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)
256 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
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 * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
268 * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
271 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
272 memcg_kmem_uncharge_page(vm->pages[i], 0);
276 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
278 struct vm_struct *vm;
282 for (i = 0; i < NR_CACHED_STACKS; i++) {
285 s = this_cpu_xchg(cached_stacks[i], NULL);
290 /* Reset stack metadata. */
291 kasan_unpoison_range(s->addr, THREAD_SIZE);
293 stack = kasan_reset_tag(s->addr);
295 /* Clear stale pointers from reused stack. */
296 memset(stack, 0, THREAD_SIZE);
298 if (memcg_charge_kernel_stack(s)) {
303 tsk->stack_vm_area = s;
309 * Allocated stacks are cached and later reused by new threads,
310 * so memcg accounting is performed manually on assigning/releasing
311 * stacks to tasks. Drop __GFP_ACCOUNT.
313 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
314 VMALLOC_START, VMALLOC_END,
315 THREADINFO_GFP & ~__GFP_ACCOUNT,
317 0, node, __builtin_return_address(0));
321 vm = find_vm_area(stack);
322 if (memcg_charge_kernel_stack(vm)) {
327 * We can't call find_vm_area() in interrupt context, and
328 * free_thread_stack() can be called in interrupt context,
329 * so cache the vm_struct.
331 tsk->stack_vm_area = vm;
332 stack = kasan_reset_tag(stack);
337 static void free_thread_stack(struct task_struct *tsk)
339 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
340 thread_stack_delayed_free(tsk);
343 tsk->stack_vm_area = NULL;
346 # else /* !CONFIG_VMAP_STACK */
348 static void thread_stack_free_rcu(struct rcu_head *rh)
350 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
353 static void thread_stack_delayed_free(struct task_struct *tsk)
355 struct rcu_head *rh = tsk->stack;
357 call_rcu(rh, thread_stack_free_rcu);
360 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
362 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
366 tsk->stack = kasan_reset_tag(page_address(page));
372 static void free_thread_stack(struct task_struct *tsk)
374 thread_stack_delayed_free(tsk);
378 # endif /* CONFIG_VMAP_STACK */
379 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
381 static struct kmem_cache *thread_stack_cache;
383 static void thread_stack_free_rcu(struct rcu_head *rh)
385 kmem_cache_free(thread_stack_cache, rh);
388 static void thread_stack_delayed_free(struct task_struct *tsk)
390 struct rcu_head *rh = tsk->stack;
392 call_rcu(rh, thread_stack_free_rcu);
395 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
397 unsigned long *stack;
398 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
399 stack = kasan_reset_tag(stack);
401 return stack ? 0 : -ENOMEM;
404 static void free_thread_stack(struct task_struct *tsk)
406 thread_stack_delayed_free(tsk);
410 void thread_stack_cache_init(void)
412 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
413 THREAD_SIZE, THREAD_SIZE, 0, 0,
415 BUG_ON(thread_stack_cache == NULL);
418 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
419 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
421 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
423 unsigned long *stack;
425 stack = arch_alloc_thread_stack_node(tsk, node);
427 return stack ? 0 : -ENOMEM;
430 static void free_thread_stack(struct task_struct *tsk)
432 arch_free_thread_stack(tsk);
436 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
438 /* SLAB cache for signal_struct structures (tsk->signal) */
439 static struct kmem_cache *signal_cachep;
441 /* SLAB cache for sighand_struct structures (tsk->sighand) */
442 struct kmem_cache *sighand_cachep;
444 /* SLAB cache for files_struct structures (tsk->files) */
445 struct kmem_cache *files_cachep;
447 /* SLAB cache for fs_struct structures (tsk->fs) */
448 struct kmem_cache *fs_cachep;
450 /* SLAB cache for vm_area_struct structures */
451 static struct kmem_cache *vm_area_cachep;
453 /* SLAB cache for mm_struct structures (tsk->mm) */
454 static struct kmem_cache *mm_cachep;
456 #ifdef CONFIG_PER_VMA_LOCK
458 /* SLAB cache for vm_area_struct.lock */
459 static struct kmem_cache *vma_lock_cachep;
461 static bool vma_lock_alloc(struct vm_area_struct *vma)
463 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
467 init_rwsem(&vma->vm_lock->lock);
468 vma->vm_lock_seq = -1;
473 static inline void vma_lock_free(struct vm_area_struct *vma)
475 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
478 #else /* CONFIG_PER_VMA_LOCK */
480 static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
481 static inline void vma_lock_free(struct vm_area_struct *vma) {}
483 #endif /* CONFIG_PER_VMA_LOCK */
485 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
487 struct vm_area_struct *vma;
489 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
494 if (!vma_lock_alloc(vma)) {
495 kmem_cache_free(vm_area_cachep, vma);
502 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
504 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
509 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
510 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
512 * orig->shared.rb may be modified concurrently, but the clone
513 * will be reinitialized.
515 data_race(memcpy(new, orig, sizeof(*new)));
516 if (!vma_lock_alloc(new)) {
517 kmem_cache_free(vm_area_cachep, new);
520 INIT_LIST_HEAD(&new->anon_vma_chain);
521 vma_numab_state_init(new);
522 dup_anon_vma_name(orig, new);
527 void __vm_area_free(struct vm_area_struct *vma)
529 vma_numab_state_free(vma);
530 free_anon_vma_name(vma);
532 kmem_cache_free(vm_area_cachep, vma);
535 #ifdef CONFIG_PER_VMA_LOCK
536 static void vm_area_free_rcu_cb(struct rcu_head *head)
538 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
541 /* The vma should not be locked while being destroyed. */
542 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
547 void vm_area_free(struct vm_area_struct *vma)
549 #ifdef CONFIG_PER_VMA_LOCK
550 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
556 static void account_kernel_stack(struct task_struct *tsk, int account)
558 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
559 struct vm_struct *vm = task_stack_vm_area(tsk);
562 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
563 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
564 account * (PAGE_SIZE / 1024));
566 void *stack = task_stack_page(tsk);
568 /* All stack pages are in the same node. */
569 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
570 account * (THREAD_SIZE / 1024));
574 void exit_task_stack_account(struct task_struct *tsk)
576 account_kernel_stack(tsk, -1);
578 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
579 struct vm_struct *vm;
582 vm = task_stack_vm_area(tsk);
583 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
584 memcg_kmem_uncharge_page(vm->pages[i], 0);
588 static void release_task_stack(struct task_struct *tsk)
590 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
591 return; /* Better to leak the stack than to free prematurely */
593 free_thread_stack(tsk);
596 #ifdef CONFIG_THREAD_INFO_IN_TASK
597 void put_task_stack(struct task_struct *tsk)
599 if (refcount_dec_and_test(&tsk->stack_refcount))
600 release_task_stack(tsk);
604 void free_task(struct task_struct *tsk)
606 #ifdef CONFIG_SECCOMP
607 WARN_ON_ONCE(tsk->seccomp.filter);
609 release_user_cpus_ptr(tsk);
612 #ifndef CONFIG_THREAD_INFO_IN_TASK
614 * The task is finally done with both the stack and thread_info,
617 release_task_stack(tsk);
620 * If the task had a separate stack allocation, it should be gone
623 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
625 rt_mutex_debug_task_free(tsk);
626 ftrace_graph_exit_task(tsk);
627 arch_release_task_struct(tsk);
628 if (tsk->flags & PF_KTHREAD)
629 free_kthread_struct(tsk);
630 bpf_task_storage_free(tsk);
631 free_task_struct(tsk);
633 EXPORT_SYMBOL(free_task);
635 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
637 struct file *exe_file;
639 exe_file = get_mm_exe_file(oldmm);
640 RCU_INIT_POINTER(mm->exe_file, exe_file);
642 * We depend on the oldmm having properly denied write access to the
645 if (exe_file && deny_write_access(exe_file))
646 pr_warn_once("deny_write_access() failed in %s\n", __func__);
650 static __latent_entropy int dup_mmap(struct mm_struct *mm,
651 struct mm_struct *oldmm)
653 struct vm_area_struct *mpnt, *tmp;
655 unsigned long charge = 0;
657 VMA_ITERATOR(old_vmi, oldmm, 0);
658 VMA_ITERATOR(vmi, mm, 0);
660 uprobe_start_dup_mmap();
661 if (mmap_write_lock_killable(oldmm)) {
663 goto fail_uprobe_end;
665 flush_cache_dup_mm(oldmm);
666 uprobe_dup_mmap(oldmm, mm);
668 * Not linked in yet - no deadlock potential:
670 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
672 /* No ordering required: file already has been exposed. */
673 dup_mm_exe_file(mm, oldmm);
675 mm->total_vm = oldmm->total_vm;
676 mm->data_vm = oldmm->data_vm;
677 mm->exec_vm = oldmm->exec_vm;
678 mm->stack_vm = oldmm->stack_vm;
680 retval = ksm_fork(mm, oldmm);
683 khugepaged_fork(mm, oldmm);
685 retval = vma_iter_bulk_alloc(&vmi, oldmm->map_count);
689 mt_clear_in_rcu(vmi.mas.tree);
690 for_each_vma(old_vmi, mpnt) {
693 if (mpnt->vm_flags & VM_DONTCOPY) {
694 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
699 * Don't duplicate many vmas if we've been oom-killed (for
702 if (fatal_signal_pending(current)) {
706 if (mpnt->vm_flags & VM_ACCOUNT) {
707 unsigned long len = vma_pages(mpnt);
709 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
713 tmp = vm_area_dup(mpnt);
716 retval = vma_dup_policy(mpnt, tmp);
718 goto fail_nomem_policy;
720 retval = dup_userfaultfd(tmp, &uf);
722 goto fail_nomem_anon_vma_fork;
723 if (tmp->vm_flags & VM_WIPEONFORK) {
725 * VM_WIPEONFORK gets a clean slate in the child.
726 * Don't prepare anon_vma until fault since we don't
727 * copy page for current vma.
729 tmp->anon_vma = NULL;
730 } else if (anon_vma_fork(tmp, mpnt))
731 goto fail_nomem_anon_vma_fork;
732 vm_flags_clear(tmp, VM_LOCKED_MASK);
735 struct address_space *mapping = file->f_mapping;
738 i_mmap_lock_write(mapping);
739 if (tmp->vm_flags & VM_SHARED)
740 mapping_allow_writable(mapping);
741 flush_dcache_mmap_lock(mapping);
742 /* insert tmp into the share list, just after mpnt */
743 vma_interval_tree_insert_after(tmp, mpnt,
745 flush_dcache_mmap_unlock(mapping);
746 i_mmap_unlock_write(mapping);
750 * Copy/update hugetlb private vma information.
752 if (is_vm_hugetlb_page(tmp))
753 hugetlb_dup_vma_private(tmp);
755 /* Link the vma into the MT */
756 if (vma_iter_bulk_store(&vmi, tmp))
757 goto fail_nomem_vmi_store;
760 if (!(tmp->vm_flags & VM_WIPEONFORK))
761 retval = copy_page_range(tmp, mpnt);
763 if (tmp->vm_ops && tmp->vm_ops->open)
764 tmp->vm_ops->open(tmp);
769 /* a new mm has just been created */
770 retval = arch_dup_mmap(oldmm, mm);
774 mt_set_in_rcu(vmi.mas.tree);
776 mmap_write_unlock(mm);
778 mmap_write_unlock(oldmm);
779 dup_userfaultfd_complete(&uf);
781 uprobe_end_dup_mmap();
784 fail_nomem_vmi_store:
785 unlink_anon_vmas(tmp);
786 fail_nomem_anon_vma_fork:
787 mpol_put(vma_policy(tmp));
792 vm_unacct_memory(charge);
796 static inline int mm_alloc_pgd(struct mm_struct *mm)
798 mm->pgd = pgd_alloc(mm);
799 if (unlikely(!mm->pgd))
804 static inline void mm_free_pgd(struct mm_struct *mm)
806 pgd_free(mm, mm->pgd);
809 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
811 mmap_write_lock(oldmm);
812 dup_mm_exe_file(mm, oldmm);
813 mmap_write_unlock(oldmm);
816 #define mm_alloc_pgd(mm) (0)
817 #define mm_free_pgd(mm)
818 #endif /* CONFIG_MMU */
820 static void check_mm(struct mm_struct *mm)
824 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
825 "Please make sure 'struct resident_page_types[]' is updated as well");
827 for (i = 0; i < NR_MM_COUNTERS; i++) {
828 long x = percpu_counter_sum(&mm->rss_stat[i]);
831 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
832 mm, resident_page_types[i], x);
835 if (mm_pgtables_bytes(mm))
836 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
837 mm_pgtables_bytes(mm));
839 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
840 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
844 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
845 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
847 static void do_check_lazy_tlb(void *arg)
849 struct mm_struct *mm = arg;
851 WARN_ON_ONCE(current->active_mm == mm);
854 static void do_shoot_lazy_tlb(void *arg)
856 struct mm_struct *mm = arg;
858 if (current->active_mm == mm) {
859 WARN_ON_ONCE(current->mm);
860 current->active_mm = &init_mm;
861 switch_mm(mm, &init_mm, current);
865 static void cleanup_lazy_tlbs(struct mm_struct *mm)
867 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
869 * In this case, lazy tlb mms are refounted and would not reach
870 * __mmdrop until all CPUs have switched away and mmdrop()ed.
876 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
877 * requires lazy mm users to switch to another mm when the refcount
878 * drops to zero, before the mm is freed. This requires IPIs here to
879 * switch kernel threads to init_mm.
881 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
882 * switch with the final userspace teardown TLB flush which leaves the
883 * mm lazy on this CPU but no others, reducing the need for additional
884 * IPIs here. There are cases where a final IPI is still required here,
885 * such as the final mmdrop being performed on a different CPU than the
886 * one exiting, or kernel threads using the mm when userspace exits.
888 * IPI overheads have not found to be expensive, but they could be
889 * reduced in a number of possible ways, for example (roughly
890 * increasing order of complexity):
891 * - The last lazy reference created by exit_mm() could instead switch
892 * to init_mm, however it's probable this will run on the same CPU
893 * immediately afterwards, so this may not reduce IPIs much.
894 * - A batch of mms requiring IPIs could be gathered and freed at once.
895 * - CPUs store active_mm where it can be remotely checked without a
896 * lock, to filter out false-positives in the cpumask.
897 * - After mm_users or mm_count reaches zero, switching away from the
898 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
899 * with some batching or delaying of the final IPIs.
900 * - A delayed freeing and RCU-like quiescing sequence based on mm
901 * switching to avoid IPIs completely.
903 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
904 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
905 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
909 * Called when the last reference to the mm
910 * is dropped: either by a lazy thread or by
911 * mmput. Free the page directory and the mm.
913 void __mmdrop(struct mm_struct *mm)
917 BUG_ON(mm == &init_mm);
918 WARN_ON_ONCE(mm == current->mm);
920 /* Ensure no CPUs are using this as their lazy tlb mm */
921 cleanup_lazy_tlbs(mm);
923 WARN_ON_ONCE(mm == current->active_mm);
926 mmu_notifier_subscriptions_destroy(mm);
928 put_user_ns(mm->user_ns);
932 for (i = 0; i < NR_MM_COUNTERS; i++)
933 percpu_counter_destroy(&mm->rss_stat[i]);
936 EXPORT_SYMBOL_GPL(__mmdrop);
938 static void mmdrop_async_fn(struct work_struct *work)
940 struct mm_struct *mm;
942 mm = container_of(work, struct mm_struct, async_put_work);
946 static void mmdrop_async(struct mm_struct *mm)
948 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
949 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
950 schedule_work(&mm->async_put_work);
954 static inline void free_signal_struct(struct signal_struct *sig)
956 taskstats_tgid_free(sig);
957 sched_autogroup_exit(sig);
959 * __mmdrop is not safe to call from softirq context on x86 due to
960 * pgd_dtor so postpone it to the async context
963 mmdrop_async(sig->oom_mm);
964 kmem_cache_free(signal_cachep, sig);
967 static inline void put_signal_struct(struct signal_struct *sig)
969 if (refcount_dec_and_test(&sig->sigcnt))
970 free_signal_struct(sig);
973 void __put_task_struct(struct task_struct *tsk)
975 WARN_ON(!tsk->exit_state);
976 WARN_ON(refcount_read(&tsk->usage));
977 WARN_ON(tsk == current);
981 task_numa_free(tsk, true);
982 security_task_free(tsk);
984 delayacct_tsk_free(tsk);
985 put_signal_struct(tsk->signal);
986 sched_core_free(tsk);
989 EXPORT_SYMBOL_GPL(__put_task_struct);
991 void __init __weak arch_task_cache_init(void) { }
996 static void set_max_threads(unsigned int max_threads_suggested)
999 unsigned long nr_pages = totalram_pages();
1002 * The number of threads shall be limited such that the thread
1003 * structures may only consume a small part of the available memory.
1005 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1006 threads = MAX_THREADS;
1008 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1009 (u64) THREAD_SIZE * 8UL);
1011 if (threads > max_threads_suggested)
1012 threads = max_threads_suggested;
1014 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1017 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1018 /* Initialized by the architecture: */
1019 int arch_task_struct_size __read_mostly;
1022 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1023 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1025 /* Fetch thread_struct whitelist for the architecture. */
1026 arch_thread_struct_whitelist(offset, size);
1029 * Handle zero-sized whitelist or empty thread_struct, otherwise
1030 * adjust offset to position of thread_struct in task_struct.
1032 if (unlikely(*size == 0))
1035 *offset += offsetof(struct task_struct, thread);
1037 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
1039 void __init fork_init(void)
1042 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1043 #ifndef ARCH_MIN_TASKALIGN
1044 #define ARCH_MIN_TASKALIGN 0
1046 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1047 unsigned long useroffset, usersize;
1049 /* create a slab on which task_structs can be allocated */
1050 task_struct_whitelist(&useroffset, &usersize);
1051 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1052 arch_task_struct_size, align,
1053 SLAB_PANIC|SLAB_ACCOUNT,
1054 useroffset, usersize, NULL);
1057 /* do the arch specific task caches init */
1058 arch_task_cache_init();
1060 set_max_threads(MAX_THREADS);
1062 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1063 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1064 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1065 init_task.signal->rlim[RLIMIT_NPROC];
1067 for (i = 0; i < UCOUNT_COUNTS; i++)
1068 init_user_ns.ucount_max[i] = max_threads/2;
1070 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1071 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1072 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1073 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1075 #ifdef CONFIG_VMAP_STACK
1076 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1077 NULL, free_vm_stack_cache);
1082 lockdep_init_task(&init_task);
1086 int __weak arch_dup_task_struct(struct task_struct *dst,
1087 struct task_struct *src)
1093 void set_task_stack_end_magic(struct task_struct *tsk)
1095 unsigned long *stackend;
1097 stackend = end_of_stack(tsk);
1098 *stackend = STACK_END_MAGIC; /* for overflow detection */
1101 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1103 struct task_struct *tsk;
1106 if (node == NUMA_NO_NODE)
1107 node = tsk_fork_get_node(orig);
1108 tsk = alloc_task_struct_node(node);
1112 err = arch_dup_task_struct(tsk, orig);
1116 err = alloc_thread_stack_node(tsk, node);
1120 #ifdef CONFIG_THREAD_INFO_IN_TASK
1121 refcount_set(&tsk->stack_refcount, 1);
1123 account_kernel_stack(tsk, 1);
1125 err = scs_prepare(tsk, node);
1129 #ifdef CONFIG_SECCOMP
1131 * We must handle setting up seccomp filters once we're under
1132 * the sighand lock in case orig has changed between now and
1133 * then. Until then, filter must be NULL to avoid messing up
1134 * the usage counts on the error path calling free_task.
1136 tsk->seccomp.filter = NULL;
1139 setup_thread_stack(tsk, orig);
1140 clear_user_return_notifier(tsk);
1141 clear_tsk_need_resched(tsk);
1142 set_task_stack_end_magic(tsk);
1143 clear_syscall_work_syscall_user_dispatch(tsk);
1145 #ifdef CONFIG_STACKPROTECTOR
1146 tsk->stack_canary = get_random_canary();
1148 if (orig->cpus_ptr == &orig->cpus_mask)
1149 tsk->cpus_ptr = &tsk->cpus_mask;
1150 dup_user_cpus_ptr(tsk, orig, node);
1153 * One for the user space visible state that goes away when reaped.
1154 * One for the scheduler.
1156 refcount_set(&tsk->rcu_users, 2);
1157 /* One for the rcu users */
1158 refcount_set(&tsk->usage, 1);
1159 #ifdef CONFIG_BLK_DEV_IO_TRACE
1160 tsk->btrace_seq = 0;
1162 tsk->splice_pipe = NULL;
1163 tsk->task_frag.page = NULL;
1164 tsk->wake_q.next = NULL;
1165 tsk->worker_private = NULL;
1167 kcov_task_init(tsk);
1168 kmsan_task_create(tsk);
1169 kmap_local_fork(tsk);
1171 #ifdef CONFIG_FAULT_INJECTION
1175 #ifdef CONFIG_BLK_CGROUP
1176 tsk->throttle_disk = NULL;
1177 tsk->use_memdelay = 0;
1180 #ifdef CONFIG_IOMMU_SVA
1181 tsk->pasid_activated = 0;
1185 tsk->active_memcg = NULL;
1188 #ifdef CONFIG_CPU_SUP_INTEL
1189 tsk->reported_split_lock = 0;
1192 #ifdef CONFIG_SCHED_MM_CID
1194 tsk->last_mm_cid = -1;
1195 tsk->mm_cid_active = 0;
1196 tsk->migrate_from_cpu = -1;
1201 exit_task_stack_account(tsk);
1202 free_thread_stack(tsk);
1204 free_task_struct(tsk);
1208 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1210 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1212 static int __init coredump_filter_setup(char *s)
1214 default_dump_filter =
1215 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1216 MMF_DUMP_FILTER_MASK;
1220 __setup("coredump_filter=", coredump_filter_setup);
1222 #include <linux/init_task.h>
1224 static void mm_init_aio(struct mm_struct *mm)
1227 spin_lock_init(&mm->ioctx_lock);
1228 mm->ioctx_table = NULL;
1232 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1233 struct task_struct *p)
1237 WRITE_ONCE(mm->owner, NULL);
1241 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1248 static void mm_init_uprobes_state(struct mm_struct *mm)
1250 #ifdef CONFIG_UPROBES
1251 mm->uprobes_state.xol_area = NULL;
1255 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1256 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 = current->mm->flags & MMF_INIT_MASK;
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 for (i = 0; i < NR_MM_COUNTERS; i++)
1308 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1311 mm->user_ns = get_user_ns(user_ns);
1312 lru_gen_init_mm(mm);
1317 percpu_counter_destroy(&mm->rss_stat[--i]);
1320 destroy_context(mm);
1329 * Allocate and initialize an mm_struct.
1331 struct mm_struct *mm_alloc(void)
1333 struct mm_struct *mm;
1339 memset(mm, 0, sizeof(*mm));
1340 return mm_init(mm, current, current_user_ns());
1343 static inline void __mmput(struct mm_struct *mm)
1345 VM_BUG_ON(atomic_read(&mm->mm_users));
1347 uprobe_clear_state(mm);
1350 khugepaged_exit(mm); /* must run before exit_mmap */
1352 mm_put_huge_zero_page(mm);
1353 set_mm_exe_file(mm, NULL);
1354 if (!list_empty(&mm->mmlist)) {
1355 spin_lock(&mmlist_lock);
1356 list_del(&mm->mmlist);
1357 spin_unlock(&mmlist_lock);
1360 module_put(mm->binfmt->module);
1366 * Decrement the use count and release all resources for an mm.
1368 void mmput(struct mm_struct *mm)
1372 if (atomic_dec_and_test(&mm->mm_users))
1375 EXPORT_SYMBOL_GPL(mmput);
1378 static void mmput_async_fn(struct work_struct *work)
1380 struct mm_struct *mm = container_of(work, struct mm_struct,
1386 void mmput_async(struct mm_struct *mm)
1388 if (atomic_dec_and_test(&mm->mm_users)) {
1389 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1390 schedule_work(&mm->async_put_work);
1393 EXPORT_SYMBOL_GPL(mmput_async);
1397 * set_mm_exe_file - change a reference to the mm's executable file
1399 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1401 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1402 * invocations: in mmput() nobody alive left, in execve task is single
1405 * Can only fail if new_exe_file != NULL.
1407 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1409 struct file *old_exe_file;
1412 * It is safe to dereference the exe_file without RCU as
1413 * this function is only called if nobody else can access
1414 * this mm -- see comment above for justification.
1416 old_exe_file = rcu_dereference_raw(mm->exe_file);
1420 * We expect the caller (i.e., sys_execve) to already denied
1421 * write access, so this is unlikely to fail.
1423 if (unlikely(deny_write_access(new_exe_file)))
1425 get_file(new_exe_file);
1427 rcu_assign_pointer(mm->exe_file, new_exe_file);
1429 allow_write_access(old_exe_file);
1436 * replace_mm_exe_file - replace a reference to the mm's executable file
1438 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1439 * dealing with concurrent invocation and without grabbing the mmap lock in
1442 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1444 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1446 struct vm_area_struct *vma;
1447 struct file *old_exe_file;
1450 /* Forbid mm->exe_file change if old file still mapped. */
1451 old_exe_file = get_mm_exe_file(mm);
1453 VMA_ITERATOR(vmi, mm, 0);
1455 for_each_vma(vmi, vma) {
1458 if (path_equal(&vma->vm_file->f_path,
1459 &old_exe_file->f_path)) {
1464 mmap_read_unlock(mm);
1470 /* set the new file, lockless */
1471 ret = deny_write_access(new_exe_file);
1474 get_file(new_exe_file);
1476 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1479 * Don't race with dup_mmap() getting the file and disallowing
1480 * write access while someone might open the file writable.
1483 allow_write_access(old_exe_file);
1485 mmap_read_unlock(mm);
1491 * get_mm_exe_file - acquire a reference to the mm's executable file
1493 * Returns %NULL if mm has no associated executable file.
1494 * User must release file via fput().
1496 struct file *get_mm_exe_file(struct mm_struct *mm)
1498 struct file *exe_file;
1501 exe_file = rcu_dereference(mm->exe_file);
1502 if (exe_file && !get_file_rcu(exe_file))
1509 * get_task_exe_file - acquire a reference to the task's executable file
1511 * Returns %NULL if task's mm (if any) has no associated executable file or
1512 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1513 * User must release file via fput().
1515 struct file *get_task_exe_file(struct task_struct *task)
1517 struct file *exe_file = NULL;
1518 struct mm_struct *mm;
1523 if (!(task->flags & PF_KTHREAD))
1524 exe_file = get_mm_exe_file(mm);
1531 * get_task_mm - acquire a reference to the task's mm
1533 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1534 * this kernel workthread has transiently adopted a user mm with use_mm,
1535 * to do its AIO) is not set and if so returns a reference to it, after
1536 * bumping up the use count. User must release the mm via mmput()
1537 * after use. Typically used by /proc and ptrace.
1539 struct mm_struct *get_task_mm(struct task_struct *task)
1541 struct mm_struct *mm;
1546 if (task->flags & PF_KTHREAD)
1554 EXPORT_SYMBOL_GPL(get_task_mm);
1556 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1558 struct mm_struct *mm;
1561 err = down_read_killable(&task->signal->exec_update_lock);
1563 return ERR_PTR(err);
1565 mm = get_task_mm(task);
1566 if (mm && mm != current->mm &&
1567 !ptrace_may_access(task, mode)) {
1569 mm = ERR_PTR(-EACCES);
1571 up_read(&task->signal->exec_update_lock);
1576 static void complete_vfork_done(struct task_struct *tsk)
1578 struct completion *vfork;
1581 vfork = tsk->vfork_done;
1582 if (likely(vfork)) {
1583 tsk->vfork_done = NULL;
1589 static int wait_for_vfork_done(struct task_struct *child,
1590 struct completion *vfork)
1592 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1595 cgroup_enter_frozen();
1596 killed = wait_for_completion_state(vfork, state);
1597 cgroup_leave_frozen(false);
1601 child->vfork_done = NULL;
1605 put_task_struct(child);
1609 /* Please note the differences between mmput and mm_release.
1610 * mmput is called whenever we stop holding onto a mm_struct,
1611 * error success whatever.
1613 * mm_release is called after a mm_struct has been removed
1614 * from the current process.
1616 * This difference is important for error handling, when we
1617 * only half set up a mm_struct for a new process and need to restore
1618 * the old one. Because we mmput the new mm_struct before
1619 * restoring the old one. . .
1620 * Eric Biederman 10 January 1998
1622 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1624 uprobe_free_utask(tsk);
1626 /* Get rid of any cached register state */
1627 deactivate_mm(tsk, mm);
1630 * Signal userspace if we're not exiting with a core dump
1631 * because we want to leave the value intact for debugging
1634 if (tsk->clear_child_tid) {
1635 if (atomic_read(&mm->mm_users) > 1) {
1637 * We don't check the error code - if userspace has
1638 * not set up a proper pointer then tough luck.
1640 put_user(0, tsk->clear_child_tid);
1641 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1642 1, NULL, NULL, 0, 0);
1644 tsk->clear_child_tid = NULL;
1648 * All done, finally we can wake up parent and return this mm to him.
1649 * Also kthread_stop() uses this completion for synchronization.
1651 if (tsk->vfork_done)
1652 complete_vfork_done(tsk);
1655 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1657 futex_exit_release(tsk);
1658 mm_release(tsk, mm);
1661 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1663 futex_exec_release(tsk);
1664 mm_release(tsk, mm);
1668 * dup_mm() - duplicates an existing mm structure
1669 * @tsk: the task_struct with which the new mm will be associated.
1670 * @oldmm: the mm to duplicate.
1672 * Allocates a new mm structure and duplicates the provided @oldmm structure
1675 * Return: the duplicated mm or NULL on failure.
1677 static struct mm_struct *dup_mm(struct task_struct *tsk,
1678 struct mm_struct *oldmm)
1680 struct mm_struct *mm;
1687 memcpy(mm, oldmm, sizeof(*mm));
1689 if (!mm_init(mm, tsk, mm->user_ns))
1692 err = dup_mmap(mm, oldmm);
1696 mm->hiwater_rss = get_mm_rss(mm);
1697 mm->hiwater_vm = mm->total_vm;
1699 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1705 /* don't put binfmt in mmput, we haven't got module yet */
1707 mm_init_owner(mm, NULL);
1714 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1716 struct mm_struct *mm, *oldmm;
1718 tsk->min_flt = tsk->maj_flt = 0;
1719 tsk->nvcsw = tsk->nivcsw = 0;
1720 #ifdef CONFIG_DETECT_HUNG_TASK
1721 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1722 tsk->last_switch_time = 0;
1726 tsk->active_mm = NULL;
1729 * Are we cloning a kernel thread?
1731 * We need to steal a active VM for that..
1733 oldmm = current->mm;
1737 if (clone_flags & CLONE_VM) {
1741 mm = dup_mm(tsk, current->mm);
1747 tsk->active_mm = mm;
1748 sched_mm_cid_fork(tsk);
1752 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1754 struct fs_struct *fs = current->fs;
1755 if (clone_flags & CLONE_FS) {
1756 /* tsk->fs is already what we want */
1757 spin_lock(&fs->lock);
1759 spin_unlock(&fs->lock);
1763 spin_unlock(&fs->lock);
1766 tsk->fs = copy_fs_struct(fs);
1772 static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1775 struct files_struct *oldf, *newf;
1779 * A background process may not have any files ...
1781 oldf = current->files;
1790 if (clone_flags & CLONE_FILES) {
1791 atomic_inc(&oldf->count);
1795 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1805 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1807 struct sighand_struct *sig;
1809 if (clone_flags & CLONE_SIGHAND) {
1810 refcount_inc(¤t->sighand->count);
1813 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1814 RCU_INIT_POINTER(tsk->sighand, sig);
1818 refcount_set(&sig->count, 1);
1819 spin_lock_irq(¤t->sighand->siglock);
1820 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1821 spin_unlock_irq(¤t->sighand->siglock);
1823 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1824 if (clone_flags & CLONE_CLEAR_SIGHAND)
1825 flush_signal_handlers(tsk, 0);
1830 void __cleanup_sighand(struct sighand_struct *sighand)
1832 if (refcount_dec_and_test(&sighand->count)) {
1833 signalfd_cleanup(sighand);
1835 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1836 * without an RCU grace period, see __lock_task_sighand().
1838 kmem_cache_free(sighand_cachep, sighand);
1843 * Initialize POSIX timer handling for a thread group.
1845 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1847 struct posix_cputimers *pct = &sig->posix_cputimers;
1848 unsigned long cpu_limit;
1850 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1851 posix_cputimers_group_init(pct, cpu_limit);
1854 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1856 struct signal_struct *sig;
1858 if (clone_flags & CLONE_THREAD)
1861 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1866 sig->nr_threads = 1;
1867 sig->quick_threads = 1;
1868 atomic_set(&sig->live, 1);
1869 refcount_set(&sig->sigcnt, 1);
1871 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1872 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1873 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1875 init_waitqueue_head(&sig->wait_chldexit);
1876 sig->curr_target = tsk;
1877 init_sigpending(&sig->shared_pending);
1878 INIT_HLIST_HEAD(&sig->multiprocess);
1879 seqlock_init(&sig->stats_lock);
1880 prev_cputime_init(&sig->prev_cputime);
1882 #ifdef CONFIG_POSIX_TIMERS
1883 INIT_LIST_HEAD(&sig->posix_timers);
1884 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1885 sig->real_timer.function = it_real_fn;
1888 task_lock(current->group_leader);
1889 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1890 task_unlock(current->group_leader);
1892 posix_cpu_timers_init_group(sig);
1894 tty_audit_fork(sig);
1895 sched_autogroup_fork(sig);
1897 sig->oom_score_adj = current->signal->oom_score_adj;
1898 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1900 mutex_init(&sig->cred_guard_mutex);
1901 init_rwsem(&sig->exec_update_lock);
1906 static void copy_seccomp(struct task_struct *p)
1908 #ifdef CONFIG_SECCOMP
1910 * Must be called with sighand->lock held, which is common to
1911 * all threads in the group. Holding cred_guard_mutex is not
1912 * needed because this new task is not yet running and cannot
1915 assert_spin_locked(¤t->sighand->siglock);
1917 /* Ref-count the new filter user, and assign it. */
1918 get_seccomp_filter(current);
1919 p->seccomp = current->seccomp;
1922 * Explicitly enable no_new_privs here in case it got set
1923 * between the task_struct being duplicated and holding the
1924 * sighand lock. The seccomp state and nnp must be in sync.
1926 if (task_no_new_privs(current))
1927 task_set_no_new_privs(p);
1930 * If the parent gained a seccomp mode after copying thread
1931 * flags and between before we held the sighand lock, we have
1932 * to manually enable the seccomp thread flag here.
1934 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1935 set_task_syscall_work(p, SECCOMP);
1939 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1941 current->clear_child_tid = tidptr;
1943 return task_pid_vnr(current);
1946 static void rt_mutex_init_task(struct task_struct *p)
1948 raw_spin_lock_init(&p->pi_lock);
1949 #ifdef CONFIG_RT_MUTEXES
1950 p->pi_waiters = RB_ROOT_CACHED;
1951 p->pi_top_task = NULL;
1952 p->pi_blocked_on = NULL;
1956 static inline void init_task_pid_links(struct task_struct *task)
1960 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1961 INIT_HLIST_NODE(&task->pid_links[type]);
1965 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1967 if (type == PIDTYPE_PID)
1968 task->thread_pid = pid;
1970 task->signal->pids[type] = pid;
1973 static inline void rcu_copy_process(struct task_struct *p)
1975 #ifdef CONFIG_PREEMPT_RCU
1976 p->rcu_read_lock_nesting = 0;
1977 p->rcu_read_unlock_special.s = 0;
1978 p->rcu_blocked_node = NULL;
1979 INIT_LIST_HEAD(&p->rcu_node_entry);
1980 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1981 #ifdef CONFIG_TASKS_RCU
1982 p->rcu_tasks_holdout = false;
1983 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1984 p->rcu_tasks_idle_cpu = -1;
1985 #endif /* #ifdef CONFIG_TASKS_RCU */
1986 #ifdef CONFIG_TASKS_TRACE_RCU
1987 p->trc_reader_nesting = 0;
1988 p->trc_reader_special.s = 0;
1989 INIT_LIST_HEAD(&p->trc_holdout_list);
1990 INIT_LIST_HEAD(&p->trc_blkd_node);
1991 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1994 struct pid *pidfd_pid(const struct file *file)
1996 if (file->f_op == &pidfd_fops)
1997 return file->private_data;
1999 return ERR_PTR(-EBADF);
2002 static int pidfd_release(struct inode *inode, struct file *file)
2004 struct pid *pid = file->private_data;
2006 file->private_data = NULL;
2011 #ifdef CONFIG_PROC_FS
2013 * pidfd_show_fdinfo - print information about a pidfd
2014 * @m: proc fdinfo file
2015 * @f: file referencing a pidfd
2018 * This function will print the pid that a given pidfd refers to in the
2019 * pid namespace of the procfs instance.
2020 * If the pid namespace of the process is not a descendant of the pid
2021 * namespace of the procfs instance 0 will be shown as its pid. This is
2022 * similar to calling getppid() on a process whose parent is outside of
2023 * its pid namespace.
2026 * If pid namespaces are supported then this function will also print
2027 * the pid of a given pidfd refers to for all descendant pid namespaces
2028 * starting from the current pid namespace of the instance, i.e. the
2029 * Pid field and the first entry in the NSpid field will be identical.
2030 * If the pid namespace of the process is not a descendant of the pid
2031 * namespace of the procfs instance 0 will be shown as its first NSpid
2032 * entry and no others will be shown.
2033 * Note that this differs from the Pid and NSpid fields in
2034 * /proc/<pid>/status where Pid and NSpid are always shown relative to
2035 * the pid namespace of the procfs instance. The difference becomes
2036 * obvious when sending around a pidfd between pid namespaces from a
2037 * different branch of the tree, i.e. where no ancestral relation is
2038 * present between the pid namespaces:
2039 * - create two new pid namespaces ns1 and ns2 in the initial pid
2040 * namespace (also take care to create new mount namespaces in the
2041 * new pid namespace and mount procfs)
2042 * - create a process with a pidfd in ns1
2043 * - send pidfd from ns1 to ns2
2044 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
2045 * have exactly one entry, which is 0
2047 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
2049 struct pid *pid = f->private_data;
2050 struct pid_namespace *ns;
2053 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
2054 ns = proc_pid_ns(file_inode(m->file)->i_sb);
2055 nr = pid_nr_ns(pid, ns);
2058 seq_put_decimal_ll(m, "Pid:\t", nr);
2060 #ifdef CONFIG_PID_NS
2061 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
2065 /* If nr is non-zero it means that 'pid' is valid and that
2066 * ns, i.e. the pid namespace associated with the procfs
2067 * instance, is in the pid namespace hierarchy of pid.
2068 * Start at one below the already printed level.
2070 for (i = ns->level + 1; i <= pid->level; i++)
2071 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2079 * Poll support for process exit notification.
2081 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2083 struct pid *pid = file->private_data;
2084 __poll_t poll_flags = 0;
2086 poll_wait(file, &pid->wait_pidfd, pts);
2089 * Inform pollers only when the whole thread group exits.
2090 * If the thread group leader exits before all other threads in the
2091 * group, then poll(2) should block, similar to the wait(2) family.
2093 if (thread_group_exited(pid))
2094 poll_flags = EPOLLIN | EPOLLRDNORM;
2099 const struct file_operations pidfd_fops = {
2100 .release = pidfd_release,
2102 #ifdef CONFIG_PROC_FS
2103 .show_fdinfo = pidfd_show_fdinfo,
2108 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2109 * @pid: the struct pid for which to create a pidfd
2110 * @flags: flags of the new @pidfd
2111 * @pidfd: the pidfd to return
2113 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2114 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2116 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2117 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2118 * pidfd file are prepared.
2120 * If this function returns successfully the caller is responsible to either
2121 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2122 * order to install the pidfd into its file descriptor table or they must use
2123 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2126 * This function is useful when a pidfd must already be reserved but there
2127 * might still be points of failure afterwards and the caller wants to ensure
2128 * that no pidfd is leaked into its file descriptor table.
2130 * Return: On success, a reserved pidfd is returned from the function and a new
2131 * pidfd file is returned in the last argument to the function. On
2132 * error, a negative error code is returned from the function and the
2133 * last argument remains unchanged.
2135 static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2138 struct file *pidfd_file;
2140 if (flags & ~(O_NONBLOCK | O_RDWR | O_CLOEXEC))
2143 pidfd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2147 pidfd_file = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2148 flags | O_RDWR | O_CLOEXEC);
2149 if (IS_ERR(pidfd_file)) {
2150 put_unused_fd(pidfd);
2151 return PTR_ERR(pidfd_file);
2153 get_pid(pid); /* held by pidfd_file now */
2159 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2160 * @pid: the struct pid for which to create a pidfd
2161 * @flags: flags of the new @pidfd
2162 * @pidfd: the pidfd to return
2164 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2165 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2167 * The helper verifies that @pid is used as a thread group leader.
2169 * If this function returns successfully the caller is responsible to either
2170 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2171 * order to install the pidfd into its file descriptor table or they must use
2172 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2175 * This function is useful when a pidfd must already be reserved but there
2176 * might still be points of failure afterwards and the caller wants to ensure
2177 * that no pidfd is leaked into its file descriptor table.
2179 * Return: On success, a reserved pidfd is returned from the function and a new
2180 * pidfd file is returned in the last argument to the function. On
2181 * error, a negative error code is returned from the function and the
2182 * last argument remains unchanged.
2184 int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2186 if (!pid || !pid_has_task(pid, PIDTYPE_TGID))
2189 return __pidfd_prepare(pid, flags, ret);
2192 static void __delayed_free_task(struct rcu_head *rhp)
2194 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2199 static __always_inline void delayed_free_task(struct task_struct *tsk)
2201 if (IS_ENABLED(CONFIG_MEMCG))
2202 call_rcu(&tsk->rcu, __delayed_free_task);
2207 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2209 /* Skip if kernel thread */
2213 /* Skip if spawning a thread or using vfork */
2214 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2217 /* We need to synchronize with __set_oom_adj */
2218 mutex_lock(&oom_adj_mutex);
2219 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2220 /* Update the values in case they were changed after copy_signal */
2221 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2222 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2223 mutex_unlock(&oom_adj_mutex);
2227 static void rv_task_fork(struct task_struct *p)
2231 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2232 p->rv[i].da_mon.monitoring = false;
2235 #define rv_task_fork(p) do {} while (0)
2239 * This creates a new process as a copy of the old one,
2240 * but does not actually start it yet.
2242 * It copies the registers, and all the appropriate
2243 * parts of the process environment (as per the clone
2244 * flags). The actual kick-off is left to the caller.
2246 __latent_entropy struct task_struct *copy_process(
2250 struct kernel_clone_args *args)
2252 int pidfd = -1, retval;
2253 struct task_struct *p;
2254 struct multiprocess_signals delayed;
2255 struct file *pidfile = NULL;
2256 const u64 clone_flags = args->flags;
2257 struct nsproxy *nsp = current->nsproxy;
2260 * Don't allow sharing the root directory with processes in a different
2263 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2264 return ERR_PTR(-EINVAL);
2266 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2267 return ERR_PTR(-EINVAL);
2270 * Thread groups must share signals as well, and detached threads
2271 * can only be started up within the thread group.
2273 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2274 return ERR_PTR(-EINVAL);
2277 * Shared signal handlers imply shared VM. By way of the above,
2278 * thread groups also imply shared VM. Blocking this case allows
2279 * for various simplifications in other code.
2281 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2282 return ERR_PTR(-EINVAL);
2285 * Siblings of global init remain as zombies on exit since they are
2286 * not reaped by their parent (swapper). To solve this and to avoid
2287 * multi-rooted process trees, prevent global and container-inits
2288 * from creating siblings.
2290 if ((clone_flags & CLONE_PARENT) &&
2291 current->signal->flags & SIGNAL_UNKILLABLE)
2292 return ERR_PTR(-EINVAL);
2295 * If the new process will be in a different pid or user namespace
2296 * do not allow it to share a thread group with the forking task.
2298 if (clone_flags & CLONE_THREAD) {
2299 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2300 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2301 return ERR_PTR(-EINVAL);
2304 if (clone_flags & CLONE_PIDFD) {
2306 * - CLONE_DETACHED is blocked so that we can potentially
2307 * reuse it later for CLONE_PIDFD.
2308 * - CLONE_THREAD is blocked until someone really needs it.
2310 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2311 return ERR_PTR(-EINVAL);
2315 * Force any signals received before this point to be delivered
2316 * before the fork happens. Collect up signals sent to multiple
2317 * processes that happen during the fork and delay them so that
2318 * they appear to happen after the fork.
2320 sigemptyset(&delayed.signal);
2321 INIT_HLIST_NODE(&delayed.node);
2323 spin_lock_irq(¤t->sighand->siglock);
2324 if (!(clone_flags & CLONE_THREAD))
2325 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2326 recalc_sigpending();
2327 spin_unlock_irq(¤t->sighand->siglock);
2328 retval = -ERESTARTNOINTR;
2329 if (task_sigpending(current))
2333 p = dup_task_struct(current, node);
2336 p->flags &= ~PF_KTHREAD;
2338 p->flags |= PF_KTHREAD;
2339 if (args->user_worker) {
2341 * Mark us a user worker, and block any signal that isn't
2344 p->flags |= PF_USER_WORKER;
2345 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2347 if (args->io_thread)
2348 p->flags |= PF_IO_WORKER;
2351 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2353 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2355 * Clear TID on mm_release()?
2357 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2359 ftrace_graph_init_task(p);
2361 rt_mutex_init_task(p);
2363 lockdep_assert_irqs_enabled();
2364 #ifdef CONFIG_PROVE_LOCKING
2365 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2367 retval = copy_creds(p, clone_flags);
2372 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2373 if (p->real_cred->user != INIT_USER &&
2374 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2375 goto bad_fork_cleanup_count;
2377 current->flags &= ~PF_NPROC_EXCEEDED;
2380 * If multiple threads are within copy_process(), then this check
2381 * triggers too late. This doesn't hurt, the check is only there
2382 * to stop root fork bombs.
2385 if (data_race(nr_threads >= max_threads))
2386 goto bad_fork_cleanup_count;
2388 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2389 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2390 p->flags |= PF_FORKNOEXEC;
2391 INIT_LIST_HEAD(&p->children);
2392 INIT_LIST_HEAD(&p->sibling);
2393 rcu_copy_process(p);
2394 p->vfork_done = NULL;
2395 spin_lock_init(&p->alloc_lock);
2397 init_sigpending(&p->pending);
2399 p->utime = p->stime = p->gtime = 0;
2400 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2401 p->utimescaled = p->stimescaled = 0;
2403 prev_cputime_init(&p->prev_cputime);
2405 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2406 seqcount_init(&p->vtime.seqcount);
2407 p->vtime.starttime = 0;
2408 p->vtime.state = VTIME_INACTIVE;
2411 #ifdef CONFIG_IO_URING
2415 #if defined(SPLIT_RSS_COUNTING)
2416 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2419 p->default_timer_slack_ns = current->timer_slack_ns;
2425 task_io_accounting_init(&p->ioac);
2426 acct_clear_integrals(p);
2428 posix_cputimers_init(&p->posix_cputimers);
2430 p->io_context = NULL;
2431 audit_set_context(p, NULL);
2433 if (args->kthread) {
2434 if (!set_kthread_struct(p))
2435 goto bad_fork_cleanup_delayacct;
2438 p->mempolicy = mpol_dup(p->mempolicy);
2439 if (IS_ERR(p->mempolicy)) {
2440 retval = PTR_ERR(p->mempolicy);
2441 p->mempolicy = NULL;
2442 goto bad_fork_cleanup_delayacct;
2445 #ifdef CONFIG_CPUSETS
2446 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2447 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2448 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2450 #ifdef CONFIG_TRACE_IRQFLAGS
2451 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2452 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2453 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2454 p->softirqs_enabled = 1;
2455 p->softirq_context = 0;
2458 p->pagefault_disabled = 0;
2460 #ifdef CONFIG_LOCKDEP
2461 lockdep_init_task(p);
2464 #ifdef CONFIG_DEBUG_MUTEXES
2465 p->blocked_on = NULL; /* not blocked yet */
2467 #ifdef CONFIG_BCACHE
2468 p->sequential_io = 0;
2469 p->sequential_io_avg = 0;
2471 #ifdef CONFIG_BPF_SYSCALL
2472 RCU_INIT_POINTER(p->bpf_storage, NULL);
2476 /* Perform scheduler related setup. Assign this task to a CPU. */
2477 retval = sched_fork(clone_flags, p);
2479 goto bad_fork_cleanup_policy;
2481 retval = perf_event_init_task(p, clone_flags);
2483 goto bad_fork_cleanup_policy;
2484 retval = audit_alloc(p);
2486 goto bad_fork_cleanup_perf;
2487 /* copy all the process information */
2489 retval = security_task_alloc(p, clone_flags);
2491 goto bad_fork_cleanup_audit;
2492 retval = copy_semundo(clone_flags, p);
2494 goto bad_fork_cleanup_security;
2495 retval = copy_files(clone_flags, p, args->no_files);
2497 goto bad_fork_cleanup_semundo;
2498 retval = copy_fs(clone_flags, p);
2500 goto bad_fork_cleanup_files;
2501 retval = copy_sighand(clone_flags, p);
2503 goto bad_fork_cleanup_fs;
2504 retval = copy_signal(clone_flags, p);
2506 goto bad_fork_cleanup_sighand;
2507 retval = copy_mm(clone_flags, p);
2509 goto bad_fork_cleanup_signal;
2510 retval = copy_namespaces(clone_flags, p);
2512 goto bad_fork_cleanup_mm;
2513 retval = copy_io(clone_flags, p);
2515 goto bad_fork_cleanup_namespaces;
2516 retval = copy_thread(p, args);
2518 goto bad_fork_cleanup_io;
2520 stackleak_task_init(p);
2522 if (pid != &init_struct_pid) {
2523 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2524 args->set_tid_size);
2526 retval = PTR_ERR(pid);
2527 goto bad_fork_cleanup_thread;
2532 * This has to happen after we've potentially unshared the file
2533 * descriptor table (so that the pidfd doesn't leak into the child
2534 * if the fd table isn't shared).
2536 if (clone_flags & CLONE_PIDFD) {
2537 /* Note that no task has been attached to @pid yet. */
2538 retval = __pidfd_prepare(pid, O_RDWR | O_CLOEXEC, &pidfile);
2540 goto bad_fork_free_pid;
2543 retval = put_user(pidfd, args->pidfd);
2545 goto bad_fork_put_pidfd;
2554 * sigaltstack should be cleared when sharing the same VM
2556 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2560 * Syscall tracing and stepping should be turned off in the
2561 * child regardless of CLONE_PTRACE.
2563 user_disable_single_step(p);
2564 clear_task_syscall_work(p, SYSCALL_TRACE);
2565 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2566 clear_task_syscall_work(p, SYSCALL_EMU);
2568 clear_tsk_latency_tracing(p);
2570 /* ok, now we should be set up.. */
2571 p->pid = pid_nr(pid);
2572 if (clone_flags & CLONE_THREAD) {
2573 p->group_leader = current->group_leader;
2574 p->tgid = current->tgid;
2576 p->group_leader = p;
2581 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2582 p->dirty_paused_when = 0;
2584 p->pdeath_signal = 0;
2585 INIT_LIST_HEAD(&p->thread_group);
2586 p->task_works = NULL;
2587 clear_posix_cputimers_work(p);
2589 #ifdef CONFIG_KRETPROBES
2590 p->kretprobe_instances.first = NULL;
2592 #ifdef CONFIG_RETHOOK
2593 p->rethooks.first = NULL;
2597 * Ensure that the cgroup subsystem policies allow the new process to be
2598 * forked. It should be noted that the new process's css_set can be changed
2599 * between here and cgroup_post_fork() if an organisation operation is in
2602 retval = cgroup_can_fork(p, args);
2604 goto bad_fork_put_pidfd;
2607 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2608 * the new task on the correct runqueue. All this *before* the task
2611 * This isn't part of ->can_fork() because while the re-cloning is
2612 * cgroup specific, it unconditionally needs to place the task on a
2615 sched_cgroup_fork(p, args);
2618 * From this point on we must avoid any synchronous user-space
2619 * communication until we take the tasklist-lock. In particular, we do
2620 * not want user-space to be able to predict the process start-time by
2621 * stalling fork(2) after we recorded the start_time but before it is
2622 * visible to the system.
2625 p->start_time = ktime_get_ns();
2626 p->start_boottime = ktime_get_boottime_ns();
2629 * Make it visible to the rest of the system, but dont wake it up yet.
2630 * Need tasklist lock for parent etc handling!
2632 write_lock_irq(&tasklist_lock);
2634 /* CLONE_PARENT re-uses the old parent */
2635 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2636 p->real_parent = current->real_parent;
2637 p->parent_exec_id = current->parent_exec_id;
2638 if (clone_flags & CLONE_THREAD)
2639 p->exit_signal = -1;
2641 p->exit_signal = current->group_leader->exit_signal;
2643 p->real_parent = current;
2644 p->parent_exec_id = current->self_exec_id;
2645 p->exit_signal = args->exit_signal;
2648 klp_copy_process(p);
2652 spin_lock(¤t->sighand->siglock);
2656 rseq_fork(p, clone_flags);
2658 /* Don't start children in a dying pid namespace */
2659 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2661 goto bad_fork_cancel_cgroup;
2664 /* Let kill terminate clone/fork in the middle */
2665 if (fatal_signal_pending(current)) {
2667 goto bad_fork_cancel_cgroup;
2670 /* No more failure paths after this point. */
2673 * Copy seccomp details explicitly here, in case they were changed
2674 * before holding sighand lock.
2678 init_task_pid_links(p);
2679 if (likely(p->pid)) {
2680 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2682 init_task_pid(p, PIDTYPE_PID, pid);
2683 if (thread_group_leader(p)) {
2684 init_task_pid(p, PIDTYPE_TGID, pid);
2685 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2686 init_task_pid(p, PIDTYPE_SID, task_session(current));
2688 if (is_child_reaper(pid)) {
2689 ns_of_pid(pid)->child_reaper = p;
2690 p->signal->flags |= SIGNAL_UNKILLABLE;
2692 p->signal->shared_pending.signal = delayed.signal;
2693 p->signal->tty = tty_kref_get(current->signal->tty);
2695 * Inherit has_child_subreaper flag under the same
2696 * tasklist_lock with adding child to the process tree
2697 * for propagate_has_child_subreaper optimization.
2699 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2700 p->real_parent->signal->is_child_subreaper;
2701 list_add_tail(&p->sibling, &p->real_parent->children);
2702 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2703 attach_pid(p, PIDTYPE_TGID);
2704 attach_pid(p, PIDTYPE_PGID);
2705 attach_pid(p, PIDTYPE_SID);
2706 __this_cpu_inc(process_counts);
2708 current->signal->nr_threads++;
2709 current->signal->quick_threads++;
2710 atomic_inc(¤t->signal->live);
2711 refcount_inc(¤t->signal->sigcnt);
2712 task_join_group_stop(p);
2713 list_add_tail_rcu(&p->thread_group,
2714 &p->group_leader->thread_group);
2715 list_add_tail_rcu(&p->thread_node,
2716 &p->signal->thread_head);
2718 attach_pid(p, PIDTYPE_PID);
2722 hlist_del_init(&delayed.node);
2723 spin_unlock(¤t->sighand->siglock);
2724 syscall_tracepoint_update(p);
2725 write_unlock_irq(&tasklist_lock);
2728 fd_install(pidfd, pidfile);
2730 proc_fork_connector(p);
2732 cgroup_post_fork(p, args);
2735 trace_task_newtask(p, clone_flags);
2736 uprobe_copy_process(p, clone_flags);
2737 user_events_fork(p, clone_flags);
2739 copy_oom_score_adj(clone_flags, p);
2743 bad_fork_cancel_cgroup:
2745 spin_unlock(¤t->sighand->siglock);
2746 write_unlock_irq(&tasklist_lock);
2747 cgroup_cancel_fork(p, args);
2749 if (clone_flags & CLONE_PIDFD) {
2751 put_unused_fd(pidfd);
2754 if (pid != &init_struct_pid)
2756 bad_fork_cleanup_thread:
2758 bad_fork_cleanup_io:
2761 bad_fork_cleanup_namespaces:
2762 exit_task_namespaces(p);
2763 bad_fork_cleanup_mm:
2765 mm_clear_owner(p->mm, p);
2768 bad_fork_cleanup_signal:
2769 if (!(clone_flags & CLONE_THREAD))
2770 free_signal_struct(p->signal);
2771 bad_fork_cleanup_sighand:
2772 __cleanup_sighand(p->sighand);
2773 bad_fork_cleanup_fs:
2774 exit_fs(p); /* blocking */
2775 bad_fork_cleanup_files:
2776 exit_files(p); /* blocking */
2777 bad_fork_cleanup_semundo:
2779 bad_fork_cleanup_security:
2780 security_task_free(p);
2781 bad_fork_cleanup_audit:
2783 bad_fork_cleanup_perf:
2784 perf_event_free_task(p);
2785 bad_fork_cleanup_policy:
2786 lockdep_free_task(p);
2788 mpol_put(p->mempolicy);
2790 bad_fork_cleanup_delayacct:
2791 delayacct_tsk_free(p);
2792 bad_fork_cleanup_count:
2793 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2796 WRITE_ONCE(p->__state, TASK_DEAD);
2797 exit_task_stack_account(p);
2799 delayed_free_task(p);
2801 spin_lock_irq(¤t->sighand->siglock);
2802 hlist_del_init(&delayed.node);
2803 spin_unlock_irq(¤t->sighand->siglock);
2804 return ERR_PTR(retval);
2807 static inline void init_idle_pids(struct task_struct *idle)
2811 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2812 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2813 init_task_pid(idle, type, &init_struct_pid);
2817 static int idle_dummy(void *dummy)
2819 /* This function is never called */
2823 struct task_struct * __init fork_idle(int cpu)
2825 struct task_struct *task;
2826 struct kernel_clone_args args = {
2834 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2835 if (!IS_ERR(task)) {
2836 init_idle_pids(task);
2837 init_idle(task, cpu);
2844 * This is like kernel_clone(), but shaved down and tailored to just
2845 * creating io_uring workers. It returns a created task, or an error pointer.
2846 * The returned task is inactive, and the caller must fire it up through
2847 * wake_up_new_task(p). All signals are blocked in the created task.
2849 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2851 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2853 struct kernel_clone_args args = {
2854 .flags = ((lower_32_bits(flags) | CLONE_VM |
2855 CLONE_UNTRACED) & ~CSIGNAL),
2856 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2863 return copy_process(NULL, 0, node, &args);
2867 * Ok, this is the main fork-routine.
2869 * It copies the process, and if successful kick-starts
2870 * it and waits for it to finish using the VM if required.
2872 * args->exit_signal is expected to be checked for sanity by the caller.
2874 pid_t kernel_clone(struct kernel_clone_args *args)
2876 u64 clone_flags = args->flags;
2877 struct completion vfork;
2879 struct task_struct *p;
2884 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2885 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2886 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2887 * field in struct clone_args and it still doesn't make sense to have
2888 * them both point at the same memory location. Performing this check
2889 * here has the advantage that we don't need to have a separate helper
2890 * to check for legacy clone().
2892 if ((args->flags & CLONE_PIDFD) &&
2893 (args->flags & CLONE_PARENT_SETTID) &&
2894 (args->pidfd == args->parent_tid))
2898 * Determine whether and which event to report to ptracer. When
2899 * called from kernel_thread or CLONE_UNTRACED is explicitly
2900 * requested, no event is reported; otherwise, report if the event
2901 * for the type of forking is enabled.
2903 if (!(clone_flags & CLONE_UNTRACED)) {
2904 if (clone_flags & CLONE_VFORK)
2905 trace = PTRACE_EVENT_VFORK;
2906 else if (args->exit_signal != SIGCHLD)
2907 trace = PTRACE_EVENT_CLONE;
2909 trace = PTRACE_EVENT_FORK;
2911 if (likely(!ptrace_event_enabled(current, trace)))
2915 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2916 add_latent_entropy();
2922 * Do this prior waking up the new thread - the thread pointer
2923 * might get invalid after that point, if the thread exits quickly.
2925 trace_sched_process_fork(current, p);
2927 pid = get_task_pid(p, PIDTYPE_PID);
2930 if (clone_flags & CLONE_PARENT_SETTID)
2931 put_user(nr, args->parent_tid);
2933 if (clone_flags & CLONE_VFORK) {
2934 p->vfork_done = &vfork;
2935 init_completion(&vfork);
2939 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2940 /* lock the task to synchronize with memcg migration */
2942 lru_gen_add_mm(p->mm);
2946 wake_up_new_task(p);
2948 /* forking complete and child started to run, tell ptracer */
2949 if (unlikely(trace))
2950 ptrace_event_pid(trace, pid);
2952 if (clone_flags & CLONE_VFORK) {
2953 if (!wait_for_vfork_done(p, &vfork))
2954 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2962 * Create a kernel thread.
2964 pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2965 unsigned long flags)
2967 struct kernel_clone_args args = {
2968 .flags = ((lower_32_bits(flags) | CLONE_VM |
2969 CLONE_UNTRACED) & ~CSIGNAL),
2970 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2977 return kernel_clone(&args);
2981 * Create a user mode thread.
2983 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2985 struct kernel_clone_args args = {
2986 .flags = ((lower_32_bits(flags) | CLONE_VM |
2987 CLONE_UNTRACED) & ~CSIGNAL),
2988 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2993 return kernel_clone(&args);
2996 #ifdef __ARCH_WANT_SYS_FORK
2997 SYSCALL_DEFINE0(fork)
3000 struct kernel_clone_args args = {
3001 .exit_signal = SIGCHLD,
3004 return kernel_clone(&args);
3006 /* can not support in nommu mode */
3012 #ifdef __ARCH_WANT_SYS_VFORK
3013 SYSCALL_DEFINE0(vfork)
3015 struct kernel_clone_args args = {
3016 .flags = CLONE_VFORK | CLONE_VM,
3017 .exit_signal = SIGCHLD,
3020 return kernel_clone(&args);
3024 #ifdef __ARCH_WANT_SYS_CLONE
3025 #ifdef CONFIG_CLONE_BACKWARDS
3026 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3027 int __user *, parent_tidptr,
3029 int __user *, child_tidptr)
3030 #elif defined(CONFIG_CLONE_BACKWARDS2)
3031 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
3032 int __user *, parent_tidptr,
3033 int __user *, child_tidptr,
3035 #elif defined(CONFIG_CLONE_BACKWARDS3)
3036 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
3038 int __user *, parent_tidptr,
3039 int __user *, child_tidptr,
3042 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3043 int __user *, parent_tidptr,
3044 int __user *, child_tidptr,
3048 struct kernel_clone_args args = {
3049 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
3050 .pidfd = parent_tidptr,
3051 .child_tid = child_tidptr,
3052 .parent_tid = parent_tidptr,
3053 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
3058 return kernel_clone(&args);
3062 #ifdef __ARCH_WANT_SYS_CLONE3
3064 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
3065 struct clone_args __user *uargs,
3069 struct clone_args args;
3070 pid_t *kset_tid = kargs->set_tid;
3072 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
3073 CLONE_ARGS_SIZE_VER0);
3074 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
3075 CLONE_ARGS_SIZE_VER1);
3076 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
3077 CLONE_ARGS_SIZE_VER2);
3078 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
3080 if (unlikely(usize > PAGE_SIZE))
3082 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
3085 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
3089 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
3092 if (unlikely(!args.set_tid && args.set_tid_size > 0))
3095 if (unlikely(args.set_tid && args.set_tid_size == 0))
3099 * Verify that higher 32bits of exit_signal are unset and that
3100 * it is a valid signal
3102 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3103 !valid_signal(args.exit_signal)))
3106 if ((args.flags & CLONE_INTO_CGROUP) &&
3107 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3110 *kargs = (struct kernel_clone_args){
3111 .flags = args.flags,
3112 .pidfd = u64_to_user_ptr(args.pidfd),
3113 .child_tid = u64_to_user_ptr(args.child_tid),
3114 .parent_tid = u64_to_user_ptr(args.parent_tid),
3115 .exit_signal = args.exit_signal,
3116 .stack = args.stack,
3117 .stack_size = args.stack_size,
3119 .set_tid_size = args.set_tid_size,
3120 .cgroup = args.cgroup,
3124 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3125 (kargs->set_tid_size * sizeof(pid_t))))
3128 kargs->set_tid = kset_tid;
3134 * clone3_stack_valid - check and prepare stack
3135 * @kargs: kernel clone args
3137 * Verify that the stack arguments userspace gave us are sane.
3138 * In addition, set the stack direction for userspace since it's easy for us to
3141 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3143 if (kargs->stack == 0) {
3144 if (kargs->stack_size > 0)
3147 if (kargs->stack_size == 0)
3150 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3153 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
3154 kargs->stack += kargs->stack_size;
3161 static bool clone3_args_valid(struct kernel_clone_args *kargs)
3163 /* Verify that no unknown flags are passed along. */
3165 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3169 * - make the CLONE_DETACHED bit reusable for clone3
3170 * - make the CSIGNAL bits reusable for clone3
3172 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3175 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3176 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3179 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3183 if (!clone3_stack_valid(kargs))
3190 * clone3 - create a new process with specific properties
3191 * @uargs: argument structure
3192 * @size: size of @uargs
3194 * clone3() is the extensible successor to clone()/clone2().
3195 * It takes a struct as argument that is versioned by its size.
3197 * Return: On success, a positive PID for the child process.
3198 * On error, a negative errno number.
3200 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3204 struct kernel_clone_args kargs;
3205 pid_t set_tid[MAX_PID_NS_LEVEL];
3207 kargs.set_tid = set_tid;
3209 err = copy_clone_args_from_user(&kargs, uargs, size);
3213 if (!clone3_args_valid(&kargs))
3216 return kernel_clone(&kargs);
3220 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3222 struct task_struct *leader, *parent, *child;
3225 read_lock(&tasklist_lock);
3226 leader = top = top->group_leader;
3228 for_each_thread(leader, parent) {
3229 list_for_each_entry(child, &parent->children, sibling) {
3230 res = visitor(child, data);
3242 if (leader != top) {
3244 parent = child->real_parent;
3245 leader = parent->group_leader;
3249 read_unlock(&tasklist_lock);
3252 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3253 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3256 static void sighand_ctor(void *data)
3258 struct sighand_struct *sighand = data;
3260 spin_lock_init(&sighand->siglock);
3261 init_waitqueue_head(&sighand->signalfd_wqh);
3264 void __init mm_cache_init(void)
3266 unsigned int mm_size;
3269 * The mm_cpumask is located at the end of mm_struct, and is
3270 * dynamically sized based on the maximum CPU number this system
3271 * can have, taking hotplug into account (nr_cpu_ids).
3273 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3275 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3276 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3277 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3278 offsetof(struct mm_struct, saved_auxv),
3279 sizeof_field(struct mm_struct, saved_auxv),
3283 void __init proc_caches_init(void)
3285 sighand_cachep = kmem_cache_create("sighand_cache",
3286 sizeof(struct sighand_struct), 0,
3287 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3288 SLAB_ACCOUNT, sighand_ctor);
3289 signal_cachep = kmem_cache_create("signal_cache",
3290 sizeof(struct signal_struct), 0,
3291 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3293 files_cachep = kmem_cache_create("files_cache",
3294 sizeof(struct files_struct), 0,
3295 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3297 fs_cachep = kmem_cache_create("fs_cache",
3298 sizeof(struct fs_struct), 0,
3299 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3302 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3303 #ifdef CONFIG_PER_VMA_LOCK
3304 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3307 nsproxy_cache_init();
3311 * Check constraints on flags passed to the unshare system call.
3313 static int check_unshare_flags(unsigned long unshare_flags)
3315 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3316 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3317 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3318 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3322 * Not implemented, but pretend it works if there is nothing
3323 * to unshare. Note that unsharing the address space or the
3324 * signal handlers also need to unshare the signal queues (aka
3327 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3328 if (!thread_group_empty(current))
3331 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3332 if (refcount_read(¤t->sighand->count) > 1)
3335 if (unshare_flags & CLONE_VM) {
3336 if (!current_is_single_threaded())
3344 * Unshare the filesystem structure if it is being shared
3346 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3348 struct fs_struct *fs = current->fs;
3350 if (!(unshare_flags & CLONE_FS) || !fs)
3353 /* don't need lock here; in the worst case we'll do useless copy */
3357 *new_fsp = copy_fs_struct(fs);
3365 * Unshare file descriptor table if it is being shared
3367 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3368 struct files_struct **new_fdp)
3370 struct files_struct *fd = current->files;
3373 if ((unshare_flags & CLONE_FILES) &&
3374 (fd && atomic_read(&fd->count) > 1)) {
3375 *new_fdp = dup_fd(fd, max_fds, &error);
3384 * unshare allows a process to 'unshare' part of the process
3385 * context which was originally shared using clone. copy_*
3386 * functions used by kernel_clone() cannot be used here directly
3387 * because they modify an inactive task_struct that is being
3388 * constructed. Here we are modifying the current, active,
3391 int ksys_unshare(unsigned long unshare_flags)
3393 struct fs_struct *fs, *new_fs = NULL;
3394 struct files_struct *new_fd = NULL;
3395 struct cred *new_cred = NULL;
3396 struct nsproxy *new_nsproxy = NULL;
3401 * If unsharing a user namespace must also unshare the thread group
3402 * and unshare the filesystem root and working directories.
3404 if (unshare_flags & CLONE_NEWUSER)
3405 unshare_flags |= CLONE_THREAD | CLONE_FS;
3407 * If unsharing vm, must also unshare signal handlers.
3409 if (unshare_flags & CLONE_VM)
3410 unshare_flags |= CLONE_SIGHAND;
3412 * If unsharing a signal handlers, must also unshare the signal queues.
3414 if (unshare_flags & CLONE_SIGHAND)
3415 unshare_flags |= CLONE_THREAD;
3417 * If unsharing namespace, must also unshare filesystem information.
3419 if (unshare_flags & CLONE_NEWNS)
3420 unshare_flags |= CLONE_FS;
3422 err = check_unshare_flags(unshare_flags);
3424 goto bad_unshare_out;
3426 * CLONE_NEWIPC must also detach from the undolist: after switching
3427 * to a new ipc namespace, the semaphore arrays from the old
3428 * namespace are unreachable.
3430 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3432 err = unshare_fs(unshare_flags, &new_fs);
3434 goto bad_unshare_out;
3435 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3437 goto bad_unshare_cleanup_fs;
3438 err = unshare_userns(unshare_flags, &new_cred);
3440 goto bad_unshare_cleanup_fd;
3441 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3444 goto bad_unshare_cleanup_cred;
3447 err = set_cred_ucounts(new_cred);
3449 goto bad_unshare_cleanup_cred;
3452 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3455 * CLONE_SYSVSEM is equivalent to sys_exit().
3459 if (unshare_flags & CLONE_NEWIPC) {
3460 /* Orphan segments in old ns (see sem above). */
3462 shm_init_task(current);
3466 switch_task_namespaces(current, new_nsproxy);
3472 spin_lock(&fs->lock);
3473 current->fs = new_fs;
3478 spin_unlock(&fs->lock);
3482 swap(current->files, new_fd);
3484 task_unlock(current);
3487 /* Install the new user namespace */
3488 commit_creds(new_cred);
3493 perf_event_namespaces(current);
3495 bad_unshare_cleanup_cred:
3498 bad_unshare_cleanup_fd:
3500 put_files_struct(new_fd);
3502 bad_unshare_cleanup_fs:
3504 free_fs_struct(new_fs);
3510 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3512 return ksys_unshare(unshare_flags);
3516 * Helper to unshare the files of the current task.
3517 * We don't want to expose copy_files internals to
3518 * the exec layer of the kernel.
3521 int unshare_files(void)
3523 struct task_struct *task = current;
3524 struct files_struct *old, *copy = NULL;
3527 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3535 put_files_struct(old);
3539 int sysctl_max_threads(struct ctl_table *table, int write,
3540 void *buffer, size_t *lenp, loff_t *ppos)
3544 int threads = max_threads;
3546 int max = MAX_THREADS;
3553 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3557 max_threads = threads;