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/vmacache.h>
47 #include <linux/nsproxy.h>
48 #include <linux/capability.h>
49 #include <linux/cpu.h>
50 #include <linux/cgroup.h>
51 #include <linux/security.h>
52 #include <linux/hugetlb.h>
53 #include <linux/seccomp.h>
54 #include <linux/swap.h>
55 #include <linux/syscalls.h>
56 #include <linux/jiffies.h>
57 #include <linux/futex.h>
58 #include <linux/compat.h>
59 #include <linux/kthread.h>
60 #include <linux/task_io_accounting_ops.h>
61 #include <linux/rcupdate.h>
62 #include <linux/ptrace.h>
63 #include <linux/mount.h>
64 #include <linux/audit.h>
65 #include <linux/memcontrol.h>
66 #include <linux/ftrace.h>
67 #include <linux/proc_fs.h>
68 #include <linux/profile.h>
69 #include <linux/rmap.h>
70 #include <linux/ksm.h>
71 #include <linux/acct.h>
72 #include <linux/userfaultfd_k.h>
73 #include <linux/tsacct_kern.h>
74 #include <linux/cn_proc.h>
75 #include <linux/freezer.h>
76 #include <linux/delayacct.h>
77 #include <linux/taskstats_kern.h>
78 #include <linux/random.h>
79 #include <linux/tty.h>
80 #include <linux/fs_struct.h>
81 #include <linux/magic.h>
82 #include <linux/perf_event.h>
83 #include <linux/posix-timers.h>
84 #include <linux/user-return-notifier.h>
85 #include <linux/oom.h>
86 #include <linux/khugepaged.h>
87 #include <linux/signalfd.h>
88 #include <linux/uprobes.h>
89 #include <linux/aio.h>
90 #include <linux/compiler.h>
91 #include <linux/sysctl.h>
92 #include <linux/kcov.h>
93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
97 #include <linux/scs.h>
98 #include <linux/io_uring.h>
99 #include <linux/bpf.h>
100 #include <linux/sched/mm.h>
102 #include <asm/pgalloc.h>
103 #include <linux/uaccess.h>
104 #include <asm/mmu_context.h>
105 #include <asm/cacheflush.h>
106 #include <asm/tlbflush.h>
108 #include <trace/events/sched.h>
110 #define CREATE_TRACE_POINTS
111 #include <trace/events/task.h>
114 * Minimum number of threads to boot the kernel
116 #define MIN_THREADS 20
119 * Maximum number of threads
121 #define MAX_THREADS FUTEX_TID_MASK
124 * Protected counters by write_lock_irq(&tasklist_lock)
126 unsigned long total_forks; /* Handle normal Linux uptimes. */
127 int nr_threads; /* The idle threads do not count.. */
129 static int max_threads; /* tunable limit on nr_threads */
131 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
133 static const char * const resident_page_types[] = {
134 NAMED_ARRAY_INDEX(MM_FILEPAGES),
135 NAMED_ARRAY_INDEX(MM_ANONPAGES),
136 NAMED_ARRAY_INDEX(MM_SWAPENTS),
137 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
140 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
142 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
144 #ifdef CONFIG_PROVE_RCU
145 int lockdep_tasklist_lock_is_held(void)
147 return lockdep_is_held(&tasklist_lock);
149 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
150 #endif /* #ifdef CONFIG_PROVE_RCU */
152 int nr_processes(void)
157 for_each_possible_cpu(cpu)
158 total += per_cpu(process_counts, cpu);
163 void __weak arch_release_task_struct(struct task_struct *tsk)
167 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
168 static struct kmem_cache *task_struct_cachep;
170 static inline struct task_struct *alloc_task_struct_node(int node)
172 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
175 static inline void free_task_struct(struct task_struct *tsk)
177 kmem_cache_free(task_struct_cachep, tsk);
181 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
184 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
185 * kmemcache based allocator.
187 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
189 # ifdef CONFIG_VMAP_STACK
191 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
192 * flush. Try to minimize the number of calls by caching stacks.
194 #define NR_CACHED_STACKS 2
195 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
199 struct vm_struct *stack_vm_area;
202 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
206 for (i = 0; i < NR_CACHED_STACKS; i++) {
207 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
214 static void thread_stack_free_rcu(struct rcu_head *rh)
216 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
218 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
224 static void thread_stack_delayed_free(struct task_struct *tsk)
226 struct vm_stack *vm_stack = tsk->stack;
228 vm_stack->stack_vm_area = tsk->stack_vm_area;
229 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
232 static int free_vm_stack_cache(unsigned int cpu)
234 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
237 for (i = 0; i < NR_CACHED_STACKS; i++) {
238 struct vm_struct *vm_stack = cached_vm_stacks[i];
243 vfree(vm_stack->addr);
244 cached_vm_stacks[i] = NULL;
250 static int memcg_charge_kernel_stack(struct vm_struct *vm)
255 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
256 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
258 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
259 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
266 * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
267 * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
270 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
271 memcg_kmem_uncharge_page(vm->pages[i], 0);
275 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
277 struct vm_struct *vm;
281 for (i = 0; i < NR_CACHED_STACKS; i++) {
284 s = this_cpu_xchg(cached_stacks[i], NULL);
289 /* Reset stack metadata. */
290 kasan_unpoison_range(s->addr, THREAD_SIZE);
292 stack = kasan_reset_tag(s->addr);
294 /* Clear stale pointers from reused stack. */
295 memset(stack, 0, THREAD_SIZE);
297 if (memcg_charge_kernel_stack(s)) {
302 tsk->stack_vm_area = s;
308 * Allocated stacks are cached and later reused by new threads,
309 * so memcg accounting is performed manually on assigning/releasing
310 * stacks to tasks. Drop __GFP_ACCOUNT.
312 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
313 VMALLOC_START, VMALLOC_END,
314 THREADINFO_GFP & ~__GFP_ACCOUNT,
316 0, node, __builtin_return_address(0));
320 vm = find_vm_area(stack);
321 if (memcg_charge_kernel_stack(vm)) {
326 * We can't call find_vm_area() in interrupt context, and
327 * free_thread_stack() can be called in interrupt context,
328 * so cache the vm_struct.
330 tsk->stack_vm_area = vm;
331 stack = kasan_reset_tag(stack);
336 static void free_thread_stack(struct task_struct *tsk)
338 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
339 thread_stack_delayed_free(tsk);
342 tsk->stack_vm_area = NULL;
345 # else /* !CONFIG_VMAP_STACK */
347 static void thread_stack_free_rcu(struct rcu_head *rh)
349 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
352 static void thread_stack_delayed_free(struct task_struct *tsk)
354 struct rcu_head *rh = tsk->stack;
356 call_rcu(rh, thread_stack_free_rcu);
359 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
361 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
365 tsk->stack = kasan_reset_tag(page_address(page));
371 static void free_thread_stack(struct task_struct *tsk)
373 thread_stack_delayed_free(tsk);
377 # endif /* CONFIG_VMAP_STACK */
378 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
380 static struct kmem_cache *thread_stack_cache;
382 static void thread_stack_free_rcu(struct rcu_head *rh)
384 kmem_cache_free(thread_stack_cache, rh);
387 static void thread_stack_delayed_free(struct task_struct *tsk)
389 struct rcu_head *rh = tsk->stack;
391 call_rcu(rh, thread_stack_free_rcu);
394 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
396 unsigned long *stack;
397 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
398 stack = kasan_reset_tag(stack);
400 return stack ? 0 : -ENOMEM;
403 static void free_thread_stack(struct task_struct *tsk)
405 thread_stack_delayed_free(tsk);
409 void thread_stack_cache_init(void)
411 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
412 THREAD_SIZE, THREAD_SIZE, 0, 0,
414 BUG_ON(thread_stack_cache == NULL);
417 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
418 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
420 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
422 unsigned long *stack;
424 stack = arch_alloc_thread_stack_node(tsk, node);
426 return stack ? 0 : -ENOMEM;
429 static void free_thread_stack(struct task_struct *tsk)
431 arch_free_thread_stack(tsk);
435 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
437 /* SLAB cache for signal_struct structures (tsk->signal) */
438 static struct kmem_cache *signal_cachep;
440 /* SLAB cache for sighand_struct structures (tsk->sighand) */
441 struct kmem_cache *sighand_cachep;
443 /* SLAB cache for files_struct structures (tsk->files) */
444 struct kmem_cache *files_cachep;
446 /* SLAB cache for fs_struct structures (tsk->fs) */
447 struct kmem_cache *fs_cachep;
449 /* SLAB cache for vm_area_struct structures */
450 static struct kmem_cache *vm_area_cachep;
452 /* SLAB cache for mm_struct structures (tsk->mm) */
453 static struct kmem_cache *mm_cachep;
455 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
457 struct vm_area_struct *vma;
459 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
465 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
467 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
470 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
471 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
473 * orig->shared.rb may be modified concurrently, but the clone
474 * will be reinitialized.
476 *new = data_race(*orig);
477 INIT_LIST_HEAD(&new->anon_vma_chain);
478 new->vm_next = new->vm_prev = NULL;
479 dup_anon_vma_name(orig, new);
484 void vm_area_free(struct vm_area_struct *vma)
486 free_anon_vma_name(vma);
487 kmem_cache_free(vm_area_cachep, vma);
490 static void account_kernel_stack(struct task_struct *tsk, int account)
492 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
493 struct vm_struct *vm = task_stack_vm_area(tsk);
496 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
497 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
498 account * (PAGE_SIZE / 1024));
500 void *stack = task_stack_page(tsk);
502 /* All stack pages are in the same node. */
503 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
504 account * (THREAD_SIZE / 1024));
508 void exit_task_stack_account(struct task_struct *tsk)
510 account_kernel_stack(tsk, -1);
512 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
513 struct vm_struct *vm;
516 vm = task_stack_vm_area(tsk);
517 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
518 memcg_kmem_uncharge_page(vm->pages[i], 0);
522 static void release_task_stack(struct task_struct *tsk)
524 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
525 return; /* Better to leak the stack than to free prematurely */
527 free_thread_stack(tsk);
530 #ifdef CONFIG_THREAD_INFO_IN_TASK
531 void put_task_stack(struct task_struct *tsk)
533 if (refcount_dec_and_test(&tsk->stack_refcount))
534 release_task_stack(tsk);
538 void free_task(struct task_struct *tsk)
540 release_user_cpus_ptr(tsk);
543 #ifndef CONFIG_THREAD_INFO_IN_TASK
545 * The task is finally done with both the stack and thread_info,
548 release_task_stack(tsk);
551 * If the task had a separate stack allocation, it should be gone
554 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
556 rt_mutex_debug_task_free(tsk);
557 ftrace_graph_exit_task(tsk);
558 arch_release_task_struct(tsk);
559 if (tsk->flags & PF_KTHREAD)
560 free_kthread_struct(tsk);
561 free_task_struct(tsk);
563 EXPORT_SYMBOL(free_task);
565 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
567 struct file *exe_file;
569 exe_file = get_mm_exe_file(oldmm);
570 RCU_INIT_POINTER(mm->exe_file, exe_file);
572 * We depend on the oldmm having properly denied write access to the
575 if (exe_file && deny_write_access(exe_file))
576 pr_warn_once("deny_write_access() failed in %s\n", __func__);
580 static __latent_entropy int dup_mmap(struct mm_struct *mm,
581 struct mm_struct *oldmm)
583 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
584 struct rb_node **rb_link, *rb_parent;
586 unsigned long charge;
589 uprobe_start_dup_mmap();
590 if (mmap_write_lock_killable(oldmm)) {
592 goto fail_uprobe_end;
594 flush_cache_dup_mm(oldmm);
595 uprobe_dup_mmap(oldmm, mm);
597 * Not linked in yet - no deadlock potential:
599 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
601 /* No ordering required: file already has been exposed. */
602 dup_mm_exe_file(mm, oldmm);
604 mm->total_vm = oldmm->total_vm;
605 mm->data_vm = oldmm->data_vm;
606 mm->exec_vm = oldmm->exec_vm;
607 mm->stack_vm = oldmm->stack_vm;
609 rb_link = &mm->mm_rb.rb_node;
612 retval = ksm_fork(mm, oldmm);
615 retval = khugepaged_fork(mm, oldmm);
620 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
623 if (mpnt->vm_flags & VM_DONTCOPY) {
624 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
629 * Don't duplicate many vmas if we've been oom-killed (for
632 if (fatal_signal_pending(current)) {
636 if (mpnt->vm_flags & VM_ACCOUNT) {
637 unsigned long len = vma_pages(mpnt);
639 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
643 tmp = vm_area_dup(mpnt);
646 retval = vma_dup_policy(mpnt, tmp);
648 goto fail_nomem_policy;
650 retval = dup_userfaultfd(tmp, &uf);
652 goto fail_nomem_anon_vma_fork;
653 if (tmp->vm_flags & VM_WIPEONFORK) {
655 * VM_WIPEONFORK gets a clean slate in the child.
656 * Don't prepare anon_vma until fault since we don't
657 * copy page for current vma.
659 tmp->anon_vma = NULL;
660 } else if (anon_vma_fork(tmp, mpnt))
661 goto fail_nomem_anon_vma_fork;
662 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
665 struct address_space *mapping = file->f_mapping;
668 i_mmap_lock_write(mapping);
669 if (tmp->vm_flags & VM_SHARED)
670 mapping_allow_writable(mapping);
671 flush_dcache_mmap_lock(mapping);
672 /* insert tmp into the share list, just after mpnt */
673 vma_interval_tree_insert_after(tmp, mpnt,
675 flush_dcache_mmap_unlock(mapping);
676 i_mmap_unlock_write(mapping);
680 * Clear hugetlb-related page reserves for children. This only
681 * affects MAP_PRIVATE mappings. Faults generated by the child
682 * are not guaranteed to succeed, even if read-only
684 if (is_vm_hugetlb_page(tmp))
685 reset_vma_resv_huge_pages(tmp);
688 * Link in the new vma and copy the page table entries.
691 pprev = &tmp->vm_next;
695 __vma_link_rb(mm, tmp, rb_link, rb_parent);
696 rb_link = &tmp->vm_rb.rb_right;
697 rb_parent = &tmp->vm_rb;
700 if (!(tmp->vm_flags & VM_WIPEONFORK))
701 retval = copy_page_range(tmp, mpnt);
703 if (tmp->vm_ops && tmp->vm_ops->open)
704 tmp->vm_ops->open(tmp);
709 /* a new mm has just been created */
710 retval = arch_dup_mmap(oldmm, mm);
712 mmap_write_unlock(mm);
714 mmap_write_unlock(oldmm);
715 dup_userfaultfd_complete(&uf);
717 uprobe_end_dup_mmap();
719 fail_nomem_anon_vma_fork:
720 mpol_put(vma_policy(tmp));
725 vm_unacct_memory(charge);
729 static inline int mm_alloc_pgd(struct mm_struct *mm)
731 mm->pgd = pgd_alloc(mm);
732 if (unlikely(!mm->pgd))
737 static inline void mm_free_pgd(struct mm_struct *mm)
739 pgd_free(mm, mm->pgd);
742 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
744 mmap_write_lock(oldmm);
745 dup_mm_exe_file(mm, oldmm);
746 mmap_write_unlock(oldmm);
749 #define mm_alloc_pgd(mm) (0)
750 #define mm_free_pgd(mm)
751 #endif /* CONFIG_MMU */
753 static void check_mm(struct mm_struct *mm)
757 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
758 "Please make sure 'struct resident_page_types[]' is updated as well");
760 for (i = 0; i < NR_MM_COUNTERS; i++) {
761 long x = atomic_long_read(&mm->rss_stat.count[i]);
764 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
765 mm, resident_page_types[i], x);
768 if (mm_pgtables_bytes(mm))
769 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
770 mm_pgtables_bytes(mm));
772 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
773 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
777 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
778 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
781 * Called when the last reference to the mm
782 * is dropped: either by a lazy thread or by
783 * mmput. Free the page directory and the mm.
785 void __mmdrop(struct mm_struct *mm)
787 BUG_ON(mm == &init_mm);
788 WARN_ON_ONCE(mm == current->mm);
789 WARN_ON_ONCE(mm == current->active_mm);
792 mmu_notifier_subscriptions_destroy(mm);
794 put_user_ns(mm->user_ns);
797 EXPORT_SYMBOL_GPL(__mmdrop);
799 static void mmdrop_async_fn(struct work_struct *work)
801 struct mm_struct *mm;
803 mm = container_of(work, struct mm_struct, async_put_work);
807 static void mmdrop_async(struct mm_struct *mm)
809 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
810 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
811 schedule_work(&mm->async_put_work);
815 static inline void free_signal_struct(struct signal_struct *sig)
817 taskstats_tgid_free(sig);
818 sched_autogroup_exit(sig);
820 * __mmdrop is not safe to call from softirq context on x86 due to
821 * pgd_dtor so postpone it to the async context
824 mmdrop_async(sig->oom_mm);
825 kmem_cache_free(signal_cachep, sig);
828 static inline void put_signal_struct(struct signal_struct *sig)
830 if (refcount_dec_and_test(&sig->sigcnt))
831 free_signal_struct(sig);
834 void __put_task_struct(struct task_struct *tsk)
836 WARN_ON(!tsk->exit_state);
837 WARN_ON(refcount_read(&tsk->usage));
838 WARN_ON(tsk == current);
842 task_numa_free(tsk, true);
843 security_task_free(tsk);
844 bpf_task_storage_free(tsk);
846 delayacct_tsk_free(tsk);
847 put_signal_struct(tsk->signal);
848 sched_core_free(tsk);
851 EXPORT_SYMBOL_GPL(__put_task_struct);
853 void __init __weak arch_task_cache_init(void) { }
858 static void set_max_threads(unsigned int max_threads_suggested)
861 unsigned long nr_pages = totalram_pages();
864 * The number of threads shall be limited such that the thread
865 * structures may only consume a small part of the available memory.
867 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
868 threads = MAX_THREADS;
870 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
871 (u64) THREAD_SIZE * 8UL);
873 if (threads > max_threads_suggested)
874 threads = max_threads_suggested;
876 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
879 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
880 /* Initialized by the architecture: */
881 int arch_task_struct_size __read_mostly;
884 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
885 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
887 /* Fetch thread_struct whitelist for the architecture. */
888 arch_thread_struct_whitelist(offset, size);
891 * Handle zero-sized whitelist or empty thread_struct, otherwise
892 * adjust offset to position of thread_struct in task_struct.
894 if (unlikely(*size == 0))
897 *offset += offsetof(struct task_struct, thread);
899 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
901 void __init fork_init(void)
904 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
905 #ifndef ARCH_MIN_TASKALIGN
906 #define ARCH_MIN_TASKALIGN 0
908 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
909 unsigned long useroffset, usersize;
911 /* create a slab on which task_structs can be allocated */
912 task_struct_whitelist(&useroffset, &usersize);
913 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
914 arch_task_struct_size, align,
915 SLAB_PANIC|SLAB_ACCOUNT,
916 useroffset, usersize, NULL);
919 /* do the arch specific task caches init */
920 arch_task_cache_init();
922 set_max_threads(MAX_THREADS);
924 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
925 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
926 init_task.signal->rlim[RLIMIT_SIGPENDING] =
927 init_task.signal->rlim[RLIMIT_NPROC];
929 for (i = 0; i < MAX_PER_NAMESPACE_UCOUNTS; i++)
930 init_user_ns.ucount_max[i] = max_threads/2;
932 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
933 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
934 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
935 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
937 #ifdef CONFIG_VMAP_STACK
938 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
939 NULL, free_vm_stack_cache);
944 lockdep_init_task(&init_task);
948 int __weak arch_dup_task_struct(struct task_struct *dst,
949 struct task_struct *src)
955 void set_task_stack_end_magic(struct task_struct *tsk)
957 unsigned long *stackend;
959 stackend = end_of_stack(tsk);
960 *stackend = STACK_END_MAGIC; /* for overflow detection */
963 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
965 struct task_struct *tsk;
968 if (node == NUMA_NO_NODE)
969 node = tsk_fork_get_node(orig);
970 tsk = alloc_task_struct_node(node);
974 err = arch_dup_task_struct(tsk, orig);
978 err = alloc_thread_stack_node(tsk, node);
982 #ifdef CONFIG_THREAD_INFO_IN_TASK
983 refcount_set(&tsk->stack_refcount, 1);
985 account_kernel_stack(tsk, 1);
987 err = scs_prepare(tsk, node);
991 #ifdef CONFIG_SECCOMP
993 * We must handle setting up seccomp filters once we're under
994 * the sighand lock in case orig has changed between now and
995 * then. Until then, filter must be NULL to avoid messing up
996 * the usage counts on the error path calling free_task.
998 tsk->seccomp.filter = NULL;
1001 setup_thread_stack(tsk, orig);
1002 clear_user_return_notifier(tsk);
1003 clear_tsk_need_resched(tsk);
1004 set_task_stack_end_magic(tsk);
1005 clear_syscall_work_syscall_user_dispatch(tsk);
1007 #ifdef CONFIG_STACKPROTECTOR
1008 tsk->stack_canary = get_random_canary();
1010 if (orig->cpus_ptr == &orig->cpus_mask)
1011 tsk->cpus_ptr = &tsk->cpus_mask;
1012 dup_user_cpus_ptr(tsk, orig, node);
1015 * One for the user space visible state that goes away when reaped.
1016 * One for the scheduler.
1018 refcount_set(&tsk->rcu_users, 2);
1019 /* One for the rcu users */
1020 refcount_set(&tsk->usage, 1);
1021 #ifdef CONFIG_BLK_DEV_IO_TRACE
1022 tsk->btrace_seq = 0;
1024 tsk->splice_pipe = NULL;
1025 tsk->task_frag.page = NULL;
1026 tsk->wake_q.next = NULL;
1027 tsk->worker_private = NULL;
1029 kcov_task_init(tsk);
1030 kmap_local_fork(tsk);
1032 #ifdef CONFIG_FAULT_INJECTION
1036 #ifdef CONFIG_BLK_CGROUP
1037 tsk->throttle_queue = NULL;
1038 tsk->use_memdelay = 0;
1041 #ifdef CONFIG_IOMMU_SVA
1042 tsk->pasid_activated = 0;
1046 tsk->active_memcg = NULL;
1051 exit_task_stack_account(tsk);
1052 free_thread_stack(tsk);
1054 free_task_struct(tsk);
1058 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1060 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1062 static int __init coredump_filter_setup(char *s)
1064 default_dump_filter =
1065 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1066 MMF_DUMP_FILTER_MASK;
1070 __setup("coredump_filter=", coredump_filter_setup);
1072 #include <linux/init_task.h>
1074 static void mm_init_aio(struct mm_struct *mm)
1077 spin_lock_init(&mm->ioctx_lock);
1078 mm->ioctx_table = NULL;
1082 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1083 struct task_struct *p)
1087 WRITE_ONCE(mm->owner, NULL);
1091 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1098 static void mm_init_uprobes_state(struct mm_struct *mm)
1100 #ifdef CONFIG_UPROBES
1101 mm->uprobes_state.xol_area = NULL;
1105 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1106 struct user_namespace *user_ns)
1109 mm->mm_rb = RB_ROOT;
1110 mm->vmacache_seqnum = 0;
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);
1161 * Allocate and initialize an mm_struct.
1163 struct mm_struct *mm_alloc(void)
1165 struct mm_struct *mm;
1171 memset(mm, 0, sizeof(*mm));
1172 return mm_init(mm, current, current_user_ns());
1175 static inline void __mmput(struct mm_struct *mm)
1177 VM_BUG_ON(atomic_read(&mm->mm_users));
1179 uprobe_clear_state(mm);
1182 khugepaged_exit(mm); /* must run before exit_mmap */
1184 mm_put_huge_zero_page(mm);
1185 set_mm_exe_file(mm, NULL);
1186 if (!list_empty(&mm->mmlist)) {
1187 spin_lock(&mmlist_lock);
1188 list_del(&mm->mmlist);
1189 spin_unlock(&mmlist_lock);
1192 module_put(mm->binfmt->module);
1198 * Decrement the use count and release all resources for an mm.
1200 void mmput(struct mm_struct *mm)
1204 if (atomic_dec_and_test(&mm->mm_users))
1207 EXPORT_SYMBOL_GPL(mmput);
1210 static void mmput_async_fn(struct work_struct *work)
1212 struct mm_struct *mm = container_of(work, struct mm_struct,
1218 void mmput_async(struct mm_struct *mm)
1220 if (atomic_dec_and_test(&mm->mm_users)) {
1221 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1222 schedule_work(&mm->async_put_work);
1228 * set_mm_exe_file - change a reference to the mm's executable file
1230 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1232 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1233 * invocations: in mmput() nobody alive left, in execve task is single
1236 * Can only fail if new_exe_file != NULL.
1238 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1240 struct file *old_exe_file;
1243 * It is safe to dereference the exe_file without RCU as
1244 * this function is only called if nobody else can access
1245 * this mm -- see comment above for justification.
1247 old_exe_file = rcu_dereference_raw(mm->exe_file);
1251 * We expect the caller (i.e., sys_execve) to already denied
1252 * write access, so this is unlikely to fail.
1254 if (unlikely(deny_write_access(new_exe_file)))
1256 get_file(new_exe_file);
1258 rcu_assign_pointer(mm->exe_file, new_exe_file);
1260 allow_write_access(old_exe_file);
1267 * replace_mm_exe_file - replace a reference to the mm's executable file
1269 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1270 * dealing with concurrent invocation and without grabbing the mmap lock in
1273 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1275 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1277 struct vm_area_struct *vma;
1278 struct file *old_exe_file;
1281 /* Forbid mm->exe_file change if old file still mapped. */
1282 old_exe_file = get_mm_exe_file(mm);
1285 for (vma = mm->mmap; vma && !ret; vma = vma->vm_next) {
1288 if (path_equal(&vma->vm_file->f_path,
1289 &old_exe_file->f_path))
1292 mmap_read_unlock(mm);
1298 /* set the new file, lockless */
1299 ret = deny_write_access(new_exe_file);
1302 get_file(new_exe_file);
1304 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1307 * Don't race with dup_mmap() getting the file and disallowing
1308 * write access while someone might open the file writable.
1311 allow_write_access(old_exe_file);
1313 mmap_read_unlock(mm);
1319 * get_mm_exe_file - acquire a reference to the mm's executable file
1321 * Returns %NULL if mm has no associated executable file.
1322 * User must release file via fput().
1324 struct file *get_mm_exe_file(struct mm_struct *mm)
1326 struct file *exe_file;
1329 exe_file = rcu_dereference(mm->exe_file);
1330 if (exe_file && !get_file_rcu(exe_file))
1337 * get_task_exe_file - acquire a reference to the task's executable file
1339 * Returns %NULL if task's mm (if any) has no associated executable file or
1340 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1341 * User must release file via fput().
1343 struct file *get_task_exe_file(struct task_struct *task)
1345 struct file *exe_file = NULL;
1346 struct mm_struct *mm;
1351 if (!(task->flags & PF_KTHREAD))
1352 exe_file = get_mm_exe_file(mm);
1359 * get_task_mm - acquire a reference to the task's mm
1361 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1362 * this kernel workthread has transiently adopted a user mm with use_mm,
1363 * to do its AIO) is not set and if so returns a reference to it, after
1364 * bumping up the use count. User must release the mm via mmput()
1365 * after use. Typically used by /proc and ptrace.
1367 struct mm_struct *get_task_mm(struct task_struct *task)
1369 struct mm_struct *mm;
1374 if (task->flags & PF_KTHREAD)
1382 EXPORT_SYMBOL_GPL(get_task_mm);
1384 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1386 struct mm_struct *mm;
1389 err = down_read_killable(&task->signal->exec_update_lock);
1391 return ERR_PTR(err);
1393 mm = get_task_mm(task);
1394 if (mm && mm != current->mm &&
1395 !ptrace_may_access(task, mode)) {
1397 mm = ERR_PTR(-EACCES);
1399 up_read(&task->signal->exec_update_lock);
1404 static void complete_vfork_done(struct task_struct *tsk)
1406 struct completion *vfork;
1409 vfork = tsk->vfork_done;
1410 if (likely(vfork)) {
1411 tsk->vfork_done = NULL;
1417 static int wait_for_vfork_done(struct task_struct *child,
1418 struct completion *vfork)
1422 freezer_do_not_count();
1423 cgroup_enter_frozen();
1424 killed = wait_for_completion_killable(vfork);
1425 cgroup_leave_frozen(false);
1430 child->vfork_done = NULL;
1434 put_task_struct(child);
1438 /* Please note the differences between mmput and mm_release.
1439 * mmput is called whenever we stop holding onto a mm_struct,
1440 * error success whatever.
1442 * mm_release is called after a mm_struct has been removed
1443 * from the current process.
1445 * This difference is important for error handling, when we
1446 * only half set up a mm_struct for a new process and need to restore
1447 * the old one. Because we mmput the new mm_struct before
1448 * restoring the old one. . .
1449 * Eric Biederman 10 January 1998
1451 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1453 uprobe_free_utask(tsk);
1455 /* Get rid of any cached register state */
1456 deactivate_mm(tsk, mm);
1459 * Signal userspace if we're not exiting with a core dump
1460 * because we want to leave the value intact for debugging
1463 if (tsk->clear_child_tid) {
1464 if (atomic_read(&mm->mm_users) > 1) {
1466 * We don't check the error code - if userspace has
1467 * not set up a proper pointer then tough luck.
1469 put_user(0, tsk->clear_child_tid);
1470 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1471 1, NULL, NULL, 0, 0);
1473 tsk->clear_child_tid = NULL;
1477 * All done, finally we can wake up parent and return this mm to him.
1478 * Also kthread_stop() uses this completion for synchronization.
1480 if (tsk->vfork_done)
1481 complete_vfork_done(tsk);
1484 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1486 futex_exit_release(tsk);
1487 mm_release(tsk, mm);
1490 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1492 futex_exec_release(tsk);
1493 mm_release(tsk, mm);
1497 * dup_mm() - duplicates an existing mm structure
1498 * @tsk: the task_struct with which the new mm will be associated.
1499 * @oldmm: the mm to duplicate.
1501 * Allocates a new mm structure and duplicates the provided @oldmm structure
1504 * Return: the duplicated mm or NULL on failure.
1506 static struct mm_struct *dup_mm(struct task_struct *tsk,
1507 struct mm_struct *oldmm)
1509 struct mm_struct *mm;
1516 memcpy(mm, oldmm, sizeof(*mm));
1518 if (!mm_init(mm, tsk, mm->user_ns))
1521 err = dup_mmap(mm, oldmm);
1525 mm->hiwater_rss = get_mm_rss(mm);
1526 mm->hiwater_vm = mm->total_vm;
1528 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1534 /* don't put binfmt in mmput, we haven't got module yet */
1536 mm_init_owner(mm, NULL);
1543 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1545 struct mm_struct *mm, *oldmm;
1547 tsk->min_flt = tsk->maj_flt = 0;
1548 tsk->nvcsw = tsk->nivcsw = 0;
1549 #ifdef CONFIG_DETECT_HUNG_TASK
1550 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1551 tsk->last_switch_time = 0;
1555 tsk->active_mm = NULL;
1558 * Are we cloning a kernel thread?
1560 * We need to steal a active VM for that..
1562 oldmm = current->mm;
1566 /* initialize the new vmacache entries */
1567 vmacache_flush(tsk);
1569 if (clone_flags & CLONE_VM) {
1573 mm = dup_mm(tsk, current->mm);
1579 tsk->active_mm = mm;
1583 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1585 struct fs_struct *fs = current->fs;
1586 if (clone_flags & CLONE_FS) {
1587 /* tsk->fs is already what we want */
1588 spin_lock(&fs->lock);
1590 spin_unlock(&fs->lock);
1594 spin_unlock(&fs->lock);
1597 tsk->fs = copy_fs_struct(fs);
1603 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1605 struct files_struct *oldf, *newf;
1609 * A background process may not have any files ...
1611 oldf = current->files;
1615 if (clone_flags & CLONE_FILES) {
1616 atomic_inc(&oldf->count);
1620 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1630 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1632 struct sighand_struct *sig;
1634 if (clone_flags & CLONE_SIGHAND) {
1635 refcount_inc(¤t->sighand->count);
1638 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1639 RCU_INIT_POINTER(tsk->sighand, sig);
1643 refcount_set(&sig->count, 1);
1644 spin_lock_irq(¤t->sighand->siglock);
1645 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1646 spin_unlock_irq(¤t->sighand->siglock);
1648 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1649 if (clone_flags & CLONE_CLEAR_SIGHAND)
1650 flush_signal_handlers(tsk, 0);
1655 void __cleanup_sighand(struct sighand_struct *sighand)
1657 if (refcount_dec_and_test(&sighand->count)) {
1658 signalfd_cleanup(sighand);
1660 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1661 * without an RCU grace period, see __lock_task_sighand().
1663 kmem_cache_free(sighand_cachep, sighand);
1668 * Initialize POSIX timer handling for a thread group.
1670 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1672 struct posix_cputimers *pct = &sig->posix_cputimers;
1673 unsigned long cpu_limit;
1675 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1676 posix_cputimers_group_init(pct, cpu_limit);
1679 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1681 struct signal_struct *sig;
1683 if (clone_flags & CLONE_THREAD)
1686 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1691 sig->nr_threads = 1;
1692 atomic_set(&sig->live, 1);
1693 refcount_set(&sig->sigcnt, 1);
1695 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1696 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1697 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1699 init_waitqueue_head(&sig->wait_chldexit);
1700 sig->curr_target = tsk;
1701 init_sigpending(&sig->shared_pending);
1702 INIT_HLIST_HEAD(&sig->multiprocess);
1703 seqlock_init(&sig->stats_lock);
1704 prev_cputime_init(&sig->prev_cputime);
1706 #ifdef CONFIG_POSIX_TIMERS
1707 INIT_LIST_HEAD(&sig->posix_timers);
1708 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1709 sig->real_timer.function = it_real_fn;
1712 task_lock(current->group_leader);
1713 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1714 task_unlock(current->group_leader);
1716 posix_cpu_timers_init_group(sig);
1718 tty_audit_fork(sig);
1719 sched_autogroup_fork(sig);
1721 sig->oom_score_adj = current->signal->oom_score_adj;
1722 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1724 mutex_init(&sig->cred_guard_mutex);
1725 init_rwsem(&sig->exec_update_lock);
1730 static void copy_seccomp(struct task_struct *p)
1732 #ifdef CONFIG_SECCOMP
1734 * Must be called with sighand->lock held, which is common to
1735 * all threads in the group. Holding cred_guard_mutex is not
1736 * needed because this new task is not yet running and cannot
1739 assert_spin_locked(¤t->sighand->siglock);
1741 /* Ref-count the new filter user, and assign it. */
1742 get_seccomp_filter(current);
1743 p->seccomp = current->seccomp;
1746 * Explicitly enable no_new_privs here in case it got set
1747 * between the task_struct being duplicated and holding the
1748 * sighand lock. The seccomp state and nnp must be in sync.
1750 if (task_no_new_privs(current))
1751 task_set_no_new_privs(p);
1754 * If the parent gained a seccomp mode after copying thread
1755 * flags and between before we held the sighand lock, we have
1756 * to manually enable the seccomp thread flag here.
1758 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1759 set_task_syscall_work(p, SECCOMP);
1763 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1765 current->clear_child_tid = tidptr;
1767 return task_pid_vnr(current);
1770 static void rt_mutex_init_task(struct task_struct *p)
1772 raw_spin_lock_init(&p->pi_lock);
1773 #ifdef CONFIG_RT_MUTEXES
1774 p->pi_waiters = RB_ROOT_CACHED;
1775 p->pi_top_task = NULL;
1776 p->pi_blocked_on = NULL;
1780 static inline void init_task_pid_links(struct task_struct *task)
1784 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1785 INIT_HLIST_NODE(&task->pid_links[type]);
1789 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1791 if (type == PIDTYPE_PID)
1792 task->thread_pid = pid;
1794 task->signal->pids[type] = pid;
1797 static inline void rcu_copy_process(struct task_struct *p)
1799 #ifdef CONFIG_PREEMPT_RCU
1800 p->rcu_read_lock_nesting = 0;
1801 p->rcu_read_unlock_special.s = 0;
1802 p->rcu_blocked_node = NULL;
1803 INIT_LIST_HEAD(&p->rcu_node_entry);
1804 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1805 #ifdef CONFIG_TASKS_RCU
1806 p->rcu_tasks_holdout = false;
1807 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1808 p->rcu_tasks_idle_cpu = -1;
1809 #endif /* #ifdef CONFIG_TASKS_RCU */
1810 #ifdef CONFIG_TASKS_TRACE_RCU
1811 p->trc_reader_nesting = 0;
1812 p->trc_reader_special.s = 0;
1813 INIT_LIST_HEAD(&p->trc_holdout_list);
1814 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1817 struct pid *pidfd_pid(const struct file *file)
1819 if (file->f_op == &pidfd_fops)
1820 return file->private_data;
1822 return ERR_PTR(-EBADF);
1825 static int pidfd_release(struct inode *inode, struct file *file)
1827 struct pid *pid = file->private_data;
1829 file->private_data = NULL;
1834 #ifdef CONFIG_PROC_FS
1836 * pidfd_show_fdinfo - print information about a pidfd
1837 * @m: proc fdinfo file
1838 * @f: file referencing a pidfd
1841 * This function will print the pid that a given pidfd refers to in the
1842 * pid namespace of the procfs instance.
1843 * If the pid namespace of the process is not a descendant of the pid
1844 * namespace of the procfs instance 0 will be shown as its pid. This is
1845 * similar to calling getppid() on a process whose parent is outside of
1846 * its pid namespace.
1849 * If pid namespaces are supported then this function will also print
1850 * the pid of a given pidfd refers to for all descendant pid namespaces
1851 * starting from the current pid namespace of the instance, i.e. the
1852 * Pid field and the first entry in the NSpid field will be identical.
1853 * If the pid namespace of the process is not a descendant of the pid
1854 * namespace of the procfs instance 0 will be shown as its first NSpid
1855 * entry and no others will be shown.
1856 * Note that this differs from the Pid and NSpid fields in
1857 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1858 * the pid namespace of the procfs instance. The difference becomes
1859 * obvious when sending around a pidfd between pid namespaces from a
1860 * different branch of the tree, i.e. where no ancestral relation is
1861 * present between the pid namespaces:
1862 * - create two new pid namespaces ns1 and ns2 in the initial pid
1863 * namespace (also take care to create new mount namespaces in the
1864 * new pid namespace and mount procfs)
1865 * - create a process with a pidfd in ns1
1866 * - send pidfd from ns1 to ns2
1867 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1868 * have exactly one entry, which is 0
1870 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1872 struct pid *pid = f->private_data;
1873 struct pid_namespace *ns;
1876 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1877 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1878 nr = pid_nr_ns(pid, ns);
1881 seq_put_decimal_ll(m, "Pid:\t", nr);
1883 #ifdef CONFIG_PID_NS
1884 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1888 /* If nr is non-zero it means that 'pid' is valid and that
1889 * ns, i.e. the pid namespace associated with the procfs
1890 * instance, is in the pid namespace hierarchy of pid.
1891 * Start at one below the already printed level.
1893 for (i = ns->level + 1; i <= pid->level; i++)
1894 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1902 * Poll support for process exit notification.
1904 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1906 struct pid *pid = file->private_data;
1907 __poll_t poll_flags = 0;
1909 poll_wait(file, &pid->wait_pidfd, pts);
1912 * Inform pollers only when the whole thread group exits.
1913 * If the thread group leader exits before all other threads in the
1914 * group, then poll(2) should block, similar to the wait(2) family.
1916 if (thread_group_exited(pid))
1917 poll_flags = EPOLLIN | EPOLLRDNORM;
1922 const struct file_operations pidfd_fops = {
1923 .release = pidfd_release,
1925 #ifdef CONFIG_PROC_FS
1926 .show_fdinfo = pidfd_show_fdinfo,
1930 static void __delayed_free_task(struct rcu_head *rhp)
1932 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1937 static __always_inline void delayed_free_task(struct task_struct *tsk)
1939 if (IS_ENABLED(CONFIG_MEMCG))
1940 call_rcu(&tsk->rcu, __delayed_free_task);
1945 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1947 /* Skip if kernel thread */
1951 /* Skip if spawning a thread or using vfork */
1952 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1955 /* We need to synchronize with __set_oom_adj */
1956 mutex_lock(&oom_adj_mutex);
1957 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1958 /* Update the values in case they were changed after copy_signal */
1959 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1960 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1961 mutex_unlock(&oom_adj_mutex);
1965 * This creates a new process as a copy of the old one,
1966 * but does not actually start it yet.
1968 * It copies the registers, and all the appropriate
1969 * parts of the process environment (as per the clone
1970 * flags). The actual kick-off is left to the caller.
1972 static __latent_entropy struct task_struct *copy_process(
1976 struct kernel_clone_args *args)
1978 int pidfd = -1, retval;
1979 struct task_struct *p;
1980 struct multiprocess_signals delayed;
1981 struct file *pidfile = NULL;
1982 u64 clone_flags = args->flags;
1983 struct nsproxy *nsp = current->nsproxy;
1986 * Don't allow sharing the root directory with processes in a different
1989 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1990 return ERR_PTR(-EINVAL);
1992 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1993 return ERR_PTR(-EINVAL);
1996 * Thread groups must share signals as well, and detached threads
1997 * can only be started up within the thread group.
1999 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2000 return ERR_PTR(-EINVAL);
2003 * Shared signal handlers imply shared VM. By way of the above,
2004 * thread groups also imply shared VM. Blocking this case allows
2005 * for various simplifications in other code.
2007 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2008 return ERR_PTR(-EINVAL);
2011 * Siblings of global init remain as zombies on exit since they are
2012 * not reaped by their parent (swapper). To solve this and to avoid
2013 * multi-rooted process trees, prevent global and container-inits
2014 * from creating siblings.
2016 if ((clone_flags & CLONE_PARENT) &&
2017 current->signal->flags & SIGNAL_UNKILLABLE)
2018 return ERR_PTR(-EINVAL);
2021 * If the new process will be in a different pid or user namespace
2022 * do not allow it to share a thread group with the forking task.
2024 if (clone_flags & CLONE_THREAD) {
2025 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2026 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2027 return ERR_PTR(-EINVAL);
2031 * If the new process will be in a different time namespace
2032 * do not allow it to share VM or a thread group with the forking task.
2034 if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
2035 if (nsp->time_ns != nsp->time_ns_for_children)
2036 return ERR_PTR(-EINVAL);
2039 if (clone_flags & CLONE_PIDFD) {
2041 * - CLONE_DETACHED is blocked so that we can potentially
2042 * reuse it later for CLONE_PIDFD.
2043 * - CLONE_THREAD is blocked until someone really needs it.
2045 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2046 return ERR_PTR(-EINVAL);
2050 * Force any signals received before this point to be delivered
2051 * before the fork happens. Collect up signals sent to multiple
2052 * processes that happen during the fork and delay them so that
2053 * they appear to happen after the fork.
2055 sigemptyset(&delayed.signal);
2056 INIT_HLIST_NODE(&delayed.node);
2058 spin_lock_irq(¤t->sighand->siglock);
2059 if (!(clone_flags & CLONE_THREAD))
2060 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2061 recalc_sigpending();
2062 spin_unlock_irq(¤t->sighand->siglock);
2063 retval = -ERESTARTNOINTR;
2064 if (task_sigpending(current))
2068 p = dup_task_struct(current, node);
2071 if (args->io_thread) {
2073 * Mark us an IO worker, and block any signal that isn't
2076 p->flags |= PF_IO_WORKER;
2077 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2080 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2082 * Clear TID on mm_release()?
2084 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2086 ftrace_graph_init_task(p);
2088 rt_mutex_init_task(p);
2090 lockdep_assert_irqs_enabled();
2091 #ifdef CONFIG_PROVE_LOCKING
2092 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2094 retval = copy_creds(p, clone_flags);
2099 if (is_ucounts_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2100 if (p->real_cred->user != INIT_USER &&
2101 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2102 goto bad_fork_cleanup_count;
2104 current->flags &= ~PF_NPROC_EXCEEDED;
2107 * If multiple threads are within copy_process(), then this check
2108 * triggers too late. This doesn't hurt, the check is only there
2109 * to stop root fork bombs.
2112 if (data_race(nr_threads >= max_threads))
2113 goto bad_fork_cleanup_count;
2115 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2116 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2117 p->flags |= PF_FORKNOEXEC;
2118 INIT_LIST_HEAD(&p->children);
2119 INIT_LIST_HEAD(&p->sibling);
2120 rcu_copy_process(p);
2121 p->vfork_done = NULL;
2122 spin_lock_init(&p->alloc_lock);
2124 init_sigpending(&p->pending);
2126 p->utime = p->stime = p->gtime = 0;
2127 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2128 p->utimescaled = p->stimescaled = 0;
2130 prev_cputime_init(&p->prev_cputime);
2132 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2133 seqcount_init(&p->vtime.seqcount);
2134 p->vtime.starttime = 0;
2135 p->vtime.state = VTIME_INACTIVE;
2138 #ifdef CONFIG_IO_URING
2142 #if defined(SPLIT_RSS_COUNTING)
2143 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2146 p->default_timer_slack_ns = current->timer_slack_ns;
2152 task_io_accounting_init(&p->ioac);
2153 acct_clear_integrals(p);
2155 posix_cputimers_init(&p->posix_cputimers);
2157 p->io_context = NULL;
2158 audit_set_context(p, NULL);
2160 if (p->flags & PF_KTHREAD) {
2161 if (!set_kthread_struct(p))
2162 goto bad_fork_cleanup_delayacct;
2165 p->mempolicy = mpol_dup(p->mempolicy);
2166 if (IS_ERR(p->mempolicy)) {
2167 retval = PTR_ERR(p->mempolicy);
2168 p->mempolicy = NULL;
2169 goto bad_fork_cleanup_delayacct;
2172 #ifdef CONFIG_CPUSETS
2173 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2174 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2175 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2177 #ifdef CONFIG_TRACE_IRQFLAGS
2178 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2179 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2180 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2181 p->softirqs_enabled = 1;
2182 p->softirq_context = 0;
2185 p->pagefault_disabled = 0;
2187 #ifdef CONFIG_LOCKDEP
2188 lockdep_init_task(p);
2191 #ifdef CONFIG_DEBUG_MUTEXES
2192 p->blocked_on = NULL; /* not blocked yet */
2194 #ifdef CONFIG_BCACHE
2195 p->sequential_io = 0;
2196 p->sequential_io_avg = 0;
2198 #ifdef CONFIG_BPF_SYSCALL
2199 RCU_INIT_POINTER(p->bpf_storage, NULL);
2203 /* Perform scheduler related setup. Assign this task to a CPU. */
2204 retval = sched_fork(clone_flags, p);
2206 goto bad_fork_cleanup_policy;
2208 retval = perf_event_init_task(p, clone_flags);
2210 goto bad_fork_cleanup_policy;
2211 retval = audit_alloc(p);
2213 goto bad_fork_cleanup_perf;
2214 /* copy all the process information */
2216 retval = security_task_alloc(p, clone_flags);
2218 goto bad_fork_cleanup_audit;
2219 retval = copy_semundo(clone_flags, p);
2221 goto bad_fork_cleanup_security;
2222 retval = copy_files(clone_flags, p);
2224 goto bad_fork_cleanup_semundo;
2225 retval = copy_fs(clone_flags, p);
2227 goto bad_fork_cleanup_files;
2228 retval = copy_sighand(clone_flags, p);
2230 goto bad_fork_cleanup_fs;
2231 retval = copy_signal(clone_flags, p);
2233 goto bad_fork_cleanup_sighand;
2234 retval = copy_mm(clone_flags, p);
2236 goto bad_fork_cleanup_signal;
2237 retval = copy_namespaces(clone_flags, p);
2239 goto bad_fork_cleanup_mm;
2240 retval = copy_io(clone_flags, p);
2242 goto bad_fork_cleanup_namespaces;
2243 retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2245 goto bad_fork_cleanup_io;
2247 stackleak_task_init(p);
2249 if (pid != &init_struct_pid) {
2250 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2251 args->set_tid_size);
2253 retval = PTR_ERR(pid);
2254 goto bad_fork_cleanup_thread;
2259 * This has to happen after we've potentially unshared the file
2260 * descriptor table (so that the pidfd doesn't leak into the child
2261 * if the fd table isn't shared).
2263 if (clone_flags & CLONE_PIDFD) {
2264 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2266 goto bad_fork_free_pid;
2270 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2271 O_RDWR | O_CLOEXEC);
2272 if (IS_ERR(pidfile)) {
2273 put_unused_fd(pidfd);
2274 retval = PTR_ERR(pidfile);
2275 goto bad_fork_free_pid;
2277 get_pid(pid); /* held by pidfile now */
2279 retval = put_user(pidfd, args->pidfd);
2281 goto bad_fork_put_pidfd;
2290 * sigaltstack should be cleared when sharing the same VM
2292 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2296 * Syscall tracing and stepping should be turned off in the
2297 * child regardless of CLONE_PTRACE.
2299 user_disable_single_step(p);
2300 clear_task_syscall_work(p, SYSCALL_TRACE);
2301 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2302 clear_task_syscall_work(p, SYSCALL_EMU);
2304 clear_tsk_latency_tracing(p);
2306 /* ok, now we should be set up.. */
2307 p->pid = pid_nr(pid);
2308 if (clone_flags & CLONE_THREAD) {
2309 p->group_leader = current->group_leader;
2310 p->tgid = current->tgid;
2312 p->group_leader = p;
2317 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2318 p->dirty_paused_when = 0;
2320 p->pdeath_signal = 0;
2321 INIT_LIST_HEAD(&p->thread_group);
2322 p->task_works = NULL;
2323 clear_posix_cputimers_work(p);
2325 #ifdef CONFIG_KRETPROBES
2326 p->kretprobe_instances.first = NULL;
2328 #ifdef CONFIG_RETHOOK
2329 p->rethooks.first = NULL;
2333 * Ensure that the cgroup subsystem policies allow the new process to be
2334 * forked. It should be noted that the new process's css_set can be changed
2335 * between here and cgroup_post_fork() if an organisation operation is in
2338 retval = cgroup_can_fork(p, args);
2340 goto bad_fork_put_pidfd;
2343 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2344 * the new task on the correct runqueue. All this *before* the task
2347 * This isn't part of ->can_fork() because while the re-cloning is
2348 * cgroup specific, it unconditionally needs to place the task on a
2351 sched_cgroup_fork(p, args);
2354 * From this point on we must avoid any synchronous user-space
2355 * communication until we take the tasklist-lock. In particular, we do
2356 * not want user-space to be able to predict the process start-time by
2357 * stalling fork(2) after we recorded the start_time but before it is
2358 * visible to the system.
2361 p->start_time = ktime_get_ns();
2362 p->start_boottime = ktime_get_boottime_ns();
2365 * Make it visible to the rest of the system, but dont wake it up yet.
2366 * Need tasklist lock for parent etc handling!
2368 write_lock_irq(&tasklist_lock);
2370 /* CLONE_PARENT re-uses the old parent */
2371 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2372 p->real_parent = current->real_parent;
2373 p->parent_exec_id = current->parent_exec_id;
2374 if (clone_flags & CLONE_THREAD)
2375 p->exit_signal = -1;
2377 p->exit_signal = current->group_leader->exit_signal;
2379 p->real_parent = current;
2380 p->parent_exec_id = current->self_exec_id;
2381 p->exit_signal = args->exit_signal;
2384 klp_copy_process(p);
2388 spin_lock(¤t->sighand->siglock);
2391 * Copy seccomp details explicitly here, in case they were changed
2392 * before holding sighand lock.
2396 rseq_fork(p, clone_flags);
2398 /* Don't start children in a dying pid namespace */
2399 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2401 goto bad_fork_cancel_cgroup;
2404 /* Let kill terminate clone/fork in the middle */
2405 if (fatal_signal_pending(current)) {
2407 goto bad_fork_cancel_cgroup;
2410 init_task_pid_links(p);
2411 if (likely(p->pid)) {
2412 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2414 init_task_pid(p, PIDTYPE_PID, pid);
2415 if (thread_group_leader(p)) {
2416 init_task_pid(p, PIDTYPE_TGID, pid);
2417 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2418 init_task_pid(p, PIDTYPE_SID, task_session(current));
2420 if (is_child_reaper(pid)) {
2421 ns_of_pid(pid)->child_reaper = p;
2422 p->signal->flags |= SIGNAL_UNKILLABLE;
2424 p->signal->shared_pending.signal = delayed.signal;
2425 p->signal->tty = tty_kref_get(current->signal->tty);
2427 * Inherit has_child_subreaper flag under the same
2428 * tasklist_lock with adding child to the process tree
2429 * for propagate_has_child_subreaper optimization.
2431 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2432 p->real_parent->signal->is_child_subreaper;
2433 list_add_tail(&p->sibling, &p->real_parent->children);
2434 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2435 attach_pid(p, PIDTYPE_TGID);
2436 attach_pid(p, PIDTYPE_PGID);
2437 attach_pid(p, PIDTYPE_SID);
2438 __this_cpu_inc(process_counts);
2440 current->signal->nr_threads++;
2441 atomic_inc(¤t->signal->live);
2442 refcount_inc(¤t->signal->sigcnt);
2443 task_join_group_stop(p);
2444 list_add_tail_rcu(&p->thread_group,
2445 &p->group_leader->thread_group);
2446 list_add_tail_rcu(&p->thread_node,
2447 &p->signal->thread_head);
2449 attach_pid(p, PIDTYPE_PID);
2453 hlist_del_init(&delayed.node);
2454 spin_unlock(¤t->sighand->siglock);
2455 syscall_tracepoint_update(p);
2456 write_unlock_irq(&tasklist_lock);
2459 fd_install(pidfd, pidfile);
2461 proc_fork_connector(p);
2463 cgroup_post_fork(p, args);
2466 trace_task_newtask(p, clone_flags);
2467 uprobe_copy_process(p, clone_flags);
2469 copy_oom_score_adj(clone_flags, p);
2473 bad_fork_cancel_cgroup:
2475 spin_unlock(¤t->sighand->siglock);
2476 write_unlock_irq(&tasklist_lock);
2477 cgroup_cancel_fork(p, args);
2479 if (clone_flags & CLONE_PIDFD) {
2481 put_unused_fd(pidfd);
2484 if (pid != &init_struct_pid)
2486 bad_fork_cleanup_thread:
2488 bad_fork_cleanup_io:
2491 bad_fork_cleanup_namespaces:
2492 exit_task_namespaces(p);
2493 bad_fork_cleanup_mm:
2495 mm_clear_owner(p->mm, p);
2498 bad_fork_cleanup_signal:
2499 if (!(clone_flags & CLONE_THREAD))
2500 free_signal_struct(p->signal);
2501 bad_fork_cleanup_sighand:
2502 __cleanup_sighand(p->sighand);
2503 bad_fork_cleanup_fs:
2504 exit_fs(p); /* blocking */
2505 bad_fork_cleanup_files:
2506 exit_files(p); /* blocking */
2507 bad_fork_cleanup_semundo:
2509 bad_fork_cleanup_security:
2510 security_task_free(p);
2511 bad_fork_cleanup_audit:
2513 bad_fork_cleanup_perf:
2514 perf_event_free_task(p);
2515 bad_fork_cleanup_policy:
2516 lockdep_free_task(p);
2518 mpol_put(p->mempolicy);
2520 bad_fork_cleanup_delayacct:
2521 delayacct_tsk_free(p);
2522 bad_fork_cleanup_count:
2523 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2526 WRITE_ONCE(p->__state, TASK_DEAD);
2527 exit_task_stack_account(p);
2529 delayed_free_task(p);
2531 spin_lock_irq(¤t->sighand->siglock);
2532 hlist_del_init(&delayed.node);
2533 spin_unlock_irq(¤t->sighand->siglock);
2534 return ERR_PTR(retval);
2537 static inline void init_idle_pids(struct task_struct *idle)
2541 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2542 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2543 init_task_pid(idle, type, &init_struct_pid);
2547 struct task_struct * __init fork_idle(int cpu)
2549 struct task_struct *task;
2550 struct kernel_clone_args args = {
2554 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2555 if (!IS_ERR(task)) {
2556 init_idle_pids(task);
2557 init_idle(task, cpu);
2563 struct mm_struct *copy_init_mm(void)
2565 return dup_mm(NULL, &init_mm);
2569 * This is like kernel_clone(), but shaved down and tailored to just
2570 * creating io_uring workers. It returns a created task, or an error pointer.
2571 * The returned task is inactive, and the caller must fire it up through
2572 * wake_up_new_task(p). All signals are blocked in the created task.
2574 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2576 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2578 struct kernel_clone_args args = {
2579 .flags = ((lower_32_bits(flags) | CLONE_VM |
2580 CLONE_UNTRACED) & ~CSIGNAL),
2581 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2582 .stack = (unsigned long)fn,
2583 .stack_size = (unsigned long)arg,
2587 return copy_process(NULL, 0, node, &args);
2591 * Ok, this is the main fork-routine.
2593 * It copies the process, and if successful kick-starts
2594 * it and waits for it to finish using the VM if required.
2596 * args->exit_signal is expected to be checked for sanity by the caller.
2598 pid_t kernel_clone(struct kernel_clone_args *args)
2600 u64 clone_flags = args->flags;
2601 struct completion vfork;
2603 struct task_struct *p;
2608 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2609 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2610 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2611 * field in struct clone_args and it still doesn't make sense to have
2612 * them both point at the same memory location. Performing this check
2613 * here has the advantage that we don't need to have a separate helper
2614 * to check for legacy clone().
2616 if ((args->flags & CLONE_PIDFD) &&
2617 (args->flags & CLONE_PARENT_SETTID) &&
2618 (args->pidfd == args->parent_tid))
2622 * Determine whether and which event to report to ptracer. When
2623 * called from kernel_thread or CLONE_UNTRACED is explicitly
2624 * requested, no event is reported; otherwise, report if the event
2625 * for the type of forking is enabled.
2627 if (!(clone_flags & CLONE_UNTRACED)) {
2628 if (clone_flags & CLONE_VFORK)
2629 trace = PTRACE_EVENT_VFORK;
2630 else if (args->exit_signal != SIGCHLD)
2631 trace = PTRACE_EVENT_CLONE;
2633 trace = PTRACE_EVENT_FORK;
2635 if (likely(!ptrace_event_enabled(current, trace)))
2639 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2640 add_latent_entropy();
2646 * Do this prior waking up the new thread - the thread pointer
2647 * might get invalid after that point, if the thread exits quickly.
2649 trace_sched_process_fork(current, p);
2651 pid = get_task_pid(p, PIDTYPE_PID);
2654 if (clone_flags & CLONE_PARENT_SETTID)
2655 put_user(nr, args->parent_tid);
2657 if (clone_flags & CLONE_VFORK) {
2658 p->vfork_done = &vfork;
2659 init_completion(&vfork);
2663 wake_up_new_task(p);
2665 /* forking complete and child started to run, tell ptracer */
2666 if (unlikely(trace))
2667 ptrace_event_pid(trace, pid);
2669 if (clone_flags & CLONE_VFORK) {
2670 if (!wait_for_vfork_done(p, &vfork))
2671 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2679 * Create a kernel thread.
2681 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2683 struct kernel_clone_args args = {
2684 .flags = ((lower_32_bits(flags) | CLONE_VM |
2685 CLONE_UNTRACED) & ~CSIGNAL),
2686 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2687 .stack = (unsigned long)fn,
2688 .stack_size = (unsigned long)arg,
2691 return kernel_clone(&args);
2694 #ifdef __ARCH_WANT_SYS_FORK
2695 SYSCALL_DEFINE0(fork)
2698 struct kernel_clone_args args = {
2699 .exit_signal = SIGCHLD,
2702 return kernel_clone(&args);
2704 /* can not support in nommu mode */
2710 #ifdef __ARCH_WANT_SYS_VFORK
2711 SYSCALL_DEFINE0(vfork)
2713 struct kernel_clone_args args = {
2714 .flags = CLONE_VFORK | CLONE_VM,
2715 .exit_signal = SIGCHLD,
2718 return kernel_clone(&args);
2722 #ifdef __ARCH_WANT_SYS_CLONE
2723 #ifdef CONFIG_CLONE_BACKWARDS
2724 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2725 int __user *, parent_tidptr,
2727 int __user *, child_tidptr)
2728 #elif defined(CONFIG_CLONE_BACKWARDS2)
2729 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2730 int __user *, parent_tidptr,
2731 int __user *, child_tidptr,
2733 #elif defined(CONFIG_CLONE_BACKWARDS3)
2734 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2736 int __user *, parent_tidptr,
2737 int __user *, child_tidptr,
2740 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2741 int __user *, parent_tidptr,
2742 int __user *, child_tidptr,
2746 struct kernel_clone_args args = {
2747 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2748 .pidfd = parent_tidptr,
2749 .child_tid = child_tidptr,
2750 .parent_tid = parent_tidptr,
2751 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2756 return kernel_clone(&args);
2760 #ifdef __ARCH_WANT_SYS_CLONE3
2762 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2763 struct clone_args __user *uargs,
2767 struct clone_args args;
2768 pid_t *kset_tid = kargs->set_tid;
2770 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2771 CLONE_ARGS_SIZE_VER0);
2772 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2773 CLONE_ARGS_SIZE_VER1);
2774 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2775 CLONE_ARGS_SIZE_VER2);
2776 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2778 if (unlikely(usize > PAGE_SIZE))
2780 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2783 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2787 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2790 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2793 if (unlikely(args.set_tid && args.set_tid_size == 0))
2797 * Verify that higher 32bits of exit_signal are unset and that
2798 * it is a valid signal
2800 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2801 !valid_signal(args.exit_signal)))
2804 if ((args.flags & CLONE_INTO_CGROUP) &&
2805 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2808 *kargs = (struct kernel_clone_args){
2809 .flags = args.flags,
2810 .pidfd = u64_to_user_ptr(args.pidfd),
2811 .child_tid = u64_to_user_ptr(args.child_tid),
2812 .parent_tid = u64_to_user_ptr(args.parent_tid),
2813 .exit_signal = args.exit_signal,
2814 .stack = args.stack,
2815 .stack_size = args.stack_size,
2817 .set_tid_size = args.set_tid_size,
2818 .cgroup = args.cgroup,
2822 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2823 (kargs->set_tid_size * sizeof(pid_t))))
2826 kargs->set_tid = kset_tid;
2832 * clone3_stack_valid - check and prepare stack
2833 * @kargs: kernel clone args
2835 * Verify that the stack arguments userspace gave us are sane.
2836 * In addition, set the stack direction for userspace since it's easy for us to
2839 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2841 if (kargs->stack == 0) {
2842 if (kargs->stack_size > 0)
2845 if (kargs->stack_size == 0)
2848 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2851 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2852 kargs->stack += kargs->stack_size;
2859 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2861 /* Verify that no unknown flags are passed along. */
2863 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2867 * - make the CLONE_DETACHED bit reusable for clone3
2868 * - make the CSIGNAL bits reusable for clone3
2870 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2873 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2874 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2877 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2881 if (!clone3_stack_valid(kargs))
2888 * clone3 - create a new process with specific properties
2889 * @uargs: argument structure
2890 * @size: size of @uargs
2892 * clone3() is the extensible successor to clone()/clone2().
2893 * It takes a struct as argument that is versioned by its size.
2895 * Return: On success, a positive PID for the child process.
2896 * On error, a negative errno number.
2898 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2902 struct kernel_clone_args kargs;
2903 pid_t set_tid[MAX_PID_NS_LEVEL];
2905 kargs.set_tid = set_tid;
2907 err = copy_clone_args_from_user(&kargs, uargs, size);
2911 if (!clone3_args_valid(&kargs))
2914 return kernel_clone(&kargs);
2918 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2920 struct task_struct *leader, *parent, *child;
2923 read_lock(&tasklist_lock);
2924 leader = top = top->group_leader;
2926 for_each_thread(leader, parent) {
2927 list_for_each_entry(child, &parent->children, sibling) {
2928 res = visitor(child, data);
2940 if (leader != top) {
2942 parent = child->real_parent;
2943 leader = parent->group_leader;
2947 read_unlock(&tasklist_lock);
2950 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2951 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2954 static void sighand_ctor(void *data)
2956 struct sighand_struct *sighand = data;
2958 spin_lock_init(&sighand->siglock);
2959 init_waitqueue_head(&sighand->signalfd_wqh);
2962 void __init proc_caches_init(void)
2964 unsigned int mm_size;
2966 sighand_cachep = kmem_cache_create("sighand_cache",
2967 sizeof(struct sighand_struct), 0,
2968 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2969 SLAB_ACCOUNT, sighand_ctor);
2970 signal_cachep = kmem_cache_create("signal_cache",
2971 sizeof(struct signal_struct), 0,
2972 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2974 files_cachep = kmem_cache_create("files_cache",
2975 sizeof(struct files_struct), 0,
2976 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2978 fs_cachep = kmem_cache_create("fs_cache",
2979 sizeof(struct fs_struct), 0,
2980 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2984 * The mm_cpumask is located at the end of mm_struct, and is
2985 * dynamically sized based on the maximum CPU number this system
2986 * can have, taking hotplug into account (nr_cpu_ids).
2988 mm_size = sizeof(struct mm_struct) + cpumask_size();
2990 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2991 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2992 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2993 offsetof(struct mm_struct, saved_auxv),
2994 sizeof_field(struct mm_struct, saved_auxv),
2996 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2998 nsproxy_cache_init();
3002 * Check constraints on flags passed to the unshare system call.
3004 static int check_unshare_flags(unsigned long unshare_flags)
3006 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3007 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3008 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3009 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3013 * Not implemented, but pretend it works if there is nothing
3014 * to unshare. Note that unsharing the address space or the
3015 * signal handlers also need to unshare the signal queues (aka
3018 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3019 if (!thread_group_empty(current))
3022 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3023 if (refcount_read(¤t->sighand->count) > 1)
3026 if (unshare_flags & CLONE_VM) {
3027 if (!current_is_single_threaded())
3035 * Unshare the filesystem structure if it is being shared
3037 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3039 struct fs_struct *fs = current->fs;
3041 if (!(unshare_flags & CLONE_FS) || !fs)
3044 /* don't need lock here; in the worst case we'll do useless copy */
3048 *new_fsp = copy_fs_struct(fs);
3056 * Unshare file descriptor table if it is being shared
3058 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3059 struct files_struct **new_fdp)
3061 struct files_struct *fd = current->files;
3064 if ((unshare_flags & CLONE_FILES) &&
3065 (fd && atomic_read(&fd->count) > 1)) {
3066 *new_fdp = dup_fd(fd, max_fds, &error);
3075 * unshare allows a process to 'unshare' part of the process
3076 * context which was originally shared using clone. copy_*
3077 * functions used by kernel_clone() cannot be used here directly
3078 * because they modify an inactive task_struct that is being
3079 * constructed. Here we are modifying the current, active,
3082 int ksys_unshare(unsigned long unshare_flags)
3084 struct fs_struct *fs, *new_fs = NULL;
3085 struct files_struct *new_fd = NULL;
3086 struct cred *new_cred = NULL;
3087 struct nsproxy *new_nsproxy = NULL;
3092 * If unsharing a user namespace must also unshare the thread group
3093 * and unshare the filesystem root and working directories.
3095 if (unshare_flags & CLONE_NEWUSER)
3096 unshare_flags |= CLONE_THREAD | CLONE_FS;
3098 * If unsharing vm, must also unshare signal handlers.
3100 if (unshare_flags & CLONE_VM)
3101 unshare_flags |= CLONE_SIGHAND;
3103 * If unsharing a signal handlers, must also unshare the signal queues.
3105 if (unshare_flags & CLONE_SIGHAND)
3106 unshare_flags |= CLONE_THREAD;
3108 * If unsharing namespace, must also unshare filesystem information.
3110 if (unshare_flags & CLONE_NEWNS)
3111 unshare_flags |= CLONE_FS;
3113 err = check_unshare_flags(unshare_flags);
3115 goto bad_unshare_out;
3117 * CLONE_NEWIPC must also detach from the undolist: after switching
3118 * to a new ipc namespace, the semaphore arrays from the old
3119 * namespace are unreachable.
3121 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3123 err = unshare_fs(unshare_flags, &new_fs);
3125 goto bad_unshare_out;
3126 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3128 goto bad_unshare_cleanup_fs;
3129 err = unshare_userns(unshare_flags, &new_cred);
3131 goto bad_unshare_cleanup_fd;
3132 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3135 goto bad_unshare_cleanup_cred;
3138 err = set_cred_ucounts(new_cred);
3140 goto bad_unshare_cleanup_cred;
3143 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3146 * CLONE_SYSVSEM is equivalent to sys_exit().
3150 if (unshare_flags & CLONE_NEWIPC) {
3151 /* Orphan segments in old ns (see sem above). */
3153 shm_init_task(current);
3157 switch_task_namespaces(current, new_nsproxy);
3163 spin_lock(&fs->lock);
3164 current->fs = new_fs;
3169 spin_unlock(&fs->lock);
3173 swap(current->files, new_fd);
3175 task_unlock(current);
3178 /* Install the new user namespace */
3179 commit_creds(new_cred);
3184 perf_event_namespaces(current);
3186 bad_unshare_cleanup_cred:
3189 bad_unshare_cleanup_fd:
3191 put_files_struct(new_fd);
3193 bad_unshare_cleanup_fs:
3195 free_fs_struct(new_fs);
3201 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3203 return ksys_unshare(unshare_flags);
3207 * Helper to unshare the files of the current task.
3208 * We don't want to expose copy_files internals to
3209 * the exec layer of the kernel.
3212 int unshare_files(void)
3214 struct task_struct *task = current;
3215 struct files_struct *old, *copy = NULL;
3218 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3226 put_files_struct(old);
3230 int sysctl_max_threads(struct ctl_table *table, int write,
3231 void *buffer, size_t *lenp, loff_t *ppos)
3235 int threads = max_threads;
3237 int max = MAX_THREADS;
3244 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3248 max_threads = threads;