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>
102 #include <asm/pgalloc.h>
103 #include <linux/uaccess.h>
104 #include <asm/mmu_context.h>
105 #include <asm/cacheflush.h>
106 #include <asm/tlbflush.h>
108 #include <trace/events/sched.h>
110 #define CREATE_TRACE_POINTS
111 #include <trace/events/task.h>
114 * Minimum number of threads to boot the kernel
116 #define MIN_THREADS 20
119 * Maximum number of threads
121 #define MAX_THREADS FUTEX_TID_MASK
124 * Protected counters by write_lock_irq(&tasklist_lock)
126 unsigned long total_forks; /* Handle normal Linux uptimes. */
127 int nr_threads; /* The idle threads do not count.. */
129 static int max_threads; /* tunable limit on nr_threads */
131 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
133 static const char * const resident_page_types[] = {
134 NAMED_ARRAY_INDEX(MM_FILEPAGES),
135 NAMED_ARRAY_INDEX(MM_ANONPAGES),
136 NAMED_ARRAY_INDEX(MM_SWAPENTS),
137 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
140 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
142 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
144 #ifdef CONFIG_PROVE_RCU
145 int lockdep_tasklist_lock_is_held(void)
147 return lockdep_is_held(&tasklist_lock);
149 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
150 #endif /* #ifdef CONFIG_PROVE_RCU */
152 int nr_processes(void)
157 for_each_possible_cpu(cpu)
158 total += per_cpu(process_counts, cpu);
163 void __weak arch_release_task_struct(struct task_struct *tsk)
167 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
168 static struct kmem_cache *task_struct_cachep;
170 static inline struct task_struct *alloc_task_struct_node(int node)
172 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
175 static inline void free_task_struct(struct task_struct *tsk)
177 kmem_cache_free(task_struct_cachep, tsk);
181 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
184 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
185 * kmemcache based allocator.
187 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
189 # ifdef CONFIG_VMAP_STACK
191 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
192 * flush. Try to minimize the number of calls by caching stacks.
194 #define NR_CACHED_STACKS 2
195 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
199 struct vm_struct *stack_vm_area;
202 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
206 for (i = 0; i < NR_CACHED_STACKS; i++) {
207 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
214 static void thread_stack_free_rcu(struct rcu_head *rh)
216 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
218 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
224 static void thread_stack_delayed_free(struct task_struct *tsk)
226 struct vm_stack *vm_stack = tsk->stack;
228 vm_stack->stack_vm_area = tsk->stack_vm_area;
229 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
232 static int free_vm_stack_cache(unsigned int cpu)
234 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
237 for (i = 0; i < NR_CACHED_STACKS; i++) {
238 struct vm_struct *vm_stack = cached_vm_stacks[i];
243 vfree(vm_stack->addr);
244 cached_vm_stacks[i] = NULL;
250 static int memcg_charge_kernel_stack(struct vm_struct *vm)
255 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
256 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
258 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
259 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
266 * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
267 * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
270 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
271 memcg_kmem_uncharge_page(vm->pages[i], 0);
275 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
277 struct vm_struct *vm;
281 for (i = 0; i < NR_CACHED_STACKS; i++) {
284 s = this_cpu_xchg(cached_stacks[i], NULL);
289 /* Reset stack metadata. */
290 kasan_unpoison_range(s->addr, THREAD_SIZE);
292 stack = kasan_reset_tag(s->addr);
294 /* Clear stale pointers from reused stack. */
295 memset(stack, 0, THREAD_SIZE);
297 if (memcg_charge_kernel_stack(s)) {
302 tsk->stack_vm_area = s;
308 * Allocated stacks are cached and later reused by new threads,
309 * so memcg accounting is performed manually on assigning/releasing
310 * stacks to tasks. Drop __GFP_ACCOUNT.
312 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
313 VMALLOC_START, VMALLOC_END,
314 THREADINFO_GFP & ~__GFP_ACCOUNT,
316 0, node, __builtin_return_address(0));
320 vm = find_vm_area(stack);
321 if (memcg_charge_kernel_stack(vm)) {
326 * We can't call find_vm_area() in interrupt context, and
327 * free_thread_stack() can be called in interrupt context,
328 * so cache the vm_struct.
330 tsk->stack_vm_area = vm;
331 stack = kasan_reset_tag(stack);
336 static void free_thread_stack(struct task_struct *tsk)
338 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
339 thread_stack_delayed_free(tsk);
342 tsk->stack_vm_area = NULL;
345 # else /* !CONFIG_VMAP_STACK */
347 static void thread_stack_free_rcu(struct rcu_head *rh)
349 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
352 static void thread_stack_delayed_free(struct task_struct *tsk)
354 struct rcu_head *rh = tsk->stack;
356 call_rcu(rh, thread_stack_free_rcu);
359 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
361 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
365 tsk->stack = kasan_reset_tag(page_address(page));
371 static void free_thread_stack(struct task_struct *tsk)
373 thread_stack_delayed_free(tsk);
377 # endif /* CONFIG_VMAP_STACK */
378 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
380 static struct kmem_cache *thread_stack_cache;
382 static void thread_stack_free_rcu(struct rcu_head *rh)
384 kmem_cache_free(thread_stack_cache, rh);
387 static void thread_stack_delayed_free(struct task_struct *tsk)
389 struct rcu_head *rh = tsk->stack;
391 call_rcu(rh, thread_stack_free_rcu);
394 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
396 unsigned long *stack;
397 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
398 stack = kasan_reset_tag(stack);
400 return stack ? 0 : -ENOMEM;
403 static void free_thread_stack(struct task_struct *tsk)
405 thread_stack_delayed_free(tsk);
409 void thread_stack_cache_init(void)
411 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
412 THREAD_SIZE, THREAD_SIZE, 0, 0,
414 BUG_ON(thread_stack_cache == NULL);
417 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
418 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
420 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
422 unsigned long *stack;
424 stack = arch_alloc_thread_stack_node(tsk, node);
426 return stack ? 0 : -ENOMEM;
429 static void free_thread_stack(struct task_struct *tsk)
431 arch_free_thread_stack(tsk);
435 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
437 /* SLAB cache for signal_struct structures (tsk->signal) */
438 static struct kmem_cache *signal_cachep;
440 /* SLAB cache for sighand_struct structures (tsk->sighand) */
441 struct kmem_cache *sighand_cachep;
443 /* SLAB cache for files_struct structures (tsk->files) */
444 struct kmem_cache *files_cachep;
446 /* SLAB cache for fs_struct structures (tsk->fs) */
447 struct kmem_cache *fs_cachep;
449 /* SLAB cache for vm_area_struct structures */
450 static struct kmem_cache *vm_area_cachep;
452 /* SLAB cache for mm_struct structures (tsk->mm) */
453 static struct kmem_cache *mm_cachep;
455 #ifdef CONFIG_PER_VMA_LOCK
457 /* SLAB cache for vm_area_struct.lock */
458 static struct kmem_cache *vma_lock_cachep;
460 static bool vma_lock_alloc(struct vm_area_struct *vma)
462 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
466 init_rwsem(&vma->vm_lock->lock);
467 vma->vm_lock_seq = -1;
472 static inline void vma_lock_free(struct vm_area_struct *vma)
474 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
477 #else /* CONFIG_PER_VMA_LOCK */
479 static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
480 static inline void vma_lock_free(struct vm_area_struct *vma) {}
482 #endif /* CONFIG_PER_VMA_LOCK */
484 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
486 struct vm_area_struct *vma;
488 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
493 if (!vma_lock_alloc(vma)) {
494 kmem_cache_free(vm_area_cachep, vma);
501 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
503 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
508 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
509 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
511 * orig->shared.rb may be modified concurrently, but the clone
512 * will be reinitialized.
514 data_race(memcpy(new, orig, sizeof(*new)));
515 if (!vma_lock_alloc(new)) {
516 kmem_cache_free(vm_area_cachep, new);
519 INIT_LIST_HEAD(&new->anon_vma_chain);
520 vma_numab_state_init(new);
521 dup_anon_vma_name(orig, new);
526 void __vm_area_free(struct vm_area_struct *vma)
528 vma_numab_state_free(vma);
529 free_anon_vma_name(vma);
531 kmem_cache_free(vm_area_cachep, vma);
534 #ifdef CONFIG_PER_VMA_LOCK
535 static void vm_area_free_rcu_cb(struct rcu_head *head)
537 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
540 /* The vma should not be locked while being destroyed. */
541 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
546 void vm_area_free(struct vm_area_struct *vma)
548 #ifdef CONFIG_PER_VMA_LOCK
549 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
555 static void account_kernel_stack(struct task_struct *tsk, int account)
557 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
558 struct vm_struct *vm = task_stack_vm_area(tsk);
561 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
562 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
563 account * (PAGE_SIZE / 1024));
565 void *stack = task_stack_page(tsk);
567 /* All stack pages are in the same node. */
568 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
569 account * (THREAD_SIZE / 1024));
573 void exit_task_stack_account(struct task_struct *tsk)
575 account_kernel_stack(tsk, -1);
577 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
578 struct vm_struct *vm;
581 vm = task_stack_vm_area(tsk);
582 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
583 memcg_kmem_uncharge_page(vm->pages[i], 0);
587 static void release_task_stack(struct task_struct *tsk)
589 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
590 return; /* Better to leak the stack than to free prematurely */
592 free_thread_stack(tsk);
595 #ifdef CONFIG_THREAD_INFO_IN_TASK
596 void put_task_stack(struct task_struct *tsk)
598 if (refcount_dec_and_test(&tsk->stack_refcount))
599 release_task_stack(tsk);
603 void free_task(struct task_struct *tsk)
605 #ifdef CONFIG_SECCOMP
606 WARN_ON_ONCE(tsk->seccomp.filter);
608 release_user_cpus_ptr(tsk);
611 #ifndef CONFIG_THREAD_INFO_IN_TASK
613 * The task is finally done with both the stack and thread_info,
616 release_task_stack(tsk);
619 * If the task had a separate stack allocation, it should be gone
622 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
624 rt_mutex_debug_task_free(tsk);
625 ftrace_graph_exit_task(tsk);
626 arch_release_task_struct(tsk);
627 if (tsk->flags & PF_KTHREAD)
628 free_kthread_struct(tsk);
629 free_task_struct(tsk);
631 EXPORT_SYMBOL(free_task);
633 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
635 struct file *exe_file;
637 exe_file = get_mm_exe_file(oldmm);
638 RCU_INIT_POINTER(mm->exe_file, exe_file);
640 * We depend on the oldmm having properly denied write access to the
643 if (exe_file && deny_write_access(exe_file))
644 pr_warn_once("deny_write_access() failed in %s\n", __func__);
648 static __latent_entropy int dup_mmap(struct mm_struct *mm,
649 struct mm_struct *oldmm)
651 struct vm_area_struct *mpnt, *tmp;
653 unsigned long charge = 0;
655 VMA_ITERATOR(old_vmi, oldmm, 0);
656 VMA_ITERATOR(vmi, mm, 0);
658 uprobe_start_dup_mmap();
659 if (mmap_write_lock_killable(oldmm)) {
661 goto fail_uprobe_end;
663 flush_cache_dup_mm(oldmm);
664 uprobe_dup_mmap(oldmm, mm);
666 * Not linked in yet - no deadlock potential:
668 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
670 /* No ordering required: file already has been exposed. */
671 dup_mm_exe_file(mm, oldmm);
673 mm->total_vm = oldmm->total_vm;
674 mm->data_vm = oldmm->data_vm;
675 mm->exec_vm = oldmm->exec_vm;
676 mm->stack_vm = oldmm->stack_vm;
678 retval = ksm_fork(mm, oldmm);
681 khugepaged_fork(mm, oldmm);
683 retval = vma_iter_bulk_alloc(&vmi, oldmm->map_count);
687 mt_clear_in_rcu(vmi.mas.tree);
688 for_each_vma(old_vmi, mpnt) {
691 if (mpnt->vm_flags & VM_DONTCOPY) {
692 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
697 * Don't duplicate many vmas if we've been oom-killed (for
700 if (fatal_signal_pending(current)) {
704 if (mpnt->vm_flags & VM_ACCOUNT) {
705 unsigned long len = vma_pages(mpnt);
707 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
711 tmp = vm_area_dup(mpnt);
714 retval = vma_dup_policy(mpnt, tmp);
716 goto fail_nomem_policy;
718 retval = dup_userfaultfd(tmp, &uf);
720 goto fail_nomem_anon_vma_fork;
721 if (tmp->vm_flags & VM_WIPEONFORK) {
723 * VM_WIPEONFORK gets a clean slate in the child.
724 * Don't prepare anon_vma until fault since we don't
725 * copy page for current vma.
727 tmp->anon_vma = NULL;
728 } else if (anon_vma_fork(tmp, mpnt))
729 goto fail_nomem_anon_vma_fork;
730 vm_flags_clear(tmp, VM_LOCKED_MASK);
733 struct address_space *mapping = file->f_mapping;
736 i_mmap_lock_write(mapping);
737 if (tmp->vm_flags & VM_SHARED)
738 mapping_allow_writable(mapping);
739 flush_dcache_mmap_lock(mapping);
740 /* insert tmp into the share list, just after mpnt */
741 vma_interval_tree_insert_after(tmp, mpnt,
743 flush_dcache_mmap_unlock(mapping);
744 i_mmap_unlock_write(mapping);
748 * Copy/update hugetlb private vma information.
750 if (is_vm_hugetlb_page(tmp))
751 hugetlb_dup_vma_private(tmp);
753 /* Link the vma into the MT */
754 if (vma_iter_bulk_store(&vmi, tmp))
755 goto fail_nomem_vmi_store;
758 if (!(tmp->vm_flags & VM_WIPEONFORK))
759 retval = copy_page_range(tmp, mpnt);
761 if (tmp->vm_ops && tmp->vm_ops->open)
762 tmp->vm_ops->open(tmp);
767 /* a new mm has just been created */
768 retval = arch_dup_mmap(oldmm, mm);
772 mt_set_in_rcu(vmi.mas.tree);
774 mmap_write_unlock(mm);
776 mmap_write_unlock(oldmm);
777 dup_userfaultfd_complete(&uf);
779 uprobe_end_dup_mmap();
782 fail_nomem_vmi_store:
783 unlink_anon_vmas(tmp);
784 fail_nomem_anon_vma_fork:
785 mpol_put(vma_policy(tmp));
790 vm_unacct_memory(charge);
794 static inline int mm_alloc_pgd(struct mm_struct *mm)
796 mm->pgd = pgd_alloc(mm);
797 if (unlikely(!mm->pgd))
802 static inline void mm_free_pgd(struct mm_struct *mm)
804 pgd_free(mm, mm->pgd);
807 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
809 mmap_write_lock(oldmm);
810 dup_mm_exe_file(mm, oldmm);
811 mmap_write_unlock(oldmm);
814 #define mm_alloc_pgd(mm) (0)
815 #define mm_free_pgd(mm)
816 #endif /* CONFIG_MMU */
818 static void check_mm(struct mm_struct *mm)
822 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
823 "Please make sure 'struct resident_page_types[]' is updated as well");
825 for (i = 0; i < NR_MM_COUNTERS; i++) {
826 long x = percpu_counter_sum(&mm->rss_stat[i]);
829 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
830 mm, resident_page_types[i], x);
833 if (mm_pgtables_bytes(mm))
834 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
835 mm_pgtables_bytes(mm));
837 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
838 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
842 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
843 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
845 static void do_check_lazy_tlb(void *arg)
847 struct mm_struct *mm = arg;
849 WARN_ON_ONCE(current->active_mm == mm);
852 static void do_shoot_lazy_tlb(void *arg)
854 struct mm_struct *mm = arg;
856 if (current->active_mm == mm) {
857 WARN_ON_ONCE(current->mm);
858 current->active_mm = &init_mm;
859 switch_mm(mm, &init_mm, current);
863 static void cleanup_lazy_tlbs(struct mm_struct *mm)
865 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
867 * In this case, lazy tlb mms are refounted and would not reach
868 * __mmdrop until all CPUs have switched away and mmdrop()ed.
874 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
875 * requires lazy mm users to switch to another mm when the refcount
876 * drops to zero, before the mm is freed. This requires IPIs here to
877 * switch kernel threads to init_mm.
879 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
880 * switch with the final userspace teardown TLB flush which leaves the
881 * mm lazy on this CPU but no others, reducing the need for additional
882 * IPIs here. There are cases where a final IPI is still required here,
883 * such as the final mmdrop being performed on a different CPU than the
884 * one exiting, or kernel threads using the mm when userspace exits.
886 * IPI overheads have not found to be expensive, but they could be
887 * reduced in a number of possible ways, for example (roughly
888 * increasing order of complexity):
889 * - The last lazy reference created by exit_mm() could instead switch
890 * to init_mm, however it's probable this will run on the same CPU
891 * immediately afterwards, so this may not reduce IPIs much.
892 * - A batch of mms requiring IPIs could be gathered and freed at once.
893 * - CPUs store active_mm where it can be remotely checked without a
894 * lock, to filter out false-positives in the cpumask.
895 * - After mm_users or mm_count reaches zero, switching away from the
896 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
897 * with some batching or delaying of the final IPIs.
898 * - A delayed freeing and RCU-like quiescing sequence based on mm
899 * switching to avoid IPIs completely.
901 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
902 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
903 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
907 * Called when the last reference to the mm
908 * is dropped: either by a lazy thread or by
909 * mmput. Free the page directory and the mm.
911 void __mmdrop(struct mm_struct *mm)
915 BUG_ON(mm == &init_mm);
916 WARN_ON_ONCE(mm == current->mm);
918 /* Ensure no CPUs are using this as their lazy tlb mm */
919 cleanup_lazy_tlbs(mm);
921 WARN_ON_ONCE(mm == current->active_mm);
924 mmu_notifier_subscriptions_destroy(mm);
926 put_user_ns(mm->user_ns);
930 for (i = 0; i < NR_MM_COUNTERS; i++)
931 percpu_counter_destroy(&mm->rss_stat[i]);
934 EXPORT_SYMBOL_GPL(__mmdrop);
936 static void mmdrop_async_fn(struct work_struct *work)
938 struct mm_struct *mm;
940 mm = container_of(work, struct mm_struct, async_put_work);
944 static void mmdrop_async(struct mm_struct *mm)
946 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
947 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
948 schedule_work(&mm->async_put_work);
952 static inline void free_signal_struct(struct signal_struct *sig)
954 taskstats_tgid_free(sig);
955 sched_autogroup_exit(sig);
957 * __mmdrop is not safe to call from softirq context on x86 due to
958 * pgd_dtor so postpone it to the async context
961 mmdrop_async(sig->oom_mm);
962 kmem_cache_free(signal_cachep, sig);
965 static inline void put_signal_struct(struct signal_struct *sig)
967 if (refcount_dec_and_test(&sig->sigcnt))
968 free_signal_struct(sig);
971 void __put_task_struct(struct task_struct *tsk)
973 WARN_ON(!tsk->exit_state);
974 WARN_ON(refcount_read(&tsk->usage));
975 WARN_ON(tsk == current);
979 task_numa_free(tsk, true);
980 security_task_free(tsk);
981 bpf_task_storage_free(tsk);
983 delayacct_tsk_free(tsk);
984 put_signal_struct(tsk->signal);
985 sched_core_free(tsk);
988 EXPORT_SYMBOL_GPL(__put_task_struct);
990 void __init __weak arch_task_cache_init(void) { }
995 static void set_max_threads(unsigned int max_threads_suggested)
998 unsigned long nr_pages = totalram_pages();
1001 * The number of threads shall be limited such that the thread
1002 * structures may only consume a small part of the available memory.
1004 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1005 threads = MAX_THREADS;
1007 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1008 (u64) THREAD_SIZE * 8UL);
1010 if (threads > max_threads_suggested)
1011 threads = max_threads_suggested;
1013 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1016 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1017 /* Initialized by the architecture: */
1018 int arch_task_struct_size __read_mostly;
1021 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1022 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1024 /* Fetch thread_struct whitelist for the architecture. */
1025 arch_thread_struct_whitelist(offset, size);
1028 * Handle zero-sized whitelist or empty thread_struct, otherwise
1029 * adjust offset to position of thread_struct in task_struct.
1031 if (unlikely(*size == 0))
1034 *offset += offsetof(struct task_struct, thread);
1036 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
1038 void __init fork_init(void)
1041 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1042 #ifndef ARCH_MIN_TASKALIGN
1043 #define ARCH_MIN_TASKALIGN 0
1045 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1046 unsigned long useroffset, usersize;
1048 /* create a slab on which task_structs can be allocated */
1049 task_struct_whitelist(&useroffset, &usersize);
1050 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1051 arch_task_struct_size, align,
1052 SLAB_PANIC|SLAB_ACCOUNT,
1053 useroffset, usersize, NULL);
1056 /* do the arch specific task caches init */
1057 arch_task_cache_init();
1059 set_max_threads(MAX_THREADS);
1061 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1062 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1063 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1064 init_task.signal->rlim[RLIMIT_NPROC];
1066 for (i = 0; i < UCOUNT_COUNTS; i++)
1067 init_user_ns.ucount_max[i] = max_threads/2;
1069 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1070 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1071 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1072 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1074 #ifdef CONFIG_VMAP_STACK
1075 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1076 NULL, free_vm_stack_cache);
1081 lockdep_init_task(&init_task);
1085 int __weak arch_dup_task_struct(struct task_struct *dst,
1086 struct task_struct *src)
1092 void set_task_stack_end_magic(struct task_struct *tsk)
1094 unsigned long *stackend;
1096 stackend = end_of_stack(tsk);
1097 *stackend = STACK_END_MAGIC; /* for overflow detection */
1100 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1102 struct task_struct *tsk;
1105 if (node == NUMA_NO_NODE)
1106 node = tsk_fork_get_node(orig);
1107 tsk = alloc_task_struct_node(node);
1111 err = arch_dup_task_struct(tsk, orig);
1115 err = alloc_thread_stack_node(tsk, node);
1119 #ifdef CONFIG_THREAD_INFO_IN_TASK
1120 refcount_set(&tsk->stack_refcount, 1);
1122 account_kernel_stack(tsk, 1);
1124 err = scs_prepare(tsk, node);
1128 #ifdef CONFIG_SECCOMP
1130 * We must handle setting up seccomp filters once we're under
1131 * the sighand lock in case orig has changed between now and
1132 * then. Until then, filter must be NULL to avoid messing up
1133 * the usage counts on the error path calling free_task.
1135 tsk->seccomp.filter = NULL;
1138 setup_thread_stack(tsk, orig);
1139 clear_user_return_notifier(tsk);
1140 clear_tsk_need_resched(tsk);
1141 set_task_stack_end_magic(tsk);
1142 clear_syscall_work_syscall_user_dispatch(tsk);
1144 #ifdef CONFIG_STACKPROTECTOR
1145 tsk->stack_canary = get_random_canary();
1147 if (orig->cpus_ptr == &orig->cpus_mask)
1148 tsk->cpus_ptr = &tsk->cpus_mask;
1149 dup_user_cpus_ptr(tsk, orig, node);
1152 * One for the user space visible state that goes away when reaped.
1153 * One for the scheduler.
1155 refcount_set(&tsk->rcu_users, 2);
1156 /* One for the rcu users */
1157 refcount_set(&tsk->usage, 1);
1158 #ifdef CONFIG_BLK_DEV_IO_TRACE
1159 tsk->btrace_seq = 0;
1161 tsk->splice_pipe = NULL;
1162 tsk->task_frag.page = NULL;
1163 tsk->wake_q.next = NULL;
1164 tsk->worker_private = NULL;
1166 kcov_task_init(tsk);
1167 kmsan_task_create(tsk);
1168 kmap_local_fork(tsk);
1170 #ifdef CONFIG_FAULT_INJECTION
1174 #ifdef CONFIG_BLK_CGROUP
1175 tsk->throttle_disk = NULL;
1176 tsk->use_memdelay = 0;
1179 #ifdef CONFIG_IOMMU_SVA
1180 tsk->pasid_activated = 0;
1184 tsk->active_memcg = NULL;
1187 #ifdef CONFIG_CPU_SUP_INTEL
1188 tsk->reported_split_lock = 0;
1191 #ifdef CONFIG_SCHED_MM_CID
1193 tsk->last_mm_cid = -1;
1194 tsk->mm_cid_active = 0;
1195 tsk->migrate_from_cpu = -1;
1200 exit_task_stack_account(tsk);
1201 free_thread_stack(tsk);
1203 free_task_struct(tsk);
1207 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1209 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1211 static int __init coredump_filter_setup(char *s)
1213 default_dump_filter =
1214 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1215 MMF_DUMP_FILTER_MASK;
1219 __setup("coredump_filter=", coredump_filter_setup);
1221 #include <linux/init_task.h>
1223 static void mm_init_aio(struct mm_struct *mm)
1226 spin_lock_init(&mm->ioctx_lock);
1227 mm->ioctx_table = NULL;
1231 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1232 struct task_struct *p)
1236 WRITE_ONCE(mm->owner, NULL);
1240 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1247 static void mm_init_uprobes_state(struct mm_struct *mm)
1249 #ifdef CONFIG_UPROBES
1250 mm->uprobes_state.xol_area = NULL;
1254 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1255 struct user_namespace *user_ns)
1259 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1260 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1261 atomic_set(&mm->mm_users, 1);
1262 atomic_set(&mm->mm_count, 1);
1263 seqcount_init(&mm->write_protect_seq);
1265 INIT_LIST_HEAD(&mm->mmlist);
1266 #ifdef CONFIG_PER_VMA_LOCK
1267 mm->mm_lock_seq = 0;
1269 mm_pgtables_bytes_init(mm);
1272 atomic64_set(&mm->pinned_vm, 0);
1273 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1274 spin_lock_init(&mm->page_table_lock);
1275 spin_lock_init(&mm->arg_lock);
1276 mm_init_cpumask(mm);
1278 mm_init_owner(mm, p);
1280 RCU_INIT_POINTER(mm->exe_file, NULL);
1281 mmu_notifier_subscriptions_init(mm);
1282 init_tlb_flush_pending(mm);
1283 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1284 mm->pmd_huge_pte = NULL;
1286 mm_init_uprobes_state(mm);
1287 hugetlb_count_init(mm);
1290 mm->flags = current->mm->flags & MMF_INIT_MASK;
1291 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1293 mm->flags = default_dump_filter;
1297 if (mm_alloc_pgd(mm))
1300 if (init_new_context(p, mm))
1301 goto fail_nocontext;
1303 if (mm_alloc_cid(mm))
1306 for (i = 0; i < NR_MM_COUNTERS; i++)
1307 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1310 mm->user_ns = get_user_ns(user_ns);
1311 lru_gen_init_mm(mm);
1316 percpu_counter_destroy(&mm->rss_stat[--i]);
1319 destroy_context(mm);
1328 * Allocate and initialize an mm_struct.
1330 struct mm_struct *mm_alloc(void)
1332 struct mm_struct *mm;
1338 memset(mm, 0, sizeof(*mm));
1339 return mm_init(mm, current, current_user_ns());
1342 static inline void __mmput(struct mm_struct *mm)
1344 VM_BUG_ON(atomic_read(&mm->mm_users));
1346 uprobe_clear_state(mm);
1349 khugepaged_exit(mm); /* must run before exit_mmap */
1351 mm_put_huge_zero_page(mm);
1352 set_mm_exe_file(mm, NULL);
1353 if (!list_empty(&mm->mmlist)) {
1354 spin_lock(&mmlist_lock);
1355 list_del(&mm->mmlist);
1356 spin_unlock(&mmlist_lock);
1359 module_put(mm->binfmt->module);
1365 * Decrement the use count and release all resources for an mm.
1367 void mmput(struct mm_struct *mm)
1371 if (atomic_dec_and_test(&mm->mm_users))
1374 EXPORT_SYMBOL_GPL(mmput);
1377 static void mmput_async_fn(struct work_struct *work)
1379 struct mm_struct *mm = container_of(work, struct mm_struct,
1385 void mmput_async(struct mm_struct *mm)
1387 if (atomic_dec_and_test(&mm->mm_users)) {
1388 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1389 schedule_work(&mm->async_put_work);
1392 EXPORT_SYMBOL_GPL(mmput_async);
1396 * set_mm_exe_file - change a reference to the mm's executable file
1398 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1400 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1401 * invocations: in mmput() nobody alive left, in execve task is single
1404 * Can only fail if new_exe_file != NULL.
1406 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1408 struct file *old_exe_file;
1411 * It is safe to dereference the exe_file without RCU as
1412 * this function is only called if nobody else can access
1413 * this mm -- see comment above for justification.
1415 old_exe_file = rcu_dereference_raw(mm->exe_file);
1419 * We expect the caller (i.e., sys_execve) to already denied
1420 * write access, so this is unlikely to fail.
1422 if (unlikely(deny_write_access(new_exe_file)))
1424 get_file(new_exe_file);
1426 rcu_assign_pointer(mm->exe_file, new_exe_file);
1428 allow_write_access(old_exe_file);
1435 * replace_mm_exe_file - replace a reference to the mm's executable file
1437 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1438 * dealing with concurrent invocation and without grabbing the mmap lock in
1441 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1443 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1445 struct vm_area_struct *vma;
1446 struct file *old_exe_file;
1449 /* Forbid mm->exe_file change if old file still mapped. */
1450 old_exe_file = get_mm_exe_file(mm);
1452 VMA_ITERATOR(vmi, mm, 0);
1454 for_each_vma(vmi, vma) {
1457 if (path_equal(&vma->vm_file->f_path,
1458 &old_exe_file->f_path)) {
1463 mmap_read_unlock(mm);
1469 /* set the new file, lockless */
1470 ret = deny_write_access(new_exe_file);
1473 get_file(new_exe_file);
1475 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1478 * Don't race with dup_mmap() getting the file and disallowing
1479 * write access while someone might open the file writable.
1482 allow_write_access(old_exe_file);
1484 mmap_read_unlock(mm);
1490 * get_mm_exe_file - acquire a reference to the mm's executable file
1492 * Returns %NULL if mm has no associated executable file.
1493 * User must release file via fput().
1495 struct file *get_mm_exe_file(struct mm_struct *mm)
1497 struct file *exe_file;
1500 exe_file = rcu_dereference(mm->exe_file);
1501 if (exe_file && !get_file_rcu(exe_file))
1508 * get_task_exe_file - acquire a reference to the task's executable file
1510 * Returns %NULL if task's mm (if any) has no associated executable file or
1511 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1512 * User must release file via fput().
1514 struct file *get_task_exe_file(struct task_struct *task)
1516 struct file *exe_file = NULL;
1517 struct mm_struct *mm;
1522 if (!(task->flags & PF_KTHREAD))
1523 exe_file = get_mm_exe_file(mm);
1530 * get_task_mm - acquire a reference to the task's mm
1532 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1533 * this kernel workthread has transiently adopted a user mm with use_mm,
1534 * to do its AIO) is not set and if so returns a reference to it, after
1535 * bumping up the use count. User must release the mm via mmput()
1536 * after use. Typically used by /proc and ptrace.
1538 struct mm_struct *get_task_mm(struct task_struct *task)
1540 struct mm_struct *mm;
1545 if (task->flags & PF_KTHREAD)
1553 EXPORT_SYMBOL_GPL(get_task_mm);
1555 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1557 struct mm_struct *mm;
1560 err = down_read_killable(&task->signal->exec_update_lock);
1562 return ERR_PTR(err);
1564 mm = get_task_mm(task);
1565 if (mm && mm != current->mm &&
1566 !ptrace_may_access(task, mode)) {
1568 mm = ERR_PTR(-EACCES);
1570 up_read(&task->signal->exec_update_lock);
1575 static void complete_vfork_done(struct task_struct *tsk)
1577 struct completion *vfork;
1580 vfork = tsk->vfork_done;
1581 if (likely(vfork)) {
1582 tsk->vfork_done = NULL;
1588 static int wait_for_vfork_done(struct task_struct *child,
1589 struct completion *vfork)
1591 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1594 cgroup_enter_frozen();
1595 killed = wait_for_completion_state(vfork, state);
1596 cgroup_leave_frozen(false);
1600 child->vfork_done = NULL;
1604 put_task_struct(child);
1608 /* Please note the differences between mmput and mm_release.
1609 * mmput is called whenever we stop holding onto a mm_struct,
1610 * error success whatever.
1612 * mm_release is called after a mm_struct has been removed
1613 * from the current process.
1615 * This difference is important for error handling, when we
1616 * only half set up a mm_struct for a new process and need to restore
1617 * the old one. Because we mmput the new mm_struct before
1618 * restoring the old one. . .
1619 * Eric Biederman 10 January 1998
1621 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1623 uprobe_free_utask(tsk);
1625 /* Get rid of any cached register state */
1626 deactivate_mm(tsk, mm);
1629 * Signal userspace if we're not exiting with a core dump
1630 * because we want to leave the value intact for debugging
1633 if (tsk->clear_child_tid) {
1634 if (atomic_read(&mm->mm_users) > 1) {
1636 * We don't check the error code - if userspace has
1637 * not set up a proper pointer then tough luck.
1639 put_user(0, tsk->clear_child_tid);
1640 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1641 1, NULL, NULL, 0, 0);
1643 tsk->clear_child_tid = NULL;
1647 * All done, finally we can wake up parent and return this mm to him.
1648 * Also kthread_stop() uses this completion for synchronization.
1650 if (tsk->vfork_done)
1651 complete_vfork_done(tsk);
1654 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1656 futex_exit_release(tsk);
1657 mm_release(tsk, mm);
1660 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1662 futex_exec_release(tsk);
1663 mm_release(tsk, mm);
1667 * dup_mm() - duplicates an existing mm structure
1668 * @tsk: the task_struct with which the new mm will be associated.
1669 * @oldmm: the mm to duplicate.
1671 * Allocates a new mm structure and duplicates the provided @oldmm structure
1674 * Return: the duplicated mm or NULL on failure.
1676 static struct mm_struct *dup_mm(struct task_struct *tsk,
1677 struct mm_struct *oldmm)
1679 struct mm_struct *mm;
1686 memcpy(mm, oldmm, sizeof(*mm));
1688 if (!mm_init(mm, tsk, mm->user_ns))
1691 err = dup_mmap(mm, oldmm);
1695 mm->hiwater_rss = get_mm_rss(mm);
1696 mm->hiwater_vm = mm->total_vm;
1698 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1704 /* don't put binfmt in mmput, we haven't got module yet */
1706 mm_init_owner(mm, NULL);
1713 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1715 struct mm_struct *mm, *oldmm;
1717 tsk->min_flt = tsk->maj_flt = 0;
1718 tsk->nvcsw = tsk->nivcsw = 0;
1719 #ifdef CONFIG_DETECT_HUNG_TASK
1720 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1721 tsk->last_switch_time = 0;
1725 tsk->active_mm = NULL;
1728 * Are we cloning a kernel thread?
1730 * We need to steal a active VM for that..
1732 oldmm = current->mm;
1736 if (clone_flags & CLONE_VM) {
1740 mm = dup_mm(tsk, current->mm);
1746 tsk->active_mm = mm;
1747 sched_mm_cid_fork(tsk);
1751 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1753 struct fs_struct *fs = current->fs;
1754 if (clone_flags & CLONE_FS) {
1755 /* tsk->fs is already what we want */
1756 spin_lock(&fs->lock);
1758 spin_unlock(&fs->lock);
1762 spin_unlock(&fs->lock);
1765 tsk->fs = copy_fs_struct(fs);
1771 static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1774 struct files_struct *oldf, *newf;
1778 * A background process may not have any files ...
1780 oldf = current->files;
1789 if (clone_flags & CLONE_FILES) {
1790 atomic_inc(&oldf->count);
1794 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1804 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1806 struct sighand_struct *sig;
1808 if (clone_flags & CLONE_SIGHAND) {
1809 refcount_inc(¤t->sighand->count);
1812 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1813 RCU_INIT_POINTER(tsk->sighand, sig);
1817 refcount_set(&sig->count, 1);
1818 spin_lock_irq(¤t->sighand->siglock);
1819 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1820 spin_unlock_irq(¤t->sighand->siglock);
1822 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1823 if (clone_flags & CLONE_CLEAR_SIGHAND)
1824 flush_signal_handlers(tsk, 0);
1829 void __cleanup_sighand(struct sighand_struct *sighand)
1831 if (refcount_dec_and_test(&sighand->count)) {
1832 signalfd_cleanup(sighand);
1834 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1835 * without an RCU grace period, see __lock_task_sighand().
1837 kmem_cache_free(sighand_cachep, sighand);
1842 * Initialize POSIX timer handling for a thread group.
1844 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1846 struct posix_cputimers *pct = &sig->posix_cputimers;
1847 unsigned long cpu_limit;
1849 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1850 posix_cputimers_group_init(pct, cpu_limit);
1853 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1855 struct signal_struct *sig;
1857 if (clone_flags & CLONE_THREAD)
1860 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1865 sig->nr_threads = 1;
1866 sig->quick_threads = 1;
1867 atomic_set(&sig->live, 1);
1868 refcount_set(&sig->sigcnt, 1);
1870 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1871 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1872 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1874 init_waitqueue_head(&sig->wait_chldexit);
1875 sig->curr_target = tsk;
1876 init_sigpending(&sig->shared_pending);
1877 INIT_HLIST_HEAD(&sig->multiprocess);
1878 seqlock_init(&sig->stats_lock);
1879 prev_cputime_init(&sig->prev_cputime);
1881 #ifdef CONFIG_POSIX_TIMERS
1882 INIT_LIST_HEAD(&sig->posix_timers);
1883 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1884 sig->real_timer.function = it_real_fn;
1887 task_lock(current->group_leader);
1888 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1889 task_unlock(current->group_leader);
1891 posix_cpu_timers_init_group(sig);
1893 tty_audit_fork(sig);
1894 sched_autogroup_fork(sig);
1896 sig->oom_score_adj = current->signal->oom_score_adj;
1897 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1899 mutex_init(&sig->cred_guard_mutex);
1900 init_rwsem(&sig->exec_update_lock);
1905 static void copy_seccomp(struct task_struct *p)
1907 #ifdef CONFIG_SECCOMP
1909 * Must be called with sighand->lock held, which is common to
1910 * all threads in the group. Holding cred_guard_mutex is not
1911 * needed because this new task is not yet running and cannot
1914 assert_spin_locked(¤t->sighand->siglock);
1916 /* Ref-count the new filter user, and assign it. */
1917 get_seccomp_filter(current);
1918 p->seccomp = current->seccomp;
1921 * Explicitly enable no_new_privs here in case it got set
1922 * between the task_struct being duplicated and holding the
1923 * sighand lock. The seccomp state and nnp must be in sync.
1925 if (task_no_new_privs(current))
1926 task_set_no_new_privs(p);
1929 * If the parent gained a seccomp mode after copying thread
1930 * flags and between before we held the sighand lock, we have
1931 * to manually enable the seccomp thread flag here.
1933 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1934 set_task_syscall_work(p, SECCOMP);
1938 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1940 current->clear_child_tid = tidptr;
1942 return task_pid_vnr(current);
1945 static void rt_mutex_init_task(struct task_struct *p)
1947 raw_spin_lock_init(&p->pi_lock);
1948 #ifdef CONFIG_RT_MUTEXES
1949 p->pi_waiters = RB_ROOT_CACHED;
1950 p->pi_top_task = NULL;
1951 p->pi_blocked_on = NULL;
1955 static inline void init_task_pid_links(struct task_struct *task)
1959 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1960 INIT_HLIST_NODE(&task->pid_links[type]);
1964 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1966 if (type == PIDTYPE_PID)
1967 task->thread_pid = pid;
1969 task->signal->pids[type] = pid;
1972 static inline void rcu_copy_process(struct task_struct *p)
1974 #ifdef CONFIG_PREEMPT_RCU
1975 p->rcu_read_lock_nesting = 0;
1976 p->rcu_read_unlock_special.s = 0;
1977 p->rcu_blocked_node = NULL;
1978 INIT_LIST_HEAD(&p->rcu_node_entry);
1979 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1980 #ifdef CONFIG_TASKS_RCU
1981 p->rcu_tasks_holdout = false;
1982 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1983 p->rcu_tasks_idle_cpu = -1;
1984 #endif /* #ifdef CONFIG_TASKS_RCU */
1985 #ifdef CONFIG_TASKS_TRACE_RCU
1986 p->trc_reader_nesting = 0;
1987 p->trc_reader_special.s = 0;
1988 INIT_LIST_HEAD(&p->trc_holdout_list);
1989 INIT_LIST_HEAD(&p->trc_blkd_node);
1990 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1993 struct pid *pidfd_pid(const struct file *file)
1995 if (file->f_op == &pidfd_fops)
1996 return file->private_data;
1998 return ERR_PTR(-EBADF);
2001 static int pidfd_release(struct inode *inode, struct file *file)
2003 struct pid *pid = file->private_data;
2005 file->private_data = NULL;
2010 #ifdef CONFIG_PROC_FS
2012 * pidfd_show_fdinfo - print information about a pidfd
2013 * @m: proc fdinfo file
2014 * @f: file referencing a pidfd
2017 * This function will print the pid that a given pidfd refers to in the
2018 * pid namespace of the procfs instance.
2019 * If the pid namespace of the process is not a descendant of the pid
2020 * namespace of the procfs instance 0 will be shown as its pid. This is
2021 * similar to calling getppid() on a process whose parent is outside of
2022 * its pid namespace.
2025 * If pid namespaces are supported then this function will also print
2026 * the pid of a given pidfd refers to for all descendant pid namespaces
2027 * starting from the current pid namespace of the instance, i.e. the
2028 * Pid field and the first entry in the NSpid field will be identical.
2029 * If the pid namespace of the process is not a descendant of the pid
2030 * namespace of the procfs instance 0 will be shown as its first NSpid
2031 * entry and no others will be shown.
2032 * Note that this differs from the Pid and NSpid fields in
2033 * /proc/<pid>/status where Pid and NSpid are always shown relative to
2034 * the pid namespace of the procfs instance. The difference becomes
2035 * obvious when sending around a pidfd between pid namespaces from a
2036 * different branch of the tree, i.e. where no ancestral relation is
2037 * present between the pid namespaces:
2038 * - create two new pid namespaces ns1 and ns2 in the initial pid
2039 * namespace (also take care to create new mount namespaces in the
2040 * new pid namespace and mount procfs)
2041 * - create a process with a pidfd in ns1
2042 * - send pidfd from ns1 to ns2
2043 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
2044 * have exactly one entry, which is 0
2046 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
2048 struct pid *pid = f->private_data;
2049 struct pid_namespace *ns;
2052 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
2053 ns = proc_pid_ns(file_inode(m->file)->i_sb);
2054 nr = pid_nr_ns(pid, ns);
2057 seq_put_decimal_ll(m, "Pid:\t", nr);
2059 #ifdef CONFIG_PID_NS
2060 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
2064 /* If nr is non-zero it means that 'pid' is valid and that
2065 * ns, i.e. the pid namespace associated with the procfs
2066 * instance, is in the pid namespace hierarchy of pid.
2067 * Start at one below the already printed level.
2069 for (i = ns->level + 1; i <= pid->level; i++)
2070 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2078 * Poll support for process exit notification.
2080 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2082 struct pid *pid = file->private_data;
2083 __poll_t poll_flags = 0;
2085 poll_wait(file, &pid->wait_pidfd, pts);
2088 * Inform pollers only when the whole thread group exits.
2089 * If the thread group leader exits before all other threads in the
2090 * group, then poll(2) should block, similar to the wait(2) family.
2092 if (thread_group_exited(pid))
2093 poll_flags = EPOLLIN | EPOLLRDNORM;
2098 const struct file_operations pidfd_fops = {
2099 .release = pidfd_release,
2101 #ifdef CONFIG_PROC_FS
2102 .show_fdinfo = pidfd_show_fdinfo,
2107 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2108 * @pid: the struct pid for which to create a pidfd
2109 * @flags: flags of the new @pidfd
2110 * @pidfd: the pidfd to return
2112 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2113 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2115 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2116 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2117 * pidfd file are prepared.
2119 * If this function returns successfully the caller is responsible to either
2120 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2121 * order to install the pidfd into its file descriptor table or they must use
2122 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2125 * This function is useful when a pidfd must already be reserved but there
2126 * might still be points of failure afterwards and the caller wants to ensure
2127 * that no pidfd is leaked into its file descriptor table.
2129 * Return: On success, a reserved pidfd is returned from the function and a new
2130 * pidfd file is returned in the last argument to the function. On
2131 * error, a negative error code is returned from the function and the
2132 * last argument remains unchanged.
2134 static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2137 struct file *pidfd_file;
2139 if (flags & ~(O_NONBLOCK | O_RDWR | O_CLOEXEC))
2142 pidfd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2146 pidfd_file = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2147 flags | O_RDWR | O_CLOEXEC);
2148 if (IS_ERR(pidfd_file)) {
2149 put_unused_fd(pidfd);
2150 return PTR_ERR(pidfd_file);
2152 get_pid(pid); /* held by pidfd_file now */
2158 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2159 * @pid: the struct pid for which to create a pidfd
2160 * @flags: flags of the new @pidfd
2161 * @pidfd: the pidfd to return
2163 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2164 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2166 * The helper verifies that @pid is used as a thread group leader.
2168 * If this function returns successfully the caller is responsible to either
2169 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2170 * order to install the pidfd into its file descriptor table or they must use
2171 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2174 * This function is useful when a pidfd must already be reserved but there
2175 * might still be points of failure afterwards and the caller wants to ensure
2176 * that no pidfd is leaked into its file descriptor table.
2178 * Return: On success, a reserved pidfd is returned from the function and a new
2179 * pidfd file is returned in the last argument to the function. On
2180 * error, a negative error code is returned from the function and the
2181 * last argument remains unchanged.
2183 int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2185 if (!pid || !pid_has_task(pid, PIDTYPE_TGID))
2188 return __pidfd_prepare(pid, flags, ret);
2191 static void __delayed_free_task(struct rcu_head *rhp)
2193 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2198 static __always_inline void delayed_free_task(struct task_struct *tsk)
2200 if (IS_ENABLED(CONFIG_MEMCG))
2201 call_rcu(&tsk->rcu, __delayed_free_task);
2206 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2208 /* Skip if kernel thread */
2212 /* Skip if spawning a thread or using vfork */
2213 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2216 /* We need to synchronize with __set_oom_adj */
2217 mutex_lock(&oom_adj_mutex);
2218 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2219 /* Update the values in case they were changed after copy_signal */
2220 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2221 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2222 mutex_unlock(&oom_adj_mutex);
2226 static void rv_task_fork(struct task_struct *p)
2230 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2231 p->rv[i].da_mon.monitoring = false;
2234 #define rv_task_fork(p) do {} while (0)
2238 * This creates a new process as a copy of the old one,
2239 * but does not actually start it yet.
2241 * It copies the registers, and all the appropriate
2242 * parts of the process environment (as per the clone
2243 * flags). The actual kick-off is left to the caller.
2245 __latent_entropy struct task_struct *copy_process(
2249 struct kernel_clone_args *args)
2251 int pidfd = -1, retval;
2252 struct task_struct *p;
2253 struct multiprocess_signals delayed;
2254 struct file *pidfile = NULL;
2255 const u64 clone_flags = args->flags;
2256 struct nsproxy *nsp = current->nsproxy;
2259 * Don't allow sharing the root directory with processes in a different
2262 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2263 return ERR_PTR(-EINVAL);
2265 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2266 return ERR_PTR(-EINVAL);
2269 * Thread groups must share signals as well, and detached threads
2270 * can only be started up within the thread group.
2272 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2273 return ERR_PTR(-EINVAL);
2276 * Shared signal handlers imply shared VM. By way of the above,
2277 * thread groups also imply shared VM. Blocking this case allows
2278 * for various simplifications in other code.
2280 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2281 return ERR_PTR(-EINVAL);
2284 * Siblings of global init remain as zombies on exit since they are
2285 * not reaped by their parent (swapper). To solve this and to avoid
2286 * multi-rooted process trees, prevent global and container-inits
2287 * from creating siblings.
2289 if ((clone_flags & CLONE_PARENT) &&
2290 current->signal->flags & SIGNAL_UNKILLABLE)
2291 return ERR_PTR(-EINVAL);
2294 * If the new process will be in a different pid or user namespace
2295 * do not allow it to share a thread group with the forking task.
2297 if (clone_flags & CLONE_THREAD) {
2298 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2299 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2300 return ERR_PTR(-EINVAL);
2303 if (clone_flags & CLONE_PIDFD) {
2305 * - CLONE_DETACHED is blocked so that we can potentially
2306 * reuse it later for CLONE_PIDFD.
2307 * - CLONE_THREAD is blocked until someone really needs it.
2309 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2310 return ERR_PTR(-EINVAL);
2314 * Force any signals received before this point to be delivered
2315 * before the fork happens. Collect up signals sent to multiple
2316 * processes that happen during the fork and delay them so that
2317 * they appear to happen after the fork.
2319 sigemptyset(&delayed.signal);
2320 INIT_HLIST_NODE(&delayed.node);
2322 spin_lock_irq(¤t->sighand->siglock);
2323 if (!(clone_flags & CLONE_THREAD))
2324 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2325 recalc_sigpending();
2326 spin_unlock_irq(¤t->sighand->siglock);
2327 retval = -ERESTARTNOINTR;
2328 if (task_sigpending(current))
2332 p = dup_task_struct(current, node);
2335 p->flags &= ~PF_KTHREAD;
2337 p->flags |= PF_KTHREAD;
2338 if (args->user_worker)
2339 p->flags |= PF_USER_WORKER;
2340 if (args->io_thread) {
2342 * Mark us an IO worker, and block any signal that isn't
2345 p->flags |= PF_IO_WORKER;
2346 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2350 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2352 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2354 * Clear TID on mm_release()?
2356 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2358 ftrace_graph_init_task(p);
2360 rt_mutex_init_task(p);
2362 lockdep_assert_irqs_enabled();
2363 #ifdef CONFIG_PROVE_LOCKING
2364 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2366 retval = copy_creds(p, clone_flags);
2371 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2372 if (p->real_cred->user != INIT_USER &&
2373 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2374 goto bad_fork_cleanup_count;
2376 current->flags &= ~PF_NPROC_EXCEEDED;
2379 * If multiple threads are within copy_process(), then this check
2380 * triggers too late. This doesn't hurt, the check is only there
2381 * to stop root fork bombs.
2384 if (data_race(nr_threads >= max_threads))
2385 goto bad_fork_cleanup_count;
2387 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2388 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2389 p->flags |= PF_FORKNOEXEC;
2390 INIT_LIST_HEAD(&p->children);
2391 INIT_LIST_HEAD(&p->sibling);
2392 rcu_copy_process(p);
2393 p->vfork_done = NULL;
2394 spin_lock_init(&p->alloc_lock);
2396 init_sigpending(&p->pending);
2398 p->utime = p->stime = p->gtime = 0;
2399 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2400 p->utimescaled = p->stimescaled = 0;
2402 prev_cputime_init(&p->prev_cputime);
2404 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2405 seqcount_init(&p->vtime.seqcount);
2406 p->vtime.starttime = 0;
2407 p->vtime.state = VTIME_INACTIVE;
2410 #ifdef CONFIG_IO_URING
2414 #if defined(SPLIT_RSS_COUNTING)
2415 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2418 p->default_timer_slack_ns = current->timer_slack_ns;
2424 task_io_accounting_init(&p->ioac);
2425 acct_clear_integrals(p);
2427 posix_cputimers_init(&p->posix_cputimers);
2429 p->io_context = NULL;
2430 audit_set_context(p, NULL);
2432 if (args->kthread) {
2433 if (!set_kthread_struct(p))
2434 goto bad_fork_cleanup_delayacct;
2437 p->mempolicy = mpol_dup(p->mempolicy);
2438 if (IS_ERR(p->mempolicy)) {
2439 retval = PTR_ERR(p->mempolicy);
2440 p->mempolicy = NULL;
2441 goto bad_fork_cleanup_delayacct;
2444 #ifdef CONFIG_CPUSETS
2445 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2446 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2447 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2449 #ifdef CONFIG_TRACE_IRQFLAGS
2450 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2451 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2452 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2453 p->softirqs_enabled = 1;
2454 p->softirq_context = 0;
2457 p->pagefault_disabled = 0;
2459 #ifdef CONFIG_LOCKDEP
2460 lockdep_init_task(p);
2463 #ifdef CONFIG_DEBUG_MUTEXES
2464 p->blocked_on = NULL; /* not blocked yet */
2466 #ifdef CONFIG_BCACHE
2467 p->sequential_io = 0;
2468 p->sequential_io_avg = 0;
2470 #ifdef CONFIG_BPF_SYSCALL
2471 RCU_INIT_POINTER(p->bpf_storage, NULL);
2475 /* Perform scheduler related setup. Assign this task to a CPU. */
2476 retval = sched_fork(clone_flags, p);
2478 goto bad_fork_cleanup_policy;
2480 retval = perf_event_init_task(p, clone_flags);
2482 goto bad_fork_cleanup_policy;
2483 retval = audit_alloc(p);
2485 goto bad_fork_cleanup_perf;
2486 /* copy all the process information */
2488 retval = security_task_alloc(p, clone_flags);
2490 goto bad_fork_cleanup_audit;
2491 retval = copy_semundo(clone_flags, p);
2493 goto bad_fork_cleanup_security;
2494 retval = copy_files(clone_flags, p, args->no_files);
2496 goto bad_fork_cleanup_semundo;
2497 retval = copy_fs(clone_flags, p);
2499 goto bad_fork_cleanup_files;
2500 retval = copy_sighand(clone_flags, p);
2502 goto bad_fork_cleanup_fs;
2503 retval = copy_signal(clone_flags, p);
2505 goto bad_fork_cleanup_sighand;
2506 retval = copy_mm(clone_flags, p);
2508 goto bad_fork_cleanup_signal;
2509 retval = copy_namespaces(clone_flags, p);
2511 goto bad_fork_cleanup_mm;
2512 retval = copy_io(clone_flags, p);
2514 goto bad_fork_cleanup_namespaces;
2515 retval = copy_thread(p, args);
2517 goto bad_fork_cleanup_io;
2519 if (args->ignore_signals)
2522 stackleak_task_init(p);
2524 if (pid != &init_struct_pid) {
2525 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2526 args->set_tid_size);
2528 retval = PTR_ERR(pid);
2529 goto bad_fork_cleanup_thread;
2534 * This has to happen after we've potentially unshared the file
2535 * descriptor table (so that the pidfd doesn't leak into the child
2536 * if the fd table isn't shared).
2538 if (clone_flags & CLONE_PIDFD) {
2539 /* Note that no task has been attached to @pid yet. */
2540 retval = __pidfd_prepare(pid, O_RDWR | O_CLOEXEC, &pidfile);
2542 goto bad_fork_free_pid;
2545 retval = put_user(pidfd, args->pidfd);
2547 goto bad_fork_put_pidfd;
2556 * sigaltstack should be cleared when sharing the same VM
2558 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2562 * Syscall tracing and stepping should be turned off in the
2563 * child regardless of CLONE_PTRACE.
2565 user_disable_single_step(p);
2566 clear_task_syscall_work(p, SYSCALL_TRACE);
2567 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2568 clear_task_syscall_work(p, SYSCALL_EMU);
2570 clear_tsk_latency_tracing(p);
2572 /* ok, now we should be set up.. */
2573 p->pid = pid_nr(pid);
2574 if (clone_flags & CLONE_THREAD) {
2575 p->group_leader = current->group_leader;
2576 p->tgid = current->tgid;
2578 p->group_leader = p;
2583 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2584 p->dirty_paused_when = 0;
2586 p->pdeath_signal = 0;
2587 INIT_LIST_HEAD(&p->thread_group);
2588 p->task_works = NULL;
2589 clear_posix_cputimers_work(p);
2591 #ifdef CONFIG_KRETPROBES
2592 p->kretprobe_instances.first = NULL;
2594 #ifdef CONFIG_RETHOOK
2595 p->rethooks.first = NULL;
2599 * Ensure that the cgroup subsystem policies allow the new process to be
2600 * forked. It should be noted that the new process's css_set can be changed
2601 * between here and cgroup_post_fork() if an organisation operation is in
2604 retval = cgroup_can_fork(p, args);
2606 goto bad_fork_put_pidfd;
2609 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2610 * the new task on the correct runqueue. All this *before* the task
2613 * This isn't part of ->can_fork() because while the re-cloning is
2614 * cgroup specific, it unconditionally needs to place the task on a
2617 sched_cgroup_fork(p, args);
2620 * From this point on we must avoid any synchronous user-space
2621 * communication until we take the tasklist-lock. In particular, we do
2622 * not want user-space to be able to predict the process start-time by
2623 * stalling fork(2) after we recorded the start_time but before it is
2624 * visible to the system.
2627 p->start_time = ktime_get_ns();
2628 p->start_boottime = ktime_get_boottime_ns();
2631 * Make it visible to the rest of the system, but dont wake it up yet.
2632 * Need tasklist lock for parent etc handling!
2634 write_lock_irq(&tasklist_lock);
2636 /* CLONE_PARENT re-uses the old parent */
2637 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2638 p->real_parent = current->real_parent;
2639 p->parent_exec_id = current->parent_exec_id;
2640 if (clone_flags & CLONE_THREAD)
2641 p->exit_signal = -1;
2643 p->exit_signal = current->group_leader->exit_signal;
2645 p->real_parent = current;
2646 p->parent_exec_id = current->self_exec_id;
2647 p->exit_signal = args->exit_signal;
2650 klp_copy_process(p);
2654 spin_lock(¤t->sighand->siglock);
2658 rseq_fork(p, clone_flags);
2660 /* Don't start children in a dying pid namespace */
2661 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2663 goto bad_fork_cancel_cgroup;
2666 /* Let kill terminate clone/fork in the middle */
2667 if (fatal_signal_pending(current)) {
2669 goto bad_fork_cancel_cgroup;
2672 /* No more failure paths after this point. */
2675 * Copy seccomp details explicitly here, in case they were changed
2676 * before holding sighand lock.
2680 init_task_pid_links(p);
2681 if (likely(p->pid)) {
2682 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2684 init_task_pid(p, PIDTYPE_PID, pid);
2685 if (thread_group_leader(p)) {
2686 init_task_pid(p, PIDTYPE_TGID, pid);
2687 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2688 init_task_pid(p, PIDTYPE_SID, task_session(current));
2690 if (is_child_reaper(pid)) {
2691 ns_of_pid(pid)->child_reaper = p;
2692 p->signal->flags |= SIGNAL_UNKILLABLE;
2694 p->signal->shared_pending.signal = delayed.signal;
2695 p->signal->tty = tty_kref_get(current->signal->tty);
2697 * Inherit has_child_subreaper flag under the same
2698 * tasklist_lock with adding child to the process tree
2699 * for propagate_has_child_subreaper optimization.
2701 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2702 p->real_parent->signal->is_child_subreaper;
2703 list_add_tail(&p->sibling, &p->real_parent->children);
2704 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2705 attach_pid(p, PIDTYPE_TGID);
2706 attach_pid(p, PIDTYPE_PGID);
2707 attach_pid(p, PIDTYPE_SID);
2708 __this_cpu_inc(process_counts);
2710 current->signal->nr_threads++;
2711 current->signal->quick_threads++;
2712 atomic_inc(¤t->signal->live);
2713 refcount_inc(¤t->signal->sigcnt);
2714 task_join_group_stop(p);
2715 list_add_tail_rcu(&p->thread_group,
2716 &p->group_leader->thread_group);
2717 list_add_tail_rcu(&p->thread_node,
2718 &p->signal->thread_head);
2720 attach_pid(p, PIDTYPE_PID);
2724 hlist_del_init(&delayed.node);
2725 spin_unlock(¤t->sighand->siglock);
2726 syscall_tracepoint_update(p);
2727 write_unlock_irq(&tasklist_lock);
2730 fd_install(pidfd, pidfile);
2732 proc_fork_connector(p);
2734 cgroup_post_fork(p, args);
2737 trace_task_newtask(p, clone_flags);
2738 uprobe_copy_process(p, clone_flags);
2739 user_events_fork(p, clone_flags);
2741 copy_oom_score_adj(clone_flags, p);
2745 bad_fork_cancel_cgroup:
2747 spin_unlock(¤t->sighand->siglock);
2748 write_unlock_irq(&tasklist_lock);
2749 cgroup_cancel_fork(p, args);
2751 if (clone_flags & CLONE_PIDFD) {
2753 put_unused_fd(pidfd);
2756 if (pid != &init_struct_pid)
2758 bad_fork_cleanup_thread:
2760 bad_fork_cleanup_io:
2763 bad_fork_cleanup_namespaces:
2764 exit_task_namespaces(p);
2765 bad_fork_cleanup_mm:
2767 mm_clear_owner(p->mm, p);
2770 bad_fork_cleanup_signal:
2771 if (!(clone_flags & CLONE_THREAD))
2772 free_signal_struct(p->signal);
2773 bad_fork_cleanup_sighand:
2774 __cleanup_sighand(p->sighand);
2775 bad_fork_cleanup_fs:
2776 exit_fs(p); /* blocking */
2777 bad_fork_cleanup_files:
2778 exit_files(p); /* blocking */
2779 bad_fork_cleanup_semundo:
2781 bad_fork_cleanup_security:
2782 security_task_free(p);
2783 bad_fork_cleanup_audit:
2785 bad_fork_cleanup_perf:
2786 perf_event_free_task(p);
2787 bad_fork_cleanup_policy:
2788 lockdep_free_task(p);
2790 mpol_put(p->mempolicy);
2792 bad_fork_cleanup_delayacct:
2793 delayacct_tsk_free(p);
2794 bad_fork_cleanup_count:
2795 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2798 WRITE_ONCE(p->__state, TASK_DEAD);
2799 exit_task_stack_account(p);
2801 delayed_free_task(p);
2803 spin_lock_irq(¤t->sighand->siglock);
2804 hlist_del_init(&delayed.node);
2805 spin_unlock_irq(¤t->sighand->siglock);
2806 return ERR_PTR(retval);
2809 static inline void init_idle_pids(struct task_struct *idle)
2813 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2814 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2815 init_task_pid(idle, type, &init_struct_pid);
2819 static int idle_dummy(void *dummy)
2821 /* This function is never called */
2825 struct task_struct * __init fork_idle(int cpu)
2827 struct task_struct *task;
2828 struct kernel_clone_args args = {
2836 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2837 if (!IS_ERR(task)) {
2838 init_idle_pids(task);
2839 init_idle(task, cpu);
2846 * This is like kernel_clone(), but shaved down and tailored to just
2847 * creating io_uring workers. It returns a created task, or an error pointer.
2848 * The returned task is inactive, and the caller must fire it up through
2849 * wake_up_new_task(p). All signals are blocked in the created task.
2851 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2853 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2855 struct kernel_clone_args args = {
2856 .flags = ((lower_32_bits(flags) | CLONE_VM |
2857 CLONE_UNTRACED) & ~CSIGNAL),
2858 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2865 return copy_process(NULL, 0, node, &args);
2869 * Ok, this is the main fork-routine.
2871 * It copies the process, and if successful kick-starts
2872 * it and waits for it to finish using the VM if required.
2874 * args->exit_signal is expected to be checked for sanity by the caller.
2876 pid_t kernel_clone(struct kernel_clone_args *args)
2878 u64 clone_flags = args->flags;
2879 struct completion vfork;
2881 struct task_struct *p;
2886 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2887 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2888 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2889 * field in struct clone_args and it still doesn't make sense to have
2890 * them both point at the same memory location. Performing this check
2891 * here has the advantage that we don't need to have a separate helper
2892 * to check for legacy clone().
2894 if ((args->flags & CLONE_PIDFD) &&
2895 (args->flags & CLONE_PARENT_SETTID) &&
2896 (args->pidfd == args->parent_tid))
2900 * Determine whether and which event to report to ptracer. When
2901 * called from kernel_thread or CLONE_UNTRACED is explicitly
2902 * requested, no event is reported; otherwise, report if the event
2903 * for the type of forking is enabled.
2905 if (!(clone_flags & CLONE_UNTRACED)) {
2906 if (clone_flags & CLONE_VFORK)
2907 trace = PTRACE_EVENT_VFORK;
2908 else if (args->exit_signal != SIGCHLD)
2909 trace = PTRACE_EVENT_CLONE;
2911 trace = PTRACE_EVENT_FORK;
2913 if (likely(!ptrace_event_enabled(current, trace)))
2917 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2918 add_latent_entropy();
2924 * Do this prior waking up the new thread - the thread pointer
2925 * might get invalid after that point, if the thread exits quickly.
2927 trace_sched_process_fork(current, p);
2929 pid = get_task_pid(p, PIDTYPE_PID);
2932 if (clone_flags & CLONE_PARENT_SETTID)
2933 put_user(nr, args->parent_tid);
2935 if (clone_flags & CLONE_VFORK) {
2936 p->vfork_done = &vfork;
2937 init_completion(&vfork);
2941 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2942 /* lock the task to synchronize with memcg migration */
2944 lru_gen_add_mm(p->mm);
2948 wake_up_new_task(p);
2950 /* forking complete and child started to run, tell ptracer */
2951 if (unlikely(trace))
2952 ptrace_event_pid(trace, pid);
2954 if (clone_flags & CLONE_VFORK) {
2955 if (!wait_for_vfork_done(p, &vfork))
2956 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2964 * Create a kernel thread.
2966 pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2967 unsigned long flags)
2969 struct kernel_clone_args args = {
2970 .flags = ((lower_32_bits(flags) | CLONE_VM |
2971 CLONE_UNTRACED) & ~CSIGNAL),
2972 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2979 return kernel_clone(&args);
2983 * Create a user mode thread.
2985 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2987 struct kernel_clone_args args = {
2988 .flags = ((lower_32_bits(flags) | CLONE_VM |
2989 CLONE_UNTRACED) & ~CSIGNAL),
2990 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2995 return kernel_clone(&args);
2998 #ifdef __ARCH_WANT_SYS_FORK
2999 SYSCALL_DEFINE0(fork)
3002 struct kernel_clone_args args = {
3003 .exit_signal = SIGCHLD,
3006 return kernel_clone(&args);
3008 /* can not support in nommu mode */
3014 #ifdef __ARCH_WANT_SYS_VFORK
3015 SYSCALL_DEFINE0(vfork)
3017 struct kernel_clone_args args = {
3018 .flags = CLONE_VFORK | CLONE_VM,
3019 .exit_signal = SIGCHLD,
3022 return kernel_clone(&args);
3026 #ifdef __ARCH_WANT_SYS_CLONE
3027 #ifdef CONFIG_CLONE_BACKWARDS
3028 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3029 int __user *, parent_tidptr,
3031 int __user *, child_tidptr)
3032 #elif defined(CONFIG_CLONE_BACKWARDS2)
3033 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
3034 int __user *, parent_tidptr,
3035 int __user *, child_tidptr,
3037 #elif defined(CONFIG_CLONE_BACKWARDS3)
3038 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
3040 int __user *, parent_tidptr,
3041 int __user *, child_tidptr,
3044 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3045 int __user *, parent_tidptr,
3046 int __user *, child_tidptr,
3050 struct kernel_clone_args args = {
3051 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
3052 .pidfd = parent_tidptr,
3053 .child_tid = child_tidptr,
3054 .parent_tid = parent_tidptr,
3055 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
3060 return kernel_clone(&args);
3064 #ifdef __ARCH_WANT_SYS_CLONE3
3066 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
3067 struct clone_args __user *uargs,
3071 struct clone_args args;
3072 pid_t *kset_tid = kargs->set_tid;
3074 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
3075 CLONE_ARGS_SIZE_VER0);
3076 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
3077 CLONE_ARGS_SIZE_VER1);
3078 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
3079 CLONE_ARGS_SIZE_VER2);
3080 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
3082 if (unlikely(usize > PAGE_SIZE))
3084 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
3087 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
3091 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
3094 if (unlikely(!args.set_tid && args.set_tid_size > 0))
3097 if (unlikely(args.set_tid && args.set_tid_size == 0))
3101 * Verify that higher 32bits of exit_signal are unset and that
3102 * it is a valid signal
3104 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3105 !valid_signal(args.exit_signal)))
3108 if ((args.flags & CLONE_INTO_CGROUP) &&
3109 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3112 *kargs = (struct kernel_clone_args){
3113 .flags = args.flags,
3114 .pidfd = u64_to_user_ptr(args.pidfd),
3115 .child_tid = u64_to_user_ptr(args.child_tid),
3116 .parent_tid = u64_to_user_ptr(args.parent_tid),
3117 .exit_signal = args.exit_signal,
3118 .stack = args.stack,
3119 .stack_size = args.stack_size,
3121 .set_tid_size = args.set_tid_size,
3122 .cgroup = args.cgroup,
3126 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3127 (kargs->set_tid_size * sizeof(pid_t))))
3130 kargs->set_tid = kset_tid;
3136 * clone3_stack_valid - check and prepare stack
3137 * @kargs: kernel clone args
3139 * Verify that the stack arguments userspace gave us are sane.
3140 * In addition, set the stack direction for userspace since it's easy for us to
3143 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3145 if (kargs->stack == 0) {
3146 if (kargs->stack_size > 0)
3149 if (kargs->stack_size == 0)
3152 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3155 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
3156 kargs->stack += kargs->stack_size;
3163 static bool clone3_args_valid(struct kernel_clone_args *kargs)
3165 /* Verify that no unknown flags are passed along. */
3167 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3171 * - make the CLONE_DETACHED bit reusable for clone3
3172 * - make the CSIGNAL bits reusable for clone3
3174 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3177 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3178 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3181 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3185 if (!clone3_stack_valid(kargs))
3192 * clone3 - create a new process with specific properties
3193 * @uargs: argument structure
3194 * @size: size of @uargs
3196 * clone3() is the extensible successor to clone()/clone2().
3197 * It takes a struct as argument that is versioned by its size.
3199 * Return: On success, a positive PID for the child process.
3200 * On error, a negative errno number.
3202 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3206 struct kernel_clone_args kargs;
3207 pid_t set_tid[MAX_PID_NS_LEVEL];
3209 kargs.set_tid = set_tid;
3211 err = copy_clone_args_from_user(&kargs, uargs, size);
3215 if (!clone3_args_valid(&kargs))
3218 return kernel_clone(&kargs);
3222 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3224 struct task_struct *leader, *parent, *child;
3227 read_lock(&tasklist_lock);
3228 leader = top = top->group_leader;
3230 for_each_thread(leader, parent) {
3231 list_for_each_entry(child, &parent->children, sibling) {
3232 res = visitor(child, data);
3244 if (leader != top) {
3246 parent = child->real_parent;
3247 leader = parent->group_leader;
3251 read_unlock(&tasklist_lock);
3254 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3255 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3258 static void sighand_ctor(void *data)
3260 struct sighand_struct *sighand = data;
3262 spin_lock_init(&sighand->siglock);
3263 init_waitqueue_head(&sighand->signalfd_wqh);
3266 void __init mm_cache_init(void)
3268 unsigned int mm_size;
3271 * The mm_cpumask is located at the end of mm_struct, and is
3272 * dynamically sized based on the maximum CPU number this system
3273 * can have, taking hotplug into account (nr_cpu_ids).
3275 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3277 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3278 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3279 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3280 offsetof(struct mm_struct, saved_auxv),
3281 sizeof_field(struct mm_struct, saved_auxv),
3285 void __init proc_caches_init(void)
3287 sighand_cachep = kmem_cache_create("sighand_cache",
3288 sizeof(struct sighand_struct), 0,
3289 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3290 SLAB_ACCOUNT, sighand_ctor);
3291 signal_cachep = kmem_cache_create("signal_cache",
3292 sizeof(struct signal_struct), 0,
3293 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3295 files_cachep = kmem_cache_create("files_cache",
3296 sizeof(struct files_struct), 0,
3297 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3299 fs_cachep = kmem_cache_create("fs_cache",
3300 sizeof(struct fs_struct), 0,
3301 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3304 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3305 #ifdef CONFIG_PER_VMA_LOCK
3306 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3309 nsproxy_cache_init();
3313 * Check constraints on flags passed to the unshare system call.
3315 static int check_unshare_flags(unsigned long unshare_flags)
3317 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3318 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3319 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3320 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3324 * Not implemented, but pretend it works if there is nothing
3325 * to unshare. Note that unsharing the address space or the
3326 * signal handlers also need to unshare the signal queues (aka
3329 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3330 if (!thread_group_empty(current))
3333 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3334 if (refcount_read(¤t->sighand->count) > 1)
3337 if (unshare_flags & CLONE_VM) {
3338 if (!current_is_single_threaded())
3346 * Unshare the filesystem structure if it is being shared
3348 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3350 struct fs_struct *fs = current->fs;
3352 if (!(unshare_flags & CLONE_FS) || !fs)
3355 /* don't need lock here; in the worst case we'll do useless copy */
3359 *new_fsp = copy_fs_struct(fs);
3367 * Unshare file descriptor table if it is being shared
3369 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3370 struct files_struct **new_fdp)
3372 struct files_struct *fd = current->files;
3375 if ((unshare_flags & CLONE_FILES) &&
3376 (fd && atomic_read(&fd->count) > 1)) {
3377 *new_fdp = dup_fd(fd, max_fds, &error);
3386 * unshare allows a process to 'unshare' part of the process
3387 * context which was originally shared using clone. copy_*
3388 * functions used by kernel_clone() cannot be used here directly
3389 * because they modify an inactive task_struct that is being
3390 * constructed. Here we are modifying the current, active,
3393 int ksys_unshare(unsigned long unshare_flags)
3395 struct fs_struct *fs, *new_fs = NULL;
3396 struct files_struct *new_fd = NULL;
3397 struct cred *new_cred = NULL;
3398 struct nsproxy *new_nsproxy = NULL;
3403 * If unsharing a user namespace must also unshare the thread group
3404 * and unshare the filesystem root and working directories.
3406 if (unshare_flags & CLONE_NEWUSER)
3407 unshare_flags |= CLONE_THREAD | CLONE_FS;
3409 * If unsharing vm, must also unshare signal handlers.
3411 if (unshare_flags & CLONE_VM)
3412 unshare_flags |= CLONE_SIGHAND;
3414 * If unsharing a signal handlers, must also unshare the signal queues.
3416 if (unshare_flags & CLONE_SIGHAND)
3417 unshare_flags |= CLONE_THREAD;
3419 * If unsharing namespace, must also unshare filesystem information.
3421 if (unshare_flags & CLONE_NEWNS)
3422 unshare_flags |= CLONE_FS;
3424 err = check_unshare_flags(unshare_flags);
3426 goto bad_unshare_out;
3428 * CLONE_NEWIPC must also detach from the undolist: after switching
3429 * to a new ipc namespace, the semaphore arrays from the old
3430 * namespace are unreachable.
3432 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3434 err = unshare_fs(unshare_flags, &new_fs);
3436 goto bad_unshare_out;
3437 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3439 goto bad_unshare_cleanup_fs;
3440 err = unshare_userns(unshare_flags, &new_cred);
3442 goto bad_unshare_cleanup_fd;
3443 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3446 goto bad_unshare_cleanup_cred;
3449 err = set_cred_ucounts(new_cred);
3451 goto bad_unshare_cleanup_cred;
3454 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3457 * CLONE_SYSVSEM is equivalent to sys_exit().
3461 if (unshare_flags & CLONE_NEWIPC) {
3462 /* Orphan segments in old ns (see sem above). */
3464 shm_init_task(current);
3468 switch_task_namespaces(current, new_nsproxy);
3474 spin_lock(&fs->lock);
3475 current->fs = new_fs;
3480 spin_unlock(&fs->lock);
3484 swap(current->files, new_fd);
3486 task_unlock(current);
3489 /* Install the new user namespace */
3490 commit_creds(new_cred);
3495 perf_event_namespaces(current);
3497 bad_unshare_cleanup_cred:
3500 bad_unshare_cleanup_fd:
3502 put_files_struct(new_fd);
3504 bad_unshare_cleanup_fs:
3506 free_fs_struct(new_fs);
3512 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3514 return ksys_unshare(unshare_flags);
3518 * Helper to unshare the files of the current task.
3519 * We don't want to expose copy_files internals to
3520 * the exec layer of the kernel.
3523 int unshare_files(void)
3525 struct task_struct *task = current;
3526 struct files_struct *old, *copy = NULL;
3529 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3537 put_files_struct(old);
3541 int sysctl_max_threads(struct ctl_table *table, int write,
3542 void *buffer, size_t *lenp, loff_t *ppos)
3546 int threads = max_threads;
3548 int max = MAX_THREADS;
3555 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3559 max_threads = threads;