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/kmsan.h>
41 #include <linux/binfmts.h>
42 #include <linux/mman.h>
43 #include <linux/mmu_notifier.h>
46 #include <linux/mm_inline.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/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/stackprotector.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 data_race(memcpy(new, orig, sizeof(*new)));
476 INIT_LIST_HEAD(&new->anon_vma_chain);
477 dup_anon_vma_name(orig, new);
482 static void __vm_area_free(struct vm_area_struct *vma)
484 free_anon_vma_name(vma);
485 kmem_cache_free(vm_area_cachep, vma);
488 #ifdef CONFIG_PER_VMA_LOCK
489 static void vm_area_free_rcu_cb(struct rcu_head *head)
491 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
497 void vm_area_free(struct vm_area_struct *vma)
499 #ifdef CONFIG_PER_VMA_LOCK
500 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
506 static void account_kernel_stack(struct task_struct *tsk, int account)
508 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
509 struct vm_struct *vm = task_stack_vm_area(tsk);
512 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
513 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
514 account * (PAGE_SIZE / 1024));
516 void *stack = task_stack_page(tsk);
518 /* All stack pages are in the same node. */
519 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
520 account * (THREAD_SIZE / 1024));
524 void exit_task_stack_account(struct task_struct *tsk)
526 account_kernel_stack(tsk, -1);
528 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
529 struct vm_struct *vm;
532 vm = task_stack_vm_area(tsk);
533 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
534 memcg_kmem_uncharge_page(vm->pages[i], 0);
538 static void release_task_stack(struct task_struct *tsk)
540 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
541 return; /* Better to leak the stack than to free prematurely */
543 free_thread_stack(tsk);
546 #ifdef CONFIG_THREAD_INFO_IN_TASK
547 void put_task_stack(struct task_struct *tsk)
549 if (refcount_dec_and_test(&tsk->stack_refcount))
550 release_task_stack(tsk);
554 void free_task(struct task_struct *tsk)
556 #ifdef CONFIG_SECCOMP
557 WARN_ON_ONCE(tsk->seccomp.filter);
559 release_user_cpus_ptr(tsk);
562 #ifndef CONFIG_THREAD_INFO_IN_TASK
564 * The task is finally done with both the stack and thread_info,
567 release_task_stack(tsk);
570 * If the task had a separate stack allocation, it should be gone
573 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
575 rt_mutex_debug_task_free(tsk);
576 ftrace_graph_exit_task(tsk);
577 arch_release_task_struct(tsk);
578 if (tsk->flags & PF_KTHREAD)
579 free_kthread_struct(tsk);
580 free_task_struct(tsk);
582 EXPORT_SYMBOL(free_task);
584 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
586 struct file *exe_file;
588 exe_file = get_mm_exe_file(oldmm);
589 RCU_INIT_POINTER(mm->exe_file, exe_file);
591 * We depend on the oldmm having properly denied write access to the
594 if (exe_file && deny_write_access(exe_file))
595 pr_warn_once("deny_write_access() failed in %s\n", __func__);
599 static __latent_entropy int dup_mmap(struct mm_struct *mm,
600 struct mm_struct *oldmm)
602 struct vm_area_struct *mpnt, *tmp;
604 unsigned long charge = 0;
606 VMA_ITERATOR(old_vmi, oldmm, 0);
607 VMA_ITERATOR(vmi, mm, 0);
609 uprobe_start_dup_mmap();
610 if (mmap_write_lock_killable(oldmm)) {
612 goto fail_uprobe_end;
614 flush_cache_dup_mm(oldmm);
615 uprobe_dup_mmap(oldmm, mm);
617 * Not linked in yet - no deadlock potential:
619 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
621 /* No ordering required: file already has been exposed. */
622 dup_mm_exe_file(mm, oldmm);
624 mm->total_vm = oldmm->total_vm;
625 mm->data_vm = oldmm->data_vm;
626 mm->exec_vm = oldmm->exec_vm;
627 mm->stack_vm = oldmm->stack_vm;
629 retval = ksm_fork(mm, oldmm);
632 khugepaged_fork(mm, oldmm);
634 retval = vma_iter_bulk_alloc(&vmi, oldmm->map_count);
638 for_each_vma(old_vmi, mpnt) {
641 if (mpnt->vm_flags & VM_DONTCOPY) {
642 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
647 * Don't duplicate many vmas if we've been oom-killed (for
650 if (fatal_signal_pending(current)) {
654 if (mpnt->vm_flags & VM_ACCOUNT) {
655 unsigned long len = vma_pages(mpnt);
657 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
661 tmp = vm_area_dup(mpnt);
664 retval = vma_dup_policy(mpnt, tmp);
666 goto fail_nomem_policy;
668 retval = dup_userfaultfd(tmp, &uf);
670 goto fail_nomem_anon_vma_fork;
671 if (tmp->vm_flags & VM_WIPEONFORK) {
673 * VM_WIPEONFORK gets a clean slate in the child.
674 * Don't prepare anon_vma until fault since we don't
675 * copy page for current vma.
677 tmp->anon_vma = NULL;
678 } else if (anon_vma_fork(tmp, mpnt))
679 goto fail_nomem_anon_vma_fork;
680 vm_flags_clear(tmp, VM_LOCKED_MASK);
683 struct address_space *mapping = file->f_mapping;
686 i_mmap_lock_write(mapping);
687 if (tmp->vm_flags & VM_SHARED)
688 mapping_allow_writable(mapping);
689 flush_dcache_mmap_lock(mapping);
690 /* insert tmp into the share list, just after mpnt */
691 vma_interval_tree_insert_after(tmp, mpnt,
693 flush_dcache_mmap_unlock(mapping);
694 i_mmap_unlock_write(mapping);
698 * Copy/update hugetlb private vma information.
700 if (is_vm_hugetlb_page(tmp))
701 hugetlb_dup_vma_private(tmp);
703 /* Link the vma into the MT */
704 if (vma_iter_bulk_store(&vmi, tmp))
705 goto fail_nomem_vmi_store;
708 if (!(tmp->vm_flags & VM_WIPEONFORK))
709 retval = copy_page_range(tmp, mpnt);
711 if (tmp->vm_ops && tmp->vm_ops->open)
712 tmp->vm_ops->open(tmp);
717 /* a new mm has just been created */
718 retval = arch_dup_mmap(oldmm, mm);
722 mmap_write_unlock(mm);
724 mmap_write_unlock(oldmm);
725 dup_userfaultfd_complete(&uf);
727 uprobe_end_dup_mmap();
730 fail_nomem_vmi_store:
731 unlink_anon_vmas(tmp);
732 fail_nomem_anon_vma_fork:
733 mpol_put(vma_policy(tmp));
738 vm_unacct_memory(charge);
742 static inline int mm_alloc_pgd(struct mm_struct *mm)
744 mm->pgd = pgd_alloc(mm);
745 if (unlikely(!mm->pgd))
750 static inline void mm_free_pgd(struct mm_struct *mm)
752 pgd_free(mm, mm->pgd);
755 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
757 mmap_write_lock(oldmm);
758 dup_mm_exe_file(mm, oldmm);
759 mmap_write_unlock(oldmm);
762 #define mm_alloc_pgd(mm) (0)
763 #define mm_free_pgd(mm)
764 #endif /* CONFIG_MMU */
766 static void check_mm(struct mm_struct *mm)
770 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
771 "Please make sure 'struct resident_page_types[]' is updated as well");
773 for (i = 0; i < NR_MM_COUNTERS; i++) {
774 long x = percpu_counter_sum(&mm->rss_stat[i]);
777 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
778 mm, resident_page_types[i], x);
781 if (mm_pgtables_bytes(mm))
782 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
783 mm_pgtables_bytes(mm));
785 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
786 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
790 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
791 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
793 static void do_check_lazy_tlb(void *arg)
795 struct mm_struct *mm = arg;
797 WARN_ON_ONCE(current->active_mm == mm);
800 static void do_shoot_lazy_tlb(void *arg)
802 struct mm_struct *mm = arg;
804 if (current->active_mm == mm) {
805 WARN_ON_ONCE(current->mm);
806 current->active_mm = &init_mm;
807 switch_mm(mm, &init_mm, current);
811 static void cleanup_lazy_tlbs(struct mm_struct *mm)
813 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
815 * In this case, lazy tlb mms are refounted and would not reach
816 * __mmdrop until all CPUs have switched away and mmdrop()ed.
822 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
823 * requires lazy mm users to switch to another mm when the refcount
824 * drops to zero, before the mm is freed. This requires IPIs here to
825 * switch kernel threads to init_mm.
827 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
828 * switch with the final userspace teardown TLB flush which leaves the
829 * mm lazy on this CPU but no others, reducing the need for additional
830 * IPIs here. There are cases where a final IPI is still required here,
831 * such as the final mmdrop being performed on a different CPU than the
832 * one exiting, or kernel threads using the mm when userspace exits.
834 * IPI overheads have not found to be expensive, but they could be
835 * reduced in a number of possible ways, for example (roughly
836 * increasing order of complexity):
837 * - The last lazy reference created by exit_mm() could instead switch
838 * to init_mm, however it's probable this will run on the same CPU
839 * immediately afterwards, so this may not reduce IPIs much.
840 * - A batch of mms requiring IPIs could be gathered and freed at once.
841 * - CPUs store active_mm where it can be remotely checked without a
842 * lock, to filter out false-positives in the cpumask.
843 * - After mm_users or mm_count reaches zero, switching away from the
844 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
845 * with some batching or delaying of the final IPIs.
846 * - A delayed freeing and RCU-like quiescing sequence based on mm
847 * switching to avoid IPIs completely.
849 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
850 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
851 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
855 * Called when the last reference to the mm
856 * is dropped: either by a lazy thread or by
857 * mmput. Free the page directory and the mm.
859 void __mmdrop(struct mm_struct *mm)
863 BUG_ON(mm == &init_mm);
864 WARN_ON_ONCE(mm == current->mm);
866 /* Ensure no CPUs are using this as their lazy tlb mm */
867 cleanup_lazy_tlbs(mm);
869 WARN_ON_ONCE(mm == current->active_mm);
872 mmu_notifier_subscriptions_destroy(mm);
874 put_user_ns(mm->user_ns);
877 for (i = 0; i < NR_MM_COUNTERS; i++)
878 percpu_counter_destroy(&mm->rss_stat[i]);
881 EXPORT_SYMBOL_GPL(__mmdrop);
883 static void mmdrop_async_fn(struct work_struct *work)
885 struct mm_struct *mm;
887 mm = container_of(work, struct mm_struct, async_put_work);
891 static void mmdrop_async(struct mm_struct *mm)
893 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
894 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
895 schedule_work(&mm->async_put_work);
899 static inline void free_signal_struct(struct signal_struct *sig)
901 taskstats_tgid_free(sig);
902 sched_autogroup_exit(sig);
904 * __mmdrop is not safe to call from softirq context on x86 due to
905 * pgd_dtor so postpone it to the async context
908 mmdrop_async(sig->oom_mm);
909 kmem_cache_free(signal_cachep, sig);
912 static inline void put_signal_struct(struct signal_struct *sig)
914 if (refcount_dec_and_test(&sig->sigcnt))
915 free_signal_struct(sig);
918 void __put_task_struct(struct task_struct *tsk)
920 WARN_ON(!tsk->exit_state);
921 WARN_ON(refcount_read(&tsk->usage));
922 WARN_ON(tsk == current);
926 task_numa_free(tsk, true);
927 security_task_free(tsk);
928 bpf_task_storage_free(tsk);
930 delayacct_tsk_free(tsk);
931 put_signal_struct(tsk->signal);
932 sched_core_free(tsk);
935 EXPORT_SYMBOL_GPL(__put_task_struct);
937 void __init __weak arch_task_cache_init(void) { }
942 static void set_max_threads(unsigned int max_threads_suggested)
945 unsigned long nr_pages = totalram_pages();
948 * The number of threads shall be limited such that the thread
949 * structures may only consume a small part of the available memory.
951 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
952 threads = MAX_THREADS;
954 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
955 (u64) THREAD_SIZE * 8UL);
957 if (threads > max_threads_suggested)
958 threads = max_threads_suggested;
960 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
963 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
964 /* Initialized by the architecture: */
965 int arch_task_struct_size __read_mostly;
968 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
969 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
971 /* Fetch thread_struct whitelist for the architecture. */
972 arch_thread_struct_whitelist(offset, size);
975 * Handle zero-sized whitelist or empty thread_struct, otherwise
976 * adjust offset to position of thread_struct in task_struct.
978 if (unlikely(*size == 0))
981 *offset += offsetof(struct task_struct, thread);
983 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
985 void __init fork_init(void)
988 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
989 #ifndef ARCH_MIN_TASKALIGN
990 #define ARCH_MIN_TASKALIGN 0
992 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
993 unsigned long useroffset, usersize;
995 /* create a slab on which task_structs can be allocated */
996 task_struct_whitelist(&useroffset, &usersize);
997 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
998 arch_task_struct_size, align,
999 SLAB_PANIC|SLAB_ACCOUNT,
1000 useroffset, usersize, NULL);
1003 /* do the arch specific task caches init */
1004 arch_task_cache_init();
1006 set_max_threads(MAX_THREADS);
1008 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1009 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1010 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1011 init_task.signal->rlim[RLIMIT_NPROC];
1013 for (i = 0; i < UCOUNT_COUNTS; i++)
1014 init_user_ns.ucount_max[i] = max_threads/2;
1016 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1017 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1018 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1019 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1021 #ifdef CONFIG_VMAP_STACK
1022 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1023 NULL, free_vm_stack_cache);
1028 lockdep_init_task(&init_task);
1032 int __weak arch_dup_task_struct(struct task_struct *dst,
1033 struct task_struct *src)
1039 void set_task_stack_end_magic(struct task_struct *tsk)
1041 unsigned long *stackend;
1043 stackend = end_of_stack(tsk);
1044 *stackend = STACK_END_MAGIC; /* for overflow detection */
1047 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1049 struct task_struct *tsk;
1052 if (node == NUMA_NO_NODE)
1053 node = tsk_fork_get_node(orig);
1054 tsk = alloc_task_struct_node(node);
1058 err = arch_dup_task_struct(tsk, orig);
1062 err = alloc_thread_stack_node(tsk, node);
1066 #ifdef CONFIG_THREAD_INFO_IN_TASK
1067 refcount_set(&tsk->stack_refcount, 1);
1069 account_kernel_stack(tsk, 1);
1071 err = scs_prepare(tsk, node);
1075 #ifdef CONFIG_SECCOMP
1077 * We must handle setting up seccomp filters once we're under
1078 * the sighand lock in case orig has changed between now and
1079 * then. Until then, filter must be NULL to avoid messing up
1080 * the usage counts on the error path calling free_task.
1082 tsk->seccomp.filter = NULL;
1085 setup_thread_stack(tsk, orig);
1086 clear_user_return_notifier(tsk);
1087 clear_tsk_need_resched(tsk);
1088 set_task_stack_end_magic(tsk);
1089 clear_syscall_work_syscall_user_dispatch(tsk);
1091 #ifdef CONFIG_STACKPROTECTOR
1092 tsk->stack_canary = get_random_canary();
1094 if (orig->cpus_ptr == &orig->cpus_mask)
1095 tsk->cpus_ptr = &tsk->cpus_mask;
1096 dup_user_cpus_ptr(tsk, orig, node);
1099 * One for the user space visible state that goes away when reaped.
1100 * One for the scheduler.
1102 refcount_set(&tsk->rcu_users, 2);
1103 /* One for the rcu users */
1104 refcount_set(&tsk->usage, 1);
1105 #ifdef CONFIG_BLK_DEV_IO_TRACE
1106 tsk->btrace_seq = 0;
1108 tsk->splice_pipe = NULL;
1109 tsk->task_frag.page = NULL;
1110 tsk->wake_q.next = NULL;
1111 tsk->worker_private = NULL;
1113 kcov_task_init(tsk);
1114 kmsan_task_create(tsk);
1115 kmap_local_fork(tsk);
1117 #ifdef CONFIG_FAULT_INJECTION
1121 #ifdef CONFIG_BLK_CGROUP
1122 tsk->throttle_disk = NULL;
1123 tsk->use_memdelay = 0;
1126 #ifdef CONFIG_IOMMU_SVA
1127 tsk->pasid_activated = 0;
1131 tsk->active_memcg = NULL;
1134 #ifdef CONFIG_CPU_SUP_INTEL
1135 tsk->reported_split_lock = 0;
1138 #ifdef CONFIG_SCHED_MM_CID
1140 tsk->mm_cid_active = 0;
1145 exit_task_stack_account(tsk);
1146 free_thread_stack(tsk);
1148 free_task_struct(tsk);
1152 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1154 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1156 static int __init coredump_filter_setup(char *s)
1158 default_dump_filter =
1159 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1160 MMF_DUMP_FILTER_MASK;
1164 __setup("coredump_filter=", coredump_filter_setup);
1166 #include <linux/init_task.h>
1168 static void mm_init_aio(struct mm_struct *mm)
1171 spin_lock_init(&mm->ioctx_lock);
1172 mm->ioctx_table = NULL;
1176 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1177 struct task_struct *p)
1181 WRITE_ONCE(mm->owner, NULL);
1185 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1192 static void mm_init_uprobes_state(struct mm_struct *mm)
1194 #ifdef CONFIG_UPROBES
1195 mm->uprobes_state.xol_area = NULL;
1199 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1200 struct user_namespace *user_ns)
1204 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1205 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1206 atomic_set(&mm->mm_users, 1);
1207 atomic_set(&mm->mm_count, 1);
1208 seqcount_init(&mm->write_protect_seq);
1210 INIT_LIST_HEAD(&mm->mmlist);
1211 mm_pgtables_bytes_init(mm);
1214 atomic64_set(&mm->pinned_vm, 0);
1215 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1216 spin_lock_init(&mm->page_table_lock);
1217 spin_lock_init(&mm->arg_lock);
1218 mm_init_cpumask(mm);
1220 mm_init_owner(mm, p);
1222 RCU_INIT_POINTER(mm->exe_file, NULL);
1223 mmu_notifier_subscriptions_init(mm);
1224 init_tlb_flush_pending(mm);
1225 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1226 mm->pmd_huge_pte = NULL;
1228 mm_init_uprobes_state(mm);
1229 hugetlb_count_init(mm);
1232 mm->flags = current->mm->flags & MMF_INIT_MASK;
1233 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1235 mm->flags = default_dump_filter;
1239 if (mm_alloc_pgd(mm))
1242 if (init_new_context(p, mm))
1243 goto fail_nocontext;
1245 for (i = 0; i < NR_MM_COUNTERS; i++)
1246 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1249 mm->user_ns = get_user_ns(user_ns);
1250 lru_gen_init_mm(mm);
1256 percpu_counter_destroy(&mm->rss_stat[--i]);
1265 * Allocate and initialize an mm_struct.
1267 struct mm_struct *mm_alloc(void)
1269 struct mm_struct *mm;
1275 memset(mm, 0, sizeof(*mm));
1276 return mm_init(mm, current, current_user_ns());
1279 static inline void __mmput(struct mm_struct *mm)
1281 VM_BUG_ON(atomic_read(&mm->mm_users));
1283 uprobe_clear_state(mm);
1286 khugepaged_exit(mm); /* must run before exit_mmap */
1288 mm_put_huge_zero_page(mm);
1289 set_mm_exe_file(mm, NULL);
1290 if (!list_empty(&mm->mmlist)) {
1291 spin_lock(&mmlist_lock);
1292 list_del(&mm->mmlist);
1293 spin_unlock(&mmlist_lock);
1296 module_put(mm->binfmt->module);
1302 * Decrement the use count and release all resources for an mm.
1304 void mmput(struct mm_struct *mm)
1308 if (atomic_dec_and_test(&mm->mm_users))
1311 EXPORT_SYMBOL_GPL(mmput);
1314 static void mmput_async_fn(struct work_struct *work)
1316 struct mm_struct *mm = container_of(work, struct mm_struct,
1322 void mmput_async(struct mm_struct *mm)
1324 if (atomic_dec_and_test(&mm->mm_users)) {
1325 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1326 schedule_work(&mm->async_put_work);
1329 EXPORT_SYMBOL_GPL(mmput_async);
1333 * set_mm_exe_file - change a reference to the mm's executable file
1335 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1337 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1338 * invocations: in mmput() nobody alive left, in execve task is single
1341 * Can only fail if new_exe_file != NULL.
1343 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1345 struct file *old_exe_file;
1348 * It is safe to dereference the exe_file without RCU as
1349 * this function is only called if nobody else can access
1350 * this mm -- see comment above for justification.
1352 old_exe_file = rcu_dereference_raw(mm->exe_file);
1356 * We expect the caller (i.e., sys_execve) to already denied
1357 * write access, so this is unlikely to fail.
1359 if (unlikely(deny_write_access(new_exe_file)))
1361 get_file(new_exe_file);
1363 rcu_assign_pointer(mm->exe_file, new_exe_file);
1365 allow_write_access(old_exe_file);
1372 * replace_mm_exe_file - replace a reference to the mm's executable file
1374 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1375 * dealing with concurrent invocation and without grabbing the mmap lock in
1378 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1380 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1382 struct vm_area_struct *vma;
1383 struct file *old_exe_file;
1386 /* Forbid mm->exe_file change if old file still mapped. */
1387 old_exe_file = get_mm_exe_file(mm);
1389 VMA_ITERATOR(vmi, mm, 0);
1391 for_each_vma(vmi, vma) {
1394 if (path_equal(&vma->vm_file->f_path,
1395 &old_exe_file->f_path)) {
1400 mmap_read_unlock(mm);
1406 /* set the new file, lockless */
1407 ret = deny_write_access(new_exe_file);
1410 get_file(new_exe_file);
1412 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1415 * Don't race with dup_mmap() getting the file and disallowing
1416 * write access while someone might open the file writable.
1419 allow_write_access(old_exe_file);
1421 mmap_read_unlock(mm);
1427 * get_mm_exe_file - acquire a reference to the mm's executable file
1429 * Returns %NULL if mm has no associated executable file.
1430 * User must release file via fput().
1432 struct file *get_mm_exe_file(struct mm_struct *mm)
1434 struct file *exe_file;
1437 exe_file = rcu_dereference(mm->exe_file);
1438 if (exe_file && !get_file_rcu(exe_file))
1445 * get_task_exe_file - acquire a reference to the task's executable file
1447 * Returns %NULL if task's mm (if any) has no associated executable file or
1448 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1449 * User must release file via fput().
1451 struct file *get_task_exe_file(struct task_struct *task)
1453 struct file *exe_file = NULL;
1454 struct mm_struct *mm;
1459 if (!(task->flags & PF_KTHREAD))
1460 exe_file = get_mm_exe_file(mm);
1467 * get_task_mm - acquire a reference to the task's mm
1469 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1470 * this kernel workthread has transiently adopted a user mm with use_mm,
1471 * to do its AIO) is not set and if so returns a reference to it, after
1472 * bumping up the use count. User must release the mm via mmput()
1473 * after use. Typically used by /proc and ptrace.
1475 struct mm_struct *get_task_mm(struct task_struct *task)
1477 struct mm_struct *mm;
1482 if (task->flags & PF_KTHREAD)
1490 EXPORT_SYMBOL_GPL(get_task_mm);
1492 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1494 struct mm_struct *mm;
1497 err = down_read_killable(&task->signal->exec_update_lock);
1499 return ERR_PTR(err);
1501 mm = get_task_mm(task);
1502 if (mm && mm != current->mm &&
1503 !ptrace_may_access(task, mode)) {
1505 mm = ERR_PTR(-EACCES);
1507 up_read(&task->signal->exec_update_lock);
1512 static void complete_vfork_done(struct task_struct *tsk)
1514 struct completion *vfork;
1517 vfork = tsk->vfork_done;
1518 if (likely(vfork)) {
1519 tsk->vfork_done = NULL;
1525 static int wait_for_vfork_done(struct task_struct *child,
1526 struct completion *vfork)
1528 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1531 cgroup_enter_frozen();
1532 killed = wait_for_completion_state(vfork, state);
1533 cgroup_leave_frozen(false);
1537 child->vfork_done = NULL;
1541 put_task_struct(child);
1545 /* Please note the differences between mmput and mm_release.
1546 * mmput is called whenever we stop holding onto a mm_struct,
1547 * error success whatever.
1549 * mm_release is called after a mm_struct has been removed
1550 * from the current process.
1552 * This difference is important for error handling, when we
1553 * only half set up a mm_struct for a new process and need to restore
1554 * the old one. Because we mmput the new mm_struct before
1555 * restoring the old one. . .
1556 * Eric Biederman 10 January 1998
1558 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1560 uprobe_free_utask(tsk);
1562 /* Get rid of any cached register state */
1563 deactivate_mm(tsk, mm);
1566 * Signal userspace if we're not exiting with a core dump
1567 * because we want to leave the value intact for debugging
1570 if (tsk->clear_child_tid) {
1571 if (atomic_read(&mm->mm_users) > 1) {
1573 * We don't check the error code - if userspace has
1574 * not set up a proper pointer then tough luck.
1576 put_user(0, tsk->clear_child_tid);
1577 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1578 1, NULL, NULL, 0, 0);
1580 tsk->clear_child_tid = NULL;
1584 * All done, finally we can wake up parent and return this mm to him.
1585 * Also kthread_stop() uses this completion for synchronization.
1587 if (tsk->vfork_done)
1588 complete_vfork_done(tsk);
1591 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1593 futex_exit_release(tsk);
1594 mm_release(tsk, mm);
1597 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1599 futex_exec_release(tsk);
1600 mm_release(tsk, mm);
1604 * dup_mm() - duplicates an existing mm structure
1605 * @tsk: the task_struct with which the new mm will be associated.
1606 * @oldmm: the mm to duplicate.
1608 * Allocates a new mm structure and duplicates the provided @oldmm structure
1611 * Return: the duplicated mm or NULL on failure.
1613 static struct mm_struct *dup_mm(struct task_struct *tsk,
1614 struct mm_struct *oldmm)
1616 struct mm_struct *mm;
1623 memcpy(mm, oldmm, sizeof(*mm));
1625 if (!mm_init(mm, tsk, mm->user_ns))
1628 err = dup_mmap(mm, oldmm);
1632 mm->hiwater_rss = get_mm_rss(mm);
1633 mm->hiwater_vm = mm->total_vm;
1635 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1641 /* don't put binfmt in mmput, we haven't got module yet */
1643 mm_init_owner(mm, NULL);
1650 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1652 struct mm_struct *mm, *oldmm;
1654 tsk->min_flt = tsk->maj_flt = 0;
1655 tsk->nvcsw = tsk->nivcsw = 0;
1656 #ifdef CONFIG_DETECT_HUNG_TASK
1657 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1658 tsk->last_switch_time = 0;
1662 tsk->active_mm = NULL;
1665 * Are we cloning a kernel thread?
1667 * We need to steal a active VM for that..
1669 oldmm = current->mm;
1673 if (clone_flags & CLONE_VM) {
1677 mm = dup_mm(tsk, current->mm);
1683 tsk->active_mm = mm;
1684 sched_mm_cid_fork(tsk);
1688 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1690 struct fs_struct *fs = current->fs;
1691 if (clone_flags & CLONE_FS) {
1692 /* tsk->fs is already what we want */
1693 spin_lock(&fs->lock);
1695 spin_unlock(&fs->lock);
1699 spin_unlock(&fs->lock);
1702 tsk->fs = copy_fs_struct(fs);
1708 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1710 struct files_struct *oldf, *newf;
1714 * A background process may not have any files ...
1716 oldf = current->files;
1720 if (clone_flags & CLONE_FILES) {
1721 atomic_inc(&oldf->count);
1725 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1735 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1737 struct sighand_struct *sig;
1739 if (clone_flags & CLONE_SIGHAND) {
1740 refcount_inc(¤t->sighand->count);
1743 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1744 RCU_INIT_POINTER(tsk->sighand, sig);
1748 refcount_set(&sig->count, 1);
1749 spin_lock_irq(¤t->sighand->siglock);
1750 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1751 spin_unlock_irq(¤t->sighand->siglock);
1753 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1754 if (clone_flags & CLONE_CLEAR_SIGHAND)
1755 flush_signal_handlers(tsk, 0);
1760 void __cleanup_sighand(struct sighand_struct *sighand)
1762 if (refcount_dec_and_test(&sighand->count)) {
1763 signalfd_cleanup(sighand);
1765 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1766 * without an RCU grace period, see __lock_task_sighand().
1768 kmem_cache_free(sighand_cachep, sighand);
1773 * Initialize POSIX timer handling for a thread group.
1775 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1777 struct posix_cputimers *pct = &sig->posix_cputimers;
1778 unsigned long cpu_limit;
1780 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1781 posix_cputimers_group_init(pct, cpu_limit);
1784 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1786 struct signal_struct *sig;
1788 if (clone_flags & CLONE_THREAD)
1791 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1796 sig->nr_threads = 1;
1797 sig->quick_threads = 1;
1798 atomic_set(&sig->live, 1);
1799 refcount_set(&sig->sigcnt, 1);
1801 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1802 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1803 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1805 init_waitqueue_head(&sig->wait_chldexit);
1806 sig->curr_target = tsk;
1807 init_sigpending(&sig->shared_pending);
1808 INIT_HLIST_HEAD(&sig->multiprocess);
1809 seqlock_init(&sig->stats_lock);
1810 prev_cputime_init(&sig->prev_cputime);
1812 #ifdef CONFIG_POSIX_TIMERS
1813 INIT_LIST_HEAD(&sig->posix_timers);
1814 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1815 sig->real_timer.function = it_real_fn;
1818 task_lock(current->group_leader);
1819 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1820 task_unlock(current->group_leader);
1822 posix_cpu_timers_init_group(sig);
1824 tty_audit_fork(sig);
1825 sched_autogroup_fork(sig);
1827 sig->oom_score_adj = current->signal->oom_score_adj;
1828 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1830 mutex_init(&sig->cred_guard_mutex);
1831 init_rwsem(&sig->exec_update_lock);
1836 static void copy_seccomp(struct task_struct *p)
1838 #ifdef CONFIG_SECCOMP
1840 * Must be called with sighand->lock held, which is common to
1841 * all threads in the group. Holding cred_guard_mutex is not
1842 * needed because this new task is not yet running and cannot
1845 assert_spin_locked(¤t->sighand->siglock);
1847 /* Ref-count the new filter user, and assign it. */
1848 get_seccomp_filter(current);
1849 p->seccomp = current->seccomp;
1852 * Explicitly enable no_new_privs here in case it got set
1853 * between the task_struct being duplicated and holding the
1854 * sighand lock. The seccomp state and nnp must be in sync.
1856 if (task_no_new_privs(current))
1857 task_set_no_new_privs(p);
1860 * If the parent gained a seccomp mode after copying thread
1861 * flags and between before we held the sighand lock, we have
1862 * to manually enable the seccomp thread flag here.
1864 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1865 set_task_syscall_work(p, SECCOMP);
1869 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1871 current->clear_child_tid = tidptr;
1873 return task_pid_vnr(current);
1876 static void rt_mutex_init_task(struct task_struct *p)
1878 raw_spin_lock_init(&p->pi_lock);
1879 #ifdef CONFIG_RT_MUTEXES
1880 p->pi_waiters = RB_ROOT_CACHED;
1881 p->pi_top_task = NULL;
1882 p->pi_blocked_on = NULL;
1886 static inline void init_task_pid_links(struct task_struct *task)
1890 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1891 INIT_HLIST_NODE(&task->pid_links[type]);
1895 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1897 if (type == PIDTYPE_PID)
1898 task->thread_pid = pid;
1900 task->signal->pids[type] = pid;
1903 static inline void rcu_copy_process(struct task_struct *p)
1905 #ifdef CONFIG_PREEMPT_RCU
1906 p->rcu_read_lock_nesting = 0;
1907 p->rcu_read_unlock_special.s = 0;
1908 p->rcu_blocked_node = NULL;
1909 INIT_LIST_HEAD(&p->rcu_node_entry);
1910 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1911 #ifdef CONFIG_TASKS_RCU
1912 p->rcu_tasks_holdout = false;
1913 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1914 p->rcu_tasks_idle_cpu = -1;
1915 #endif /* #ifdef CONFIG_TASKS_RCU */
1916 #ifdef CONFIG_TASKS_TRACE_RCU
1917 p->trc_reader_nesting = 0;
1918 p->trc_reader_special.s = 0;
1919 INIT_LIST_HEAD(&p->trc_holdout_list);
1920 INIT_LIST_HEAD(&p->trc_blkd_node);
1921 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1924 struct pid *pidfd_pid(const struct file *file)
1926 if (file->f_op == &pidfd_fops)
1927 return file->private_data;
1929 return ERR_PTR(-EBADF);
1932 static int pidfd_release(struct inode *inode, struct file *file)
1934 struct pid *pid = file->private_data;
1936 file->private_data = NULL;
1941 #ifdef CONFIG_PROC_FS
1943 * pidfd_show_fdinfo - print information about a pidfd
1944 * @m: proc fdinfo file
1945 * @f: file referencing a pidfd
1948 * This function will print the pid that a given pidfd refers to in the
1949 * pid namespace of the procfs instance.
1950 * If the pid namespace of the process is not a descendant of the pid
1951 * namespace of the procfs instance 0 will be shown as its pid. This is
1952 * similar to calling getppid() on a process whose parent is outside of
1953 * its pid namespace.
1956 * If pid namespaces are supported then this function will also print
1957 * the pid of a given pidfd refers to for all descendant pid namespaces
1958 * starting from the current pid namespace of the instance, i.e. the
1959 * Pid field and the first entry in the NSpid field will be identical.
1960 * If the pid namespace of the process is not a descendant of the pid
1961 * namespace of the procfs instance 0 will be shown as its first NSpid
1962 * entry and no others will be shown.
1963 * Note that this differs from the Pid and NSpid fields in
1964 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1965 * the pid namespace of the procfs instance. The difference becomes
1966 * obvious when sending around a pidfd between pid namespaces from a
1967 * different branch of the tree, i.e. where no ancestral relation is
1968 * present between the pid namespaces:
1969 * - create two new pid namespaces ns1 and ns2 in the initial pid
1970 * namespace (also take care to create new mount namespaces in the
1971 * new pid namespace and mount procfs)
1972 * - create a process with a pidfd in ns1
1973 * - send pidfd from ns1 to ns2
1974 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1975 * have exactly one entry, which is 0
1977 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1979 struct pid *pid = f->private_data;
1980 struct pid_namespace *ns;
1983 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1984 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1985 nr = pid_nr_ns(pid, ns);
1988 seq_put_decimal_ll(m, "Pid:\t", nr);
1990 #ifdef CONFIG_PID_NS
1991 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1995 /* If nr is non-zero it means that 'pid' is valid and that
1996 * ns, i.e. the pid namespace associated with the procfs
1997 * instance, is in the pid namespace hierarchy of pid.
1998 * Start at one below the already printed level.
2000 for (i = ns->level + 1; i <= pid->level; i++)
2001 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2009 * Poll support for process exit notification.
2011 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2013 struct pid *pid = file->private_data;
2014 __poll_t poll_flags = 0;
2016 poll_wait(file, &pid->wait_pidfd, pts);
2019 * Inform pollers only when the whole thread group exits.
2020 * If the thread group leader exits before all other threads in the
2021 * group, then poll(2) should block, similar to the wait(2) family.
2023 if (thread_group_exited(pid))
2024 poll_flags = EPOLLIN | EPOLLRDNORM;
2029 const struct file_operations pidfd_fops = {
2030 .release = pidfd_release,
2032 #ifdef CONFIG_PROC_FS
2033 .show_fdinfo = pidfd_show_fdinfo,
2037 static void __delayed_free_task(struct rcu_head *rhp)
2039 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2044 static __always_inline void delayed_free_task(struct task_struct *tsk)
2046 if (IS_ENABLED(CONFIG_MEMCG))
2047 call_rcu(&tsk->rcu, __delayed_free_task);
2052 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2054 /* Skip if kernel thread */
2058 /* Skip if spawning a thread or using vfork */
2059 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2062 /* We need to synchronize with __set_oom_adj */
2063 mutex_lock(&oom_adj_mutex);
2064 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2065 /* Update the values in case they were changed after copy_signal */
2066 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2067 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2068 mutex_unlock(&oom_adj_mutex);
2072 static void rv_task_fork(struct task_struct *p)
2076 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2077 p->rv[i].da_mon.monitoring = false;
2080 #define rv_task_fork(p) do {} while (0)
2084 * This creates a new process as a copy of the old one,
2085 * but does not actually start it yet.
2087 * It copies the registers, and all the appropriate
2088 * parts of the process environment (as per the clone
2089 * flags). The actual kick-off is left to the caller.
2091 static __latent_entropy struct task_struct *copy_process(
2095 struct kernel_clone_args *args)
2097 int pidfd = -1, retval;
2098 struct task_struct *p;
2099 struct multiprocess_signals delayed;
2100 struct file *pidfile = NULL;
2101 const u64 clone_flags = args->flags;
2102 struct nsproxy *nsp = current->nsproxy;
2105 * Don't allow sharing the root directory with processes in a different
2108 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2109 return ERR_PTR(-EINVAL);
2111 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2112 return ERR_PTR(-EINVAL);
2115 * Thread groups must share signals as well, and detached threads
2116 * can only be started up within the thread group.
2118 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2119 return ERR_PTR(-EINVAL);
2122 * Shared signal handlers imply shared VM. By way of the above,
2123 * thread groups also imply shared VM. Blocking this case allows
2124 * for various simplifications in other code.
2126 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2127 return ERR_PTR(-EINVAL);
2130 * Siblings of global init remain as zombies on exit since they are
2131 * not reaped by their parent (swapper). To solve this and to avoid
2132 * multi-rooted process trees, prevent global and container-inits
2133 * from creating siblings.
2135 if ((clone_flags & CLONE_PARENT) &&
2136 current->signal->flags & SIGNAL_UNKILLABLE)
2137 return ERR_PTR(-EINVAL);
2140 * If the new process will be in a different pid or user namespace
2141 * do not allow it to share a thread group with the forking task.
2143 if (clone_flags & CLONE_THREAD) {
2144 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2145 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2146 return ERR_PTR(-EINVAL);
2149 if (clone_flags & CLONE_PIDFD) {
2151 * - CLONE_DETACHED is blocked so that we can potentially
2152 * reuse it later for CLONE_PIDFD.
2153 * - CLONE_THREAD is blocked until someone really needs it.
2155 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2156 return ERR_PTR(-EINVAL);
2160 * Force any signals received before this point to be delivered
2161 * before the fork happens. Collect up signals sent to multiple
2162 * processes that happen during the fork and delay them so that
2163 * they appear to happen after the fork.
2165 sigemptyset(&delayed.signal);
2166 INIT_HLIST_NODE(&delayed.node);
2168 spin_lock_irq(¤t->sighand->siglock);
2169 if (!(clone_flags & CLONE_THREAD))
2170 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2171 recalc_sigpending();
2172 spin_unlock_irq(¤t->sighand->siglock);
2173 retval = -ERESTARTNOINTR;
2174 if (task_sigpending(current))
2178 p = dup_task_struct(current, node);
2181 p->flags &= ~PF_KTHREAD;
2183 p->flags |= PF_KTHREAD;
2184 if (args->io_thread) {
2186 * Mark us an IO worker, and block any signal that isn't
2189 p->flags |= PF_IO_WORKER;
2190 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2193 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2195 * Clear TID on mm_release()?
2197 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2199 ftrace_graph_init_task(p);
2201 rt_mutex_init_task(p);
2203 lockdep_assert_irqs_enabled();
2204 #ifdef CONFIG_PROVE_LOCKING
2205 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2207 retval = copy_creds(p, clone_flags);
2212 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2213 if (p->real_cred->user != INIT_USER &&
2214 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2215 goto bad_fork_cleanup_count;
2217 current->flags &= ~PF_NPROC_EXCEEDED;
2220 * If multiple threads are within copy_process(), then this check
2221 * triggers too late. This doesn't hurt, the check is only there
2222 * to stop root fork bombs.
2225 if (data_race(nr_threads >= max_threads))
2226 goto bad_fork_cleanup_count;
2228 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2229 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2230 p->flags |= PF_FORKNOEXEC;
2231 INIT_LIST_HEAD(&p->children);
2232 INIT_LIST_HEAD(&p->sibling);
2233 rcu_copy_process(p);
2234 p->vfork_done = NULL;
2235 spin_lock_init(&p->alloc_lock);
2237 init_sigpending(&p->pending);
2239 p->utime = p->stime = p->gtime = 0;
2240 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2241 p->utimescaled = p->stimescaled = 0;
2243 prev_cputime_init(&p->prev_cputime);
2245 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2246 seqcount_init(&p->vtime.seqcount);
2247 p->vtime.starttime = 0;
2248 p->vtime.state = VTIME_INACTIVE;
2251 #ifdef CONFIG_IO_URING
2255 #if defined(SPLIT_RSS_COUNTING)
2256 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2259 p->default_timer_slack_ns = current->timer_slack_ns;
2265 task_io_accounting_init(&p->ioac);
2266 acct_clear_integrals(p);
2268 posix_cputimers_init(&p->posix_cputimers);
2270 p->io_context = NULL;
2271 audit_set_context(p, NULL);
2273 if (args->kthread) {
2274 if (!set_kthread_struct(p))
2275 goto bad_fork_cleanup_delayacct;
2278 p->mempolicy = mpol_dup(p->mempolicy);
2279 if (IS_ERR(p->mempolicy)) {
2280 retval = PTR_ERR(p->mempolicy);
2281 p->mempolicy = NULL;
2282 goto bad_fork_cleanup_delayacct;
2285 #ifdef CONFIG_CPUSETS
2286 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2287 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2288 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2290 #ifdef CONFIG_TRACE_IRQFLAGS
2291 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2292 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2293 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2294 p->softirqs_enabled = 1;
2295 p->softirq_context = 0;
2298 p->pagefault_disabled = 0;
2300 #ifdef CONFIG_LOCKDEP
2301 lockdep_init_task(p);
2304 #ifdef CONFIG_DEBUG_MUTEXES
2305 p->blocked_on = NULL; /* not blocked yet */
2307 #ifdef CONFIG_BCACHE
2308 p->sequential_io = 0;
2309 p->sequential_io_avg = 0;
2311 #ifdef CONFIG_BPF_SYSCALL
2312 RCU_INIT_POINTER(p->bpf_storage, NULL);
2316 /* Perform scheduler related setup. Assign this task to a CPU. */
2317 retval = sched_fork(clone_flags, p);
2319 goto bad_fork_cleanup_policy;
2321 retval = perf_event_init_task(p, clone_flags);
2323 goto bad_fork_cleanup_policy;
2324 retval = audit_alloc(p);
2326 goto bad_fork_cleanup_perf;
2327 /* copy all the process information */
2329 retval = security_task_alloc(p, clone_flags);
2331 goto bad_fork_cleanup_audit;
2332 retval = copy_semundo(clone_flags, p);
2334 goto bad_fork_cleanup_security;
2335 retval = copy_files(clone_flags, p);
2337 goto bad_fork_cleanup_semundo;
2338 retval = copy_fs(clone_flags, p);
2340 goto bad_fork_cleanup_files;
2341 retval = copy_sighand(clone_flags, p);
2343 goto bad_fork_cleanup_fs;
2344 retval = copy_signal(clone_flags, p);
2346 goto bad_fork_cleanup_sighand;
2347 retval = copy_mm(clone_flags, p);
2349 goto bad_fork_cleanup_signal;
2350 retval = copy_namespaces(clone_flags, p);
2352 goto bad_fork_cleanup_mm;
2353 retval = copy_io(clone_flags, p);
2355 goto bad_fork_cleanup_namespaces;
2356 retval = copy_thread(p, args);
2358 goto bad_fork_cleanup_io;
2360 stackleak_task_init(p);
2362 if (pid != &init_struct_pid) {
2363 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2364 args->set_tid_size);
2366 retval = PTR_ERR(pid);
2367 goto bad_fork_cleanup_thread;
2372 * This has to happen after we've potentially unshared the file
2373 * descriptor table (so that the pidfd doesn't leak into the child
2374 * if the fd table isn't shared).
2376 if (clone_flags & CLONE_PIDFD) {
2377 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2379 goto bad_fork_free_pid;
2383 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2384 O_RDWR | O_CLOEXEC);
2385 if (IS_ERR(pidfile)) {
2386 put_unused_fd(pidfd);
2387 retval = PTR_ERR(pidfile);
2388 goto bad_fork_free_pid;
2390 get_pid(pid); /* held by pidfile now */
2392 retval = put_user(pidfd, args->pidfd);
2394 goto bad_fork_put_pidfd;
2403 * sigaltstack should be cleared when sharing the same VM
2405 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2409 * Syscall tracing and stepping should be turned off in the
2410 * child regardless of CLONE_PTRACE.
2412 user_disable_single_step(p);
2413 clear_task_syscall_work(p, SYSCALL_TRACE);
2414 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2415 clear_task_syscall_work(p, SYSCALL_EMU);
2417 clear_tsk_latency_tracing(p);
2419 /* ok, now we should be set up.. */
2420 p->pid = pid_nr(pid);
2421 if (clone_flags & CLONE_THREAD) {
2422 p->group_leader = current->group_leader;
2423 p->tgid = current->tgid;
2425 p->group_leader = p;
2430 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2431 p->dirty_paused_when = 0;
2433 p->pdeath_signal = 0;
2434 INIT_LIST_HEAD(&p->thread_group);
2435 p->task_works = NULL;
2436 clear_posix_cputimers_work(p);
2438 #ifdef CONFIG_KRETPROBES
2439 p->kretprobe_instances.first = NULL;
2441 #ifdef CONFIG_RETHOOK
2442 p->rethooks.first = NULL;
2446 * Ensure that the cgroup subsystem policies allow the new process to be
2447 * forked. It should be noted that the new process's css_set can be changed
2448 * between here and cgroup_post_fork() if an organisation operation is in
2451 retval = cgroup_can_fork(p, args);
2453 goto bad_fork_put_pidfd;
2456 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2457 * the new task on the correct runqueue. All this *before* the task
2460 * This isn't part of ->can_fork() because while the re-cloning is
2461 * cgroup specific, it unconditionally needs to place the task on a
2464 sched_cgroup_fork(p, args);
2467 * From this point on we must avoid any synchronous user-space
2468 * communication until we take the tasklist-lock. In particular, we do
2469 * not want user-space to be able to predict the process start-time by
2470 * stalling fork(2) after we recorded the start_time but before it is
2471 * visible to the system.
2474 p->start_time = ktime_get_ns();
2475 p->start_boottime = ktime_get_boottime_ns();
2478 * Make it visible to the rest of the system, but dont wake it up yet.
2479 * Need tasklist lock for parent etc handling!
2481 write_lock_irq(&tasklist_lock);
2483 /* CLONE_PARENT re-uses the old parent */
2484 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2485 p->real_parent = current->real_parent;
2486 p->parent_exec_id = current->parent_exec_id;
2487 if (clone_flags & CLONE_THREAD)
2488 p->exit_signal = -1;
2490 p->exit_signal = current->group_leader->exit_signal;
2492 p->real_parent = current;
2493 p->parent_exec_id = current->self_exec_id;
2494 p->exit_signal = args->exit_signal;
2497 klp_copy_process(p);
2501 spin_lock(¤t->sighand->siglock);
2505 rseq_fork(p, clone_flags);
2507 /* Don't start children in a dying pid namespace */
2508 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2510 goto bad_fork_cancel_cgroup;
2513 /* Let kill terminate clone/fork in the middle */
2514 if (fatal_signal_pending(current)) {
2516 goto bad_fork_cancel_cgroup;
2519 /* No more failure paths after this point. */
2522 * Copy seccomp details explicitly here, in case they were changed
2523 * before holding sighand lock.
2527 init_task_pid_links(p);
2528 if (likely(p->pid)) {
2529 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2531 init_task_pid(p, PIDTYPE_PID, pid);
2532 if (thread_group_leader(p)) {
2533 init_task_pid(p, PIDTYPE_TGID, pid);
2534 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2535 init_task_pid(p, PIDTYPE_SID, task_session(current));
2537 if (is_child_reaper(pid)) {
2538 ns_of_pid(pid)->child_reaper = p;
2539 p->signal->flags |= SIGNAL_UNKILLABLE;
2541 p->signal->shared_pending.signal = delayed.signal;
2542 p->signal->tty = tty_kref_get(current->signal->tty);
2544 * Inherit has_child_subreaper flag under the same
2545 * tasklist_lock with adding child to the process tree
2546 * for propagate_has_child_subreaper optimization.
2548 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2549 p->real_parent->signal->is_child_subreaper;
2550 list_add_tail(&p->sibling, &p->real_parent->children);
2551 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2552 attach_pid(p, PIDTYPE_TGID);
2553 attach_pid(p, PIDTYPE_PGID);
2554 attach_pid(p, PIDTYPE_SID);
2555 __this_cpu_inc(process_counts);
2557 current->signal->nr_threads++;
2558 current->signal->quick_threads++;
2559 atomic_inc(¤t->signal->live);
2560 refcount_inc(¤t->signal->sigcnt);
2561 task_join_group_stop(p);
2562 list_add_tail_rcu(&p->thread_group,
2563 &p->group_leader->thread_group);
2564 list_add_tail_rcu(&p->thread_node,
2565 &p->signal->thread_head);
2567 attach_pid(p, PIDTYPE_PID);
2571 hlist_del_init(&delayed.node);
2572 spin_unlock(¤t->sighand->siglock);
2573 syscall_tracepoint_update(p);
2574 write_unlock_irq(&tasklist_lock);
2577 fd_install(pidfd, pidfile);
2579 proc_fork_connector(p);
2581 cgroup_post_fork(p, args);
2584 trace_task_newtask(p, clone_flags);
2585 uprobe_copy_process(p, clone_flags);
2587 copy_oom_score_adj(clone_flags, p);
2591 bad_fork_cancel_cgroup:
2593 spin_unlock(¤t->sighand->siglock);
2594 write_unlock_irq(&tasklist_lock);
2595 cgroup_cancel_fork(p, args);
2597 if (clone_flags & CLONE_PIDFD) {
2599 put_unused_fd(pidfd);
2602 if (pid != &init_struct_pid)
2604 bad_fork_cleanup_thread:
2606 bad_fork_cleanup_io:
2609 bad_fork_cleanup_namespaces:
2610 exit_task_namespaces(p);
2611 bad_fork_cleanup_mm:
2613 mm_clear_owner(p->mm, p);
2616 bad_fork_cleanup_signal:
2617 if (!(clone_flags & CLONE_THREAD))
2618 free_signal_struct(p->signal);
2619 bad_fork_cleanup_sighand:
2620 __cleanup_sighand(p->sighand);
2621 bad_fork_cleanup_fs:
2622 exit_fs(p); /* blocking */
2623 bad_fork_cleanup_files:
2624 exit_files(p); /* blocking */
2625 bad_fork_cleanup_semundo:
2627 bad_fork_cleanup_security:
2628 security_task_free(p);
2629 bad_fork_cleanup_audit:
2631 bad_fork_cleanup_perf:
2632 perf_event_free_task(p);
2633 bad_fork_cleanup_policy:
2634 lockdep_free_task(p);
2636 mpol_put(p->mempolicy);
2638 bad_fork_cleanup_delayacct:
2639 delayacct_tsk_free(p);
2640 bad_fork_cleanup_count:
2641 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2644 WRITE_ONCE(p->__state, TASK_DEAD);
2645 exit_task_stack_account(p);
2647 delayed_free_task(p);
2649 spin_lock_irq(¤t->sighand->siglock);
2650 hlist_del_init(&delayed.node);
2651 spin_unlock_irq(¤t->sighand->siglock);
2652 return ERR_PTR(retval);
2655 static inline void init_idle_pids(struct task_struct *idle)
2659 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2660 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2661 init_task_pid(idle, type, &init_struct_pid);
2665 static int idle_dummy(void *dummy)
2667 /* This function is never called */
2671 struct task_struct * __init fork_idle(int cpu)
2673 struct task_struct *task;
2674 struct kernel_clone_args args = {
2682 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2683 if (!IS_ERR(task)) {
2684 init_idle_pids(task);
2685 init_idle(task, cpu);
2692 * This is like kernel_clone(), but shaved down and tailored to just
2693 * creating io_uring workers. It returns a created task, or an error pointer.
2694 * The returned task is inactive, and the caller must fire it up through
2695 * wake_up_new_task(p). All signals are blocked in the created task.
2697 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2699 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2701 struct kernel_clone_args args = {
2702 .flags = ((lower_32_bits(flags) | CLONE_VM |
2703 CLONE_UNTRACED) & ~CSIGNAL),
2704 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2710 return copy_process(NULL, 0, node, &args);
2714 * Ok, this is the main fork-routine.
2716 * It copies the process, and if successful kick-starts
2717 * it and waits for it to finish using the VM if required.
2719 * args->exit_signal is expected to be checked for sanity by the caller.
2721 pid_t kernel_clone(struct kernel_clone_args *args)
2723 u64 clone_flags = args->flags;
2724 struct completion vfork;
2726 struct task_struct *p;
2731 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2732 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2733 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2734 * field in struct clone_args and it still doesn't make sense to have
2735 * them both point at the same memory location. Performing this check
2736 * here has the advantage that we don't need to have a separate helper
2737 * to check for legacy clone().
2739 if ((args->flags & CLONE_PIDFD) &&
2740 (args->flags & CLONE_PARENT_SETTID) &&
2741 (args->pidfd == args->parent_tid))
2745 * Determine whether and which event to report to ptracer. When
2746 * called from kernel_thread or CLONE_UNTRACED is explicitly
2747 * requested, no event is reported; otherwise, report if the event
2748 * for the type of forking is enabled.
2750 if (!(clone_flags & CLONE_UNTRACED)) {
2751 if (clone_flags & CLONE_VFORK)
2752 trace = PTRACE_EVENT_VFORK;
2753 else if (args->exit_signal != SIGCHLD)
2754 trace = PTRACE_EVENT_CLONE;
2756 trace = PTRACE_EVENT_FORK;
2758 if (likely(!ptrace_event_enabled(current, trace)))
2762 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2763 add_latent_entropy();
2769 * Do this prior waking up the new thread - the thread pointer
2770 * might get invalid after that point, if the thread exits quickly.
2772 trace_sched_process_fork(current, p);
2774 pid = get_task_pid(p, PIDTYPE_PID);
2777 if (clone_flags & CLONE_PARENT_SETTID)
2778 put_user(nr, args->parent_tid);
2780 if (clone_flags & CLONE_VFORK) {
2781 p->vfork_done = &vfork;
2782 init_completion(&vfork);
2786 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2787 /* lock the task to synchronize with memcg migration */
2789 lru_gen_add_mm(p->mm);
2793 wake_up_new_task(p);
2795 /* forking complete and child started to run, tell ptracer */
2796 if (unlikely(trace))
2797 ptrace_event_pid(trace, pid);
2799 if (clone_flags & CLONE_VFORK) {
2800 if (!wait_for_vfork_done(p, &vfork))
2801 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2809 * Create a kernel thread.
2811 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2813 struct kernel_clone_args args = {
2814 .flags = ((lower_32_bits(flags) | CLONE_VM |
2815 CLONE_UNTRACED) & ~CSIGNAL),
2816 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2822 return kernel_clone(&args);
2826 * Create a user mode thread.
2828 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2830 struct kernel_clone_args args = {
2831 .flags = ((lower_32_bits(flags) | CLONE_VM |
2832 CLONE_UNTRACED) & ~CSIGNAL),
2833 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2838 return kernel_clone(&args);
2841 #ifdef __ARCH_WANT_SYS_FORK
2842 SYSCALL_DEFINE0(fork)
2845 struct kernel_clone_args args = {
2846 .exit_signal = SIGCHLD,
2849 return kernel_clone(&args);
2851 /* can not support in nommu mode */
2857 #ifdef __ARCH_WANT_SYS_VFORK
2858 SYSCALL_DEFINE0(vfork)
2860 struct kernel_clone_args args = {
2861 .flags = CLONE_VFORK | CLONE_VM,
2862 .exit_signal = SIGCHLD,
2865 return kernel_clone(&args);
2869 #ifdef __ARCH_WANT_SYS_CLONE
2870 #ifdef CONFIG_CLONE_BACKWARDS
2871 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2872 int __user *, parent_tidptr,
2874 int __user *, child_tidptr)
2875 #elif defined(CONFIG_CLONE_BACKWARDS2)
2876 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2877 int __user *, parent_tidptr,
2878 int __user *, child_tidptr,
2880 #elif defined(CONFIG_CLONE_BACKWARDS3)
2881 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2883 int __user *, parent_tidptr,
2884 int __user *, child_tidptr,
2887 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2888 int __user *, parent_tidptr,
2889 int __user *, child_tidptr,
2893 struct kernel_clone_args args = {
2894 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2895 .pidfd = parent_tidptr,
2896 .child_tid = child_tidptr,
2897 .parent_tid = parent_tidptr,
2898 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2903 return kernel_clone(&args);
2907 #ifdef __ARCH_WANT_SYS_CLONE3
2909 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2910 struct clone_args __user *uargs,
2914 struct clone_args args;
2915 pid_t *kset_tid = kargs->set_tid;
2917 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2918 CLONE_ARGS_SIZE_VER0);
2919 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2920 CLONE_ARGS_SIZE_VER1);
2921 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2922 CLONE_ARGS_SIZE_VER2);
2923 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2925 if (unlikely(usize > PAGE_SIZE))
2927 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2930 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2934 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2937 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2940 if (unlikely(args.set_tid && args.set_tid_size == 0))
2944 * Verify that higher 32bits of exit_signal are unset and that
2945 * it is a valid signal
2947 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2948 !valid_signal(args.exit_signal)))
2951 if ((args.flags & CLONE_INTO_CGROUP) &&
2952 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2955 *kargs = (struct kernel_clone_args){
2956 .flags = args.flags,
2957 .pidfd = u64_to_user_ptr(args.pidfd),
2958 .child_tid = u64_to_user_ptr(args.child_tid),
2959 .parent_tid = u64_to_user_ptr(args.parent_tid),
2960 .exit_signal = args.exit_signal,
2961 .stack = args.stack,
2962 .stack_size = args.stack_size,
2964 .set_tid_size = args.set_tid_size,
2965 .cgroup = args.cgroup,
2969 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2970 (kargs->set_tid_size * sizeof(pid_t))))
2973 kargs->set_tid = kset_tid;
2979 * clone3_stack_valid - check and prepare stack
2980 * @kargs: kernel clone args
2982 * Verify that the stack arguments userspace gave us are sane.
2983 * In addition, set the stack direction for userspace since it's easy for us to
2986 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2988 if (kargs->stack == 0) {
2989 if (kargs->stack_size > 0)
2992 if (kargs->stack_size == 0)
2995 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2998 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2999 kargs->stack += kargs->stack_size;
3006 static bool clone3_args_valid(struct kernel_clone_args *kargs)
3008 /* Verify that no unknown flags are passed along. */
3010 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3014 * - make the CLONE_DETACHED bit reusable for clone3
3015 * - make the CSIGNAL bits reusable for clone3
3017 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3020 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3021 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3024 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3028 if (!clone3_stack_valid(kargs))
3035 * clone3 - create a new process with specific properties
3036 * @uargs: argument structure
3037 * @size: size of @uargs
3039 * clone3() is the extensible successor to clone()/clone2().
3040 * It takes a struct as argument that is versioned by its size.
3042 * Return: On success, a positive PID for the child process.
3043 * On error, a negative errno number.
3045 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3049 struct kernel_clone_args kargs;
3050 pid_t set_tid[MAX_PID_NS_LEVEL];
3052 kargs.set_tid = set_tid;
3054 err = copy_clone_args_from_user(&kargs, uargs, size);
3058 if (!clone3_args_valid(&kargs))
3061 return kernel_clone(&kargs);
3065 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3067 struct task_struct *leader, *parent, *child;
3070 read_lock(&tasklist_lock);
3071 leader = top = top->group_leader;
3073 for_each_thread(leader, parent) {
3074 list_for_each_entry(child, &parent->children, sibling) {
3075 res = visitor(child, data);
3087 if (leader != top) {
3089 parent = child->real_parent;
3090 leader = parent->group_leader;
3094 read_unlock(&tasklist_lock);
3097 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3098 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3101 static void sighand_ctor(void *data)
3103 struct sighand_struct *sighand = data;
3105 spin_lock_init(&sighand->siglock);
3106 init_waitqueue_head(&sighand->signalfd_wqh);
3109 void __init mm_cache_init(void)
3111 unsigned int mm_size;
3114 * The mm_cpumask is located at the end of mm_struct, and is
3115 * dynamically sized based on the maximum CPU number this system
3116 * can have, taking hotplug into account (nr_cpu_ids).
3118 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3120 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3121 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3122 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3123 offsetof(struct mm_struct, saved_auxv),
3124 sizeof_field(struct mm_struct, saved_auxv),
3128 void __init proc_caches_init(void)
3130 sighand_cachep = kmem_cache_create("sighand_cache",
3131 sizeof(struct sighand_struct), 0,
3132 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3133 SLAB_ACCOUNT, sighand_ctor);
3134 signal_cachep = kmem_cache_create("signal_cache",
3135 sizeof(struct signal_struct), 0,
3136 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3138 files_cachep = kmem_cache_create("files_cache",
3139 sizeof(struct files_struct), 0,
3140 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3142 fs_cachep = kmem_cache_create("fs_cache",
3143 sizeof(struct fs_struct), 0,
3144 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3147 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3149 nsproxy_cache_init();
3153 * Check constraints on flags passed to the unshare system call.
3155 static int check_unshare_flags(unsigned long unshare_flags)
3157 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3158 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3159 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3160 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3164 * Not implemented, but pretend it works if there is nothing
3165 * to unshare. Note that unsharing the address space or the
3166 * signal handlers also need to unshare the signal queues (aka
3169 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3170 if (!thread_group_empty(current))
3173 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3174 if (refcount_read(¤t->sighand->count) > 1)
3177 if (unshare_flags & CLONE_VM) {
3178 if (!current_is_single_threaded())
3186 * Unshare the filesystem structure if it is being shared
3188 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3190 struct fs_struct *fs = current->fs;
3192 if (!(unshare_flags & CLONE_FS) || !fs)
3195 /* don't need lock here; in the worst case we'll do useless copy */
3199 *new_fsp = copy_fs_struct(fs);
3207 * Unshare file descriptor table if it is being shared
3209 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3210 struct files_struct **new_fdp)
3212 struct files_struct *fd = current->files;
3215 if ((unshare_flags & CLONE_FILES) &&
3216 (fd && atomic_read(&fd->count) > 1)) {
3217 *new_fdp = dup_fd(fd, max_fds, &error);
3226 * unshare allows a process to 'unshare' part of the process
3227 * context which was originally shared using clone. copy_*
3228 * functions used by kernel_clone() cannot be used here directly
3229 * because they modify an inactive task_struct that is being
3230 * constructed. Here we are modifying the current, active,
3233 int ksys_unshare(unsigned long unshare_flags)
3235 struct fs_struct *fs, *new_fs = NULL;
3236 struct files_struct *new_fd = NULL;
3237 struct cred *new_cred = NULL;
3238 struct nsproxy *new_nsproxy = NULL;
3243 * If unsharing a user namespace must also unshare the thread group
3244 * and unshare the filesystem root and working directories.
3246 if (unshare_flags & CLONE_NEWUSER)
3247 unshare_flags |= CLONE_THREAD | CLONE_FS;
3249 * If unsharing vm, must also unshare signal handlers.
3251 if (unshare_flags & CLONE_VM)
3252 unshare_flags |= CLONE_SIGHAND;
3254 * If unsharing a signal handlers, must also unshare the signal queues.
3256 if (unshare_flags & CLONE_SIGHAND)
3257 unshare_flags |= CLONE_THREAD;
3259 * If unsharing namespace, must also unshare filesystem information.
3261 if (unshare_flags & CLONE_NEWNS)
3262 unshare_flags |= CLONE_FS;
3264 err = check_unshare_flags(unshare_flags);
3266 goto bad_unshare_out;
3268 * CLONE_NEWIPC must also detach from the undolist: after switching
3269 * to a new ipc namespace, the semaphore arrays from the old
3270 * namespace are unreachable.
3272 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3274 err = unshare_fs(unshare_flags, &new_fs);
3276 goto bad_unshare_out;
3277 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3279 goto bad_unshare_cleanup_fs;
3280 err = unshare_userns(unshare_flags, &new_cred);
3282 goto bad_unshare_cleanup_fd;
3283 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3286 goto bad_unshare_cleanup_cred;
3289 err = set_cred_ucounts(new_cred);
3291 goto bad_unshare_cleanup_cred;
3294 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3297 * CLONE_SYSVSEM is equivalent to sys_exit().
3301 if (unshare_flags & CLONE_NEWIPC) {
3302 /* Orphan segments in old ns (see sem above). */
3304 shm_init_task(current);
3308 switch_task_namespaces(current, new_nsproxy);
3314 spin_lock(&fs->lock);
3315 current->fs = new_fs;
3320 spin_unlock(&fs->lock);
3324 swap(current->files, new_fd);
3326 task_unlock(current);
3329 /* Install the new user namespace */
3330 commit_creds(new_cred);
3335 perf_event_namespaces(current);
3337 bad_unshare_cleanup_cred:
3340 bad_unshare_cleanup_fd:
3342 put_files_struct(new_fd);
3344 bad_unshare_cleanup_fs:
3346 free_fs_struct(new_fs);
3352 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3354 return ksys_unshare(unshare_flags);
3358 * Helper to unshare the files of the current task.
3359 * We don't want to expose copy_files internals to
3360 * the exec layer of the kernel.
3363 int unshare_files(void)
3365 struct task_struct *task = current;
3366 struct files_struct *old, *copy = NULL;
3369 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3377 put_files_struct(old);
3381 int sysctl_max_threads(struct ctl_table *table, int write,
3382 void *buffer, size_t *lenp, loff_t *ppos)
3386 int threads = max_threads;
3388 int max = MAX_THREADS;
3395 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3399 max_threads = threads;