount_read(¤t->sighand->count) > 1)
3082 if (unshare_flags & CLONE_VM) {
3083 if (!current_is_single_threaded())
3091 * Unshare the filesystem structure if it is being shared
3093 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3095 struct fs_struct *fs = current->fs;
3097 if (!(unshare_flags & CLONE_FS) || !fs)
3100 /* don't need lock here; in the worst case we'll do useless copy */
3104 *new_fsp = copy_fs_struct(fs);
3112 * Unshare file descriptor table if it is being shared
3114 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3115 struct files_struct **new_fdp)
3117 struct files_struct *fd = current->files;
3120 if ((unshare_flags & CLONE_FILES) &&
3121 (fd && atomic_read(&fd->count) > 1)) {
3122 *new_fdp = dup_fd(fd, max_fds, &error);
3131 * unshare allows a process to 'unshare' part of the process
3132 * context which was originally shared using clone. copy_*
3133 * functions used by kernel_clone() cannot be used here directly
3134 * because they modify an inactive task_struct that is being
3135 * constructed. Here we are modifying the current, active,
3138 int ksys_unshare(unsigned long unshare_flags)
3140 struct fs_struct *fs, *new_fs = NULL;
3141 struct files_struct *new_fd = NULL;
3142 struct cred *new_cred = NULL;
3143 struct nsproxy *new_nsproxy = NULL;
3148 * If unsharing a user namespace must also unshare the thread group
3149 * and unshare the filesystem root and working directories.
3151 if (unshare_flags & CLONE_NEWUSER)
3152 unshare_flags |= CLONE_THREAD | CLONE_FS;
3154 * If unsharing vm, must also unshare signal handlers.
3156 if (unshare_flags & CLONE_VM)
3157 unshare_flags |= CLONE_SIGHAND;
3159 * If unsharing a signal handlers, must also unshare the signal queues.
3161 if (unshare_flags & CLONE_SIGHAND)
3162 unshare_flags |= CLONE_THREAD;
3164 * If unsharing namespace, must also unshare filesystem information.
3166 if (unshare_flags & CLONE_NEWNS)
3167 unshare_flags |= CLONE_FS;
3169 err = check_unshare_flags(unshare_flags);
3171 goto bad_unshare_out;
3173 * CLONE_NEWIPC must also detach from the undolist: after switching
3174 * to a new ipc namespace, the semaphore arrays from the old
3175 * namespace are unreachable.
3177 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3179 err = unshare_fs(unshare_flags, &new_fs);
3181 goto bad_unshare_out;
3182 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3184 goto bad_unshare_cleanup_fs;
3185 err = unshare_userns(unshare_flags, &new_cred);
3187 goto bad_unshare_cleanup_fd;
3188 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3191 goto bad_unshare_cleanup_cred;
3194 err = set_cred_ucounts(new_cred);
3196 goto bad_unshare_cleanup_cred;
3199 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3202 * CLONE_SYSVSEM is equivalent to sys_exit().
3206 if (unshare_flags & CLONE_NEWIPC) {
3207 /* Orphan segments in old ns (see sem above). */
3209 shm_init_task(current);
3213 switch_task_namespaces(current, new_nsproxy);
3219 spin_lock(&fs->lock);
3220 current->fs = new_fs;
3225 spin_unlock(&fs->lock);
3229 swap(current->files, new_fd);
3231 task_unlock(current);
3234 /* Install the new user namespace */
3235 commit_creds(new_cred);
3240 perf_event_namespaces(current);
3242 bad_unshare_cleanup_cred:
3245 bad_unshare_cleanup_fd:
3247 put_files_struct(new_fd);
3249 bad_unshare_cleanup_fs:
3251 free_fs_struct(new_fs);
3257 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3259 return ksys_unshare(unshare_flags);
3263 * Helper to unshare the files of the current task.
3264 * We don't want to expose copy_files internals to
3265 * the exec layer of the kernel.
3268 int unshare_files(void)
3270 struct task_struct *task = current;
3271 struct files_struct *old, *copy = NULL;
3274 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3282 put_files_struct(old);
3286 int sysctl_max_threads(struct ctl_table *table, int write,
3287 void *buffer, size_t *lenp, loff_t *ppos)
3291 int threads = max_threads;
3293 int max = MAX_THREADS;
3300 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3304 max_threads = threads;