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>
100 #include <linux/user_events.h>
101 #include <linux/iommu.h>
103 #include <asm/pgalloc.h>
104 #include <linux/uaccess.h>
105 #include <asm/mmu_context.h>
106 #include <asm/cacheflush.h>
107 #include <asm/tlbflush.h>
109 #include <trace/events/sched.h>
111 #define CREATE_TRACE_POINTS
112 #include <trace/events/task.h>
115 * Minimum number of threads to boot the kernel
117 #define MIN_THREADS 20
120 * Maximum number of threads
122 #define MAX_THREADS FUTEX_TID_MASK
125 * Protected counters by write_lock_irq(&tasklist_lock)
127 unsigned long total_forks; /* Handle normal Linux uptimes. */
128 int nr_threads; /* The idle threads do not count.. */
130 static int max_threads; /* tunable limit on nr_threads */
132 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
134 static const char * const resident_page_types[] = {
135 NAMED_ARRAY_INDEX(MM_FILEPAGES),
136 NAMED_ARRAY_INDEX(MM_ANONPAGES),
137 NAMED_ARRAY_INDEX(MM_SWAPENTS),
138 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
141 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
143 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
145 #ifdef CONFIG_PROVE_RCU
146 int lockdep_tasklist_lock_is_held(void)
148 return lockdep_is_held(&tasklist_lock);
150 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
151 #endif /* #ifdef CONFIG_PROVE_RCU */
153 int nr_processes(void)
158 for_each_possible_cpu(cpu)
159 total += per_cpu(process_counts, cpu);
164 void __weak arch_release_task_struct(struct task_struct *tsk)
168 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
169 static struct kmem_cache *task_struct_cachep;
171 static inline struct task_struct *alloc_task_struct_node(int node)
173 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
176 static inline void free_task_struct(struct task_struct *tsk)
178 kmem_cache_free(task_struct_cachep, tsk);
182 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
185 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
186 * kmemcache based allocator.
188 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
190 # ifdef CONFIG_VMAP_STACK
192 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
193 * flush. Try to minimize the number of calls by caching stacks.
195 #define NR_CACHED_STACKS 2
196 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
200 struct vm_struct *stack_vm_area;
203 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
207 for (i = 0; i < NR_CACHED_STACKS; i++) {
208 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
215 static void thread_stack_free_rcu(struct rcu_head *rh)
217 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
219 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
225 static void thread_stack_delayed_free(struct task_struct *tsk)
227 struct vm_stack *vm_stack = tsk->stack;
229 vm_stack->stack_vm_area = tsk->stack_vm_area;
230 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
233 static int free_vm_stack_cache(unsigned int cpu)
235 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
238 for (i = 0; i < NR_CACHED_STACKS; i++) {
239 struct vm_struct *vm_stack = cached_vm_stacks[i];
244 vfree(vm_stack->addr);
245 cached_vm_stacks[i] = NULL;
251 static int memcg_charge_kernel_stack(struct vm_struct *vm)
257 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
259 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
260 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
267 for (i = 0; i < nr_charged; i++)
268 memcg_kmem_uncharge_page(vm->pages[i], 0);
272 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
274 struct vm_struct *vm;
278 for (i = 0; i < NR_CACHED_STACKS; i++) {
281 s = this_cpu_xchg(cached_stacks[i], NULL);
286 /* Reset stack metadata. */
287 kasan_unpoison_range(s->addr, THREAD_SIZE);
289 stack = kasan_reset_tag(s->addr);
291 /* Clear stale pointers from reused stack. */
292 memset(stack, 0, THREAD_SIZE);
294 if (memcg_charge_kernel_stack(s)) {
299 tsk->stack_vm_area = s;
305 * Allocated stacks are cached and later reused by new threads,
306 * so memcg accounting is performed manually on assigning/releasing
307 * stacks to tasks. Drop __GFP_ACCOUNT.
309 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
310 VMALLOC_START, VMALLOC_END,
311 THREADINFO_GFP & ~__GFP_ACCOUNT,
313 0, node, __builtin_return_address(0));
317 vm = find_vm_area(stack);
318 if (memcg_charge_kernel_stack(vm)) {
323 * We can't call find_vm_area() in interrupt context, and
324 * free_thread_stack() can be called in interrupt context,
325 * so cache the vm_struct.
327 tsk->stack_vm_area = vm;
328 stack = kasan_reset_tag(stack);
333 static void free_thread_stack(struct task_struct *tsk)
335 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
336 thread_stack_delayed_free(tsk);
339 tsk->stack_vm_area = NULL;
342 # else /* !CONFIG_VMAP_STACK */
344 static void thread_stack_free_rcu(struct rcu_head *rh)
346 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
349 static void thread_stack_delayed_free(struct task_struct *tsk)
351 struct rcu_head *rh = tsk->stack;
353 call_rcu(rh, thread_stack_free_rcu);
356 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
358 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
362 tsk->stack = kasan_reset_tag(page_address(page));
368 static void free_thread_stack(struct task_struct *tsk)
370 thread_stack_delayed_free(tsk);
374 # endif /* CONFIG_VMAP_STACK */
375 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
377 static struct kmem_cache *thread_stack_cache;
379 static void thread_stack_free_rcu(struct rcu_head *rh)
381 kmem_cache_free(thread_stack_cache, rh);
384 static void thread_stack_delayed_free(struct task_struct *tsk)
386 struct rcu_head *rh = tsk->stack;
388 call_rcu(rh, thread_stack_free_rcu);
391 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
393 unsigned long *stack;
394 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
395 stack = kasan_reset_tag(stack);
397 return stack ? 0 : -ENOMEM;
400 static void free_thread_stack(struct task_struct *tsk)
402 thread_stack_delayed_free(tsk);
406 void thread_stack_cache_init(void)
408 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
409 THREAD_SIZE, THREAD_SIZE, 0, 0,
411 BUG_ON(thread_stack_cache == NULL);
414 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
415 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
417 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
419 unsigned long *stack;
421 stack = arch_alloc_thread_stack_node(tsk, node);
423 return stack ? 0 : -ENOMEM;
426 static void free_thread_stack(struct task_struct *tsk)
428 arch_free_thread_stack(tsk);
432 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
434 /* SLAB cache for signal_struct structures (tsk->signal) */
435 static struct kmem_cache *signal_cachep;
437 /* SLAB cache for sighand_struct structures (tsk->sighand) */
438 struct kmem_cache *sighand_cachep;
440 /* SLAB cache for files_struct structures (tsk->files) */
441 struct kmem_cache *files_cachep;
443 /* SLAB cache for fs_struct structures (tsk->fs) */
444 struct kmem_cache *fs_cachep;
446 /* SLAB cache for vm_area_struct structures */
447 static struct kmem_cache *vm_area_cachep;
449 /* SLAB cache for mm_struct structures (tsk->mm) */
450 static struct kmem_cache *mm_cachep;
452 #ifdef CONFIG_PER_VMA_LOCK
454 /* SLAB cache for vm_area_struct.lock */
455 static struct kmem_cache *vma_lock_cachep;
457 static bool vma_lock_alloc(struct vm_area_struct *vma)
459 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
463 init_rwsem(&vma->vm_lock->lock);
464 vma->vm_lock_seq = -1;
469 static inline void vma_lock_free(struct vm_area_struct *vma)
471 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
474 #else /* CONFIG_PER_VMA_LOCK */
476 static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
477 static inline void vma_lock_free(struct vm_area_struct *vma) {}
479 #endif /* CONFIG_PER_VMA_LOCK */
481 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
483 struct vm_area_struct *vma;
485 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
490 if (!vma_lock_alloc(vma)) {
491 kmem_cache_free(vm_area_cachep, vma);
498 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
500 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
505 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
506 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
508 * orig->shared.rb may be modified concurrently, but the clone
509 * will be reinitialized.
511 data_race(memcpy(new, orig, sizeof(*new)));
512 if (!vma_lock_alloc(new)) {
513 kmem_cache_free(vm_area_cachep, new);
516 INIT_LIST_HEAD(&new->anon_vma_chain);
517 vma_numab_state_init(new);
518 dup_anon_vma_name(orig, new);
523 void __vm_area_free(struct vm_area_struct *vma)
525 vma_numab_state_free(vma);
526 free_anon_vma_name(vma);
528 kmem_cache_free(vm_area_cachep, vma);
531 #ifdef CONFIG_PER_VMA_LOCK
532 static void vm_area_free_rcu_cb(struct rcu_head *head)
534 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
537 /* The vma should not be locked while being destroyed. */
538 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
543 void vm_area_free(struct vm_area_struct *vma)
545 #ifdef CONFIG_PER_VMA_LOCK
546 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
552 static void account_kernel_stack(struct task_struct *tsk, int account)
554 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
555 struct vm_struct *vm = task_stack_vm_area(tsk);
558 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
559 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
560 account * (PAGE_SIZE / 1024));
562 void *stack = task_stack_page(tsk);
564 /* All stack pages are in the same node. */
565 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
566 account * (THREAD_SIZE / 1024));
570 void exit_task_stack_account(struct task_struct *tsk)
572 account_kernel_stack(tsk, -1);
574 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
575 struct vm_struct *vm;
578 vm = task_stack_vm_area(tsk);
579 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
580 memcg_kmem_uncharge_page(vm->pages[i], 0);
584 static void release_task_stack(struct task_struct *tsk)
586 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
587 return; /* Better to leak the stack than to free prematurely */
589 free_thread_stack(tsk);
592 #ifdef CONFIG_THREAD_INFO_IN_TASK
593 void put_task_stack(struct task_struct *tsk)
595 if (refcount_dec_and_test(&tsk->stack_refcount))
596 release_task_stack(tsk);
600 void free_task(struct task_struct *tsk)
602 #ifdef CONFIG_SECCOMP
603 WARN_ON_ONCE(tsk->seccomp.filter);
605 release_user_cpus_ptr(tsk);
608 #ifndef CONFIG_THREAD_INFO_IN_TASK
610 * The task is finally done with both the stack and thread_info,
613 release_task_stack(tsk);
616 * If the task had a separate stack allocation, it should be gone
619 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
621 rt_mutex_debug_task_free(tsk);
622 ftrace_graph_exit_task(tsk);
623 arch_release_task_struct(tsk);
624 if (tsk->flags & PF_KTHREAD)
625 free_kthread_struct(tsk);
626 bpf_task_storage_free(tsk);
627 free_task_struct(tsk);
629 EXPORT_SYMBOL(free_task);
631 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
633 struct file *exe_file;
635 exe_file = get_mm_exe_file(oldmm);
636 RCU_INIT_POINTER(mm->exe_file, exe_file);
638 * We depend on the oldmm having properly denied write access to the
641 if (exe_file && deny_write_access(exe_file))
642 pr_warn_once("deny_write_access() failed in %s\n", __func__);
646 static __latent_entropy int dup_mmap(struct mm_struct *mm,
647 struct mm_struct *oldmm)
649 struct vm_area_struct *mpnt, *tmp;
651 unsigned long charge = 0;
653 VMA_ITERATOR(old_vmi, oldmm, 0);
654 VMA_ITERATOR(vmi, mm, 0);
656 uprobe_start_dup_mmap();
657 if (mmap_write_lock_killable(oldmm)) {
659 goto fail_uprobe_end;
661 flush_cache_dup_mm(oldmm);
662 uprobe_dup_mmap(oldmm, mm);
664 * Not linked in yet - no deadlock potential:
666 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
668 /* No ordering required: file already has been exposed. */
669 dup_mm_exe_file(mm, oldmm);
671 mm->total_vm = oldmm->total_vm;
672 mm->data_vm = oldmm->data_vm;
673 mm->exec_vm = oldmm->exec_vm;
674 mm->stack_vm = oldmm->stack_vm;
676 retval = ksm_fork(mm, oldmm);
679 khugepaged_fork(mm, oldmm);
681 retval = vma_iter_bulk_alloc(&vmi, oldmm->map_count);
685 mt_clear_in_rcu(vmi.mas.tree);
686 for_each_vma(old_vmi, mpnt) {
689 vma_start_write(mpnt);
690 if (mpnt->vm_flags & VM_DONTCOPY) {
691 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
696 * Don't duplicate many vmas if we've been oom-killed (for
699 if (fatal_signal_pending(current)) {
703 if (mpnt->vm_flags & VM_ACCOUNT) {
704 unsigned long len = vma_pages(mpnt);
706 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
710 tmp = vm_area_dup(mpnt);
713 retval = vma_dup_policy(mpnt, tmp);
715 goto fail_nomem_policy;
717 retval = dup_userfaultfd(tmp, &uf);
719 goto fail_nomem_anon_vma_fork;
720 if (tmp->vm_flags & VM_WIPEONFORK) {
722 * VM_WIPEONFORK gets a clean slate in the child.
723 * Don't prepare anon_vma until fault since we don't
724 * copy page for current vma.
726 tmp->anon_vma = NULL;
727 } else if (anon_vma_fork(tmp, mpnt))
728 goto fail_nomem_anon_vma_fork;
729 vm_flags_clear(tmp, VM_LOCKED_MASK);
732 struct address_space *mapping = file->f_mapping;
735 i_mmap_lock_write(mapping);
736 if (tmp->vm_flags & VM_SHARED)
737 mapping_allow_writable(mapping);
738 flush_dcache_mmap_lock(mapping);
739 /* insert tmp into the share list, just after mpnt */
740 vma_interval_tree_insert_after(tmp, mpnt,
742 flush_dcache_mmap_unlock(mapping);
743 i_mmap_unlock_write(mapping);
747 * Copy/update hugetlb private vma information.
749 if (is_vm_hugetlb_page(tmp))
750 hugetlb_dup_vma_private(tmp);
752 /* Link the vma into the MT */
753 if (vma_iter_bulk_store(&vmi, tmp))
754 goto fail_nomem_vmi_store;
757 if (!(tmp->vm_flags & VM_WIPEONFORK))
758 retval = copy_page_range(tmp, mpnt);
760 if (tmp->vm_ops && tmp->vm_ops->open)
761 tmp->vm_ops->open(tmp);
766 /* a new mm has just been created */
767 retval = arch_dup_mmap(oldmm, mm);
771 mt_set_in_rcu(vmi.mas.tree);
773 mmap_write_unlock(mm);
775 mmap_write_unlock(oldmm);
776 dup_userfaultfd_complete(&uf);
778 uprobe_end_dup_mmap();
781 fail_nomem_vmi_store:
782 unlink_anon_vmas(tmp);
783 fail_nomem_anon_vma_fork:
784 mpol_put(vma_policy(tmp));
789 vm_unacct_memory(charge);
793 static inline int mm_alloc_pgd(struct mm_struct *mm)
795 mm->pgd = pgd_alloc(mm);
796 if (unlikely(!mm->pgd))
801 static inline void mm_free_pgd(struct mm_struct *mm)
803 pgd_free(mm, mm->pgd);
806 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
808 mmap_write_lock(oldmm);
809 dup_mm_exe_file(mm, oldmm);
810 mmap_write_unlock(oldmm);
813 #define mm_alloc_pgd(mm) (0)
814 #define mm_free_pgd(mm)
815 #endif /* CONFIG_MMU */
817 static void check_mm(struct mm_struct *mm)
821 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
822 "Please make sure 'struct resident_page_types[]' is updated as well");
824 for (i = 0; i < NR_MM_COUNTERS; i++) {
825 long x = percpu_counter_sum(&mm->rss_stat[i]);
828 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
829 mm, resident_page_types[i], x);
832 if (mm_pgtables_bytes(mm))
833 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
834 mm_pgtables_bytes(mm));
836 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
837 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
841 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
842 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
844 static void do_check_lazy_tlb(void *arg)
846 struct mm_struct *mm = arg;
848 WARN_ON_ONCE(current->active_mm == mm);
851 static void do_shoot_lazy_tlb(void *arg)
853 struct mm_struct *mm = arg;
855 if (current->active_mm == mm) {
856 WARN_ON_ONCE(current->mm);
857 current->active_mm = &init_mm;
858 switch_mm(mm, &init_mm, current);
862 static void cleanup_lazy_tlbs(struct mm_struct *mm)
864 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
866 * In this case, lazy tlb mms are refounted and would not reach
867 * __mmdrop until all CPUs have switched away and mmdrop()ed.
873 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
874 * requires lazy mm users to switch to another mm when the refcount
875 * drops to zero, before the mm is freed. This requires IPIs here to
876 * switch kernel threads to init_mm.
878 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
879 * switch with the final userspace teardown TLB flush which leaves the
880 * mm lazy on this CPU but no others, reducing the need for additional
881 * IPIs here. There are cases where a final IPI is still required here,
882 * such as the final mmdrop being performed on a different CPU than the
883 * one exiting, or kernel threads using the mm when userspace exits.
885 * IPI overheads have not found to be expensive, but they could be
886 * reduced in a number of possible ways, for example (roughly
887 * increasing order of complexity):
888 * - The last lazy reference created by exit_mm() could instead switch
889 * to init_mm, however it's probable this will run on the same CPU
890 * immediately afterwards, so this may not reduce IPIs much.
891 * - A batch of mms requiring IPIs could be gathered and freed at once.
892 * - CPUs store active_mm where it can be remotely checked without a
893 * lock, to filter out false-positives in the cpumask.
894 * - After mm_users or mm_count reaches zero, switching away from the
895 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
896 * with some batching or delaying of the final IPIs.
897 * - A delayed freeing and RCU-like quiescing sequence based on mm
898 * switching to avoid IPIs completely.
900 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
901 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
902 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
906 * Called when the last reference to the mm
907 * is dropped: either by a lazy thread or by
908 * mmput. Free the page directory and the mm.
910 void __mmdrop(struct mm_struct *mm)
912 BUG_ON(mm == &init_mm);
913 WARN_ON_ONCE(mm == current->mm);
915 /* Ensure no CPUs are using this as their lazy tlb mm */
916 cleanup_lazy_tlbs(mm);
918 WARN_ON_ONCE(mm == current->active_mm);
921 mmu_notifier_subscriptions_destroy(mm);
923 put_user_ns(mm->user_ns);
926 percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
930 EXPORT_SYMBOL_GPL(__mmdrop);
932 static void mmdrop_async_fn(struct work_struct *work)
934 struct mm_struct *mm;
936 mm = container_of(work, struct mm_struct, async_put_work);
940 static void mmdrop_async(struct mm_struct *mm)
942 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
943 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
944 schedule_work(&mm->async_put_work);
948 static inline void free_signal_struct(struct signal_struct *sig)
950 taskstats_tgid_free(sig);
951 sched_autogroup_exit(sig);
953 * __mmdrop is not safe to call from softirq context on x86 due to
954 * pgd_dtor so postpone it to the async context
957 mmdrop_async(sig->oom_mm);
958 kmem_cache_free(signal_cachep, sig);
961 static inline void put_signal_struct(struct signal_struct *sig)
963 if (refcount_dec_and_test(&sig->sigcnt))
964 free_signal_struct(sig);
967 void __put_task_struct(struct task_struct *tsk)
969 WARN_ON(!tsk->exit_state);
970 WARN_ON(refcount_read(&tsk->usage));
971 WARN_ON(tsk == current);
975 task_numa_free(tsk, true);
976 security_task_free(tsk);
978 delayacct_tsk_free(tsk);
979 put_signal_struct(tsk->signal);
980 sched_core_free(tsk);
983 EXPORT_SYMBOL_GPL(__put_task_struct);
985 void __put_task_struct_rcu_cb(struct rcu_head *rhp)
987 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
989 __put_task_struct(task);
991 EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
993 void __init __weak arch_task_cache_init(void) { }
998 static void set_max_threads(unsigned int max_threads_suggested)
1001 unsigned long nr_pages = totalram_pages();
1004 * The number of threads shall be limited such that the thread
1005 * structures may only consume a small part of the available memory.
1007 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1008 threads = MAX_THREADS;
1010 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1011 (u64) THREAD_SIZE * 8UL);
1013 if (threads > max_threads_suggested)
1014 threads = max_threads_suggested;
1016 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1019 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1020 /* Initialized by the architecture: */
1021 int arch_task_struct_size __read_mostly;
1024 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1025 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1027 /* Fetch thread_struct whitelist for the architecture. */
1028 arch_thread_struct_whitelist(offset, size);
1031 * Handle zero-sized whitelist or empty thread_struct, otherwise
1032 * adjust offset to position of thread_struct in task_struct.
1034 if (unlikely(*size == 0))
1037 *offset += offsetof(struct task_struct, thread);
1039 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
1041 void __init fork_init(void)
1044 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1045 #ifndef ARCH_MIN_TASKALIGN
1046 #define ARCH_MIN_TASKALIGN 0
1048 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1049 unsigned long useroffset, usersize;
1051 /* create a slab on which task_structs can be allocated */
1052 task_struct_whitelist(&useroffset, &usersize);
1053 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1054 arch_task_struct_size, align,
1055 SLAB_PANIC|SLAB_ACCOUNT,
1056 useroffset, usersize, NULL);
1059 /* do the arch specific task caches init */
1060 arch_task_cache_init();
1062 set_max_threads(MAX_THREADS);
1064 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1065 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1066 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1067 init_task.signal->rlim[RLIMIT_NPROC];
1069 for (i = 0; i < UCOUNT_COUNTS; i++)
1070 init_user_ns.ucount_max[i] = max_threads/2;
1072 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1073 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1074 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1075 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1077 #ifdef CONFIG_VMAP_STACK
1078 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1079 NULL, free_vm_stack_cache);
1084 lockdep_init_task(&init_task);
1088 int __weak arch_dup_task_struct(struct task_struct *dst,
1089 struct task_struct *src)
1095 void set_task_stack_end_magic(struct task_struct *tsk)
1097 unsigned long *stackend;
1099 stackend = end_of_stack(tsk);
1100 *stackend = STACK_END_MAGIC; /* for overflow detection */
1103 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1105 struct task_struct *tsk;
1108 if (node == NUMA_NO_NODE)
1109 node = tsk_fork_get_node(orig);
1110 tsk = alloc_task_struct_node(node);
1114 err = arch_dup_task_struct(tsk, orig);
1118 err = alloc_thread_stack_node(tsk, node);
1122 #ifdef CONFIG_THREAD_INFO_IN_TASK
1123 refcount_set(&tsk->stack_refcount, 1);
1125 account_kernel_stack(tsk, 1);
1127 err = scs_prepare(tsk, node);
1131 #ifdef CONFIG_SECCOMP
1133 * We must handle setting up seccomp filters once we're under
1134 * the sighand lock in case orig has changed between now and
1135 * then. Until then, filter must be NULL to avoid messing up
1136 * the usage counts on the error path calling free_task.
1138 tsk->seccomp.filter = NULL;
1141 setup_thread_stack(tsk, orig);
1142 clear_user_return_notifier(tsk);
1143 clear_tsk_need_resched(tsk);
1144 set_task_stack_end_magic(tsk);
1145 clear_syscall_work_syscall_user_dispatch(tsk);
1147 #ifdef CONFIG_STACKPROTECTOR
1148 tsk->stack_canary = get_random_canary();
1150 if (orig->cpus_ptr == &orig->cpus_mask)
1151 tsk->cpus_ptr = &tsk->cpus_mask;
1152 dup_user_cpus_ptr(tsk, orig, node);
1155 * One for the user space visible state that goes away when reaped.
1156 * One for the scheduler.
1158 refcount_set(&tsk->rcu_users, 2);
1159 /* One for the rcu users */
1160 refcount_set(&tsk->usage, 1);
1161 #ifdef CONFIG_BLK_DEV_IO_TRACE
1162 tsk->btrace_seq = 0;
1164 tsk->splice_pipe = NULL;
1165 tsk->task_frag.page = NULL;
1166 tsk->wake_q.next = NULL;
1167 tsk->worker_private = NULL;
1169 kcov_task_init(tsk);
1170 kmsan_task_create(tsk);
1171 kmap_local_fork(tsk);
1173 #ifdef CONFIG_FAULT_INJECTION
1177 #ifdef CONFIG_BLK_CGROUP
1178 tsk->throttle_disk = NULL;
1179 tsk->use_memdelay = 0;
1182 #ifdef CONFIG_IOMMU_SVA
1183 tsk->pasid_activated = 0;
1187 tsk->active_memcg = NULL;
1190 #ifdef CONFIG_CPU_SUP_INTEL
1191 tsk->reported_split_lock = 0;
1194 #ifdef CONFIG_SCHED_MM_CID
1196 tsk->last_mm_cid = -1;
1197 tsk->mm_cid_active = 0;
1198 tsk->migrate_from_cpu = -1;
1203 exit_task_stack_account(tsk);
1204 free_thread_stack(tsk);
1206 free_task_struct(tsk);
1210 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1212 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1214 static int __init coredump_filter_setup(char *s)
1216 default_dump_filter =
1217 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1218 MMF_DUMP_FILTER_MASK;
1222 __setup("coredump_filter=", coredump_filter_setup);
1224 #include <linux/init_task.h>
1226 static void mm_init_aio(struct mm_struct *mm)
1229 spin_lock_init(&mm->ioctx_lock);
1230 mm->ioctx_table = NULL;
1234 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1235 struct task_struct *p)
1239 WRITE_ONCE(mm->owner, NULL);
1243 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1250 static void mm_init_uprobes_state(struct mm_struct *mm)
1252 #ifdef CONFIG_UPROBES
1253 mm->uprobes_state.xol_area = NULL;
1257 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1258 struct user_namespace *user_ns)
1260 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1261 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1262 atomic_set(&mm->mm_users, 1);
1263 atomic_set(&mm->mm_count, 1);
1264 seqcount_init(&mm->write_protect_seq);
1266 INIT_LIST_HEAD(&mm->mmlist);
1267 #ifdef CONFIG_PER_VMA_LOCK
1268 mm->mm_lock_seq = 0;
1270 mm_pgtables_bytes_init(mm);
1273 atomic64_set(&mm->pinned_vm, 0);
1274 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1275 spin_lock_init(&mm->page_table_lock);
1276 spin_lock_init(&mm->arg_lock);
1277 mm_init_cpumask(mm);
1279 mm_init_owner(mm, p);
1281 RCU_INIT_POINTER(mm->exe_file, NULL);
1282 mmu_notifier_subscriptions_init(mm);
1283 init_tlb_flush_pending(mm);
1284 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1285 mm->pmd_huge_pte = NULL;
1287 mm_init_uprobes_state(mm);
1288 hugetlb_count_init(mm);
1291 mm->flags = mmf_init_flags(current->mm->flags);
1292 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1294 mm->flags = default_dump_filter;
1298 if (mm_alloc_pgd(mm))
1301 if (init_new_context(p, mm))
1302 goto fail_nocontext;
1304 if (mm_alloc_cid(mm))
1307 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1311 mm->user_ns = get_user_ns(user_ns);
1312 lru_gen_init_mm(mm);
1318 destroy_context(mm);
1327 * Allocate and initialize an mm_struct.
1329 struct mm_struct *mm_alloc(void)
1331 struct mm_struct *mm;
1337 memset(mm, 0, sizeof(*mm));
1338 return mm_init(mm, current, current_user_ns());
1341 static inline void __mmput(struct mm_struct *mm)
1343 VM_BUG_ON(atomic_read(&mm->mm_users));
1345 uprobe_clear_state(mm);
1348 khugepaged_exit(mm); /* must run before exit_mmap */
1350 mm_put_huge_zero_page(mm);
1351 set_mm_exe_file(mm, NULL);
1352 if (!list_empty(&mm->mmlist)) {
1353 spin_lock(&mmlist_lock);
1354 list_del(&mm->mmlist);
1355 spin_unlock(&mmlist_lock);
1358 module_put(mm->binfmt->module);
1364 * Decrement the use count and release all resources for an mm.
1366 void mmput(struct mm_struct *mm)
1370 if (atomic_dec_and_test(&mm->mm_users))
1373 EXPORT_SYMBOL_GPL(mmput);
1376 static void mmput_async_fn(struct work_struct *work)
1378 struct mm_struct *mm = container_of(work, struct mm_struct,
1384 void mmput_async(struct mm_struct *mm)
1386 if (atomic_dec_and_test(&mm->mm_users)) {
1387 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1388 schedule_work(&mm->async_put_work);
1391 EXPORT_SYMBOL_GPL(mmput_async);
1395 * set_mm_exe_file - change a reference to the mm's executable file
1397 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1399 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1400 * invocations: in mmput() nobody alive left, in execve it happens before
1401 * the new mm is made visible to anyone.
1403 * Can only fail if new_exe_file != NULL.
1405 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1407 struct file *old_exe_file;
1410 * It is safe to dereference the exe_file without RCU as
1411 * this function is only called if nobody else can access
1412 * this mm -- see comment above for justification.
1414 old_exe_file = rcu_dereference_raw(mm->exe_file);
1418 * We expect the caller (i.e., sys_execve) to already denied
1419 * write access, so this is unlikely to fail.
1421 if (unlikely(deny_write_access(new_exe_file)))
1423 get_file(new_exe_file);
1425 rcu_assign_pointer(mm->exe_file, new_exe_file);
1427 allow_write_access(old_exe_file);
1434 * replace_mm_exe_file - replace a reference to the mm's executable file
1436 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1438 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1440 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1442 struct vm_area_struct *vma;
1443 struct file *old_exe_file;
1446 /* Forbid mm->exe_file change if old file still mapped. */
1447 old_exe_file = get_mm_exe_file(mm);
1449 VMA_ITERATOR(vmi, mm, 0);
1451 for_each_vma(vmi, vma) {
1454 if (path_equal(&vma->vm_file->f_path,
1455 &old_exe_file->f_path)) {
1460 mmap_read_unlock(mm);
1466 ret = deny_write_access(new_exe_file);
1469 get_file(new_exe_file);
1471 /* set the new file */
1472 mmap_write_lock(mm);
1473 old_exe_file = rcu_dereference_raw(mm->exe_file);
1474 rcu_assign_pointer(mm->exe_file, new_exe_file);
1475 mmap_write_unlock(mm);
1478 allow_write_access(old_exe_file);
1485 * get_mm_exe_file - acquire a reference to the mm's executable file
1487 * Returns %NULL if mm has no associated executable file.
1488 * User must release file via fput().
1490 struct file *get_mm_exe_file(struct mm_struct *mm)
1492 struct file *exe_file;
1495 exe_file = rcu_dereference(mm->exe_file);
1496 if (exe_file && !get_file_rcu(exe_file))
1503 * get_task_exe_file - acquire a reference to the task's executable file
1505 * Returns %NULL if task's mm (if any) has no associated executable file or
1506 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1507 * User must release file via fput().
1509 struct file *get_task_exe_file(struct task_struct *task)
1511 struct file *exe_file = NULL;
1512 struct mm_struct *mm;
1517 if (!(task->flags & PF_KTHREAD))
1518 exe_file = get_mm_exe_file(mm);
1525 * get_task_mm - acquire a reference to the task's mm
1527 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1528 * this kernel workthread has transiently adopted a user mm with use_mm,
1529 * to do its AIO) is not set and if so returns a reference to it, after
1530 * bumping up the use count. User must release the mm via mmput()
1531 * after use. Typically used by /proc and ptrace.
1533 struct mm_struct *get_task_mm(struct task_struct *task)
1535 struct mm_struct *mm;
1540 if (task->flags & PF_KTHREAD)
1548 EXPORT_SYMBOL_GPL(get_task_mm);
1550 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1552 struct mm_struct *mm;
1555 err = down_read_killable(&task->signal->exec_update_lock);
1557 return ERR_PTR(err);
1559 mm = get_task_mm(task);
1560 if (mm && mm != current->mm &&
1561 !ptrace_may_access(task, mode)) {
1563 mm = ERR_PTR(-EACCES);
1565 up_read(&task->signal->exec_update_lock);
1570 static void complete_vfork_done(struct task_struct *tsk)
1572 struct completion *vfork;
1575 vfork = tsk->vfork_done;
1576 if (likely(vfork)) {
1577 tsk->vfork_done = NULL;
1583 static int wait_for_vfork_done(struct task_struct *child,
1584 struct completion *vfork)
1586 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1589 cgroup_enter_frozen();
1590 killed = wait_for_completion_state(vfork, state);
1591 cgroup_leave_frozen(false);
1595 child->vfork_done = NULL;
1599 put_task_struct(child);
1603 /* Please note the differences between mmput and mm_release.
1604 * mmput is called whenever we stop holding onto a mm_struct,
1605 * error success whatever.
1607 * mm_release is called after a mm_struct has been removed
1608 * from the current process.
1610 * This difference is important for error handling, when we
1611 * only half set up a mm_struct for a new process and need to restore
1612 * the old one. Because we mmput the new mm_struct before
1613 * restoring the old one. . .
1614 * Eric Biederman 10 January 1998
1616 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1618 uprobe_free_utask(tsk);
1620 /* Get rid of any cached register state */
1621 deactivate_mm(tsk, mm);
1624 * Signal userspace if we're not exiting with a core dump
1625 * because we want to leave the value intact for debugging
1628 if (tsk->clear_child_tid) {
1629 if (atomic_read(&mm->mm_users) > 1) {
1631 * We don't check the error code - if userspace has
1632 * not set up a proper pointer then tough luck.
1634 put_user(0, tsk->clear_child_tid);
1635 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1636 1, NULL, NULL, 0, 0);
1638 tsk->clear_child_tid = NULL;
1642 * All done, finally we can wake up parent and return this mm to him.
1643 * Also kthread_stop() uses this completion for synchronization.
1645 if (tsk->vfork_done)
1646 complete_vfork_done(tsk);
1649 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1651 futex_exit_release(tsk);
1652 mm_release(tsk, mm);
1655 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1657 futex_exec_release(tsk);
1658 mm_release(tsk, mm);
1662 * dup_mm() - duplicates an existing mm structure
1663 * @tsk: the task_struct with which the new mm will be associated.
1664 * @oldmm: the mm to duplicate.
1666 * Allocates a new mm structure and duplicates the provided @oldmm structure
1669 * Return: the duplicated mm or NULL on failure.
1671 static struct mm_struct *dup_mm(struct task_struct *tsk,
1672 struct mm_struct *oldmm)
1674 struct mm_struct *mm;
1681 memcpy(mm, oldmm, sizeof(*mm));
1683 if (!mm_init(mm, tsk, mm->user_ns))
1686 err = dup_mmap(mm, oldmm);
1690 mm->hiwater_rss = get_mm_rss(mm);
1691 mm->hiwater_vm = mm->total_vm;
1693 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1699 /* don't put binfmt in mmput, we haven't got module yet */
1701 mm_init_owner(mm, NULL);
1708 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1710 struct mm_struct *mm, *oldmm;
1712 tsk->min_flt = tsk->maj_flt = 0;
1713 tsk->nvcsw = tsk->nivcsw = 0;
1714 #ifdef CONFIG_DETECT_HUNG_TASK
1715 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1716 tsk->last_switch_time = 0;
1720 tsk->active_mm = NULL;
1723 * Are we cloning a kernel thread?
1725 * We need to steal a active VM for that..
1727 oldmm = current->mm;
1731 if (clone_flags & CLONE_VM) {
1735 mm = dup_mm(tsk, current->mm);
1741 tsk->active_mm = mm;
1742 sched_mm_cid_fork(tsk);
1746 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1748 struct fs_struct *fs = current->fs;
1749 if (clone_flags & CLONE_FS) {
1750 /* tsk->fs is already what we want */
1751 spin_lock(&fs->lock);
1753 spin_unlock(&fs->lock);
1757 spin_unlock(&fs->lock);
1760 tsk->fs = copy_fs_struct(fs);
1766 static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1769 struct files_struct *oldf, *newf;
1773 * A background process may not have any files ...
1775 oldf = current->files;
1784 if (clone_flags & CLONE_FILES) {
1785 atomic_inc(&oldf->count);
1789 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1799 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1801 struct sighand_struct *sig;
1803 if (clone_flags & CLONE_SIGHAND) {
1804 refcount_inc(¤t->sighand->count);
1807 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1808 RCU_INIT_POINTER(tsk->sighand, sig);
1812 refcount_set(&sig->count, 1);
1813 spin_lock_irq(¤t->sighand->siglock);
1814 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1815 spin_unlock_irq(¤t->sighand->siglock);
1817 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1818 if (clone_flags & CLONE_CLEAR_SIGHAND)
1819 flush_signal_handlers(tsk, 0);
1824 void __cleanup_sighand(struct sighand_struct *sighand)
1826 if (refcount_dec_and_test(&sighand->count)) {
1827 signalfd_cleanup(sighand);
1829 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1830 * without an RCU grace period, see __lock_task_sighand().
1832 kmem_cache_free(sighand_cachep, sighand);
1837 * Initialize POSIX timer handling for a thread group.
1839 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1841 struct posix_cputimers *pct = &sig->posix_cputimers;
1842 unsigned long cpu_limit;
1844 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1845 posix_cputimers_group_init(pct, cpu_limit);
1848 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1850 struct signal_struct *sig;
1852 if (clone_flags & CLONE_THREAD)
1855 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1860 sig->nr_threads = 1;
1861 sig->quick_threads = 1;
1862 atomic_set(&sig->live, 1);
1863 refcount_set(&sig->sigcnt, 1);
1865 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1866 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1867 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1869 init_waitqueue_head(&sig->wait_chldexit);
1870 sig->curr_target = tsk;
1871 init_sigpending(&sig->shared_pending);
1872 INIT_HLIST_HEAD(&sig->multiprocess);
1873 seqlock_init(&sig->stats_lock);
1874 prev_cputime_init(&sig->prev_cputime);
1876 #ifdef CONFIG_POSIX_TIMERS
1877 INIT_LIST_HEAD(&sig->posix_timers);
1878 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1879 sig->real_timer.function = it_real_fn;
1882 task_lock(current->group_leader);
1883 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1884 task_unlock(current->group_leader);
1886 posix_cpu_timers_init_group(sig);
1888 tty_audit_fork(sig);
1889 sched_autogroup_fork(sig);
1891 sig->oom_score_adj = current->signal->oom_score_adj;
1892 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1894 mutex_init(&sig->cred_guard_mutex);
1895 init_rwsem(&sig->exec_update_lock);
1900 static void copy_seccomp(struct task_struct *p)
1902 #ifdef CONFIG_SECCOMP
1904 * Must be called with sighand->lock held, which is common to
1905 * all threads in the group. Holding cred_guard_mutex is not
1906 * needed because this new task is not yet running and cannot
1909 assert_spin_locked(¤t->sighand->siglock);
1911 /* Ref-count the new filter user, and assign it. */
1912 get_seccomp_filter(current);
1913 p->seccomp = current->seccomp;
1916 * Explicitly enable no_new_privs here in case it got set
1917 * between the task_struct being duplicated and holding the
1918 * sighand lock. The seccomp state and nnp must be in sync.
1920 if (task_no_new_privs(current))
1921 task_set_no_new_privs(p);
1924 * If the parent gained a seccomp mode after copying thread
1925 * flags and between before we held the sighand lock, we have
1926 * to manually enable the seccomp thread flag here.
1928 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1929 set_task_syscall_work(p, SECCOMP);
1933 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1935 current->clear_child_tid = tidptr;
1937 return task_pid_vnr(current);
1940 static void rt_mutex_init_task(struct task_struct *p)
1942 raw_spin_lock_init(&p->pi_lock);
1943 #ifdef CONFIG_RT_MUTEXES
1944 p->pi_waiters = RB_ROOT_CACHED;
1945 p->pi_top_task = NULL;
1946 p->pi_blocked_on = NULL;
1950 static inline void init_task_pid_links(struct task_struct *task)
1954 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1955 INIT_HLIST_NODE(&task->pid_links[type]);
1959 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1961 if (type == PIDTYPE_PID)
1962 task->thread_pid = pid;
1964 task->signal->pids[type] = pid;
1967 static inline void rcu_copy_process(struct task_struct *p)
1969 #ifdef CONFIG_PREEMPT_RCU
1970 p->rcu_read_lock_nesting = 0;
1971 p->rcu_read_unlock_special.s = 0;
1972 p->rcu_blocked_node = NULL;
1973 INIT_LIST_HEAD(&p->rcu_node_entry);
1974 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1975 #ifdef CONFIG_TASKS_RCU
1976 p->rcu_tasks_holdout = false;
1977 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1978 p->rcu_tasks_idle_cpu = -1;
1979 #endif /* #ifdef CONFIG_TASKS_RCU */
1980 #ifdef CONFIG_TASKS_TRACE_RCU
1981 p->trc_reader_nesting = 0;
1982 p->trc_reader_special.s = 0;
1983 INIT_LIST_HEAD(&p->trc_holdout_list);
1984 INIT_LIST_HEAD(&p->trc_blkd_node);
1985 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1988 struct pid *pidfd_pid(const struct file *file)
1990 if (file->f_op == &pidfd_fops)
1991 return file->private_data;
1993 return ERR_PTR(-EBADF);
1996 static int pidfd_release(struct inode *inode, struct file *file)
1998 struct pid *pid = file->private_data;
2000 file->private_data = NULL;
2005 #ifdef CONFIG_PROC_FS
2007 * pidfd_show_fdinfo - print information about a pidfd
2008 * @m: proc fdinfo file
2009 * @f: file referencing a pidfd
2012 * This function will print the pid that a given pidfd refers to in the
2013 * pid namespace of the procfs instance.
2014 * If the pid namespace of the process is not a descendant of the pid
2015 * namespace of the procfs instance 0 will be shown as its pid. This is
2016 * similar to calling getppid() on a process whose parent is outside of
2017 * its pid namespace.
2020 * If pid namespaces are supported then this function will also print
2021 * the pid of a given pidfd refers to for all descendant pid namespaces
2022 * starting from the current pid namespace of the instance, i.e. the
2023 * Pid field and the first entry in the NSpid field will be identical.
2024 * If the pid namespace of the process is not a descendant of the pid
2025 * namespace of the procfs instance 0 will be shown as its first NSpid
2026 * entry and no others will be shown.
2027 * Note that this differs from the Pid and NSpid fields in
2028 * /proc/<pid>/status where Pid and NSpid are always shown relative to
2029 * the pid namespace of the procfs instance. The difference becomes
2030 * obvious when sending around a pidfd between pid namespaces from a
2031 * different branch of the tree, i.e. where no ancestral relation is
2032 * present between the pid namespaces:
2033 * - create two new pid namespaces ns1 and ns2 in the initial pid
2034 * namespace (also take care to create new mount namespaces in the
2035 * new pid namespace and mount procfs)
2036 * - create a process with a pidfd in ns1
2037 * - send pidfd from ns1 to ns2
2038 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
2039 * have exactly one entry, which is 0
2041 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
2043 struct pid *pid = f->private_data;
2044 struct pid_namespace *ns;
2047 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
2048 ns = proc_pid_ns(file_inode(m->file)->i_sb);
2049 nr = pid_nr_ns(pid, ns);
2052 seq_put_decimal_ll(m, "Pid:\t", nr);
2054 #ifdef CONFIG_PID_NS
2055 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
2059 /* If nr is non-zero it means that 'pid' is valid and that
2060 * ns, i.e. the pid namespace associated with the procfs
2061 * instance, is in the pid namespace hierarchy of pid.
2062 * Start at one below the already printed level.
2064 for (i = ns->level + 1; i <= pid->level; i++)
2065 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2073 * Poll support for process exit notification.
2075 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2077 struct pid *pid = file->private_data;
2078 __poll_t poll_flags = 0;
2080 poll_wait(file, &pid->wait_pidfd, pts);
2083 * Inform pollers only when the whole thread group exits.
2084 * If the thread group leader exits before all other threads in the
2085 * group, then poll(2) should block, similar to the wait(2) family.
2087 if (thread_group_exited(pid))
2088 poll_flags = EPOLLIN | EPOLLRDNORM;
2093 const struct file_operations pidfd_fops = {
2094 .release = pidfd_release,
2096 #ifdef CONFIG_PROC_FS
2097 .show_fdinfo = pidfd_show_fdinfo,
2102 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2103 * @pid: the struct pid for which to create a pidfd
2104 * @flags: flags of the new @pidfd
2105 * @pidfd: the pidfd to return
2107 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2108 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2110 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2111 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2112 * pidfd file are prepared.
2114 * If this function returns successfully the caller is responsible to either
2115 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2116 * order to install the pidfd into its file descriptor table or they must use
2117 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2120 * This function is useful when a pidfd must already be reserved but there
2121 * might still be points of failure afterwards and the caller wants to ensure
2122 * that no pidfd is leaked into its file descriptor table.
2124 * Return: On success, a reserved pidfd is returned from the function and a new
2125 * pidfd file is returned in the last argument to the function. On
2126 * error, a negative error code is returned from the function and the
2127 * last argument remains unchanged.
2129 static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2132 struct file *pidfd_file;
2134 if (flags & ~(O_NONBLOCK | O_RDWR | O_CLOEXEC))
2137 pidfd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2141 pidfd_file = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2142 flags | O_RDWR | O_CLOEXEC);
2143 if (IS_ERR(pidfd_file)) {
2144 put_unused_fd(pidfd);
2145 return PTR_ERR(pidfd_file);
2147 get_pid(pid); /* held by pidfd_file now */
2153 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2154 * @pid: the struct pid for which to create a pidfd
2155 * @flags: flags of the new @pidfd
2156 * @pidfd: the pidfd to return
2158 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2159 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2161 * The helper verifies that @pid is used as a thread group leader.
2163 * If this function returns successfully the caller is responsible to either
2164 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2165 * order to install the pidfd into its file descriptor table or they must use
2166 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2169 * This function is useful when a pidfd must already be reserved but there
2170 * might still be points of failure afterwards and the caller wants to ensure
2171 * that no pidfd is leaked into its file descriptor table.
2173 * Return: On success, a reserved pidfd is returned from the function and a new
2174 * pidfd file is returned in the last argument to the function. On
2175 * error, a negative error code is returned from the function and the
2176 * last argument remains unchanged.
2178 int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2180 if (!pid || !pid_has_task(pid, PIDTYPE_TGID))
2183 return __pidfd_prepare(pid, flags, ret);
2186 static void __delayed_free_task(struct rcu_head *rhp)
2188 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2193 static __always_inline void delayed_free_task(struct task_struct *tsk)
2195 if (IS_ENABLED(CONFIG_MEMCG))
2196 call_rcu(&tsk->rcu, __delayed_free_task);
2201 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2203 /* Skip if kernel thread */
2207 /* Skip if spawning a thread or using vfork */
2208 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2211 /* We need to synchronize with __set_oom_adj */
2212 mutex_lock(&oom_adj_mutex);
2213 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2214 /* Update the values in case they were changed after copy_signal */
2215 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2216 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2217 mutex_unlock(&oom_adj_mutex);
2221 static void rv_task_fork(struct task_struct *p)
2225 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2226 p->rv[i].da_mon.monitoring = false;
2229 #define rv_task_fork(p) do {} while (0)
2233 * This creates a new process as a copy of the old one,
2234 * but does not actually start it yet.
2236 * It copies the registers, and all the appropriate
2237 * parts of the process environment (as per the clone
2238 * flags). The actual kick-off is left to the caller.
2240 __latent_entropy struct task_struct *copy_process(
2244 struct kernel_clone_args *args)
2246 int pidfd = -1, retval;
2247 struct task_struct *p;
2248 struct multiprocess_signals delayed;
2249 struct file *pidfile = NULL;
2250 const u64 clone_flags = args->flags;
2251 struct nsproxy *nsp = current->nsproxy;
2254 * Don't allow sharing the root directory with processes in a different
2257 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2258 return ERR_PTR(-EINVAL);
2260 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2261 return ERR_PTR(-EINVAL);
2264 * Thread groups must share signals as well, and detached threads
2265 * can only be started up within the thread group.
2267 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2268 return ERR_PTR(-EINVAL);
2271 * Shared signal handlers imply shared VM. By way of the above,
2272 * thread groups also imply shared VM. Blocking this case allows
2273 * for various simplifications in other code.
2275 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2276 return ERR_PTR(-EINVAL);
2279 * Siblings of global init remain as zombies on exit since they are
2280 * not reaped by their parent (swapper). To solve this and to avoid
2281 * multi-rooted process trees, prevent global and container-inits
2282 * from creating siblings.
2284 if ((clone_flags & CLONE_PARENT) &&
2285 current->signal->flags & SIGNAL_UNKILLABLE)
2286 return ERR_PTR(-EINVAL);
2289 * If the new process will be in a different pid or user namespace
2290 * do not allow it to share a thread group with the forking task.
2292 if (clone_flags & CLONE_THREAD) {
2293 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2294 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2295 return ERR_PTR(-EINVAL);
2298 if (clone_flags & CLONE_PIDFD) {
2300 * - CLONE_DETACHED is blocked so that we can potentially
2301 * reuse it later for CLONE_PIDFD.
2302 * - CLONE_THREAD is blocked until someone really needs it.
2304 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2305 return ERR_PTR(-EINVAL);
2309 * Force any signals received before this point to be delivered
2310 * before the fork happens. Collect up signals sent to multiple
2311 * processes that happen during the fork and delay them so that
2312 * they appear to happen after the fork.
2314 sigemptyset(&delayed.signal);
2315 INIT_HLIST_NODE(&delayed.node);
2317 spin_lock_irq(¤t->sighand->siglock);
2318 if (!(clone_flags & CLONE_THREAD))
2319 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2320 recalc_sigpending();
2321 spin_unlock_irq(¤t->sighand->siglock);
2322 retval = -ERESTARTNOINTR;
2323 if (task_sigpending(current))
2327 p = dup_task_struct(current, node);
2330 p->flags &= ~PF_KTHREAD;
2332 p->flags |= PF_KTHREAD;
2333 if (args->user_worker) {
2335 * Mark us a user worker, and block any signal that isn't
2338 p->flags |= PF_USER_WORKER;
2339 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2341 if (args->io_thread)
2342 p->flags |= PF_IO_WORKER;
2345 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2347 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2349 * Clear TID on mm_release()?
2351 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2353 ftrace_graph_init_task(p);
2355 rt_mutex_init_task(p);
2357 lockdep_assert_irqs_enabled();
2358 #ifdef CONFIG_PROVE_LOCKING
2359 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2361 retval = copy_creds(p, clone_flags);
2366 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2367 if (p->real_cred->user != INIT_USER &&
2368 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2369 goto bad_fork_cleanup_count;
2371 current->flags &= ~PF_NPROC_EXCEEDED;
2374 * If multiple threads are within copy_process(), then this check
2375 * triggers too late. This doesn't hurt, the check is only there
2376 * to stop root fork bombs.
2379 if (data_race(nr_threads >= max_threads))
2380 goto bad_fork_cleanup_count;
2382 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2383 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2384 p->flags |= PF_FORKNOEXEC;
2385 INIT_LIST_HEAD(&p->children);
2386 INIT_LIST_HEAD(&p->sibling);
2387 rcu_copy_process(p);
2388 p->vfork_done = NULL;
2389 spin_lock_init(&p->alloc_lock);
2391 init_sigpending(&p->pending);
2393 p->utime = p->stime = p->gtime = 0;
2394 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2395 p->utimescaled = p->stimescaled = 0;
2397 prev_cputime_init(&p->prev_cputime);
2399 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2400 seqcount_init(&p->vtime.seqcount);
2401 p->vtime.starttime = 0;
2402 p->vtime.state = VTIME_INACTIVE;
2405 #ifdef CONFIG_IO_URING
2409 #if defined(SPLIT_RSS_COUNTING)
2410 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2413 p->default_timer_slack_ns = current->timer_slack_ns;
2419 task_io_accounting_init(&p->ioac);
2420 acct_clear_integrals(p);
2422 posix_cputimers_init(&p->posix_cputimers);
2424 p->io_context = NULL;
2425 audit_set_context(p, NULL);
2427 if (args->kthread) {
2428 if (!set_kthread_struct(p))
2429 goto bad_fork_cleanup_delayacct;
2432 p->mempolicy = mpol_dup(p->mempolicy);
2433 if (IS_ERR(p->mempolicy)) {
2434 retval = PTR_ERR(p->mempolicy);
2435 p->mempolicy = NULL;
2436 goto bad_fork_cleanup_delayacct;
2439 #ifdef CONFIG_CPUSETS
2440 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2441 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2442 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2444 #ifdef CONFIG_TRACE_IRQFLAGS
2445 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2446 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2447 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2448 p->softirqs_enabled = 1;
2449 p->softirq_context = 0;
2452 p->pagefault_disabled = 0;
2454 #ifdef CONFIG_LOCKDEP
2455 lockdep_init_task(p);
2458 #ifdef CONFIG_DEBUG_MUTEXES
2459 p->blocked_on = NULL; /* not blocked yet */
2461 #ifdef CONFIG_BCACHE
2462 p->sequential_io = 0;
2463 p->sequential_io_avg = 0;
2465 #ifdef CONFIG_BPF_SYSCALL
2466 RCU_INIT_POINTER(p->bpf_storage, NULL);
2470 /* Perform scheduler related setup. Assign this task to a CPU. */
2471 retval = sched_fork(clone_flags, p);
2473 goto bad_fork_cleanup_policy;
2475 retval = perf_event_init_task(p, clone_flags);
2477 goto bad_fork_cleanup_policy;
2478 retval = audit_alloc(p);
2480 goto bad_fork_cleanup_perf;
2481 /* copy all the process information */
2483 retval = security_task_alloc(p, clone_flags);
2485 goto bad_fork_cleanup_audit;
2486 retval = copy_semundo(clone_flags, p);
2488 goto bad_fork_cleanup_security;
2489 retval = copy_files(clone_flags, p, args->no_files);
2491 goto bad_fork_cleanup_semundo;
2492 retval = copy_fs(clone_flags, p);
2494 goto bad_fork_cleanup_files;
2495 retval = copy_sighand(clone_flags, p);
2497 goto bad_fork_cleanup_fs;
2498 retval = copy_signal(clone_flags, p);
2500 goto bad_fork_cleanup_sighand;
2501 retval = copy_mm(clone_flags, p);
2503 goto bad_fork_cleanup_signal;
2504 retval = copy_namespaces(clone_flags, p);
2506 goto bad_fork_cleanup_mm;
2507 retval = copy_io(clone_flags, p);
2509 goto bad_fork_cleanup_namespaces;
2510 retval = copy_thread(p, args);
2512 goto bad_fork_cleanup_io;
2514 stackleak_task_init(p);
2516 if (pid != &init_struct_pid) {
2517 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2518 args->set_tid_size);
2520 retval = PTR_ERR(pid);
2521 goto bad_fork_cleanup_thread;
2526 * This has to happen after we've potentially unshared the file
2527 * descriptor table (so that the pidfd doesn't leak into the child
2528 * if the fd table isn't shared).
2530 if (clone_flags & CLONE_PIDFD) {
2531 /* Note that no task has been attached to @pid yet. */
2532 retval = __pidfd_prepare(pid, O_RDWR | O_CLOEXEC, &pidfile);
2534 goto bad_fork_free_pid;
2537 retval = put_user(pidfd, args->pidfd);
2539 goto bad_fork_put_pidfd;
2548 * sigaltstack should be cleared when sharing the same VM
2550 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2554 * Syscall tracing and stepping should be turned off in the
2555 * child regardless of CLONE_PTRACE.
2557 user_disable_single_step(p);
2558 clear_task_syscall_work(p, SYSCALL_TRACE);
2559 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2560 clear_task_syscall_work(p, SYSCALL_EMU);
2562 clear_tsk_latency_tracing(p);
2564 /* ok, now we should be set up.. */
2565 p->pid = pid_nr(pid);
2566 if (clone_flags & CLONE_THREAD) {
2567 p->group_leader = current->group_leader;
2568 p->tgid = current->tgid;
2570 p->group_leader = p;
2575 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2576 p->dirty_paused_when = 0;
2578 p->pdeath_signal = 0;
2579 INIT_LIST_HEAD(&p->thread_group);
2580 p->task_works = NULL;
2581 clear_posix_cputimers_work(p);
2583 #ifdef CONFIG_KRETPROBES
2584 p->kretprobe_instances.first = NULL;
2586 #ifdef CONFIG_RETHOOK
2587 p->rethooks.first = NULL;
2591 * Ensure that the cgroup subsystem policies allow the new process to be
2592 * forked. It should be noted that the new process's css_set can be changed
2593 * between here and cgroup_post_fork() if an organisation operation is in
2596 retval = cgroup_can_fork(p, args);
2598 goto bad_fork_put_pidfd;
2601 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2602 * the new task on the correct runqueue. All this *before* the task
2605 * This isn't part of ->can_fork() because while the re-cloning is
2606 * cgroup specific, it unconditionally needs to place the task on a
2609 sched_cgroup_fork(p, args);
2612 * From this point on we must avoid any synchronous user-space
2613 * communication until we take the tasklist-lock. In particular, we do
2614 * not want user-space to be able to predict the process start-time by
2615 * stalling fork(2) after we recorded the start_time but before it is
2616 * visible to the system.
2619 p->start_time = ktime_get_ns();
2620 p->start_boottime = ktime_get_boottime_ns();
2623 * Make it visible to the rest of the system, but dont wake it up yet.
2624 * Need tasklist lock for parent etc handling!
2626 write_lock_irq(&tasklist_lock);
2628 /* CLONE_PARENT re-uses the old parent */
2629 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2630 p->real_parent = current->real_parent;
2631 p->parent_exec_id = current->parent_exec_id;
2632 if (clone_flags & CLONE_THREAD)
2633 p->exit_signal = -1;
2635 p->exit_signal = current->group_leader->exit_signal;
2637 p->real_parent = current;
2638 p->parent_exec_id = current->self_exec_id;
2639 p->exit_signal = args->exit_signal;
2642 klp_copy_process(p);
2646 spin_lock(¤t->sighand->siglock);
2650 rseq_fork(p, clone_flags);
2652 /* Don't start children in a dying pid namespace */
2653 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2655 goto bad_fork_cancel_cgroup;
2658 /* Let kill terminate clone/fork in the middle */
2659 if (fatal_signal_pending(current)) {
2661 goto bad_fork_cancel_cgroup;
2664 /* No more failure paths after this point. */
2667 * Copy seccomp details explicitly here, in case they were changed
2668 * before holding sighand lock.
2672 init_task_pid_links(p);
2673 if (likely(p->pid)) {
2674 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2676 init_task_pid(p, PIDTYPE_PID, pid);
2677 if (thread_group_leader(p)) {
2678 init_task_pid(p, PIDTYPE_TGID, pid);
2679 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2680 init_task_pid(p, PIDTYPE_SID, task_session(current));
2682 if (is_child_reaper(pid)) {
2683 ns_of_pid(pid)->child_reaper = p;
2684 p->signal->flags |= SIGNAL_UNKILLABLE;
2686 p->signal->shared_pending.signal = delayed.signal;
2687 p->signal->tty = tty_kref_get(current->signal->tty);
2689 * Inherit has_child_subreaper flag under the same
2690 * tasklist_lock with adding child to the process tree
2691 * for propagate_has_child_subreaper optimization.
2693 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2694 p->real_parent->signal->is_child_subreaper;
2695 list_add_tail(&p->sibling, &p->real_parent->children);
2696 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2697 attach_pid(p, PIDTYPE_TGID);
2698 attach_pid(p, PIDTYPE_PGID);
2699 attach_pid(p, PIDTYPE_SID);
2700 __this_cpu_inc(process_counts);
2702 current->signal->nr_threads++;
2703 current->signal->quick_threads++;
2704 atomic_inc(¤t->signal->live);
2705 refcount_inc(¤t->signal->sigcnt);
2706 task_join_group_stop(p);
2707 list_add_tail_rcu(&p->thread_group,
2708 &p->group_leader->thread_group);
2709 list_add_tail_rcu(&p->thread_node,
2710 &p->signal->thread_head);
2712 attach_pid(p, PIDTYPE_PID);
2716 hlist_del_init(&delayed.node);
2717 spin_unlock(¤t->sighand->siglock);
2718 syscall_tracepoint_update(p);
2719 write_unlock_irq(&tasklist_lock);
2722 fd_install(pidfd, pidfile);
2724 proc_fork_connector(p);
2726 cgroup_post_fork(p, args);
2729 trace_task_newtask(p, clone_flags);
2730 uprobe_copy_process(p, clone_flags);
2731 user_events_fork(p, clone_flags);
2733 copy_oom_score_adj(clone_flags, p);
2737 bad_fork_cancel_cgroup:
2739 spin_unlock(¤t->sighand->siglock);
2740 write_unlock_irq(&tasklist_lock);
2741 cgroup_cancel_fork(p, args);
2743 if (clone_flags & CLONE_PIDFD) {
2745 put_unused_fd(pidfd);
2748 if (pid != &init_struct_pid)
2750 bad_fork_cleanup_thread:
2752 bad_fork_cleanup_io:
2755 bad_fork_cleanup_namespaces:
2756 exit_task_namespaces(p);
2757 bad_fork_cleanup_mm:
2759 mm_clear_owner(p->mm, p);
2762 bad_fork_cleanup_signal:
2763 if (!(clone_flags & CLONE_THREAD))
2764 free_signal_struct(p->signal);
2765 bad_fork_cleanup_sighand:
2766 __cleanup_sighand(p->sighand);
2767 bad_fork_cleanup_fs:
2768 exit_fs(p); /* blocking */
2769 bad_fork_cleanup_files:
2770 exit_files(p); /* blocking */
2771 bad_fork_cleanup_semundo:
2773 bad_fork_cleanup_security:
2774 security_task_free(p);
2775 bad_fork_cleanup_audit:
2777 bad_fork_cleanup_perf:
2778 perf_event_free_task(p);
2779 bad_fork_cleanup_policy:
2780 lockdep_free_task(p);
2782 mpol_put(p->mempolicy);
2784 bad_fork_cleanup_delayacct:
2785 delayacct_tsk_free(p);
2786 bad_fork_cleanup_count:
2787 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2790 WRITE_ONCE(p->__state, TASK_DEAD);
2791 exit_task_stack_account(p);
2793 delayed_free_task(p);
2795 spin_lock_irq(¤t->sighand->siglock);
2796 hlist_del_init(&delayed.node);
2797 spin_unlock_irq(¤t->sighand->siglock);
2798 return ERR_PTR(retval);
2801 static inline void init_idle_pids(struct task_struct *idle)
2805 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2806 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2807 init_task_pid(idle, type, &init_struct_pid);
2811 static int idle_dummy(void *dummy)
2813 /* This function is never called */
2817 struct task_struct * __init fork_idle(int cpu)
2819 struct task_struct *task;
2820 struct kernel_clone_args args = {
2828 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2829 if (!IS_ERR(task)) {
2830 init_idle_pids(task);
2831 init_idle(task, cpu);
2838 * This is like kernel_clone(), but shaved down and tailored to just
2839 * creating io_uring workers. It returns a created task, or an error pointer.
2840 * The returned task is inactive, and the caller must fire it up through
2841 * wake_up_new_task(p). All signals are blocked in the created task.
2843 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2845 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2847 struct kernel_clone_args args = {
2848 .flags = ((lower_32_bits(flags) | CLONE_VM |
2849 CLONE_UNTRACED) & ~CSIGNAL),
2850 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2857 return copy_process(NULL, 0, node, &args);
2861 * Ok, this is the main fork-routine.
2863 * It copies the process, and if successful kick-starts
2864 * it and waits for it to finish using the VM if required.
2866 * args->exit_signal is expected to be checked for sanity by the caller.
2868 pid_t kernel_clone(struct kernel_clone_args *args)
2870 u64 clone_flags = args->flags;
2871 struct completion vfork;
2873 struct task_struct *p;
2878 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2879 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2880 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2881 * field in struct clone_args and it still doesn't make sense to have
2882 * them both point at the same memory location. Performing this check
2883 * here has the advantage that we don't need to have a separate helper
2884 * to check for legacy clone().
2886 if ((args->flags & CLONE_PIDFD) &&
2887 (args->flags & CLONE_PARENT_SETTID) &&
2888 (args->pidfd == args->parent_tid))
2892 * Determine whether and which event to report to ptracer. When
2893 * called from kernel_thread or CLONE_UNTRACED is explicitly
2894 * requested, no event is reported; otherwise, report if the event
2895 * for the type of forking is enabled.
2897 if (!(clone_flags & CLONE_UNTRACED)) {
2898 if (clone_flags & CLONE_VFORK)
2899 trace = PTRACE_EVENT_VFORK;
2900 else if (args->exit_signal != SIGCHLD)
2901 trace = PTRACE_EVENT_CLONE;
2903 trace = PTRACE_EVENT_FORK;
2905 if (likely(!ptrace_event_enabled(current, trace)))
2909 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2910 add_latent_entropy();
2916 * Do this prior waking up the new thread - the thread pointer
2917 * might get invalid after that point, if the thread exits quickly.
2919 trace_sched_process_fork(current, p);
2921 pid = get_task_pid(p, PIDTYPE_PID);
2924 if (clone_flags & CLONE_PARENT_SETTID)
2925 put_user(nr, args->parent_tid);
2927 if (clone_flags & CLONE_VFORK) {
2928 p->vfork_done = &vfork;
2929 init_completion(&vfork);
2933 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2934 /* lock the task to synchronize with memcg migration */
2936 lru_gen_add_mm(p->mm);
2940 wake_up_new_task(p);
2942 /* forking complete and child started to run, tell ptracer */
2943 if (unlikely(trace))
2944 ptrace_event_pid(trace, pid);
2946 if (clone_flags & CLONE_VFORK) {
2947 if (!wait_for_vfork_done(p, &vfork))
2948 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2956 * Create a kernel thread.
2958 pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2959 unsigned long flags)
2961 struct kernel_clone_args args = {
2962 .flags = ((lower_32_bits(flags) | CLONE_VM |
2963 CLONE_UNTRACED) & ~CSIGNAL),
2964 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2971 return kernel_clone(&args);
2975 * Create a user mode thread.
2977 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2979 struct kernel_clone_args args = {
2980 .flags = ((lower_32_bits(flags) | CLONE_VM |
2981 CLONE_UNTRACED) & ~CSIGNAL),
2982 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2987 return kernel_clone(&args);
2990 #ifdef __ARCH_WANT_SYS_FORK
2991 SYSCALL_DEFINE0(fork)
2994 struct kernel_clone_args args = {
2995 .exit_signal = SIGCHLD,
2998 return kernel_clone(&args);
3000 /* can not support in nommu mode */
3006 #ifdef __ARCH_WANT_SYS_VFORK
3007 SYSCALL_DEFINE0(vfork)
3009 struct kernel_clone_args args = {
3010 .flags = CLONE_VFORK | CLONE_VM,
3011 .exit_signal = SIGCHLD,
3014 return kernel_clone(&args);
3018 #ifdef __ARCH_WANT_SYS_CLONE
3019 #ifdef CONFIG_CLONE_BACKWARDS
3020 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3021 int __user *, parent_tidptr,
3023 int __user *, child_tidptr)
3024 #elif defined(CONFIG_CLONE_BACKWARDS2)
3025 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
3026 int __user *, parent_tidptr,
3027 int __user *, child_tidptr,
3029 #elif defined(CONFIG_CLONE_BACKWARDS3)
3030 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
3032 int __user *, parent_tidptr,
3033 int __user *, child_tidptr,
3036 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3037 int __user *, parent_tidptr,
3038 int __user *, child_tidptr,
3042 struct kernel_clone_args args = {
3043 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
3044 .pidfd = parent_tidptr,
3045 .child_tid = child_tidptr,
3046 .parent_tid = parent_tidptr,
3047 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
3052 return kernel_clone(&args);
3056 #ifdef __ARCH_WANT_SYS_CLONE3
3058 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
3059 struct clone_args __user *uargs,
3063 struct clone_args args;
3064 pid_t *kset_tid = kargs->set_tid;
3066 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
3067 CLONE_ARGS_SIZE_VER0);
3068 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
3069 CLONE_ARGS_SIZE_VER1);
3070 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
3071 CLONE_ARGS_SIZE_VER2);
3072 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
3074 if (unlikely(usize > PAGE_SIZE))
3076 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
3079 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
3083 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
3086 if (unlikely(!args.set_tid && args.set_tid_size > 0))
3089 if (unlikely(args.set_tid && args.set_tid_size == 0))
3093 * Verify that higher 32bits of exit_signal are unset and that
3094 * it is a valid signal
3096 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3097 !valid_signal(args.exit_signal)))
3100 if ((args.flags & CLONE_INTO_CGROUP) &&
3101 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3104 *kargs = (struct kernel_clone_args){
3105 .flags = args.flags,
3106 .pidfd = u64_to_user_ptr(args.pidfd),
3107 .child_tid = u64_to_user_ptr(args.child_tid),
3108 .parent_tid = u64_to_user_ptr(args.parent_tid),
3109 .exit_signal = args.exit_signal,
3110 .stack = args.stack,
3111 .stack_size = args.stack_size,
3113 .set_tid_size = args.set_tid_size,
3114 .cgroup = args.cgroup,
3118 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3119 (kargs->set_tid_size * sizeof(pid_t))))
3122 kargs->set_tid = kset_tid;
3128 * clone3_stack_valid - check and prepare stack
3129 * @kargs: kernel clone args
3131 * Verify that the stack arguments userspace gave us are sane.
3132 * In addition, set the stack direction for userspace since it's easy for us to
3135 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3137 if (kargs->stack == 0) {
3138 if (kargs->stack_size > 0)
3141 if (kargs->stack_size == 0)
3144 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3147 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
3148 kargs->stack += kargs->stack_size;
3155 static bool clone3_args_valid(struct kernel_clone_args *kargs)
3157 /* Verify that no unknown flags are passed along. */
3159 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3163 * - make the CLONE_DETACHED bit reusable for clone3
3164 * - make the CSIGNAL bits reusable for clone3
3166 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3169 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3170 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3173 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3177 if (!clone3_stack_valid(kargs))
3184 * clone3 - create a new process with specific properties
3185 * @uargs: argument structure
3186 * @size: size of @uargs
3188 * clone3() is the extensible successor to clone()/clone2().
3189 * It takes a struct as argument that is versioned by its size.
3191 * Return: On success, a positive PID for the child process.
3192 * On error, a negative errno number.
3194 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3198 struct kernel_clone_args kargs;
3199 pid_t set_tid[MAX_PID_NS_LEVEL];
3201 kargs.set_tid = set_tid;
3203 err = copy_clone_args_from_user(&kargs, uargs, size);
3207 if (!clone3_args_valid(&kargs))
3210 return kernel_clone(&kargs);
3214 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3216 struct task_struct *leader, *parent, *child;
3219 read_lock(&tasklist_lock);
3220 leader = top = top->group_leader;
3222 for_each_thread(leader, parent) {
3223 list_for_each_entry(child, &parent->children, sibling) {
3224 res = visitor(child, data);
3236 if (leader != top) {
3238 parent = child->real_parent;
3239 leader = parent->group_leader;
3243 read_unlock(&tasklist_lock);
3246 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3247 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3250 static void sighand_ctor(void *data)
3252 struct sighand_struct *sighand = data;
3254 spin_lock_init(&sighand->siglock);
3255 init_waitqueue_head(&sighand->signalfd_wqh);
3258 void __init mm_cache_init(void)
3260 unsigned int mm_size;
3263 * The mm_cpumask is located at the end of mm_struct, and is
3264 * dynamically sized based on the maximum CPU number this system
3265 * can have, taking hotplug into account (nr_cpu_ids).
3267 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3269 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3270 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3271 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3272 offsetof(struct mm_struct, saved_auxv),
3273 sizeof_field(struct mm_struct, saved_auxv),
3277 void __init proc_caches_init(void)
3279 sighand_cachep = kmem_cache_create("sighand_cache",
3280 sizeof(struct sighand_struct), 0,
3281 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3282 SLAB_ACCOUNT, sighand_ctor);
3283 signal_cachep = kmem_cache_create("signal_cache",
3284 sizeof(struct signal_struct), 0,
3285 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3287 files_cachep = kmem_cache_create("files_cache",
3288 sizeof(struct files_struct), 0,
3289 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3291 fs_cachep = kmem_cache_create("fs_cache",
3292 sizeof(struct fs_struct), 0,
3293 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3296 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3297 #ifdef CONFIG_PER_VMA_LOCK
3298 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3301 nsproxy_cache_init();
3305 * Check constraints on flags passed to the unshare system call.
3307 static int check_unshare_flags(unsigned long unshare_flags)
3309 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3310 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3311 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3312 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3316 * Not implemented, but pretend it works if there is nothing
3317 * to unshare. Note that unsharing the address space or the
3318 * signal handlers also need to unshare the signal queues (aka
3321 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3322 if (!thread_group_empty(current))
3325 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3326 if (refcount_read(¤t->sighand->count) > 1)
3329 if (unshare_flags & CLONE_VM) {
3330 if (!current_is_single_threaded())
3338 * Unshare the filesystem structure if it is being shared
3340 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3342 struct fs_struct *fs = current->fs;
3344 if (!(unshare_flags & CLONE_FS) || !fs)
3347 /* don't need lock here; in the worst case we'll do useless copy */
3351 *new_fsp = copy_fs_struct(fs);
3359 * Unshare file descriptor table if it is being shared
3361 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3362 struct files_struct **new_fdp)
3364 struct files_struct *fd = current->files;
3367 if ((unshare_flags & CLONE_FILES) &&
3368 (fd && atomic_read(&fd->count) > 1)) {
3369 *new_fdp = dup_fd(fd, max_fds, &error);
3378 * unshare allows a process to 'unshare' part of the process
3379 * context which was originally shared using clone. copy_*
3380 * functions used by kernel_clone() cannot be used here directly
3381 * because they modify an inactive task_struct that is being
3382 * constructed. Here we are modifying the current, active,
3385 int ksys_unshare(unsigned long unshare_flags)
3387 struct fs_struct *fs, *new_fs = NULL;
3388 struct files_struct *new_fd = NULL;
3389 struct cred *new_cred = NULL;
3390 struct nsproxy *new_nsproxy = NULL;
3395 * If unsharing a user namespace must also unshare the thread group
3396 * and unshare the filesystem root and working directories.
3398 if (unshare_flags & CLONE_NEWUSER)
3399 unshare_flags |= CLONE_THREAD | CLONE_FS;
3401 * If unsharing vm, must also unshare signal handlers.
3403 if (unshare_flags & CLONE_VM)
3404 unshare_flags |= CLONE_SIGHAND;
3406 * If unsharing a signal handlers, must also unshare the signal queues.
3408 if (unshare_flags & CLONE_SIGHAND)
3409 unshare_flags |= CLONE_THREAD;
3411 * If unsharing namespace, must also unshare filesystem information.
3413 if (unshare_flags & CLONE_NEWNS)
3414 unshare_flags |= CLONE_FS;
3416 err = check_unshare_flags(unshare_flags);
3418 goto bad_unshare_out;
3420 * CLONE_NEWIPC must also detach from the undolist: after switching
3421 * to a new ipc namespace, the semaphore arrays from the old
3422 * namespace are unreachable.
3424 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3426 err = unshare_fs(unshare_flags, &new_fs);
3428 goto bad_unshare_out;
3429 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3431 goto bad_unshare_cleanup_fs;
3432 err = unshare_userns(unshare_flags, &new_cred);
3434 goto bad_unshare_cleanup_fd;
3435 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3438 goto bad_unshare_cleanup_cred;
3441 err = set_cred_ucounts(new_cred);
3443 goto bad_unshare_cleanup_cred;
3446 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3449 * CLONE_SYSVSEM is equivalent to sys_exit().
3453 if (unshare_flags & CLONE_NEWIPC) {
3454 /* Orphan segments in old ns (see sem above). */
3456 shm_init_task(current);
3460 switch_task_namespaces(current, new_nsproxy);
3466 spin_lock(&fs->lock);
3467 current->fs = new_fs;
3472 spin_unlock(&fs->lock);
3476 swap(current->files, new_fd);
3478 task_unlock(current);
3481 /* Install the new user namespace */
3482 commit_creds(new_cred);
3487 perf_event_namespaces(current);
3489 bad_unshare_cleanup_cred:
3492 bad_unshare_cleanup_fd:
3494 put_files_struct(new_fd);
3496 bad_unshare_cleanup_fs:
3498 free_fs_struct(new_fs);
3504 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3506 return ksys_unshare(unshare_flags);
3510 * Helper to unshare the files of the current task.
3511 * We don't want to expose copy_files internals to
3512 * the exec layer of the kernel.
3515 int unshare_files(void)
3517 struct task_struct *task = current;
3518 struct files_struct *old, *copy = NULL;
3521 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3529 put_files_struct(old);
3533 int sysctl_max_threads(struct ctl_table *table, int write,
3534 void *buffer, size_t *lenp, loff_t *ppos)
3538 int threads = max_threads;
3540 int max = MAX_THREADS;
3547 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3551 max_threads = threads;