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/binfmts.h>
41 #include <linux/mman.h>
42 #include <linux/mmu_notifier.h>
45 #include <linux/vmacache.h>
46 #include <linux/nsproxy.h>
47 #include <linux/capability.h>
48 #include <linux/cpu.h>
49 #include <linux/cgroup.h>
50 #include <linux/security.h>
51 #include <linux/hugetlb.h>
52 #include <linux/seccomp.h>
53 #include <linux/swap.h>
54 #include <linux/syscalls.h>
55 #include <linux/jiffies.h>
56 #include <linux/futex.h>
57 #include <linux/compat.h>
58 #include <linux/kthread.h>
59 #include <linux/task_io_accounting_ops.h>
60 #include <linux/rcupdate.h>
61 #include <linux/ptrace.h>
62 #include <linux/mount.h>
63 #include <linux/audit.h>
64 #include <linux/memcontrol.h>
65 #include <linux/ftrace.h>
66 #include <linux/proc_fs.h>
67 #include <linux/profile.h>
68 #include <linux/rmap.h>
69 #include <linux/ksm.h>
70 #include <linux/acct.h>
71 #include <linux/userfaultfd_k.h>
72 #include <linux/tsacct_kern.h>
73 #include <linux/cn_proc.h>
74 #include <linux/freezer.h>
75 #include <linux/delayacct.h>
76 #include <linux/taskstats_kern.h>
77 #include <linux/random.h>
78 #include <linux/tty.h>
79 #include <linux/blkdev.h>
80 #include <linux/fs_struct.h>
81 #include <linux/magic.h>
82 #include <linux/perf_event.h>
83 #include <linux/posix-timers.h>
84 #include <linux/user-return-notifier.h>
85 #include <linux/oom.h>
86 #include <linux/khugepaged.h>
87 #include <linux/signalfd.h>
88 #include <linux/uprobes.h>
89 #include <linux/aio.h>
90 #include <linux/compiler.h>
91 #include <linux/sysctl.h>
92 #include <linux/kcov.h>
93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
98 #include <asm/pgtable.h>
99 #include <asm/pgalloc.h>
100 #include <linux/uaccess.h>
101 #include <asm/mmu_context.h>
102 #include <asm/cacheflush.h>
103 #include <asm/tlbflush.h>
105 #include <trace/events/sched.h>
107 #define CREATE_TRACE_POINTS
108 #include <trace/events/task.h>
111 * Minimum number of threads to boot the kernel
113 #define MIN_THREADS 20
116 * Maximum number of threads
118 #define MAX_THREADS FUTEX_TID_MASK
121 * Protected counters by write_lock_irq(&tasklist_lock)
123 unsigned long total_forks; /* Handle normal Linux uptimes. */
124 int nr_threads; /* The idle threads do not count.. */
126 static int max_threads; /* tunable limit on nr_threads */
128 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
130 static const char * const resident_page_types[] = {
131 NAMED_ARRAY_INDEX(MM_FILEPAGES),
132 NAMED_ARRAY_INDEX(MM_ANONPAGES),
133 NAMED_ARRAY_INDEX(MM_SWAPENTS),
134 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
137 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
139 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
141 #ifdef CONFIG_PROVE_RCU
142 int lockdep_tasklist_lock_is_held(void)
144 return lockdep_is_held(&tasklist_lock);
146 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
147 #endif /* #ifdef CONFIG_PROVE_RCU */
149 int nr_processes(void)
154 for_each_possible_cpu(cpu)
155 total += per_cpu(process_counts, cpu);
160 void __weak arch_release_task_struct(struct task_struct *tsk)
164 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
165 static struct kmem_cache *task_struct_cachep;
167 static inline struct task_struct *alloc_task_struct_node(int node)
169 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
172 static inline void free_task_struct(struct task_struct *tsk)
174 kmem_cache_free(task_struct_cachep, tsk);
178 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
181 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
182 * kmemcache based allocator.
184 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
186 #ifdef CONFIG_VMAP_STACK
188 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
189 * flush. Try to minimize the number of calls by caching stacks.
191 #define NR_CACHED_STACKS 2
192 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
194 static int free_vm_stack_cache(unsigned int cpu)
196 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
199 for (i = 0; i < NR_CACHED_STACKS; i++) {
200 struct vm_struct *vm_stack = cached_vm_stacks[i];
205 vfree(vm_stack->addr);
206 cached_vm_stacks[i] = NULL;
213 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
215 #ifdef CONFIG_VMAP_STACK
219 for (i = 0; i < NR_CACHED_STACKS; i++) {
222 s = this_cpu_xchg(cached_stacks[i], NULL);
227 /* Clear the KASAN shadow of the stack. */
228 kasan_unpoison_shadow(s->addr, THREAD_SIZE);
230 /* Clear stale pointers from reused stack. */
231 memset(s->addr, 0, THREAD_SIZE);
233 tsk->stack_vm_area = s;
234 tsk->stack = s->addr;
239 * Allocated stacks are cached and later reused by new threads,
240 * so memcg accounting is performed manually on assigning/releasing
241 * stacks to tasks. Drop __GFP_ACCOUNT.
243 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
244 VMALLOC_START, VMALLOC_END,
245 THREADINFO_GFP & ~__GFP_ACCOUNT,
247 0, node, __builtin_return_address(0));
250 * We can't call find_vm_area() in interrupt context, and
251 * free_thread_stack() can be called in interrupt context,
252 * so cache the vm_struct.
255 tsk->stack_vm_area = find_vm_area(stack);
260 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
264 tsk->stack = page_address(page);
271 static inline void free_thread_stack(struct task_struct *tsk)
273 #ifdef CONFIG_VMAP_STACK
274 struct vm_struct *vm = task_stack_vm_area(tsk);
279 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
280 mod_memcg_page_state(vm->pages[i],
281 MEMCG_KERNEL_STACK_KB,
282 -(int)(PAGE_SIZE / 1024));
284 memcg_kmem_uncharge_page(vm->pages[i], 0);
287 for (i = 0; i < NR_CACHED_STACKS; i++) {
288 if (this_cpu_cmpxchg(cached_stacks[i],
289 NULL, tsk->stack_vm_area) != NULL)
295 vfree_atomic(tsk->stack);
300 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
303 static struct kmem_cache *thread_stack_cache;
305 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
308 unsigned long *stack;
309 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
314 static void free_thread_stack(struct task_struct *tsk)
316 kmem_cache_free(thread_stack_cache, tsk->stack);
319 void thread_stack_cache_init(void)
321 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
322 THREAD_SIZE, THREAD_SIZE, 0, 0,
324 BUG_ON(thread_stack_cache == NULL);
329 /* SLAB cache for signal_struct structures (tsk->signal) */
330 static struct kmem_cache *signal_cachep;
332 /* SLAB cache for sighand_struct structures (tsk->sighand) */
333 struct kmem_cache *sighand_cachep;
335 /* SLAB cache for files_struct structures (tsk->files) */
336 struct kmem_cache *files_cachep;
338 /* SLAB cache for fs_struct structures (tsk->fs) */
339 struct kmem_cache *fs_cachep;
341 /* SLAB cache for vm_area_struct structures */
342 static struct kmem_cache *vm_area_cachep;
344 /* SLAB cache for mm_struct structures (tsk->mm) */
345 static struct kmem_cache *mm_cachep;
347 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
349 struct vm_area_struct *vma;
351 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
357 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
359 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
363 INIT_LIST_HEAD(&new->anon_vma_chain);
364 new->vm_next = new->vm_prev = NULL;
369 void vm_area_free(struct vm_area_struct *vma)
371 kmem_cache_free(vm_area_cachep, vma);
374 static void account_kernel_stack(struct task_struct *tsk, int account)
376 void *stack = task_stack_page(tsk);
377 struct vm_struct *vm = task_stack_vm_area(tsk);
379 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
384 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
386 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
387 mod_zone_page_state(page_zone(vm->pages[i]),
389 PAGE_SIZE / 1024 * account);
393 * All stack pages are in the same zone and belong to the
396 struct page *first_page = virt_to_page(stack);
398 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
399 THREAD_SIZE / 1024 * account);
401 mod_memcg_obj_state(stack, MEMCG_KERNEL_STACK_KB,
402 account * (THREAD_SIZE / 1024));
406 static int memcg_charge_kernel_stack(struct task_struct *tsk)
408 #ifdef CONFIG_VMAP_STACK
409 struct vm_struct *vm = task_stack_vm_area(tsk);
415 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
417 * If memcg_kmem_charge_page() fails, page->mem_cgroup
418 * pointer is NULL, and both memcg_kmem_uncharge_page()
419 * and mod_memcg_page_state() in free_thread_stack()
420 * will ignore this page. So it's safe.
422 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
427 mod_memcg_page_state(vm->pages[i],
428 MEMCG_KERNEL_STACK_KB,
436 static void release_task_stack(struct task_struct *tsk)
438 if (WARN_ON(tsk->state != TASK_DEAD))
439 return; /* Better to leak the stack than to free prematurely */
441 account_kernel_stack(tsk, -1);
442 free_thread_stack(tsk);
444 #ifdef CONFIG_VMAP_STACK
445 tsk->stack_vm_area = NULL;
449 #ifdef CONFIG_THREAD_INFO_IN_TASK
450 void put_task_stack(struct task_struct *tsk)
452 if (refcount_dec_and_test(&tsk->stack_refcount))
453 release_task_stack(tsk);
457 void free_task(struct task_struct *tsk)
459 #ifndef CONFIG_THREAD_INFO_IN_TASK
461 * The task is finally done with both the stack and thread_info,
464 release_task_stack(tsk);
467 * If the task had a separate stack allocation, it should be gone
470 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
472 rt_mutex_debug_task_free(tsk);
473 ftrace_graph_exit_task(tsk);
474 put_seccomp_filter(tsk);
475 arch_release_task_struct(tsk);
476 if (tsk->flags & PF_KTHREAD)
477 free_kthread_struct(tsk);
478 free_task_struct(tsk);
480 EXPORT_SYMBOL(free_task);
483 static __latent_entropy int dup_mmap(struct mm_struct *mm,
484 struct mm_struct *oldmm)
486 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
487 struct rb_node **rb_link, *rb_parent;
489 unsigned long charge;
492 uprobe_start_dup_mmap();
493 if (down_write_killable(&oldmm->mmap_sem)) {
495 goto fail_uprobe_end;
497 flush_cache_dup_mm(oldmm);
498 uprobe_dup_mmap(oldmm, mm);
500 * Not linked in yet - no deadlock potential:
502 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
504 /* No ordering required: file already has been exposed. */
505 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
507 mm->total_vm = oldmm->total_vm;
508 mm->data_vm = oldmm->data_vm;
509 mm->exec_vm = oldmm->exec_vm;
510 mm->stack_vm = oldmm->stack_vm;
512 rb_link = &mm->mm_rb.rb_node;
515 retval = ksm_fork(mm, oldmm);
518 retval = khugepaged_fork(mm, oldmm);
523 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
526 if (mpnt->vm_flags & VM_DONTCOPY) {
527 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
532 * Don't duplicate many vmas if we've been oom-killed (for
535 if (fatal_signal_pending(current)) {
539 if (mpnt->vm_flags & VM_ACCOUNT) {
540 unsigned long len = vma_pages(mpnt);
542 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
546 tmp = vm_area_dup(mpnt);
549 retval = vma_dup_policy(mpnt, tmp);
551 goto fail_nomem_policy;
553 retval = dup_userfaultfd(tmp, &uf);
555 goto fail_nomem_anon_vma_fork;
556 if (tmp->vm_flags & VM_WIPEONFORK) {
558 * VM_WIPEONFORK gets a clean slate in the child.
559 * Don't prepare anon_vma until fault since we don't
560 * copy page for current vma.
562 tmp->anon_vma = NULL;
563 } else if (anon_vma_fork(tmp, mpnt))
564 goto fail_nomem_anon_vma_fork;
565 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
568 struct inode *inode = file_inode(file);
569 struct address_space *mapping = file->f_mapping;
572 if (tmp->vm_flags & VM_DENYWRITE)
573 atomic_dec(&inode->i_writecount);
574 i_mmap_lock_write(mapping);
575 if (tmp->vm_flags & VM_SHARED)
576 atomic_inc(&mapping->i_mmap_writable);
577 flush_dcache_mmap_lock(mapping);
578 /* insert tmp into the share list, just after mpnt */
579 vma_interval_tree_insert_after(tmp, mpnt,
581 flush_dcache_mmap_unlock(mapping);
582 i_mmap_unlock_write(mapping);
586 * Clear hugetlb-related page reserves for children. This only
587 * affects MAP_PRIVATE mappings. Faults generated by the child
588 * are not guaranteed to succeed, even if read-only
590 if (is_vm_hugetlb_page(tmp))
591 reset_vma_resv_huge_pages(tmp);
594 * Link in the new vma and copy the page table entries.
597 pprev = &tmp->vm_next;
601 __vma_link_rb(mm, tmp, rb_link, rb_parent);
602 rb_link = &tmp->vm_rb.rb_right;
603 rb_parent = &tmp->vm_rb;
606 if (!(tmp->vm_flags & VM_WIPEONFORK))
607 retval = copy_page_range(mm, oldmm, mpnt);
609 if (tmp->vm_ops && tmp->vm_ops->open)
610 tmp->vm_ops->open(tmp);
615 /* a new mm has just been created */
616 retval = arch_dup_mmap(oldmm, mm);
618 up_write(&mm->mmap_sem);
620 up_write(&oldmm->mmap_sem);
621 dup_userfaultfd_complete(&uf);
623 uprobe_end_dup_mmap();
625 fail_nomem_anon_vma_fork:
626 mpol_put(vma_policy(tmp));
631 vm_unacct_memory(charge);
635 static inline int mm_alloc_pgd(struct mm_struct *mm)
637 mm->pgd = pgd_alloc(mm);
638 if (unlikely(!mm->pgd))
643 static inline void mm_free_pgd(struct mm_struct *mm)
645 pgd_free(mm, mm->pgd);
648 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
650 down_write(&oldmm->mmap_sem);
651 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
652 up_write(&oldmm->mmap_sem);
655 #define mm_alloc_pgd(mm) (0)
656 #define mm_free_pgd(mm)
657 #endif /* CONFIG_MMU */
659 static void check_mm(struct mm_struct *mm)
663 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
664 "Please make sure 'struct resident_page_types[]' is updated as well");
666 for (i = 0; i < NR_MM_COUNTERS; i++) {
667 long x = atomic_long_read(&mm->rss_stat.count[i]);
670 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
671 mm, resident_page_types[i], x);
674 if (mm_pgtables_bytes(mm))
675 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
676 mm_pgtables_bytes(mm));
678 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
679 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
683 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
684 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
687 * Called when the last reference to the mm
688 * is dropped: either by a lazy thread or by
689 * mmput. Free the page directory and the mm.
691 void __mmdrop(struct mm_struct *mm)
693 BUG_ON(mm == &init_mm);
694 WARN_ON_ONCE(mm == current->mm);
695 WARN_ON_ONCE(mm == current->active_mm);
698 mmu_notifier_subscriptions_destroy(mm);
700 put_user_ns(mm->user_ns);
703 EXPORT_SYMBOL_GPL(__mmdrop);
705 static void mmdrop_async_fn(struct work_struct *work)
707 struct mm_struct *mm;
709 mm = container_of(work, struct mm_struct, async_put_work);
713 static void mmdrop_async(struct mm_struct *mm)
715 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
716 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
717 schedule_work(&mm->async_put_work);
721 static inline void free_signal_struct(struct signal_struct *sig)
723 taskstats_tgid_free(sig);
724 sched_autogroup_exit(sig);
726 * __mmdrop is not safe to call from softirq context on x86 due to
727 * pgd_dtor so postpone it to the async context
730 mmdrop_async(sig->oom_mm);
731 kmem_cache_free(signal_cachep, sig);
734 static inline void put_signal_struct(struct signal_struct *sig)
736 if (refcount_dec_and_test(&sig->sigcnt))
737 free_signal_struct(sig);
740 void __put_task_struct(struct task_struct *tsk)
742 WARN_ON(!tsk->exit_state);
743 WARN_ON(refcount_read(&tsk->usage));
744 WARN_ON(tsk == current);
747 task_numa_free(tsk, true);
748 security_task_free(tsk);
750 delayacct_tsk_free(tsk);
751 put_signal_struct(tsk->signal);
753 if (!profile_handoff_task(tsk))
756 EXPORT_SYMBOL_GPL(__put_task_struct);
758 void __init __weak arch_task_cache_init(void) { }
763 static void set_max_threads(unsigned int max_threads_suggested)
766 unsigned long nr_pages = totalram_pages();
769 * The number of threads shall be limited such that the thread
770 * structures may only consume a small part of the available memory.
772 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
773 threads = MAX_THREADS;
775 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
776 (u64) THREAD_SIZE * 8UL);
778 if (threads > max_threads_suggested)
779 threads = max_threads_suggested;
781 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
784 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
785 /* Initialized by the architecture: */
786 int arch_task_struct_size __read_mostly;
789 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
790 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
792 /* Fetch thread_struct whitelist for the architecture. */
793 arch_thread_struct_whitelist(offset, size);
796 * Handle zero-sized whitelist or empty thread_struct, otherwise
797 * adjust offset to position of thread_struct in task_struct.
799 if (unlikely(*size == 0))
802 *offset += offsetof(struct task_struct, thread);
804 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
806 void __init fork_init(void)
809 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
810 #ifndef ARCH_MIN_TASKALIGN
811 #define ARCH_MIN_TASKALIGN 0
813 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
814 unsigned long useroffset, usersize;
816 /* create a slab on which task_structs can be allocated */
817 task_struct_whitelist(&useroffset, &usersize);
818 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
819 arch_task_struct_size, align,
820 SLAB_PANIC|SLAB_ACCOUNT,
821 useroffset, usersize, NULL);
824 /* do the arch specific task caches init */
825 arch_task_cache_init();
827 set_max_threads(MAX_THREADS);
829 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
830 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
831 init_task.signal->rlim[RLIMIT_SIGPENDING] =
832 init_task.signal->rlim[RLIMIT_NPROC];
834 for (i = 0; i < UCOUNT_COUNTS; i++) {
835 init_user_ns.ucount_max[i] = max_threads/2;
838 #ifdef CONFIG_VMAP_STACK
839 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
840 NULL, free_vm_stack_cache);
843 lockdep_init_task(&init_task);
847 int __weak arch_dup_task_struct(struct task_struct *dst,
848 struct task_struct *src)
854 void set_task_stack_end_magic(struct task_struct *tsk)
856 unsigned long *stackend;
858 stackend = end_of_stack(tsk);
859 *stackend = STACK_END_MAGIC; /* for overflow detection */
862 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
864 struct task_struct *tsk;
865 unsigned long *stack;
866 struct vm_struct *stack_vm_area __maybe_unused;
869 if (node == NUMA_NO_NODE)
870 node = tsk_fork_get_node(orig);
871 tsk = alloc_task_struct_node(node);
875 stack = alloc_thread_stack_node(tsk, node);
879 if (memcg_charge_kernel_stack(tsk))
882 stack_vm_area = task_stack_vm_area(tsk);
884 err = arch_dup_task_struct(tsk, orig);
887 * arch_dup_task_struct() clobbers the stack-related fields. Make
888 * sure they're properly initialized before using any stack-related
892 #ifdef CONFIG_VMAP_STACK
893 tsk->stack_vm_area = stack_vm_area;
895 #ifdef CONFIG_THREAD_INFO_IN_TASK
896 refcount_set(&tsk->stack_refcount, 1);
902 #ifdef CONFIG_SECCOMP
904 * We must handle setting up seccomp filters once we're under
905 * the sighand lock in case orig has changed between now and
906 * then. Until then, filter must be NULL to avoid messing up
907 * the usage counts on the error path calling free_task.
909 tsk->seccomp.filter = NULL;
912 setup_thread_stack(tsk, orig);
913 clear_user_return_notifier(tsk);
914 clear_tsk_need_resched(tsk);
915 set_task_stack_end_magic(tsk);
917 #ifdef CONFIG_STACKPROTECTOR
918 tsk->stack_canary = get_random_canary();
920 if (orig->cpus_ptr == &orig->cpus_mask)
921 tsk->cpus_ptr = &tsk->cpus_mask;
924 * One for the user space visible state that goes away when reaped.
925 * One for the scheduler.
927 refcount_set(&tsk->rcu_users, 2);
928 /* One for the rcu users */
929 refcount_set(&tsk->usage, 1);
930 #ifdef CONFIG_BLK_DEV_IO_TRACE
933 tsk->splice_pipe = NULL;
934 tsk->task_frag.page = NULL;
935 tsk->wake_q.next = NULL;
937 account_kernel_stack(tsk, 1);
941 #ifdef CONFIG_FAULT_INJECTION
945 #ifdef CONFIG_BLK_CGROUP
946 tsk->throttle_queue = NULL;
947 tsk->use_memdelay = 0;
951 tsk->active_memcg = NULL;
956 free_thread_stack(tsk);
958 free_task_struct(tsk);
962 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
964 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
966 static int __init coredump_filter_setup(char *s)
968 default_dump_filter =
969 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
970 MMF_DUMP_FILTER_MASK;
974 __setup("coredump_filter=", coredump_filter_setup);
976 #include <linux/init_task.h>
978 static void mm_init_aio(struct mm_struct *mm)
981 spin_lock_init(&mm->ioctx_lock);
982 mm->ioctx_table = NULL;
986 static __always_inline void mm_clear_owner(struct mm_struct *mm,
987 struct task_struct *p)
991 WRITE_ONCE(mm->owner, NULL);
995 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1002 static void mm_init_uprobes_state(struct mm_struct *mm)
1004 #ifdef CONFIG_UPROBES
1005 mm->uprobes_state.xol_area = NULL;
1009 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1010 struct user_namespace *user_ns)
1013 mm->mm_rb = RB_ROOT;
1014 mm->vmacache_seqnum = 0;
1015 atomic_set(&mm->mm_users, 1);
1016 atomic_set(&mm->mm_count, 1);
1017 init_rwsem(&mm->mmap_sem);
1018 INIT_LIST_HEAD(&mm->mmlist);
1019 mm->core_state = NULL;
1020 mm_pgtables_bytes_init(mm);
1023 atomic64_set(&mm->pinned_vm, 0);
1024 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1025 spin_lock_init(&mm->page_table_lock);
1026 spin_lock_init(&mm->arg_lock);
1027 mm_init_cpumask(mm);
1029 mm_init_owner(mm, p);
1030 RCU_INIT_POINTER(mm->exe_file, NULL);
1031 mmu_notifier_subscriptions_init(mm);
1032 init_tlb_flush_pending(mm);
1033 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1034 mm->pmd_huge_pte = NULL;
1036 mm_init_uprobes_state(mm);
1039 mm->flags = current->mm->flags & MMF_INIT_MASK;
1040 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1042 mm->flags = default_dump_filter;
1046 if (mm_alloc_pgd(mm))
1049 if (init_new_context(p, mm))
1050 goto fail_nocontext;
1052 mm->user_ns = get_user_ns(user_ns);
1063 * Allocate and initialize an mm_struct.
1065 struct mm_struct *mm_alloc(void)
1067 struct mm_struct *mm;
1073 memset(mm, 0, sizeof(*mm));
1074 return mm_init(mm, current, current_user_ns());
1077 static inline void __mmput(struct mm_struct *mm)
1079 VM_BUG_ON(atomic_read(&mm->mm_users));
1081 uprobe_clear_state(mm);
1084 khugepaged_exit(mm); /* must run before exit_mmap */
1086 mm_put_huge_zero_page(mm);
1087 set_mm_exe_file(mm, NULL);
1088 if (!list_empty(&mm->mmlist)) {
1089 spin_lock(&mmlist_lock);
1090 list_del(&mm->mmlist);
1091 spin_unlock(&mmlist_lock);
1094 module_put(mm->binfmt->module);
1099 * Decrement the use count and release all resources for an mm.
1101 void mmput(struct mm_struct *mm)
1105 if (atomic_dec_and_test(&mm->mm_users))
1108 EXPORT_SYMBOL_GPL(mmput);
1111 static void mmput_async_fn(struct work_struct *work)
1113 struct mm_struct *mm = container_of(work, struct mm_struct,
1119 void mmput_async(struct mm_struct *mm)
1121 if (atomic_dec_and_test(&mm->mm_users)) {
1122 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1123 schedule_work(&mm->async_put_work);
1129 * set_mm_exe_file - change a reference to the mm's executable file
1131 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1133 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1134 * invocations: in mmput() nobody alive left, in execve task is single
1135 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1136 * mm->exe_file, but does so without using set_mm_exe_file() in order
1137 * to do avoid the need for any locks.
1139 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1141 struct file *old_exe_file;
1144 * It is safe to dereference the exe_file without RCU as
1145 * this function is only called if nobody else can access
1146 * this mm -- see comment above for justification.
1148 old_exe_file = rcu_dereference_raw(mm->exe_file);
1151 get_file(new_exe_file);
1152 rcu_assign_pointer(mm->exe_file, new_exe_file);
1158 * get_mm_exe_file - acquire a reference to the mm's executable file
1160 * Returns %NULL if mm has no associated executable file.
1161 * User must release file via fput().
1163 struct file *get_mm_exe_file(struct mm_struct *mm)
1165 struct file *exe_file;
1168 exe_file = rcu_dereference(mm->exe_file);
1169 if (exe_file && !get_file_rcu(exe_file))
1174 EXPORT_SYMBOL(get_mm_exe_file);
1177 * get_task_exe_file - acquire a reference to the task's executable file
1179 * Returns %NULL if task's mm (if any) has no associated executable file or
1180 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1181 * User must release file via fput().
1183 struct file *get_task_exe_file(struct task_struct *task)
1185 struct file *exe_file = NULL;
1186 struct mm_struct *mm;
1191 if (!(task->flags & PF_KTHREAD))
1192 exe_file = get_mm_exe_file(mm);
1197 EXPORT_SYMBOL(get_task_exe_file);
1200 * get_task_mm - acquire a reference to the task's mm
1202 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1203 * this kernel workthread has transiently adopted a user mm with use_mm,
1204 * to do its AIO) is not set and if so returns a reference to it, after
1205 * bumping up the use count. User must release the mm via mmput()
1206 * after use. Typically used by /proc and ptrace.
1208 struct mm_struct *get_task_mm(struct task_struct *task)
1210 struct mm_struct *mm;
1215 if (task->flags & PF_KTHREAD)
1223 EXPORT_SYMBOL_GPL(get_task_mm);
1225 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1227 struct mm_struct *mm;
1230 err = mutex_lock_killable(&task->signal->exec_update_mutex);
1232 return ERR_PTR(err);
1234 mm = get_task_mm(task);
1235 if (mm && mm != current->mm &&
1236 !ptrace_may_access(task, mode)) {
1238 mm = ERR_PTR(-EACCES);
1240 mutex_unlock(&task->signal->exec_update_mutex);
1245 static void complete_vfork_done(struct task_struct *tsk)
1247 struct completion *vfork;
1250 vfork = tsk->vfork_done;
1251 if (likely(vfork)) {
1252 tsk->vfork_done = NULL;
1258 static int wait_for_vfork_done(struct task_struct *child,
1259 struct completion *vfork)
1263 freezer_do_not_count();
1264 cgroup_enter_frozen();
1265 killed = wait_for_completion_killable(vfork);
1266 cgroup_leave_frozen(false);
1271 child->vfork_done = NULL;
1275 put_task_struct(child);
1279 /* Please note the differences between mmput and mm_release.
1280 * mmput is called whenever we stop holding onto a mm_struct,
1281 * error success whatever.
1283 * mm_release is called after a mm_struct has been removed
1284 * from the current process.
1286 * This difference is important for error handling, when we
1287 * only half set up a mm_struct for a new process and need to restore
1288 * the old one. Because we mmput the new mm_struct before
1289 * restoring the old one. . .
1290 * Eric Biederman 10 January 1998
1292 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1294 uprobe_free_utask(tsk);
1296 /* Get rid of any cached register state */
1297 deactivate_mm(tsk, mm);
1300 * Signal userspace if we're not exiting with a core dump
1301 * because we want to leave the value intact for debugging
1304 if (tsk->clear_child_tid) {
1305 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1306 atomic_read(&mm->mm_users) > 1) {
1308 * We don't check the error code - if userspace has
1309 * not set up a proper pointer then tough luck.
1311 put_user(0, tsk->clear_child_tid);
1312 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1313 1, NULL, NULL, 0, 0);
1315 tsk->clear_child_tid = NULL;
1319 * All done, finally we can wake up parent and return this mm to him.
1320 * Also kthread_stop() uses this completion for synchronization.
1322 if (tsk->vfork_done)
1323 complete_vfork_done(tsk);
1326 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1328 futex_exit_release(tsk);
1329 mm_release(tsk, mm);
1332 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1334 futex_exec_release(tsk);
1335 mm_release(tsk, mm);
1339 * dup_mm() - duplicates an existing mm structure
1340 * @tsk: the task_struct with which the new mm will be associated.
1341 * @oldmm: the mm to duplicate.
1343 * Allocates a new mm structure and duplicates the provided @oldmm structure
1346 * Return: the duplicated mm or NULL on failure.
1348 static struct mm_struct *dup_mm(struct task_struct *tsk,
1349 struct mm_struct *oldmm)
1351 struct mm_struct *mm;
1358 memcpy(mm, oldmm, sizeof(*mm));
1360 if (!mm_init(mm, tsk, mm->user_ns))
1363 err = dup_mmap(mm, oldmm);
1367 mm->hiwater_rss = get_mm_rss(mm);
1368 mm->hiwater_vm = mm->total_vm;
1370 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1376 /* don't put binfmt in mmput, we haven't got module yet */
1378 mm_init_owner(mm, NULL);
1385 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1387 struct mm_struct *mm, *oldmm;
1390 tsk->min_flt = tsk->maj_flt = 0;
1391 tsk->nvcsw = tsk->nivcsw = 0;
1392 #ifdef CONFIG_DETECT_HUNG_TASK
1393 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1394 tsk->last_switch_time = 0;
1398 tsk->active_mm = NULL;
1401 * Are we cloning a kernel thread?
1403 * We need to steal a active VM for that..
1405 oldmm = current->mm;
1409 /* initialize the new vmacache entries */
1410 vmacache_flush(tsk);
1412 if (clone_flags & CLONE_VM) {
1419 mm = dup_mm(tsk, current->mm);
1425 tsk->active_mm = mm;
1432 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1434 struct fs_struct *fs = current->fs;
1435 if (clone_flags & CLONE_FS) {
1436 /* tsk->fs is already what we want */
1437 spin_lock(&fs->lock);
1439 spin_unlock(&fs->lock);
1443 spin_unlock(&fs->lock);
1446 tsk->fs = copy_fs_struct(fs);
1452 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1454 struct files_struct *oldf, *newf;
1458 * A background process may not have any files ...
1460 oldf = current->files;
1464 if (clone_flags & CLONE_FILES) {
1465 atomic_inc(&oldf->count);
1469 newf = dup_fd(oldf, &error);
1479 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1482 struct io_context *ioc = current->io_context;
1483 struct io_context *new_ioc;
1488 * Share io context with parent, if CLONE_IO is set
1490 if (clone_flags & CLONE_IO) {
1492 tsk->io_context = ioc;
1493 } else if (ioprio_valid(ioc->ioprio)) {
1494 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1495 if (unlikely(!new_ioc))
1498 new_ioc->ioprio = ioc->ioprio;
1499 put_io_context(new_ioc);
1505 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1507 struct sighand_struct *sig;
1509 if (clone_flags & CLONE_SIGHAND) {
1510 refcount_inc(¤t->sighand->count);
1513 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1514 RCU_INIT_POINTER(tsk->sighand, sig);
1518 refcount_set(&sig->count, 1);
1519 spin_lock_irq(¤t->sighand->siglock);
1520 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1521 spin_unlock_irq(¤t->sighand->siglock);
1523 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1524 if (clone_flags & CLONE_CLEAR_SIGHAND)
1525 flush_signal_handlers(tsk, 0);
1530 void __cleanup_sighand(struct sighand_struct *sighand)
1532 if (refcount_dec_and_test(&sighand->count)) {
1533 signalfd_cleanup(sighand);
1535 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1536 * without an RCU grace period, see __lock_task_sighand().
1538 kmem_cache_free(sighand_cachep, sighand);
1543 * Initialize POSIX timer handling for a thread group.
1545 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1547 struct posix_cputimers *pct = &sig->posix_cputimers;
1548 unsigned long cpu_limit;
1550 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1551 posix_cputimers_group_init(pct, cpu_limit);
1554 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1556 struct signal_struct *sig;
1558 if (clone_flags & CLONE_THREAD)
1561 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1566 sig->nr_threads = 1;
1567 atomic_set(&sig->live, 1);
1568 refcount_set(&sig->sigcnt, 1);
1570 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1571 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1572 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1574 init_waitqueue_head(&sig->wait_chldexit);
1575 sig->curr_target = tsk;
1576 init_sigpending(&sig->shared_pending);
1577 INIT_HLIST_HEAD(&sig->multiprocess);
1578 seqlock_init(&sig->stats_lock);
1579 prev_cputime_init(&sig->prev_cputime);
1581 #ifdef CONFIG_POSIX_TIMERS
1582 INIT_LIST_HEAD(&sig->posix_timers);
1583 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1584 sig->real_timer.function = it_real_fn;
1587 task_lock(current->group_leader);
1588 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1589 task_unlock(current->group_leader);
1591 posix_cpu_timers_init_group(sig);
1593 tty_audit_fork(sig);
1594 sched_autogroup_fork(sig);
1596 sig->oom_score_adj = current->signal->oom_score_adj;
1597 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1599 mutex_init(&sig->cred_guard_mutex);
1600 mutex_init(&sig->exec_update_mutex);
1605 static void copy_seccomp(struct task_struct *p)
1607 #ifdef CONFIG_SECCOMP
1609 * Must be called with sighand->lock held, which is common to
1610 * all threads in the group. Holding cred_guard_mutex is not
1611 * needed because this new task is not yet running and cannot
1614 assert_spin_locked(¤t->sighand->siglock);
1616 /* Ref-count the new filter user, and assign it. */
1617 get_seccomp_filter(current);
1618 p->seccomp = current->seccomp;
1621 * Explicitly enable no_new_privs here in case it got set
1622 * between the task_struct being duplicated and holding the
1623 * sighand lock. The seccomp state and nnp must be in sync.
1625 if (task_no_new_privs(current))
1626 task_set_no_new_privs(p);
1629 * If the parent gained a seccomp mode after copying thread
1630 * flags and between before we held the sighand lock, we have
1631 * to manually enable the seccomp thread flag here.
1633 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1634 set_tsk_thread_flag(p, TIF_SECCOMP);
1638 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1640 current->clear_child_tid = tidptr;
1642 return task_pid_vnr(current);
1645 static void rt_mutex_init_task(struct task_struct *p)
1647 raw_spin_lock_init(&p->pi_lock);
1648 #ifdef CONFIG_RT_MUTEXES
1649 p->pi_waiters = RB_ROOT_CACHED;
1650 p->pi_top_task = NULL;
1651 p->pi_blocked_on = NULL;
1655 static inline void init_task_pid_links(struct task_struct *task)
1659 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1660 INIT_HLIST_NODE(&task->pid_links[type]);
1665 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1667 if (type == PIDTYPE_PID)
1668 task->thread_pid = pid;
1670 task->signal->pids[type] = pid;
1673 static inline void rcu_copy_process(struct task_struct *p)
1675 #ifdef CONFIG_PREEMPT_RCU
1676 p->rcu_read_lock_nesting = 0;
1677 p->rcu_read_unlock_special.s = 0;
1678 p->rcu_blocked_node = NULL;
1679 INIT_LIST_HEAD(&p->rcu_node_entry);
1680 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1681 #ifdef CONFIG_TASKS_RCU
1682 p->rcu_tasks_holdout = false;
1683 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1684 p->rcu_tasks_idle_cpu = -1;
1685 #endif /* #ifdef CONFIG_TASKS_RCU */
1686 #ifdef CONFIG_TASKS_TRACE_RCU
1687 p->trc_reader_nesting = 0;
1688 p->trc_reader_special.s = 0;
1689 INIT_LIST_HEAD(&p->trc_holdout_list);
1690 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1693 struct pid *pidfd_pid(const struct file *file)
1695 if (file->f_op == &pidfd_fops)
1696 return file->private_data;
1698 return ERR_PTR(-EBADF);
1701 static int pidfd_release(struct inode *inode, struct file *file)
1703 struct pid *pid = file->private_data;
1705 file->private_data = NULL;
1710 #ifdef CONFIG_PROC_FS
1712 * pidfd_show_fdinfo - print information about a pidfd
1713 * @m: proc fdinfo file
1714 * @f: file referencing a pidfd
1717 * This function will print the pid that a given pidfd refers to in the
1718 * pid namespace of the procfs instance.
1719 * If the pid namespace of the process is not a descendant of the pid
1720 * namespace of the procfs instance 0 will be shown as its pid. This is
1721 * similar to calling getppid() on a process whose parent is outside of
1722 * its pid namespace.
1725 * If pid namespaces are supported then this function will also print
1726 * the pid of a given pidfd refers to for all descendant pid namespaces
1727 * starting from the current pid namespace of the instance, i.e. the
1728 * Pid field and the first entry in the NSpid field will be identical.
1729 * If the pid namespace of the process is not a descendant of the pid
1730 * namespace of the procfs instance 0 will be shown as its first NSpid
1731 * entry and no others will be shown.
1732 * Note that this differs from the Pid and NSpid fields in
1733 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1734 * the pid namespace of the procfs instance. The difference becomes
1735 * obvious when sending around a pidfd between pid namespaces from a
1736 * different branch of the tree, i.e. where no ancestoral relation is
1737 * present between the pid namespaces:
1738 * - create two new pid namespaces ns1 and ns2 in the initial pid
1739 * namespace (also take care to create new mount namespaces in the
1740 * new pid namespace and mount procfs)
1741 * - create a process with a pidfd in ns1
1742 * - send pidfd from ns1 to ns2
1743 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1744 * have exactly one entry, which is 0
1746 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1748 struct pid *pid = f->private_data;
1749 struct pid_namespace *ns;
1752 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1753 ns = proc_pid_ns(file_inode(m->file));
1754 nr = pid_nr_ns(pid, ns);
1757 seq_put_decimal_ll(m, "Pid:\t", nr);
1759 #ifdef CONFIG_PID_NS
1760 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1764 /* If nr is non-zero it means that 'pid' is valid and that
1765 * ns, i.e. the pid namespace associated with the procfs
1766 * instance, is in the pid namespace hierarchy of pid.
1767 * Start at one below the already printed level.
1769 for (i = ns->level + 1; i <= pid->level; i++)
1770 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1778 * Poll support for process exit notification.
1780 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1782 struct task_struct *task;
1783 struct pid *pid = file->private_data;
1784 __poll_t poll_flags = 0;
1786 poll_wait(file, &pid->wait_pidfd, pts);
1789 task = pid_task(pid, PIDTYPE_PID);
1791 * Inform pollers only when the whole thread group exits.
1792 * If the thread group leader exits before all other threads in the
1793 * group, then poll(2) should block, similar to the wait(2) family.
1795 if (!task || (task->exit_state && thread_group_empty(task)))
1796 poll_flags = EPOLLIN | EPOLLRDNORM;
1802 const struct file_operations pidfd_fops = {
1803 .release = pidfd_release,
1805 #ifdef CONFIG_PROC_FS
1806 .show_fdinfo = pidfd_show_fdinfo,
1810 static void __delayed_free_task(struct rcu_head *rhp)
1812 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1817 static __always_inline void delayed_free_task(struct task_struct *tsk)
1819 if (IS_ENABLED(CONFIG_MEMCG))
1820 call_rcu(&tsk->rcu, __delayed_free_task);
1826 * This creates a new process as a copy of the old one,
1827 * but does not actually start it yet.
1829 * It copies the registers, and all the appropriate
1830 * parts of the process environment (as per the clone
1831 * flags). The actual kick-off is left to the caller.
1833 static __latent_entropy struct task_struct *copy_process(
1837 struct kernel_clone_args *args)
1839 int pidfd = -1, retval;
1840 struct task_struct *p;
1841 struct multiprocess_signals delayed;
1842 struct file *pidfile = NULL;
1843 u64 clone_flags = args->flags;
1844 struct nsproxy *nsp = current->nsproxy;
1847 * Don't allow sharing the root directory with processes in a different
1850 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1851 return ERR_PTR(-EINVAL);
1853 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1854 return ERR_PTR(-EINVAL);
1857 * Thread groups must share signals as well, and detached threads
1858 * can only be started up within the thread group.
1860 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1861 return ERR_PTR(-EINVAL);
1864 * Shared signal handlers imply shared VM. By way of the above,
1865 * thread groups also imply shared VM. Blocking this case allows
1866 * for various simplifications in other code.
1868 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1869 return ERR_PTR(-EINVAL);
1872 * Siblings of global init remain as zombies on exit since they are
1873 * not reaped by their parent (swapper). To solve this and to avoid
1874 * multi-rooted process trees, prevent global and container-inits
1875 * from creating siblings.
1877 if ((clone_flags & CLONE_PARENT) &&
1878 current->signal->flags & SIGNAL_UNKILLABLE)
1879 return ERR_PTR(-EINVAL);
1882 * If the new process will be in a different pid or user namespace
1883 * do not allow it to share a thread group with the forking task.
1885 if (clone_flags & CLONE_THREAD) {
1886 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1887 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1888 return ERR_PTR(-EINVAL);
1892 * If the new process will be in a different time namespace
1893 * do not allow it to share VM or a thread group with the forking task.
1895 if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1896 if (nsp->time_ns != nsp->time_ns_for_children)
1897 return ERR_PTR(-EINVAL);
1900 if (clone_flags & CLONE_PIDFD) {
1902 * - CLONE_DETACHED is blocked so that we can potentially
1903 * reuse it later for CLONE_PIDFD.
1904 * - CLONE_THREAD is blocked until someone really needs it.
1906 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1907 return ERR_PTR(-EINVAL);
1911 * Force any signals received before this point to be delivered
1912 * before the fork happens. Collect up signals sent to multiple
1913 * processes that happen during the fork and delay them so that
1914 * they appear to happen after the fork.
1916 sigemptyset(&delayed.signal);
1917 INIT_HLIST_NODE(&delayed.node);
1919 spin_lock_irq(¤t->sighand->siglock);
1920 if (!(clone_flags & CLONE_THREAD))
1921 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1922 recalc_sigpending();
1923 spin_unlock_irq(¤t->sighand->siglock);
1924 retval = -ERESTARTNOINTR;
1925 if (signal_pending(current))
1929 p = dup_task_struct(current, node);
1934 * This _must_ happen before we call free_task(), i.e. before we jump
1935 * to any of the bad_fork_* labels. This is to avoid freeing
1936 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1937 * kernel threads (PF_KTHREAD).
1939 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1941 * Clear TID on mm_release()?
1943 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1945 ftrace_graph_init_task(p);
1947 rt_mutex_init_task(p);
1949 #ifdef CONFIG_PROVE_LOCKING
1950 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1951 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1954 if (atomic_read(&p->real_cred->user->processes) >=
1955 task_rlimit(p, RLIMIT_NPROC)) {
1956 if (p->real_cred->user != INIT_USER &&
1957 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1960 current->flags &= ~PF_NPROC_EXCEEDED;
1962 retval = copy_creds(p, clone_flags);
1967 * If multiple threads are within copy_process(), then this check
1968 * triggers too late. This doesn't hurt, the check is only there
1969 * to stop root fork bombs.
1972 if (nr_threads >= max_threads)
1973 goto bad_fork_cleanup_count;
1975 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1976 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1977 p->flags |= PF_FORKNOEXEC;
1978 INIT_LIST_HEAD(&p->children);
1979 INIT_LIST_HEAD(&p->sibling);
1980 rcu_copy_process(p);
1981 p->vfork_done = NULL;
1982 spin_lock_init(&p->alloc_lock);
1984 init_sigpending(&p->pending);
1986 p->utime = p->stime = p->gtime = 0;
1987 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1988 p->utimescaled = p->stimescaled = 0;
1990 prev_cputime_init(&p->prev_cputime);
1992 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1993 seqcount_init(&p->vtime.seqcount);
1994 p->vtime.starttime = 0;
1995 p->vtime.state = VTIME_INACTIVE;
1998 #if defined(SPLIT_RSS_COUNTING)
1999 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2002 p->default_timer_slack_ns = current->timer_slack_ns;
2008 task_io_accounting_init(&p->ioac);
2009 acct_clear_integrals(p);
2011 posix_cputimers_init(&p->posix_cputimers);
2013 p->io_context = NULL;
2014 audit_set_context(p, NULL);
2017 p->mempolicy = mpol_dup(p->mempolicy);
2018 if (IS_ERR(p->mempolicy)) {
2019 retval = PTR_ERR(p->mempolicy);
2020 p->mempolicy = NULL;
2021 goto bad_fork_cleanup_threadgroup_lock;
2024 #ifdef CONFIG_CPUSETS
2025 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2026 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2027 seqcount_init(&p->mems_allowed_seq);
2029 #ifdef CONFIG_TRACE_IRQFLAGS
2031 p->hardirqs_enabled = 0;
2032 p->hardirq_enable_ip = 0;
2033 p->hardirq_enable_event = 0;
2034 p->hardirq_disable_ip = _THIS_IP_;
2035 p->hardirq_disable_event = 0;
2036 p->softirqs_enabled = 1;
2037 p->softirq_enable_ip = _THIS_IP_;
2038 p->softirq_enable_event = 0;
2039 p->softirq_disable_ip = 0;
2040 p->softirq_disable_event = 0;
2041 p->hardirq_context = 0;
2042 p->softirq_context = 0;
2045 p->pagefault_disabled = 0;
2047 #ifdef CONFIG_LOCKDEP
2048 lockdep_init_task(p);
2051 #ifdef CONFIG_DEBUG_MUTEXES
2052 p->blocked_on = NULL; /* not blocked yet */
2054 #ifdef CONFIG_BCACHE
2055 p->sequential_io = 0;
2056 p->sequential_io_avg = 0;
2059 /* Perform scheduler related setup. Assign this task to a CPU. */
2060 retval = sched_fork(clone_flags, p);
2062 goto bad_fork_cleanup_policy;
2064 retval = perf_event_init_task(p);
2066 goto bad_fork_cleanup_policy;
2067 retval = audit_alloc(p);
2069 goto bad_fork_cleanup_perf;
2070 /* copy all the process information */
2072 retval = security_task_alloc(p, clone_flags);
2074 goto bad_fork_cleanup_audit;
2075 retval = copy_semundo(clone_flags, p);
2077 goto bad_fork_cleanup_security;
2078 retval = copy_files(clone_flags, p);
2080 goto bad_fork_cleanup_semundo;
2081 retval = copy_fs(clone_flags, p);
2083 goto bad_fork_cleanup_files;
2084 retval = copy_sighand(clone_flags, p);
2086 goto bad_fork_cleanup_fs;
2087 retval = copy_signal(clone_flags, p);
2089 goto bad_fork_cleanup_sighand;
2090 retval = copy_mm(clone_flags, p);
2092 goto bad_fork_cleanup_signal;
2093 retval = copy_namespaces(clone_flags, p);
2095 goto bad_fork_cleanup_mm;
2096 retval = copy_io(clone_flags, p);
2098 goto bad_fork_cleanup_namespaces;
2099 retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2102 goto bad_fork_cleanup_io;
2104 stackleak_task_init(p);
2106 if (pid != &init_struct_pid) {
2107 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2108 args->set_tid_size);
2110 retval = PTR_ERR(pid);
2111 goto bad_fork_cleanup_thread;
2116 * This has to happen after we've potentially unshared the file
2117 * descriptor table (so that the pidfd doesn't leak into the child
2118 * if the fd table isn't shared).
2120 if (clone_flags & CLONE_PIDFD) {
2121 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2123 goto bad_fork_free_pid;
2127 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2128 O_RDWR | O_CLOEXEC);
2129 if (IS_ERR(pidfile)) {
2130 put_unused_fd(pidfd);
2131 retval = PTR_ERR(pidfile);
2132 goto bad_fork_free_pid;
2134 get_pid(pid); /* held by pidfile now */
2136 retval = put_user(pidfd, args->pidfd);
2138 goto bad_fork_put_pidfd;
2147 * sigaltstack should be cleared when sharing the same VM
2149 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2153 * Syscall tracing and stepping should be turned off in the
2154 * child regardless of CLONE_PTRACE.
2156 user_disable_single_step(p);
2157 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2158 #ifdef TIF_SYSCALL_EMU
2159 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2161 clear_tsk_latency_tracing(p);
2163 /* ok, now we should be set up.. */
2164 p->pid = pid_nr(pid);
2165 if (clone_flags & CLONE_THREAD) {
2166 p->exit_signal = -1;
2167 p->group_leader = current->group_leader;
2168 p->tgid = current->tgid;
2170 if (clone_flags & CLONE_PARENT)
2171 p->exit_signal = current->group_leader->exit_signal;
2173 p->exit_signal = args->exit_signal;
2174 p->group_leader = p;
2179 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2180 p->dirty_paused_when = 0;
2182 p->pdeath_signal = 0;
2183 INIT_LIST_HEAD(&p->thread_group);
2184 p->task_works = NULL;
2187 * Ensure that the cgroup subsystem policies allow the new process to be
2188 * forked. It should be noted the the new process's css_set can be changed
2189 * between here and cgroup_post_fork() if an organisation operation is in
2192 retval = cgroup_can_fork(p, args);
2194 goto bad_fork_put_pidfd;
2197 * From this point on we must avoid any synchronous user-space
2198 * communication until we take the tasklist-lock. In particular, we do
2199 * not want user-space to be able to predict the process start-time by
2200 * stalling fork(2) after we recorded the start_time but before it is
2201 * visible to the system.
2204 p->start_time = ktime_get_ns();
2205 p->start_boottime = ktime_get_boottime_ns();
2208 * Make it visible to the rest of the system, but dont wake it up yet.
2209 * Need tasklist lock for parent etc handling!
2211 write_lock_irq(&tasklist_lock);
2213 /* CLONE_PARENT re-uses the old parent */
2214 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2215 p->real_parent = current->real_parent;
2216 p->parent_exec_id = current->parent_exec_id;
2218 p->real_parent = current;
2219 p->parent_exec_id = current->self_exec_id;
2222 klp_copy_process(p);
2224 spin_lock(¤t->sighand->siglock);
2227 * Copy seccomp details explicitly here, in case they were changed
2228 * before holding sighand lock.
2232 rseq_fork(p, clone_flags);
2234 /* Don't start children in a dying pid namespace */
2235 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2237 goto bad_fork_cancel_cgroup;
2240 /* Let kill terminate clone/fork in the middle */
2241 if (fatal_signal_pending(current)) {
2243 goto bad_fork_cancel_cgroup;
2246 /* past the last point of failure */
2248 fd_install(pidfd, pidfile);
2250 init_task_pid_links(p);
2251 if (likely(p->pid)) {
2252 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2254 init_task_pid(p, PIDTYPE_PID, pid);
2255 if (thread_group_leader(p)) {
2256 init_task_pid(p, PIDTYPE_TGID, pid);
2257 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2258 init_task_pid(p, PIDTYPE_SID, task_session(current));
2260 if (is_child_reaper(pid)) {
2261 ns_of_pid(pid)->child_reaper = p;
2262 p->signal->flags |= SIGNAL_UNKILLABLE;
2264 p->signal->shared_pending.signal = delayed.signal;
2265 p->signal->tty = tty_kref_get(current->signal->tty);
2267 * Inherit has_child_subreaper flag under the same
2268 * tasklist_lock with adding child to the process tree
2269 * for propagate_has_child_subreaper optimization.
2271 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2272 p->real_parent->signal->is_child_subreaper;
2273 list_add_tail(&p->sibling, &p->real_parent->children);
2274 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2275 attach_pid(p, PIDTYPE_TGID);
2276 attach_pid(p, PIDTYPE_PGID);
2277 attach_pid(p, PIDTYPE_SID);
2278 __this_cpu_inc(process_counts);
2280 current->signal->nr_threads++;
2281 atomic_inc(¤t->signal->live);
2282 refcount_inc(¤t->signal->sigcnt);
2283 task_join_group_stop(p);
2284 list_add_tail_rcu(&p->thread_group,
2285 &p->group_leader->thread_group);
2286 list_add_tail_rcu(&p->thread_node,
2287 &p->signal->thread_head);
2289 attach_pid(p, PIDTYPE_PID);
2293 hlist_del_init(&delayed.node);
2294 spin_unlock(¤t->sighand->siglock);
2295 syscall_tracepoint_update(p);
2296 write_unlock_irq(&tasklist_lock);
2298 proc_fork_connector(p);
2299 cgroup_post_fork(p, args);
2302 trace_task_newtask(p, clone_flags);
2303 uprobe_copy_process(p, clone_flags);
2307 bad_fork_cancel_cgroup:
2308 spin_unlock(¤t->sighand->siglock);
2309 write_unlock_irq(&tasklist_lock);
2310 cgroup_cancel_fork(p, args);
2312 if (clone_flags & CLONE_PIDFD) {
2314 put_unused_fd(pidfd);
2317 if (pid != &init_struct_pid)
2319 bad_fork_cleanup_thread:
2321 bad_fork_cleanup_io:
2324 bad_fork_cleanup_namespaces:
2325 exit_task_namespaces(p);
2326 bad_fork_cleanup_mm:
2328 mm_clear_owner(p->mm, p);
2331 bad_fork_cleanup_signal:
2332 if (!(clone_flags & CLONE_THREAD))
2333 free_signal_struct(p->signal);
2334 bad_fork_cleanup_sighand:
2335 __cleanup_sighand(p->sighand);
2336 bad_fork_cleanup_fs:
2337 exit_fs(p); /* blocking */
2338 bad_fork_cleanup_files:
2339 exit_files(p); /* blocking */
2340 bad_fork_cleanup_semundo:
2342 bad_fork_cleanup_security:
2343 security_task_free(p);
2344 bad_fork_cleanup_audit:
2346 bad_fork_cleanup_perf:
2347 perf_event_free_task(p);
2348 bad_fork_cleanup_policy:
2349 lockdep_free_task(p);
2351 mpol_put(p->mempolicy);
2352 bad_fork_cleanup_threadgroup_lock:
2354 delayacct_tsk_free(p);
2355 bad_fork_cleanup_count:
2356 atomic_dec(&p->cred->user->processes);
2359 p->state = TASK_DEAD;
2361 delayed_free_task(p);
2363 spin_lock_irq(¤t->sighand->siglock);
2364 hlist_del_init(&delayed.node);
2365 spin_unlock_irq(¤t->sighand->siglock);
2366 return ERR_PTR(retval);
2369 static inline void init_idle_pids(struct task_struct *idle)
2373 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2374 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2375 init_task_pid(idle, type, &init_struct_pid);
2379 struct task_struct *fork_idle(int cpu)
2381 struct task_struct *task;
2382 struct kernel_clone_args args = {
2386 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2387 if (!IS_ERR(task)) {
2388 init_idle_pids(task);
2389 init_idle(task, cpu);
2395 struct mm_struct *copy_init_mm(void)
2397 return dup_mm(NULL, &init_mm);
2401 * Ok, this is the main fork-routine.
2403 * It copies the process, and if successful kick-starts
2404 * it and waits for it to finish using the VM if required.
2406 * args->exit_signal is expected to be checked for sanity by the caller.
2408 long _do_fork(struct kernel_clone_args *args)
2410 u64 clone_flags = args->flags;
2411 struct completion vfork;
2413 struct task_struct *p;
2418 * Determine whether and which event to report to ptracer. When
2419 * called from kernel_thread or CLONE_UNTRACED is explicitly
2420 * requested, no event is reported; otherwise, report if the event
2421 * for the type of forking is enabled.
2423 if (!(clone_flags & CLONE_UNTRACED)) {
2424 if (clone_flags & CLONE_VFORK)
2425 trace = PTRACE_EVENT_VFORK;
2426 else if (args->exit_signal != SIGCHLD)
2427 trace = PTRACE_EVENT_CLONE;
2429 trace = PTRACE_EVENT_FORK;
2431 if (likely(!ptrace_event_enabled(current, trace)))
2435 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2436 add_latent_entropy();
2442 * Do this prior waking up the new thread - the thread pointer
2443 * might get invalid after that point, if the thread exits quickly.
2445 trace_sched_process_fork(current, p);
2447 pid = get_task_pid(p, PIDTYPE_PID);
2450 if (clone_flags & CLONE_PARENT_SETTID)
2451 put_user(nr, args->parent_tid);
2453 if (clone_flags & CLONE_VFORK) {
2454 p->vfork_done = &vfork;
2455 init_completion(&vfork);
2459 wake_up_new_task(p);
2461 /* forking complete and child started to run, tell ptracer */
2462 if (unlikely(trace))
2463 ptrace_event_pid(trace, pid);
2465 if (clone_flags & CLONE_VFORK) {
2466 if (!wait_for_vfork_done(p, &vfork))
2467 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2474 bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
2476 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2477 if ((kargs->flags & CLONE_PIDFD) &&
2478 (kargs->flags & CLONE_PARENT_SETTID))
2484 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2485 /* For compatibility with architectures that call do_fork directly rather than
2486 * using the syscall entry points below. */
2487 long do_fork(unsigned long clone_flags,
2488 unsigned long stack_start,
2489 unsigned long stack_size,
2490 int __user *parent_tidptr,
2491 int __user *child_tidptr)
2493 struct kernel_clone_args args = {
2494 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2495 .pidfd = parent_tidptr,
2496 .child_tid = child_tidptr,
2497 .parent_tid = parent_tidptr,
2498 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2499 .stack = stack_start,
2500 .stack_size = stack_size,
2503 if (!legacy_clone_args_valid(&args))
2506 return _do_fork(&args);
2511 * Create a kernel thread.
2513 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2515 struct kernel_clone_args args = {
2516 .flags = ((lower_32_bits(flags) | CLONE_VM |
2517 CLONE_UNTRACED) & ~CSIGNAL),
2518 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2519 .stack = (unsigned long)fn,
2520 .stack_size = (unsigned long)arg,
2523 return _do_fork(&args);
2526 #ifdef __ARCH_WANT_SYS_FORK
2527 SYSCALL_DEFINE0(fork)
2530 struct kernel_clone_args args = {
2531 .exit_signal = SIGCHLD,
2534 return _do_fork(&args);
2536 /* can not support in nommu mode */
2542 #ifdef __ARCH_WANT_SYS_VFORK
2543 SYSCALL_DEFINE0(vfork)
2545 struct kernel_clone_args args = {
2546 .flags = CLONE_VFORK | CLONE_VM,
2547 .exit_signal = SIGCHLD,
2550 return _do_fork(&args);
2554 #ifdef __ARCH_WANT_SYS_CLONE
2555 #ifdef CONFIG_CLONE_BACKWARDS
2556 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2557 int __user *, parent_tidptr,
2559 int __user *, child_tidptr)
2560 #elif defined(CONFIG_CLONE_BACKWARDS2)
2561 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2562 int __user *, parent_tidptr,
2563 int __user *, child_tidptr,
2565 #elif defined(CONFIG_CLONE_BACKWARDS3)
2566 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2568 int __user *, parent_tidptr,
2569 int __user *, child_tidptr,
2572 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2573 int __user *, parent_tidptr,
2574 int __user *, child_tidptr,
2578 struct kernel_clone_args args = {
2579 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2580 .pidfd = parent_tidptr,
2581 .child_tid = child_tidptr,
2582 .parent_tid = parent_tidptr,
2583 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2588 if (!legacy_clone_args_valid(&args))
2591 return _do_fork(&args);
2595 #ifdef __ARCH_WANT_SYS_CLONE3
2598 * copy_thread implementations handle CLONE_SETTLS by reading the TLS value from
2599 * the registers containing the syscall arguments for clone. This doesn't work
2600 * with clone3 since the TLS value is passed in clone_args instead.
2602 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2603 #error clone3 requires copy_thread_tls support in arch
2606 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2607 struct clone_args __user *uargs,
2611 struct clone_args args;
2612 pid_t *kset_tid = kargs->set_tid;
2614 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2615 CLONE_ARGS_SIZE_VER0);
2616 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2617 CLONE_ARGS_SIZE_VER1);
2618 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2619 CLONE_ARGS_SIZE_VER2);
2620 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2622 if (unlikely(usize > PAGE_SIZE))
2624 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2627 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2631 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2634 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2637 if (unlikely(args.set_tid && args.set_tid_size == 0))
2641 * Verify that higher 32bits of exit_signal are unset and that
2642 * it is a valid signal
2644 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2645 !valid_signal(args.exit_signal)))
2648 if ((args.flags & CLONE_INTO_CGROUP) &&
2649 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2652 *kargs = (struct kernel_clone_args){
2653 .flags = args.flags,
2654 .pidfd = u64_to_user_ptr(args.pidfd),
2655 .child_tid = u64_to_user_ptr(args.child_tid),
2656 .parent_tid = u64_to_user_ptr(args.parent_tid),
2657 .exit_signal = args.exit_signal,
2658 .stack = args.stack,
2659 .stack_size = args.stack_size,
2661 .set_tid_size = args.set_tid_size,
2662 .cgroup = args.cgroup,
2666 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2667 (kargs->set_tid_size * sizeof(pid_t))))
2670 kargs->set_tid = kset_tid;
2676 * clone3_stack_valid - check and prepare stack
2677 * @kargs: kernel clone args
2679 * Verify that the stack arguments userspace gave us are sane.
2680 * In addition, set the stack direction for userspace since it's easy for us to
2683 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2685 if (kargs->stack == 0) {
2686 if (kargs->stack_size > 0)
2689 if (kargs->stack_size == 0)
2692 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2695 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2696 kargs->stack += kargs->stack_size;
2703 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2705 /* Verify that no unknown flags are passed along. */
2707 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2711 * - make the CLONE_DETACHED bit reuseable for clone3
2712 * - make the CSIGNAL bits reuseable for clone3
2714 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2717 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2718 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2721 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2725 if (!clone3_stack_valid(kargs))
2732 * clone3 - create a new process with specific properties
2733 * @uargs: argument structure
2734 * @size: size of @uargs
2736 * clone3() is the extensible successor to clone()/clone2().
2737 * It takes a struct as argument that is versioned by its size.
2739 * Return: On success, a positive PID for the child process.
2740 * On error, a negative errno number.
2742 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2746 struct kernel_clone_args kargs;
2747 pid_t set_tid[MAX_PID_NS_LEVEL];
2749 kargs.set_tid = set_tid;
2751 err = copy_clone_args_from_user(&kargs, uargs, size);
2755 if (!clone3_args_valid(&kargs))
2758 return _do_fork(&kargs);
2762 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2764 struct task_struct *leader, *parent, *child;
2767 read_lock(&tasklist_lock);
2768 leader = top = top->group_leader;
2770 for_each_thread(leader, parent) {
2771 list_for_each_entry(child, &parent->children, sibling) {
2772 res = visitor(child, data);
2784 if (leader != top) {
2786 parent = child->real_parent;
2787 leader = parent->group_leader;
2791 read_unlock(&tasklist_lock);
2794 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2795 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2798 static void sighand_ctor(void *data)
2800 struct sighand_struct *sighand = data;
2802 spin_lock_init(&sighand->siglock);
2803 init_waitqueue_head(&sighand->signalfd_wqh);
2806 void __init proc_caches_init(void)
2808 unsigned int mm_size;
2810 sighand_cachep = kmem_cache_create("sighand_cache",
2811 sizeof(struct sighand_struct), 0,
2812 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2813 SLAB_ACCOUNT, sighand_ctor);
2814 signal_cachep = kmem_cache_create("signal_cache",
2815 sizeof(struct signal_struct), 0,
2816 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2818 files_cachep = kmem_cache_create("files_cache",
2819 sizeof(struct files_struct), 0,
2820 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2822 fs_cachep = kmem_cache_create("fs_cache",
2823 sizeof(struct fs_struct), 0,
2824 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2828 * The mm_cpumask is located at the end of mm_struct, and is
2829 * dynamically sized based on the maximum CPU number this system
2830 * can have, taking hotplug into account (nr_cpu_ids).
2832 mm_size = sizeof(struct mm_struct) + cpumask_size();
2834 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2835 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2836 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2837 offsetof(struct mm_struct, saved_auxv),
2838 sizeof_field(struct mm_struct, saved_auxv),
2840 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2842 nsproxy_cache_init();
2846 * Check constraints on flags passed to the unshare system call.
2848 static int check_unshare_flags(unsigned long unshare_flags)
2850 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2851 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2852 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2853 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2857 * Not implemented, but pretend it works if there is nothing
2858 * to unshare. Note that unsharing the address space or the
2859 * signal handlers also need to unshare the signal queues (aka
2862 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2863 if (!thread_group_empty(current))
2866 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2867 if (refcount_read(¤t->sighand->count) > 1)
2870 if (unshare_flags & CLONE_VM) {
2871 if (!current_is_single_threaded())
2879 * Unshare the filesystem structure if it is being shared
2881 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2883 struct fs_struct *fs = current->fs;
2885 if (!(unshare_flags & CLONE_FS) || !fs)
2888 /* don't need lock here; in the worst case we'll do useless copy */
2892 *new_fsp = copy_fs_struct(fs);
2900 * Unshare file descriptor table if it is being shared
2902 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2904 struct files_struct *fd = current->files;
2907 if ((unshare_flags & CLONE_FILES) &&
2908 (fd && atomic_read(&fd->count) > 1)) {
2909 *new_fdp = dup_fd(fd, &error);
2918 * unshare allows a process to 'unshare' part of the process
2919 * context which was originally shared using clone. copy_*
2920 * functions used by do_fork() cannot be used here directly
2921 * because they modify an inactive task_struct that is being
2922 * constructed. Here we are modifying the current, active,
2925 int ksys_unshare(unsigned long unshare_flags)
2927 struct fs_struct *fs, *new_fs = NULL;
2928 struct files_struct *fd, *new_fd = NULL;
2929 struct cred *new_cred = NULL;
2930 struct nsproxy *new_nsproxy = NULL;
2935 * If unsharing a user namespace must also unshare the thread group
2936 * and unshare the filesystem root and working directories.
2938 if (unshare_flags & CLONE_NEWUSER)
2939 unshare_flags |= CLONE_THREAD | CLONE_FS;
2941 * If unsharing vm, must also unshare signal handlers.
2943 if (unshare_flags & CLONE_VM)
2944 unshare_flags |= CLONE_SIGHAND;
2946 * If unsharing a signal handlers, must also unshare the signal queues.
2948 if (unshare_flags & CLONE_SIGHAND)
2949 unshare_flags |= CLONE_THREAD;
2951 * If unsharing namespace, must also unshare filesystem information.
2953 if (unshare_flags & CLONE_NEWNS)
2954 unshare_flags |= CLONE_FS;
2956 err = check_unshare_flags(unshare_flags);
2958 goto bad_unshare_out;
2960 * CLONE_NEWIPC must also detach from the undolist: after switching
2961 * to a new ipc namespace, the semaphore arrays from the old
2962 * namespace are unreachable.
2964 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2966 err = unshare_fs(unshare_flags, &new_fs);
2968 goto bad_unshare_out;
2969 err = unshare_fd(unshare_flags, &new_fd);
2971 goto bad_unshare_cleanup_fs;
2972 err = unshare_userns(unshare_flags, &new_cred);
2974 goto bad_unshare_cleanup_fd;
2975 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2978 goto bad_unshare_cleanup_cred;
2980 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2983 * CLONE_SYSVSEM is equivalent to sys_exit().
2987 if (unshare_flags & CLONE_NEWIPC) {
2988 /* Orphan segments in old ns (see sem above). */
2990 shm_init_task(current);
2994 switch_task_namespaces(current, new_nsproxy);
3000 spin_lock(&fs->lock);
3001 current->fs = new_fs;
3006 spin_unlock(&fs->lock);
3010 fd = current->files;
3011 current->files = new_fd;
3015 task_unlock(current);
3018 /* Install the new user namespace */
3019 commit_creds(new_cred);
3024 perf_event_namespaces(current);
3026 bad_unshare_cleanup_cred:
3029 bad_unshare_cleanup_fd:
3031 put_files_struct(new_fd);
3033 bad_unshare_cleanup_fs:
3035 free_fs_struct(new_fs);
3041 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3043 return ksys_unshare(unshare_flags);
3047 * Helper to unshare the files of the current task.
3048 * We don't want to expose copy_files internals to
3049 * the exec layer of the kernel.
3052 int unshare_files(struct files_struct **displaced)
3054 struct task_struct *task = current;
3055 struct files_struct *copy = NULL;
3058 error = unshare_fd(CLONE_FILES, ©);
3059 if (error || !copy) {
3063 *displaced = task->files;
3070 int sysctl_max_threads(struct ctl_table *table, int write,
3071 void __user *buffer, size_t *lenp, loff_t *ppos)
3075 int threads = max_threads;
3077 int max = MAX_THREADS;
3084 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3088 max_threads = threads;