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 if (mpnt->vm_flags & VM_DONTCOPY) {
690 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
695 * Don't duplicate many vmas if we've been oom-killed (for
698 if (fatal_signal_pending(current)) {
702 if (mpnt->vm_flags & VM_ACCOUNT) {
703 unsigned long len = vma_pages(mpnt);
705 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
709 tmp = vm_area_dup(mpnt);
712 retval = vma_dup_policy(mpnt, tmp);
714 goto fail_nomem_policy;
716 retval = dup_userfaultfd(tmp, &uf);
718 goto fail_nomem_anon_vma_fork;
719 if (tmp->vm_flags & VM_WIPEONFORK) {
721 * VM_WIPEONFORK gets a clean slate in the child.
722 * Don't prepare anon_vma until fault since we don't
723 * copy page for current vma.
725 tmp->anon_vma = NULL;
726 } else if (anon_vma_fork(tmp, mpnt))
727 goto fail_nomem_anon_vma_fork;
728 vm_flags_clear(tmp, VM_LOCKED_MASK);
731 struct address_space *mapping = file->f_mapping;
734 i_mmap_lock_write(mapping);
735 if (tmp->vm_flags & VM_SHARED)
736 mapping_allow_writable(mapping);
737 flush_dcache_mmap_lock(mapping);
738 /* insert tmp into the share list, just after mpnt */
739 vma_interval_tree_insert_after(tmp, mpnt,
741 flush_dcache_mmap_unlock(mapping);
742 i_mmap_unlock_write(mapping);
746 * Copy/update hugetlb private vma information.
748 if (is_vm_hugetlb_page(tmp))
749 hugetlb_dup_vma_private(tmp);
751 /* Link the vma into the MT */
752 if (vma_iter_bulk_store(&vmi, tmp))
753 goto fail_nomem_vmi_store;
756 if (!(tmp->vm_flags & VM_WIPEONFORK))
757 retval = copy_page_range(tmp, mpnt);
759 if (tmp->vm_ops && tmp->vm_ops->open)
760 tmp->vm_ops->open(tmp);
765 /* a new mm has just been created */
766 retval = arch_dup_mmap(oldmm, mm);
770 mt_set_in_rcu(vmi.mas.tree);
772 mmap_write_unlock(mm);
774 mmap_write_unlock(oldmm);
775 dup_userfaultfd_complete(&uf);
777 uprobe_end_dup_mmap();
780 fail_nomem_vmi_store:
781 unlink_anon_vmas(tmp);
782 fail_nomem_anon_vma_fork:
783 mpol_put(vma_policy(tmp));
788 vm_unacct_memory(charge);
792 static inline int mm_alloc_pgd(struct mm_struct *mm)
794 mm->pgd = pgd_alloc(mm);
795 if (unlikely(!mm->pgd))
800 static inline void mm_free_pgd(struct mm_struct *mm)
802 pgd_free(mm, mm->pgd);
805 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
807 mmap_write_lock(oldmm);
808 dup_mm_exe_file(mm, oldmm);
809 mmap_write_unlock(oldmm);
812 #define mm_alloc_pgd(mm) (0)
813 #define mm_free_pgd(mm)
814 #endif /* CONFIG_MMU */
816 static void check_mm(struct mm_struct *mm)
820 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
821 "Please make sure 'struct resident_page_types[]' is updated as well");
823 for (i = 0; i < NR_MM_COUNTERS; i++) {
824 long x = percpu_counter_sum(&mm->rss_stat[i]);
827 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
828 mm, resident_page_types[i], x);
831 if (mm_pgtables_bytes(mm))
832 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
833 mm_pgtables_bytes(mm));
835 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
836 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
840 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
841 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
843 static void do_check_lazy_tlb(void *arg)
845 struct mm_struct *mm = arg;
847 WARN_ON_ONCE(current->active_mm == mm);
850 static void do_shoot_lazy_tlb(void *arg)
852 struct mm_struct *mm = arg;
854 if (current->active_mm == mm) {
855 WARN_ON_ONCE(current->mm);
856 current->active_mm = &init_mm;
857 switch_mm(mm, &init_mm, current);
861 static void cleanup_lazy_tlbs(struct mm_struct *mm)
863 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
865 * In this case, lazy tlb mms are refounted and would not reach
866 * __mmdrop until all CPUs have switched away and mmdrop()ed.
872 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
873 * requires lazy mm users to switch to another mm when the refcount
874 * drops to zero, before the mm is freed. This requires IPIs here to
875 * switch kernel threads to init_mm.
877 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
878 * switch with the final userspace teardown TLB flush which leaves the
879 * mm lazy on this CPU but no others, reducing the need for additional
880 * IPIs here. There are cases where a final IPI is still required here,
881 * such as the final mmdrop being performed on a different CPU than the
882 * one exiting, or kernel threads using the mm when userspace exits.
884 * IPI overheads have not found to be expensive, but they could be
885 * reduced in a number of possible ways, for example (roughly
886 * increasing order of complexity):
887 * - The last lazy reference created by exit_mm() could instead switch
888 * to init_mm, however it's probable this will run on the same CPU
889 * immediately afterwards, so this may not reduce IPIs much.
890 * - A batch of mms requiring IPIs could be gathered and freed at once.
891 * - CPUs store active_mm where it can be remotely checked without a
892 * lock, to filter out false-positives in the cpumask.
893 * - After mm_users or mm_count reaches zero, switching away from the
894 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
895 * with some batching or delaying of the final IPIs.
896 * - A delayed freeing and RCU-like quiescing sequence based on mm
897 * switching to avoid IPIs completely.
899 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
900 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
901 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
905 * Called when the last reference to the mm
906 * is dropped: either by a lazy thread or by
907 * mmput. Free the page directory and the mm.
909 void __mmdrop(struct mm_struct *mm)
913 BUG_ON(mm == &init_mm);
914 WARN_ON_ONCE(mm == current->mm);
916 /* Ensure no CPUs are using this as their lazy tlb mm */
917 cleanup_lazy_tlbs(mm);
919 WARN_ON_ONCE(mm == current->active_mm);
922 mmu_notifier_subscriptions_destroy(mm);
924 put_user_ns(mm->user_ns);
928 for (i = 0; i < NR_MM_COUNTERS; i++)
929 percpu_counter_destroy(&mm->rss_stat[i]);
932 EXPORT_SYMBOL_GPL(__mmdrop);
934 static void mmdrop_async_fn(struct work_struct *work)
936 struct mm_struct *mm;
938 mm = container_of(work, struct mm_struct, async_put_work);
942 static void mmdrop_async(struct mm_struct *mm)
944 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
945 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
946 schedule_work(&mm->async_put_work);
950 static inline void free_signal_struct(struct signal_struct *sig)
952 taskstats_tgid_free(sig);
953 sched_autogroup_exit(sig);
955 * __mmdrop is not safe to call from softirq context on x86 due to
956 * pgd_dtor so postpone it to the async context
959 mmdrop_async(sig->oom_mm);
960 kmem_cache_free(signal_cachep, sig);
963 static inline void put_signal_struct(struct signal_struct *sig)
965 if (refcount_dec_and_test(&sig->sigcnt))
966 free_signal_struct(sig);
969 void __put_task_struct(struct task_struct *tsk)
971 WARN_ON(!tsk->exit_state);
972 WARN_ON(refcount_read(&tsk->usage));
973 WARN_ON(tsk == current);
977 task_numa_free(tsk, true);
978 security_task_free(tsk);
980 delayacct_tsk_free(tsk);
981 put_signal_struct(tsk->signal);
982 sched_core_free(tsk);
985 EXPORT_SYMBOL_GPL(__put_task_struct);
987 void __init __weak arch_task_cache_init(void) { }
992 static void set_max_threads(unsigned int max_threads_suggested)
995 unsigned long nr_pages = totalram_pages();
998 * The number of threads shall be limited such that the thread
999 * structures may only consume a small part of the available memory.
1001 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1002 threads = MAX_THREADS;
1004 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1005 (u64) THREAD_SIZE * 8UL);
1007 if (threads > max_threads_suggested)
1008 threads = max_threads_suggested;
1010 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1013 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1014 /* Initialized by the architecture: */
1015 int arch_task_struct_size __read_mostly;
1018 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1019 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1021 /* Fetch thread_struct whitelist for the architecture. */
1022 arch_thread_struct_whitelist(offset, size);
1025 * Handle zero-sized whitelist or empty thread_struct, otherwise
1026 * adjust offset to position of thread_struct in task_struct.
1028 if (unlikely(*size == 0))
1031 *offset += offsetof(struct task_struct, thread);
1033 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
1035 void __init fork_init(void)
1038 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1039 #ifndef ARCH_MIN_TASKALIGN
1040 #define ARCH_MIN_TASKALIGN 0
1042 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1043 unsigned long useroffset, usersize;
1045 /* create a slab on which task_structs can be allocated */
1046 task_struct_whitelist(&useroffset, &usersize);
1047 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1048 arch_task_struct_size, align,
1049 SLAB_PANIC|SLAB_ACCOUNT,
1050 useroffset, usersize, NULL);
1053 /* do the arch specific task caches init */
1054 arch_task_cache_init();
1056 set_max_threads(MAX_THREADS);
1058 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1059 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1060 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1061 init_task.signal->rlim[RLIMIT_NPROC];
1063 for (i = 0; i < UCOUNT_COUNTS; i++)
1064 init_user_ns.ucount_max[i] = max_threads/2;
1066 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1067 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1068 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1069 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1071 #ifdef CONFIG_VMAP_STACK
1072 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1073 NULL, free_vm_stack_cache);
1078 lockdep_init_task(&init_task);
1082 int __weak arch_dup_task_struct(struct task_struct *dst,
1083 struct task_struct *src)
1089 void set_task_stack_end_magic(struct task_struct *tsk)
1091 unsigned long *stackend;
1093 stackend = end_of_stack(tsk);
1094 *stackend = STACK_END_MAGIC; /* for overflow detection */
1097 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1099 struct task_struct *tsk;
1102 if (node == NUMA_NO_NODE)
1103 node = tsk_fork_get_node(orig);
1104 tsk = alloc_task_struct_node(node);
1108 err = arch_dup_task_struct(tsk, orig);
1112 err = alloc_thread_stack_node(tsk, node);
1116 #ifdef CONFIG_THREAD_INFO_IN_TASK
1117 refcount_set(&tsk->stack_refcount, 1);
1119 account_kernel_stack(tsk, 1);
1121 err = scs_prepare(tsk, node);
1125 #ifdef CONFIG_SECCOMP
1127 * We must handle setting up seccomp filters once we're under
1128 * the sighand lock in case orig has changed between now and
1129 * then. Until then, filter must be NULL to avoid messing up
1130 * the usage counts on the error path calling free_task.
1132 tsk->seccomp.filter = NULL;
1135 setup_thread_stack(tsk, orig);
1136 clear_user_return_notifier(tsk);
1137 clear_tsk_need_resched(tsk);
1138 set_task_stack_end_magic(tsk);
1139 clear_syscall_work_syscall_user_dispatch(tsk);
1141 #ifdef CONFIG_STACKPROTECTOR
1142 tsk->stack_canary = get_random_canary();
1144 if (orig->cpus_ptr == &orig->cpus_mask)
1145 tsk->cpus_ptr = &tsk->cpus_mask;
1146 dup_user_cpus_ptr(tsk, orig, node);
1149 * One for the user space visible state that goes away when reaped.
1150 * One for the scheduler.
1152 refcount_set(&tsk->rcu_users, 2);
1153 /* One for the rcu users */
1154 refcount_set(&tsk->usage, 1);
1155 #ifdef CONFIG_BLK_DEV_IO_TRACE
1156 tsk->btrace_seq = 0;
1158 tsk->splice_pipe = NULL;
1159 tsk->task_frag.page = NULL;
1160 tsk->wake_q.next = NULL;
1161 tsk->worker_private = NULL;
1163 kcov_task_init(tsk);
1164 kmsan_task_create(tsk);
1165 kmap_local_fork(tsk);
1167 #ifdef CONFIG_FAULT_INJECTION
1171 #ifdef CONFIG_BLK_CGROUP
1172 tsk->throttle_disk = NULL;
1173 tsk->use_memdelay = 0;
1176 #ifdef CONFIG_IOMMU_SVA
1177 tsk->pasid_activated = 0;
1181 tsk->active_memcg = NULL;
1184 #ifdef CONFIG_CPU_SUP_INTEL
1185 tsk->reported_split_lock = 0;
1188 #ifdef CONFIG_SCHED_MM_CID
1190 tsk->last_mm_cid = -1;
1191 tsk->mm_cid_active = 0;
1192 tsk->migrate_from_cpu = -1;
1197 exit_task_stack_account(tsk);
1198 free_thread_stack(tsk);
1200 free_task_struct(tsk);
1204 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1206 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1208 static int __init coredump_filter_setup(char *s)
1210 default_dump_filter =
1211 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1212 MMF_DUMP_FILTER_MASK;
1216 __setup("coredump_filter=", coredump_filter_setup);
1218 #include <linux/init_task.h>
1220 static void mm_init_aio(struct mm_struct *mm)
1223 spin_lock_init(&mm->ioctx_lock);
1224 mm->ioctx_table = NULL;
1228 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1229 struct task_struct *p)
1233 WRITE_ONCE(mm->owner, NULL);
1237 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1244 static void mm_init_uprobes_state(struct mm_struct *mm)
1246 #ifdef CONFIG_UPROBES
1247 mm->uprobes_state.xol_area = NULL;
1251 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1252 struct user_namespace *user_ns)
1256 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1257 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1258 atomic_set(&mm->mm_users, 1);
1259 atomic_set(&mm->mm_count, 1);
1260 seqcount_init(&mm->write_protect_seq);
1262 INIT_LIST_HEAD(&mm->mmlist);
1263 #ifdef CONFIG_PER_VMA_LOCK
1264 mm->mm_lock_seq = 0;
1266 mm_pgtables_bytes_init(mm);
1269 atomic64_set(&mm->pinned_vm, 0);
1270 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1271 spin_lock_init(&mm->page_table_lock);
1272 spin_lock_init(&mm->arg_lock);
1273 mm_init_cpumask(mm);
1275 mm_init_owner(mm, p);
1277 RCU_INIT_POINTER(mm->exe_file, NULL);
1278 mmu_notifier_subscriptions_init(mm);
1279 init_tlb_flush_pending(mm);
1280 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1281 mm->pmd_huge_pte = NULL;
1283 mm_init_uprobes_state(mm);
1284 hugetlb_count_init(mm);
1287 mm->flags = current->mm->flags & MMF_INIT_MASK;
1288 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1290 mm->flags = default_dump_filter;
1294 if (mm_alloc_pgd(mm))
1297 if (init_new_context(p, mm))
1298 goto fail_nocontext;
1300 if (mm_alloc_cid(mm))
1303 for (i = 0; i < NR_MM_COUNTERS; i++)
1304 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1307 mm->user_ns = get_user_ns(user_ns);
1308 lru_gen_init_mm(mm);
1313 percpu_counter_destroy(&mm->rss_stat[--i]);
1316 destroy_context(mm);
1325 * Allocate and initialize an mm_struct.
1327 struct mm_struct *mm_alloc(void)
1329 struct mm_struct *mm;
1335 memset(mm, 0, sizeof(*mm));
1336 return mm_init(mm, current, current_user_ns());
1339 static inline void __mmput(struct mm_struct *mm)
1341 VM_BUG_ON(atomic_read(&mm->mm_users));
1343 uprobe_clear_state(mm);
1346 khugepaged_exit(mm); /* must run before exit_mmap */
1348 mm_put_huge_zero_page(mm);
1349 set_mm_exe_file(mm, NULL);
1350 if (!list_empty(&mm->mmlist)) {
1351 spin_lock(&mmlist_lock);
1352 list_del(&mm->mmlist);
1353 spin_unlock(&mmlist_lock);
1356 module_put(mm->binfmt->module);
1362 * Decrement the use count and release all resources for an mm.
1364 void mmput(struct mm_struct *mm)
1368 if (atomic_dec_and_test(&mm->mm_users))
1371 EXPORT_SYMBOL_GPL(mmput);
1374 static void mmput_async_fn(struct work_struct *work)
1376 struct mm_struct *mm = container_of(work, struct mm_struct,
1382 void mmput_async(struct mm_struct *mm)
1384 if (atomic_dec_and_test(&mm->mm_users)) {
1385 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1386 schedule_work(&mm->async_put_work);
1389 EXPORT_SYMBOL_GPL(mmput_async);
1393 * set_mm_exe_file - change a reference to the mm's executable file
1395 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1397 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1398 * invocations: in mmput() nobody alive left, in execve task is single
1401 * Can only fail if new_exe_file != NULL.
1403 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1405 struct file *old_exe_file;
1408 * It is safe to dereference the exe_file without RCU as
1409 * this function is only called if nobody else can access
1410 * this mm -- see comment above for justification.
1412 old_exe_file = rcu_dereference_raw(mm->exe_file);
1416 * We expect the caller (i.e., sys_execve) to already denied
1417 * write access, so this is unlikely to fail.
1419 if (unlikely(deny_write_access(new_exe_file)))
1421 get_file(new_exe_file);
1423 rcu_assign_pointer(mm->exe_file, new_exe_file);
1425 allow_write_access(old_exe_file);
1432 * replace_mm_exe_file - replace a reference to the mm's executable file
1434 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1435 * dealing with concurrent invocation and without grabbing the mmap lock in
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 /* set the new file, lockless */
1467 ret = deny_write_access(new_exe_file);
1470 get_file(new_exe_file);
1472 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1475 * Don't race with dup_mmap() getting the file and disallowing
1476 * write access while someone might open the file writable.
1479 allow_write_access(old_exe_file);
1481 mmap_read_unlock(mm);
1487 * get_mm_exe_file - acquire a reference to the mm's executable file
1489 * Returns %NULL if mm has no associated executable file.
1490 * User must release file via fput().
1492 struct file *get_mm_exe_file(struct mm_struct *mm)
1494 struct file *exe_file;
1497 exe_file = rcu_dereference(mm->exe_file);
1498 if (exe_file && !get_file_rcu(exe_file))
1505 * get_task_exe_file - acquire a reference to the task's executable file
1507 * Returns %NULL if task's mm (if any) has no associated executable file or
1508 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1509 * User must release file via fput().
1511 struct file *get_task_exe_file(struct task_struct *task)
1513 struct file *exe_file = NULL;
1514 struct mm_struct *mm;
1519 if (!(task->flags & PF_KTHREAD))
1520 exe_file = get_mm_exe_file(mm);
1527 * get_task_mm - acquire a reference to the task's mm
1529 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1530 * this kernel workthread has transiently adopted a user mm with use_mm,
1531 * to do its AIO) is not set and if so returns a reference to it, after
1532 * bumping up the use count. User must release the mm via mmput()
1533 * after use. Typically used by /proc and ptrace.
1535 struct mm_struct *get_task_mm(struct task_struct *task)
1537 struct mm_struct *mm;
1542 if (task->flags & PF_KTHREAD)
1550 EXPORT_SYMBOL_GPL(get_task_mm);
1552 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1554 struct mm_struct *mm;
1557 err = down_read_killable(&task->signal->exec_update_lock);
1559 return ERR_PTR(err);
1561 mm = get_task_mm(task);
1562 if (mm && mm != current->mm &&
1563 !ptrace_may_access(task, mode)) {
1565 mm = ERR_PTR(-EACCES);
1567 up_read(&task->signal->exec_update_lock);
1572 static void complete_vfork_done(struct task_struct *tsk)
1574 struct completion *vfork;
1577 vfork = tsk->vfork_done;
1578 if (likely(vfork)) {
1579 tsk->vfork_done = NULL;
1585 static int wait_for_vfork_done(struct task_struct *child,
1586 struct completion *vfork)
1588 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1591 cgroup_enter_frozen();
1592 killed = wait_for_completion_state(vfork, state);
1593 cgroup_leave_frozen(false);
1597 child->vfork_done = NULL;
1601 put_task_struct(child);
1605 /* Please note the differences between mmput and mm_release.
1606 * mmput is called whenever we stop holding onto a mm_struct,
1607 * error success whatever.
1609 * mm_release is called after a mm_struct has been removed
1610 * from the current process.
1612 * This difference is important for error handling, when we
1613 * only half set up a mm_struct for a new process and need to restore
1614 * the old one. Because we mmput the new mm_struct before
1615 * restoring the old one. . .
1616 * Eric Biederman 10 January 1998
1618 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1620 uprobe_free_utask(tsk);
1622 /* Get rid of any cached register state */
1623 deactivate_mm(tsk, mm);
1626 * Signal userspace if we're not exiting with a core dump
1627 * because we want to leave the value intact for debugging
1630 if (tsk->clear_child_tid) {
1631 if (atomic_read(&mm->mm_users) > 1) {
1633 * We don't check the error code - if userspace has
1634 * not set up a proper pointer then tough luck.
1636 put_user(0, tsk->clear_child_tid);
1637 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1638 1, NULL, NULL, 0, 0);
1640 tsk->clear_child_tid = NULL;
1644 * All done, finally we can wake up parent and return this mm to him.
1645 * Also kthread_stop() uses this completion for synchronization.
1647 if (tsk->vfork_done)
1648 complete_vfork_done(tsk);
1651 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1653 futex_exit_release(tsk);
1654 mm_release(tsk, mm);
1657 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1659 futex_exec_release(tsk);
1660 mm_release(tsk, mm);
1664 * dup_mm() - duplicates an existing mm structure
1665 * @tsk: the task_struct with which the new mm will be associated.
1666 * @oldmm: the mm to duplicate.
1668 * Allocates a new mm structure and duplicates the provided @oldmm structure
1671 * Return: the duplicated mm or NULL on failure.
1673 static struct mm_struct *dup_mm(struct task_struct *tsk,
1674 struct mm_struct *oldmm)
1676 struct mm_struct *mm;
1683 memcpy(mm, oldmm, sizeof(*mm));
1685 if (!mm_init(mm, tsk, mm->user_ns))
1688 err = dup_mmap(mm, oldmm);
1692 mm->hiwater_rss = get_mm_rss(mm);
1693 mm->hiwater_vm = mm->total_vm;
1695 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1701 /* don't put binfmt in mmput, we haven't got module yet */
1703 mm_init_owner(mm, NULL);
1710 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1712 struct mm_struct *mm, *oldmm;
1714 tsk->min_flt = tsk->maj_flt = 0;
1715 tsk->nvcsw = tsk->nivcsw = 0;
1716 #ifdef CONFIG_DETECT_HUNG_TASK
1717 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1718 tsk->last_switch_time = 0;
1722 tsk->active_mm = NULL;
1725 * Are we cloning a kernel thread?
1727 * We need to steal a active VM for that..
1729 oldmm = current->mm;
1733 if (clone_flags & CLONE_VM) {
1737 mm = dup_mm(tsk, current->mm);
1743 tsk->active_mm = mm;
1744 sched_mm_cid_fork(tsk);
1748 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1750 struct fs_struct *fs = current->fs;
1751 if (clone_flags & CLONE_FS) {
1752 /* tsk->fs is already what we want */
1753 spin_lock(&fs->lock);
1755 spin_unlock(&fs->lock);
1759 spin_unlock(&fs->lock);
1762 tsk->fs = copy_fs_struct(fs);
1768 static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1771 struct files_struct *oldf, *newf;
1775 * A background process may not have any files ...
1777 oldf = current->files;
1786 if (clone_flags & CLONE_FILES) {
1787 atomic_inc(&oldf->count);
1791 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1801 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1803 struct sighand_struct *sig;
1805 if (clone_flags & CLONE_SIGHAND) {
1806 refcount_inc(¤t->sighand->count);
1809 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1810 RCU_INIT_POINTER(tsk->sighand, sig);
1814 refcount_set(&sig->count, 1);
1815 spin_lock_irq(¤t->sighand->siglock);
1816 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1817 spin_unlock_irq(¤t->sighand->siglock);
1819 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1820 if (clone_flags & CLONE_CLEAR_SIGHAND)
1821 flush_signal_handlers(tsk, 0);
1826 void __cleanup_sighand(struct sighand_struct *sighand)
1828 if (refcount_dec_and_test(&sighand->count)) {
1829 signalfd_cleanup(sighand);
1831 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1832 * without an RCU grace period, see __lock_task_sighand().
1834 kmem_cache_free(sighand_cachep, sighand);
1839 * Initialize POSIX timer handling for a thread group.
1841 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1843 struct posix_cputimers *pct = &sig->posix_cputimers;
1844 unsigned long cpu_limit;
1846 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1847 posix_cputimers_group_init(pct, cpu_limit);
1850 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1852 struct signal_struct *sig;
1854 if (clone_flags & CLONE_THREAD)
1857 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1862 sig->nr_threads = 1;
1863 sig->quick_threads = 1;
1864 atomic_set(&sig->live, 1);
1865 refcount_set(&sig->sigcnt, 1);
1867 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1868 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1869 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1871 init_waitqueue_head(&sig->wait_chldexit);
1872 sig->curr_target = tsk;
1873 init_sigpending(&sig->shared_pending);
1874 INIT_HLIST_HEAD(&sig->multiprocess);
1875 seqlock_init(&sig->stats_lock);
1876 prev_cputime_init(&sig->prev_cputime);
1878 #ifdef CONFIG_POSIX_TIMERS
1879 INIT_LIST_HEAD(&sig->posix_timers);
1880 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1881 sig->real_timer.function = it_real_fn;
1884 task_lock(current->group_leader);
1885 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1886 task_unlock(current->group_leader);
1888 posix_cpu_timers_init_group(sig);
1890 tty_audit_fork(sig);
1891 sched_autogroup_fork(sig);
1893 sig->oom_score_adj = current->signal->oom_score_adj;
1894 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1896 mutex_init(&sig->cred_guard_mutex);
1897 init_rwsem(&sig->exec_update_lock);
1902 static void copy_seccomp(struct task_struct *p)
1904 #ifdef CONFIG_SECCOMP
1906 * Must be called with sighand->lock held, which is common to
1907 * all threads in the group. Holding cred_guard_mutex is not
1908 * needed because this new task is not yet running and cannot
1911 assert_spin_locked(¤t->sighand->siglock);
1913 /* Ref-count the new filter user, and assign it. */
1914 get_seccomp_filter(current);
1915 p->seccomp = current->seccomp;
1918 * Explicitly enable no_new_privs here in case it got set
1919 * between the task_struct being duplicated and holding the
1920 * sighand lock. The seccomp state and nnp must be in sync.
1922 if (task_no_new_privs(current))
1923 task_set_no_new_privs(p);
1926 * If the parent gained a seccomp mode after copying thread
1927 * flags and between before we held the sighand lock, we have
1928 * to manually enable the seccomp thread flag here.
1930 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1931 set_task_syscall_work(p, SECCOMP);
1935 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1937 current->clear_child_tid = tidptr;
1939 return task_pid_vnr(current);
1942 static void rt_mutex_init_task(struct task_struct *p)
1944 raw_spin_lock_init(&p->pi_lock);
1945 #ifdef CONFIG_RT_MUTEXES
1946 p->pi_waiters = RB_ROOT_CACHED;
1947 p->pi_top_task = NULL;
1948 p->pi_blocked_on = NULL;
1952 static inline void init_task_pid_links(struct task_struct *task)
1956 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1957 INIT_HLIST_NODE(&task->pid_links[type]);
1961 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1963 if (type == PIDTYPE_PID)
1964 task->thread_pid = pid;
1966 task->signal->pids[type] = pid;
1969 static inline void rcu_copy_process(struct task_struct *p)
1971 #ifdef CONFIG_PREEMPT_RCU
1972 p->rcu_read_lock_nesting = 0;
1973 p->rcu_read_unlock_special.s = 0;
1974 p->rcu_blocked_node = NULL;
1975 INIT_LIST_HEAD(&p->rcu_node_entry);
1976 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1977 #ifdef CONFIG_TASKS_RCU
1978 p->rcu_tasks_holdout = false;
1979 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1980 p->rcu_tasks_idle_cpu = -1;
1981 #endif /* #ifdef CONFIG_TASKS_RCU */
1982 #ifdef CONFIG_TASKS_TRACE_RCU
1983 p->trc_reader_nesting = 0;
1984 p->trc_reader_special.s = 0;
1985 INIT_LIST_HEAD(&p->trc_holdout_list);
1986 INIT_LIST_HEAD(&p->trc_blkd_node);
1987 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1990 struct pid *pidfd_pid(const struct file *file)
1992 if (file->f_op == &pidfd_fops)
1993 return file->private_data;
1995 return ERR_PTR(-EBADF);
1998 static int pidfd_release(struct inode *inode, struct file *file)
2000 struct pid *pid = file->private_data;
2002 file->private_data = NULL;
2007 #ifdef CONFIG_PROC_FS
2009 * pidfd_show_fdinfo - print information about a pidfd
2010 * @m: proc fdinfo file
2011 * @f: file referencing a pidfd
2014 * This function will print the pid that a given pidfd refers to in the
2015 * pid namespace of the procfs instance.
2016 * If the pid namespace of the process is not a descendant of the pid
2017 * namespace of the procfs instance 0 will be shown as its pid. This is
2018 * similar to calling getppid() on a process whose parent is outside of
2019 * its pid namespace.
2022 * If pid namespaces are supported then this function will also print
2023 * the pid of a given pidfd refers to for all descendant pid namespaces
2024 * starting from the current pid namespace of the instance, i.e. the
2025 * Pid field and the first entry in the NSpid field will be identical.
2026 * If the pid namespace of the process is not a descendant of the pid
2027 * namespace of the procfs instance 0 will be shown as its first NSpid
2028 * entry and no others will be shown.
2029 * Note that this differs from the Pid and NSpid fields in
2030 * /proc/<pid>/status where Pid and NSpid are always shown relative to
2031 * the pid namespace of the procfs instance. The difference becomes
2032 * obvious when sending around a pidfd between pid namespaces from a
2033 * different branch of the tree, i.e. where no ancestral relation is
2034 * present between the pid namespaces:
2035 * - create two new pid namespaces ns1 and ns2 in the initial pid
2036 * namespace (also take care to create new mount namespaces in the
2037 * new pid namespace and mount procfs)
2038 * - create a process with a pidfd in ns1
2039 * - send pidfd from ns1 to ns2
2040 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
2041 * have exactly one entry, which is 0
2043 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
2045 struct pid *pid = f->private_data;
2046 struct pid_namespace *ns;
2049 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
2050 ns = proc_pid_ns(file_inode(m->file)->i_sb);
2051 nr = pid_nr_ns(pid, ns);
2054 seq_put_decimal_ll(m, "Pid:\t", nr);
2056 #ifdef CONFIG_PID_NS
2057 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
2061 /* If nr is non-zero it means that 'pid' is valid and that
2062 * ns, i.e. the pid namespace associated with the procfs
2063 * instance, is in the pid namespace hierarchy of pid.
2064 * Start at one below the already printed level.
2066 for (i = ns->level + 1; i <= pid->level; i++)
2067 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2075 * Poll support for process exit notification.
2077 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2079 struct pid *pid = file->private_data;
2080 __poll_t poll_flags = 0;
2082 poll_wait(file, &pid->wait_pidfd, pts);
2085 * Inform pollers only when the whole thread group exits.
2086 * If the thread group leader exits before all other threads in the
2087 * group, then poll(2) should block, similar to the wait(2) family.
2089 if (thread_group_exited(pid))
2090 poll_flags = EPOLLIN | EPOLLRDNORM;
2095 const struct file_operations pidfd_fops = {
2096 .release = pidfd_release,
2098 #ifdef CONFIG_PROC_FS
2099 .show_fdinfo = pidfd_show_fdinfo,
2104 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2105 * @pid: the struct pid for which to create a pidfd
2106 * @flags: flags of the new @pidfd
2107 * @pidfd: the pidfd to return
2109 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2110 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2112 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2113 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2114 * pidfd file are prepared.
2116 * If this function returns successfully the caller is responsible to either
2117 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2118 * order to install the pidfd into its file descriptor table or they must use
2119 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2122 * This function is useful when a pidfd must already be reserved but there
2123 * might still be points of failure afterwards and the caller wants to ensure
2124 * that no pidfd is leaked into its file descriptor table.
2126 * Return: On success, a reserved pidfd is returned from the function and a new
2127 * pidfd file is returned in the last argument to the function. On
2128 * error, a negative error code is returned from the function and the
2129 * last argument remains unchanged.
2131 static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2134 struct file *pidfd_file;
2136 if (flags & ~(O_NONBLOCK | O_RDWR | O_CLOEXEC))
2139 pidfd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2143 pidfd_file = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2144 flags | O_RDWR | O_CLOEXEC);
2145 if (IS_ERR(pidfd_file)) {
2146 put_unused_fd(pidfd);
2147 return PTR_ERR(pidfd_file);
2149 get_pid(pid); /* held by pidfd_file now */
2155 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2156 * @pid: the struct pid for which to create a pidfd
2157 * @flags: flags of the new @pidfd
2158 * @pidfd: the pidfd to return
2160 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2161 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2163 * The helper verifies that @pid is used as a thread group leader.
2165 * If this function returns successfully the caller is responsible to either
2166 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2167 * order to install the pidfd into its file descriptor table or they must use
2168 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2171 * This function is useful when a pidfd must already be reserved but there
2172 * might still be points of failure afterwards and the caller wants to ensure
2173 * that no pidfd is leaked into its file descriptor table.
2175 * Return: On success, a reserved pidfd is returned from the function and a new
2176 * pidfd file is returned in the last argument to the function. On
2177 * error, a negative error code is returned from the function and the
2178 * last argument remains unchanged.
2180 int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2182 if (!pid || !pid_has_task(pid, PIDTYPE_TGID))
2185 return __pidfd_prepare(pid, flags, ret);
2188 static void __delayed_free_task(struct rcu_head *rhp)
2190 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2195 static __always_inline void delayed_free_task(struct task_struct *tsk)
2197 if (IS_ENABLED(CONFIG_MEMCG))
2198 call_rcu(&tsk->rcu, __delayed_free_task);
2203 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2205 /* Skip if kernel thread */
2209 /* Skip if spawning a thread or using vfork */
2210 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2213 /* We need to synchronize with __set_oom_adj */
2214 mutex_lock(&oom_adj_mutex);
2215 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2216 /* Update the values in case they were changed after copy_signal */
2217 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2218 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2219 mutex_unlock(&oom_adj_mutex);
2223 static void rv_task_fork(struct task_struct *p)
2227 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2228 p->rv[i].da_mon.monitoring = false;
2231 #define rv_task_fork(p) do {} while (0)
2235 * This creates a new process as a copy of the old one,
2236 * but does not actually start it yet.
2238 * It copies the registers, and all the appropriate
2239 * parts of the process environment (as per the clone
2240 * flags). The actual kick-off is left to the caller.
2242 __latent_entropy struct task_struct *copy_process(
2246 struct kernel_clone_args *args)
2248 int pidfd = -1, retval;
2249 struct task_struct *p;
2250 struct multiprocess_signals delayed;
2251 struct file *pidfile = NULL;
2252 const u64 clone_flags = args->flags;
2253 struct nsproxy *nsp = current->nsproxy;
2256 * Don't allow sharing the root directory with processes in a different
2259 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2260 return ERR_PTR(-EINVAL);
2262 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2263 return ERR_PTR(-EINVAL);
2266 * Thread groups must share signals as well, and detached threads
2267 * can only be started up within the thread group.
2269 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2270 return ERR_PTR(-EINVAL);
2273 * Shared signal handlers imply shared VM. By way of the above,
2274 * thread groups also imply shared VM. Blocking this case allows
2275 * for various simplifications in other code.
2277 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2278 return ERR_PTR(-EINVAL);
2281 * Siblings of global init remain as zombies on exit since they are
2282 * not reaped by their parent (swapper). To solve this and to avoid
2283 * multi-rooted process trees, prevent global and container-inits
2284 * from creating siblings.
2286 if ((clone_flags & CLONE_PARENT) &&
2287 current->signal->flags & SIGNAL_UNKILLABLE)
2288 return ERR_PTR(-EINVAL);
2291 * If the new process will be in a different pid or user namespace
2292 * do not allow it to share a thread group with the forking task.
2294 if (clone_flags & CLONE_THREAD) {
2295 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2296 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2297 return ERR_PTR(-EINVAL);
2300 if (clone_flags & CLONE_PIDFD) {
2302 * - CLONE_DETACHED is blocked so that we can potentially
2303 * reuse it later for CLONE_PIDFD.
2304 * - CLONE_THREAD is blocked until someone really needs it.
2306 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2307 return ERR_PTR(-EINVAL);
2311 * Force any signals received before this point to be delivered
2312 * before the fork happens. Collect up signals sent to multiple
2313 * processes that happen during the fork and delay them so that
2314 * they appear to happen after the fork.
2316 sigemptyset(&delayed.signal);
2317 INIT_HLIST_NODE(&delayed.node);
2319 spin_lock_irq(¤t->sighand->siglock);
2320 if (!(clone_flags & CLONE_THREAD))
2321 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2322 recalc_sigpending();
2323 spin_unlock_irq(¤t->sighand->siglock);
2324 retval = -ERESTARTNOINTR;
2325 if (task_sigpending(current))
2329 p = dup_task_struct(current, node);
2332 p->flags &= ~PF_KTHREAD;
2334 p->flags |= PF_KTHREAD;
2335 if (args->user_worker) {
2337 * Mark us a user worker, and block any signal that isn't
2340 p->flags |= PF_USER_WORKER;
2341 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2343 if (args->io_thread)
2344 p->flags |= PF_IO_WORKER;
2347 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2349 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2351 * Clear TID on mm_release()?
2353 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2355 ftrace_graph_init_task(p);
2357 rt_mutex_init_task(p);
2359 lockdep_assert_irqs_enabled();
2360 #ifdef CONFIG_PROVE_LOCKING
2361 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2363 retval = copy_creds(p, clone_flags);
2368 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2369 if (p->real_cred->user != INIT_USER &&
2370 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2371 goto bad_fork_cleanup_count;
2373 current->flags &= ~PF_NPROC_EXCEEDED;
2376 * If multiple threads are within copy_process(), then this check
2377 * triggers too late. This doesn't hurt, the check is only there
2378 * to stop root fork bombs.
2381 if (data_race(nr_threads >= max_threads))
2382 goto bad_fork_cleanup_count;
2384 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2385 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2386 p->flags |= PF_FORKNOEXEC;
2387 INIT_LIST_HEAD(&p->children);
2388 INIT_LIST_HEAD(&p->sibling);
2389 rcu_copy_process(p);
2390 p->vfork_done = NULL;
2391 spin_lock_init(&p->alloc_lock);
2393 init_sigpending(&p->pending);
2395 p->utime = p->stime = p->gtime = 0;
2396 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2397 p->utimescaled = p->stimescaled = 0;
2399 prev_cputime_init(&p->prev_cputime);
2401 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2402 seqcount_init(&p->vtime.seqcount);
2403 p->vtime.starttime = 0;
2404 p->vtime.state = VTIME_INACTIVE;
2407 #ifdef CONFIG_IO_URING
2411 #if defined(SPLIT_RSS_COUNTING)
2412 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2415 p->default_timer_slack_ns = current->timer_slack_ns;
2421 task_io_accounting_init(&p->ioac);
2422 acct_clear_integrals(p);
2424 posix_cputimers_init(&p->posix_cputimers);
2426 p->io_context = NULL;
2427 audit_set_context(p, NULL);
2429 if (args->kthread) {
2430 if (!set_kthread_struct(p))
2431 goto bad_fork_cleanup_delayacct;
2434 p->mempolicy = mpol_dup(p->mempolicy);
2435 if (IS_ERR(p->mempolicy)) {
2436 retval = PTR_ERR(p->mempolicy);
2437 p->mempolicy = NULL;
2438 goto bad_fork_cleanup_delayacct;
2441 #ifdef CONFIG_CPUSETS
2442 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2443 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2444 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2446 #ifdef CONFIG_TRACE_IRQFLAGS
2447 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2448 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2449 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2450 p->softirqs_enabled = 1;
2451 p->softirq_context = 0;
2454 p->pagefault_disabled = 0;
2456 #ifdef CONFIG_LOCKDEP
2457 lockdep_init_task(p);
2460 #ifdef CONFIG_DEBUG_MUTEXES
2461 p->blocked_on = NULL; /* not blocked yet */
2463 #ifdef CONFIG_BCACHE
2464 p->sequential_io = 0;
2465 p->sequential_io_avg = 0;
2467 #ifdef CONFIG_BPF_SYSCALL
2468 RCU_INIT_POINTER(p->bpf_storage, NULL);
2472 /* Perform scheduler related setup. Assign this task to a CPU. */
2473 retval = sched_fork(clone_flags, p);
2475 goto bad_fork_cleanup_policy;
2477 retval = perf_event_init_task(p, clone_flags);
2479 goto bad_fork_cleanup_policy;
2480 retval = audit_alloc(p);
2482 goto bad_fork_cleanup_perf;
2483 /* copy all the process information */
2485 retval = security_task_alloc(p, clone_flags);
2487 goto bad_fork_cleanup_audit;
2488 retval = copy_semundo(clone_flags, p);
2490 goto bad_fork_cleanup_security;
2491 retval = copy_files(clone_flags, p, args->no_files);
2493 goto bad_fork_cleanup_semundo;
2494 retval = copy_fs(clone_flags, p);
2496 goto bad_fork_cleanup_files;
2497 retval = copy_sighand(clone_flags, p);
2499 goto bad_fork_cleanup_fs;
2500 retval = copy_signal(clone_flags, p);
2502 goto bad_fork_cleanup_sighand;
2503 retval = copy_mm(clone_flags, p);
2505 goto bad_fork_cleanup_signal;
2506 retval = copy_namespaces(clone_flags, p);
2508 goto bad_fork_cleanup_mm;
2509 retval = copy_io(clone_flags, p);
2511 goto bad_fork_cleanup_namespaces;
2512 retval = copy_thread(p, args);
2514 goto bad_fork_cleanup_io;
2516 stackleak_task_init(p);
2518 if (pid != &init_struct_pid) {
2519 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2520 args->set_tid_size);
2522 retval = PTR_ERR(pid);
2523 goto bad_fork_cleanup_thread;
2528 * This has to happen after we've potentially unshared the file
2529 * descriptor table (so that the pidfd doesn't leak into the child
2530 * if the fd table isn't shared).
2532 if (clone_flags & CLONE_PIDFD) {
2533 /* Note that no task has been attached to @pid yet. */
2534 retval = __pidfd_prepare(pid, O_RDWR | O_CLOEXEC, &pidfile);
2536 goto bad_fork_free_pid;
2539 retval = put_user(pidfd, args->pidfd);
2541 goto bad_fork_put_pidfd;
2550 * sigaltstack should be cleared when sharing the same VM
2552 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2556 * Syscall tracing and stepping should be turned off in the
2557 * child regardless of CLONE_PTRACE.
2559 user_disable_single_step(p);
2560 clear_task_syscall_work(p, SYSCALL_TRACE);
2561 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2562 clear_task_syscall_work(p, SYSCALL_EMU);
2564 clear_tsk_latency_tracing(p);
2566 /* ok, now we should be set up.. */
2567 p->pid = pid_nr(pid);
2568 if (clone_flags & CLONE_THREAD) {
2569 p->group_leader = current->group_leader;
2570 p->tgid = current->tgid;
2572 p->group_leader = p;
2577 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2578 p->dirty_paused_when = 0;
2580 p->pdeath_signal = 0;
2581 INIT_LIST_HEAD(&p->thread_group);
2582 p->task_works = NULL;
2583 clear_posix_cputimers_work(p);
2585 #ifdef CONFIG_KRETPROBES
2586 p->kretprobe_instances.first = NULL;
2588 #ifdef CONFIG_RETHOOK
2589 p->rethooks.first = NULL;
2593 * Ensure that the cgroup subsystem policies allow the new process to be
2594 * forked. It should be noted that the new process's css_set can be changed
2595 * between here and cgroup_post_fork() if an organisation operation is in
2598 retval = cgroup_can_fork(p, args);
2600 goto bad_fork_put_pidfd;
2603 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2604 * the new task on the correct runqueue. All this *before* the task
2607 * This isn't part of ->can_fork() because while the re-cloning is
2608 * cgroup specific, it unconditionally needs to place the task on a
2611 sched_cgroup_fork(p, args);
2614 * From this point on we must avoid any synchronous user-space
2615 * communication until we take the tasklist-lock. In particular, we do
2616 * not want user-space to be able to predict the process start-time by
2617 * stalling fork(2) after we recorded the start_time but before it is
2618 * visible to the system.
2621 p->start_time = ktime_get_ns();
2622 p->start_boottime = ktime_get_boottime_ns();
2625 * Make it visible to the rest of the system, but dont wake it up yet.
2626 * Need tasklist lock for parent etc handling!
2628 write_lock_irq(&tasklist_lock);
2630 /* CLONE_PARENT re-uses the old parent */
2631 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2632 p->real_parent = current->real_parent;
2633 p->parent_exec_id = current->parent_exec_id;
2634 if (clone_flags & CLONE_THREAD)
2635 p->exit_signal = -1;
2637 p->exit_signal = current->group_leader->exit_signal;
2639 p->real_parent = current;
2640 p->parent_exec_id = current->self_exec_id;
2641 p->exit_signal = args->exit_signal;
2644 klp_copy_process(p);
2648 spin_lock(¤t->sighand->siglock);
2652 rseq_fork(p, clone_flags);
2654 /* Don't start children in a dying pid namespace */
2655 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2657 goto bad_fork_cancel_cgroup;
2660 /* Let kill terminate clone/fork in the middle */
2661 if (fatal_signal_pending(current)) {
2663 goto bad_fork_cancel_cgroup;
2666 /* No more failure paths after this point. */
2669 * Copy seccomp details explicitly here, in case they were changed
2670 * before holding sighand lock.
2674 init_task_pid_links(p);
2675 if (likely(p->pid)) {
2676 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2678 init_task_pid(p, PIDTYPE_PID, pid);
2679 if (thread_group_leader(p)) {
2680 init_task_pid(p, PIDTYPE_TGID, pid);
2681 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2682 init_task_pid(p, PIDTYPE_SID, task_session(current));
2684 if (is_child_reaper(pid)) {
2685 ns_of_pid(pid)->child_reaper = p;
2686 p->signal->flags |= SIGNAL_UNKILLABLE;
2688 p->signal->shared_pending.signal = delayed.signal;
2689 p->signal->tty = tty_kref_get(current->signal->tty);
2691 * Inherit has_child_subreaper flag under the same
2692 * tasklist_lock with adding child to the process tree
2693 * for propagate_has_child_subreaper optimization.
2695 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2696 p->real_parent->signal->is_child_subreaper;
2697 list_add_tail(&p->sibling, &p->real_parent->children);
2698 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2699 attach_pid(p, PIDTYPE_TGID);
2700 attach_pid(p, PIDTYPE_PGID);
2701 attach_pid(p, PIDTYPE_SID);
2702 __this_cpu_inc(process_counts);
2704 current->signal->nr_threads++;
2705 current->signal->quick_threads++;
2706 atomic_inc(¤t->signal->live);
2707 refcount_inc(¤t->signal->sigcnt);
2708 task_join_group_stop(p);
2709 list_add_tail_rcu(&p->thread_group,
2710 &p->group_leader->thread_group);
2711 list_add_tail_rcu(&p->thread_node,
2712 &p->signal->thread_head);
2714 attach_pid(p, PIDTYPE_PID);
2718 hlist_del_init(&delayed.node);
2719 spin_unlock(¤t->sighand->siglock);
2720 syscall_tracepoint_update(p);
2721 write_unlock_irq(&tasklist_lock);
2724 fd_install(pidfd, pidfile);
2726 proc_fork_connector(p);
2728 cgroup_post_fork(p, args);
2731 trace_task_newtask(p, clone_flags);
2732 uprobe_copy_process(p, clone_flags);
2733 user_events_fork(p, clone_flags);
2735 copy_oom_score_adj(clone_flags, p);
2739 bad_fork_cancel_cgroup:
2741 spin_unlock(¤t->sighand->siglock);
2742 write_unlock_irq(&tasklist_lock);
2743 cgroup_cancel_fork(p, args);
2745 if (clone_flags & CLONE_PIDFD) {
2747 put_unused_fd(pidfd);
2750 if (pid != &init_struct_pid)
2752 bad_fork_cleanup_thread:
2754 bad_fork_cleanup_io:
2757 bad_fork_cleanup_namespaces:
2758 exit_task_namespaces(p);
2759 bad_fork_cleanup_mm:
2761 mm_clear_owner(p->mm, p);
2764 bad_fork_cleanup_signal:
2765 if (!(clone_flags & CLONE_THREAD))
2766 free_signal_struct(p->signal);
2767 bad_fork_cleanup_sighand:
2768 __cleanup_sighand(p->sighand);
2769 bad_fork_cleanup_fs:
2770 exit_fs(p); /* blocking */
2771 bad_fork_cleanup_files:
2772 exit_files(p); /* blocking */
2773 bad_fork_cleanup_semundo:
2775 bad_fork_cleanup_security:
2776 security_task_free(p);
2777 bad_fork_cleanup_audit:
2779 bad_fork_cleanup_perf:
2780 perf_event_free_task(p);
2781 bad_fork_cleanup_policy:
2782 lockdep_free_task(p);
2784 mpol_put(p->mempolicy);
2786 bad_fork_cleanup_delayacct:
2787 delayacct_tsk_free(p);
2788 bad_fork_cleanup_count:
2789 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2792 WRITE_ONCE(p->__state, TASK_DEAD);
2793 exit_task_stack_account(p);
2795 delayed_free_task(p);
2797 spin_lock_irq(¤t->sighand->siglock);
2798 hlist_del_init(&delayed.node);
2799 spin_unlock_irq(¤t->sighand->siglock);
2800 return ERR_PTR(retval);
2803 static inline void init_idle_pids(struct task_struct *idle)
2807 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2808 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2809 init_task_pid(idle, type, &init_struct_pid);
2813 static int idle_dummy(void *dummy)
2815 /* This function is never called */
2819 struct task_struct * __init fork_idle(int cpu)
2821 struct task_struct *task;
2822 struct kernel_clone_args args = {
2830 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2831 if (!IS_ERR(task)) {
2832 init_idle_pids(task);
2833 init_idle(task, cpu);
2840 * This is like kernel_clone(), but shaved down and tailored to just
2841 * creating io_uring workers. It returns a created task, or an error pointer.
2842 * The returned task is inactive, and the caller must fire it up through
2843 * wake_up_new_task(p). All signals are blocked in the created task.
2845 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2847 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2849 struct kernel_clone_args args = {
2850 .flags = ((lower_32_bits(flags) | CLONE_VM |
2851 CLONE_UNTRACED) & ~CSIGNAL),
2852 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2859 return copy_process(NULL, 0, node, &args);
2863 * Ok, this is the main fork-routine.
2865 * It copies the process, and if successful kick-starts
2866 * it and waits for it to finish using the VM if required.
2868 * args->exit_signal is expected to be checked for sanity by the caller.
2870 pid_t kernel_clone(struct kernel_clone_args *args)
2872 u64 clone_flags = args->flags;
2873 struct completion vfork;
2875 struct task_struct *p;
2880 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2881 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2882 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2883 * field in struct clone_args and it still doesn't make sense to have
2884 * them both point at the same memory location. Performing this check
2885 * here has the advantage that we don't need to have a separate helper
2886 * to check for legacy clone().
2888 if ((args->flags & CLONE_PIDFD) &&
2889 (args->flags & CLONE_PARENT_SETTID) &&
2890 (args->pidfd == args->parent_tid))
2894 * Determine whether and which event to report to ptracer. When
2895 * called from kernel_thread or CLONE_UNTRACED is explicitly
2896 * requested, no event is reported; otherwise, report if the event
2897 * for the type of forking is enabled.
2899 if (!(clone_flags & CLONE_UNTRACED)) {
2900 if (clone_flags & CLONE_VFORK)
2901 trace = PTRACE_EVENT_VFORK;
2902 else if (args->exit_signal != SIGCHLD)
2903 trace = PTRACE_EVENT_CLONE;
2905 trace = PTRACE_EVENT_FORK;
2907 if (likely(!ptrace_event_enabled(current, trace)))
2911 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2912 add_latent_entropy();
2918 * Do this prior waking up the new thread - the thread pointer
2919 * might get invalid after that point, if the thread exits quickly.
2921 trace_sched_process_fork(current, p);
2923 pid = get_task_pid(p, PIDTYPE_PID);
2926 if (clone_flags & CLONE_PARENT_SETTID)
2927 put_user(nr, args->parent_tid);
2929 if (clone_flags & CLONE_VFORK) {
2930 p->vfork_done = &vfork;
2931 init_completion(&vfork);
2935 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2936 /* lock the task to synchronize with memcg migration */
2938 lru_gen_add_mm(p->mm);
2942 wake_up_new_task(p);
2944 /* forking complete and child started to run, tell ptracer */
2945 if (unlikely(trace))
2946 ptrace_event_pid(trace, pid);
2948 if (clone_flags & CLONE_VFORK) {
2949 if (!wait_for_vfork_done(p, &vfork))
2950 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2958 * Create a kernel thread.
2960 pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2961 unsigned long flags)
2963 struct kernel_clone_args args = {
2964 .flags = ((lower_32_bits(flags) | CLONE_VM |
2965 CLONE_UNTRACED) & ~CSIGNAL),
2966 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2973 return kernel_clone(&args);
2977 * Create a user mode thread.
2979 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2981 struct kernel_clone_args args = {
2982 .flags = ((lower_32_bits(flags) | CLONE_VM |
2983 CLONE_UNTRACED) & ~CSIGNAL),
2984 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2989 return kernel_clone(&args);
2992 #ifdef __ARCH_WANT_SYS_FORK
2993 SYSCALL_DEFINE0(fork)
2996 struct kernel_clone_args args = {
2997 .exit_signal = SIGCHLD,
3000 return kernel_clone(&args);
3002 /* can not support in nommu mode */
3008 #ifdef __ARCH_WANT_SYS_VFORK
3009 SYSCALL_DEFINE0(vfork)
3011 struct kernel_clone_args args = {
3012 .flags = CLONE_VFORK | CLONE_VM,
3013 .exit_signal = SIGCHLD,
3016 return kernel_clone(&args);
3020 #ifdef __ARCH_WANT_SYS_CLONE
3021 #ifdef CONFIG_CLONE_BACKWARDS
3022 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3023 int __user *, parent_tidptr,
3025 int __user *, child_tidptr)
3026 #elif defined(CONFIG_CLONE_BACKWARDS2)
3027 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
3028 int __user *, parent_tidptr,
3029 int __user *, child_tidptr,
3031 #elif defined(CONFIG_CLONE_BACKWARDS3)
3032 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
3034 int __user *, parent_tidptr,
3035 int __user *, child_tidptr,
3038 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3039 int __user *, parent_tidptr,
3040 int __user *, child_tidptr,
3044 struct kernel_clone_args args = {
3045 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
3046 .pidfd = parent_tidptr,
3047 .child_tid = child_tidptr,
3048 .parent_tid = parent_tidptr,
3049 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
3054 return kernel_clone(&args);
3058 #ifdef __ARCH_WANT_SYS_CLONE3
3060 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
3061 struct clone_args __user *uargs,
3065 struct clone_args args;
3066 pid_t *kset_tid = kargs->set_tid;
3068 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
3069 CLONE_ARGS_SIZE_VER0);
3070 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
3071 CLONE_ARGS_SIZE_VER1);
3072 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
3073 CLONE_ARGS_SIZE_VER2);
3074 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
3076 if (unlikely(usize > PAGE_SIZE))
3078 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
3081 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
3085 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
3088 if (unlikely(!args.set_tid && args.set_tid_size > 0))
3091 if (unlikely(args.set_tid && args.set_tid_size == 0))
3095 * Verify that higher 32bits of exit_signal are unset and that
3096 * it is a valid signal
3098 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3099 !valid_signal(args.exit_signal)))
3102 if ((args.flags & CLONE_INTO_CGROUP) &&
3103 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3106 *kargs = (struct kernel_clone_args){
3107 .flags = args.flags,
3108 .pidfd = u64_to_user_ptr(args.pidfd),
3109 .child_tid = u64_to_user_ptr(args.child_tid),
3110 .parent_tid = u64_to_user_ptr(args.parent_tid),
3111 .exit_signal = args.exit_signal,
3112 .stack = args.stack,
3113 .stack_size = args.stack_size,
3115 .set_tid_size = args.set_tid_size,
3116 .cgroup = args.cgroup,
3120 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3121 (kargs->set_tid_size * sizeof(pid_t))))
3124 kargs->set_tid = kset_tid;
3130 * clone3_stack_valid - check and prepare stack
3131 * @kargs: kernel clone args
3133 * Verify that the stack arguments userspace gave us are sane.
3134 * In addition, set the stack direction for userspace since it's easy for us to
3137 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3139 if (kargs->stack == 0) {
3140 if (kargs->stack_size > 0)
3143 if (kargs->stack_size == 0)
3146 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3149 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
3150 kargs->stack += kargs->stack_size;
3157 static bool clone3_args_valid(struct kernel_clone_args *kargs)
3159 /* Verify that no unknown flags are passed along. */
3161 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3165 * - make the CLONE_DETACHED bit reusable for clone3
3166 * - make the CSIGNAL bits reusable for clone3
3168 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3171 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3172 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3175 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3179 if (!clone3_stack_valid(kargs))
3186 * clone3 - create a new process with specific properties
3187 * @uargs: argument structure
3188 * @size: size of @uargs
3190 * clone3() is the extensible successor to clone()/clone2().
3191 * It takes a struct as argument that is versioned by its size.
3193 * Return: On success, a positive PID for the child process.
3194 * On error, a negative errno number.
3196 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3200 struct kernel_clone_args kargs;
3201 pid_t set_tid[MAX_PID_NS_LEVEL];
3203 kargs.set_tid = set_tid;
3205 err = copy_clone_args_from_user(&kargs, uargs, size);
3209 if (!clone3_args_valid(&kargs))
3212 return kernel_clone(&kargs);
3216 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3218 struct task_struct *leader, *parent, *child;
3221 read_lock(&tasklist_lock);
3222 leader = top = top->group_leader;
3224 for_each_thread(leader, parent) {
3225 list_for_each_entry(child, &parent->children, sibling) {
3226 res = visitor(child, data);
3238 if (leader != top) {
3240 parent = child->real_parent;
3241 leader = parent->group_leader;
3245 read_unlock(&tasklist_lock);
3248 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3249 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3252 static void sighand_ctor(void *data)
3254 struct sighand_struct *sighand = data;
3256 spin_lock_init(&sighand->siglock);
3257 init_waitqueue_head(&sighand->signalfd_wqh);
3260 void __init mm_cache_init(void)
3262 unsigned int mm_size;
3265 * The mm_cpumask is located at the end of mm_struct, and is
3266 * dynamically sized based on the maximum CPU number this system
3267 * can have, taking hotplug into account (nr_cpu_ids).
3269 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3271 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3272 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3273 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3274 offsetof(struct mm_struct, saved_auxv),
3275 sizeof_field(struct mm_struct, saved_auxv),
3279 void __init proc_caches_init(void)
3281 sighand_cachep = kmem_cache_create("sighand_cache",
3282 sizeof(struct sighand_struct), 0,
3283 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3284 SLAB_ACCOUNT, sighand_ctor);
3285 signal_cachep = kmem_cache_create("signal_cache",
3286 sizeof(struct signal_struct), 0,
3287 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3289 files_cachep = kmem_cache_create("files_cache",
3290 sizeof(struct files_struct), 0,
3291 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3293 fs_cachep = kmem_cache_create("fs_cache",
3294 sizeof(struct fs_struct), 0,
3295 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3298 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3299 #ifdef CONFIG_PER_VMA_LOCK
3300 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3303 nsproxy_cache_init();
3307 * Check constraints on flags passed to the unshare system call.
3309 static int check_unshare_flags(unsigned long unshare_flags)
3311 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3312 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3313 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3314 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3318 * Not implemented, but pretend it works if there is nothing
3319 * to unshare. Note that unsharing the address space or the
3320 * signal handlers also need to unshare the signal queues (aka
3323 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3324 if (!thread_group_empty(current))
3327 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3328 if (refcount_read(¤t->sighand->count) > 1)
3331 if (unshare_flags & CLONE_VM) {
3332 if (!current_is_single_threaded())
3340 * Unshare the filesystem structure if it is being shared
3342 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3344 struct fs_struct *fs = current->fs;
3346 if (!(unshare_flags & CLONE_FS) || !fs)
3349 /* don't need lock here; in the worst case we'll do useless copy */
3353 *new_fsp = copy_fs_struct(fs);
3361 * Unshare file descriptor table if it is being shared
3363 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3364 struct files_struct **new_fdp)
3366 struct files_struct *fd = current->files;
3369 if ((unshare_flags & CLONE_FILES) &&
3370 (fd && atomic_read(&fd->count) > 1)) {
3371 *new_fdp = dup_fd(fd, max_fds, &error);
3380 * unshare allows a process to 'unshare' part of the process
3381 * context which was originally shared using clone. copy_*
3382 * functions used by kernel_clone() cannot be used here directly
3383 * because they modify an inactive task_struct that is being
3384 * constructed. Here we are modifying the current, active,
3387 int ksys_unshare(unsigned long unshare_flags)
3389 struct fs_struct *fs, *new_fs = NULL;
3390 struct files_struct *new_fd = NULL;
3391 struct cred *new_cred = NULL;
3392 struct nsproxy *new_nsproxy = NULL;
3397 * If unsharing a user namespace must also unshare the thread group
3398 * and unshare the filesystem root and working directories.
3400 if (unshare_flags & CLONE_NEWUSER)
3401 unshare_flags |= CLONE_THREAD | CLONE_FS;
3403 * If unsharing vm, must also unshare signal handlers.
3405 if (unshare_flags & CLONE_VM)
3406 unshare_flags |= CLONE_SIGHAND;
3408 * If unsharing a signal handlers, must also unshare the signal queues.
3410 if (unshare_flags & CLONE_SIGHAND)
3411 unshare_flags |= CLONE_THREAD;
3413 * If unsharing namespace, must also unshare filesystem information.
3415 if (unshare_flags & CLONE_NEWNS)
3416 unshare_flags |= CLONE_FS;
3418 err = check_unshare_flags(unshare_flags);
3420 goto bad_unshare_out;
3422 * CLONE_NEWIPC must also detach from the undolist: after switching
3423 * to a new ipc namespace, the semaphore arrays from the old
3424 * namespace are unreachable.
3426 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3428 err = unshare_fs(unshare_flags, &new_fs);
3430 goto bad_unshare_out;
3431 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3433 goto bad_unshare_cleanup_fs;
3434 err = unshare_userns(unshare_flags, &new_cred);
3436 goto bad_unshare_cleanup_fd;
3437 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3440 goto bad_unshare_cleanup_cred;
3443 err = set_cred_ucounts(new_cred);
3445 goto bad_unshare_cleanup_cred;
3448 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3451 * CLONE_SYSVSEM is equivalent to sys_exit().
3455 if (unshare_flags & CLONE_NEWIPC) {
3456 /* Orphan segments in old ns (see sem above). */
3458 shm_init_task(current);
3462 switch_task_namespaces(current, new_nsproxy);
3468 spin_lock(&fs->lock);
3469 current->fs = new_fs;
3474 spin_unlock(&fs->lock);
3478 swap(current->files, new_fd);
3480 task_unlock(current);
3483 /* Install the new user namespace */
3484 commit_creds(new_cred);
3489 perf_event_namespaces(current);
3491 bad_unshare_cleanup_cred:
3494 bad_unshare_cleanup_fd:
3496 put_files_struct(new_fd);
3498 bad_unshare_cleanup_fs:
3500 free_fs_struct(new_fs);
3506 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3508 return ksys_unshare(unshare_flags);
3512 * Helper to unshare the files of the current task.
3513 * We don't want to expose copy_files internals to
3514 * the exec layer of the kernel.
3517 int unshare_files(void)
3519 struct task_struct *task = current;
3520 struct files_struct *old, *copy = NULL;
3523 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3531 put_files_struct(old);
3535 int sysctl_max_threads(struct ctl_table *table, int write,
3536 void *buffer, size_t *lenp, loff_t *ppos)
3540 int threads = max_threads;
3542 int max = MAX_THREADS;
3549 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3553 max_threads = threads;