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>
101 #include <asm/pgalloc.h>
102 #include <linux/uaccess.h>
103 #include <asm/mmu_context.h>
104 #include <asm/cacheflush.h>
105 #include <asm/tlbflush.h>
107 #include <trace/events/sched.h>
109 #define CREATE_TRACE_POINTS
110 #include <trace/events/task.h>
113 * Minimum number of threads to boot the kernel
115 #define MIN_THREADS 20
118 * Maximum number of threads
120 #define MAX_THREADS FUTEX_TID_MASK
123 * Protected counters by write_lock_irq(&tasklist_lock)
125 unsigned long total_forks; /* Handle normal Linux uptimes. */
126 int nr_threads; /* The idle threads do not count.. */
128 static int max_threads; /* tunable limit on nr_threads */
130 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
132 static const char * const resident_page_types[] = {
133 NAMED_ARRAY_INDEX(MM_FILEPAGES),
134 NAMED_ARRAY_INDEX(MM_ANONPAGES),
135 NAMED_ARRAY_INDEX(MM_SWAPENTS),
136 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
139 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
141 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
143 #ifdef CONFIG_PROVE_RCU
144 int lockdep_tasklist_lock_is_held(void)
146 return lockdep_is_held(&tasklist_lock);
148 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
149 #endif /* #ifdef CONFIG_PROVE_RCU */
151 int nr_processes(void)
156 for_each_possible_cpu(cpu)
157 total += per_cpu(process_counts, cpu);
162 void __weak arch_release_task_struct(struct task_struct *tsk)
166 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
167 static struct kmem_cache *task_struct_cachep;
169 static inline struct task_struct *alloc_task_struct_node(int node)
171 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
174 static inline void free_task_struct(struct task_struct *tsk)
176 kmem_cache_free(task_struct_cachep, tsk);
180 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
183 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
184 * kmemcache based allocator.
186 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
188 # ifdef CONFIG_VMAP_STACK
190 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
191 * flush. Try to minimize the number of calls by caching stacks.
193 #define NR_CACHED_STACKS 2
194 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
198 struct vm_struct *stack_vm_area;
201 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
205 for (i = 0; i < NR_CACHED_STACKS; i++) {
206 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
213 static void thread_stack_free_rcu(struct rcu_head *rh)
215 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
217 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
223 static void thread_stack_delayed_free(struct task_struct *tsk)
225 struct vm_stack *vm_stack = tsk->stack;
227 vm_stack->stack_vm_area = tsk->stack_vm_area;
228 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
231 static int free_vm_stack_cache(unsigned int cpu)
233 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
236 for (i = 0; i < NR_CACHED_STACKS; i++) {
237 struct vm_struct *vm_stack = cached_vm_stacks[i];
242 vfree(vm_stack->addr);
243 cached_vm_stacks[i] = NULL;
249 static int memcg_charge_kernel_stack(struct vm_struct *vm)
254 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
255 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
257 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
258 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
265 * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
266 * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
269 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
270 memcg_kmem_uncharge_page(vm->pages[i], 0);
274 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
276 struct vm_struct *vm;
280 for (i = 0; i < NR_CACHED_STACKS; i++) {
283 s = this_cpu_xchg(cached_stacks[i], NULL);
288 /* Reset stack metadata. */
289 kasan_unpoison_range(s->addr, THREAD_SIZE);
291 stack = kasan_reset_tag(s->addr);
293 /* Clear stale pointers from reused stack. */
294 memset(stack, 0, THREAD_SIZE);
296 if (memcg_charge_kernel_stack(s)) {
301 tsk->stack_vm_area = s;
307 * Allocated stacks are cached and later reused by new threads,
308 * so memcg accounting is performed manually on assigning/releasing
309 * stacks to tasks. Drop __GFP_ACCOUNT.
311 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
312 VMALLOC_START, VMALLOC_END,
313 THREADINFO_GFP & ~__GFP_ACCOUNT,
315 0, node, __builtin_return_address(0));
319 vm = find_vm_area(stack);
320 if (memcg_charge_kernel_stack(vm)) {
325 * We can't call find_vm_area() in interrupt context, and
326 * free_thread_stack() can be called in interrupt context,
327 * so cache the vm_struct.
329 tsk->stack_vm_area = vm;
330 stack = kasan_reset_tag(stack);
335 static void free_thread_stack(struct task_struct *tsk)
337 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
338 thread_stack_delayed_free(tsk);
341 tsk->stack_vm_area = NULL;
344 # else /* !CONFIG_VMAP_STACK */
346 static void thread_stack_free_rcu(struct rcu_head *rh)
348 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
351 static void thread_stack_delayed_free(struct task_struct *tsk)
353 struct rcu_head *rh = tsk->stack;
355 call_rcu(rh, thread_stack_free_rcu);
358 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
360 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
364 tsk->stack = kasan_reset_tag(page_address(page));
370 static void free_thread_stack(struct task_struct *tsk)
372 thread_stack_delayed_free(tsk);
376 # endif /* CONFIG_VMAP_STACK */
377 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
379 static struct kmem_cache *thread_stack_cache;
381 static void thread_stack_free_rcu(struct rcu_head *rh)
383 kmem_cache_free(thread_stack_cache, rh);
386 static void thread_stack_delayed_free(struct task_struct *tsk)
388 struct rcu_head *rh = tsk->stack;
390 call_rcu(rh, thread_stack_free_rcu);
393 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
395 unsigned long *stack;
396 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
397 stack = kasan_reset_tag(stack);
399 return stack ? 0 : -ENOMEM;
402 static void free_thread_stack(struct task_struct *tsk)
404 thread_stack_delayed_free(tsk);
408 void thread_stack_cache_init(void)
410 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
411 THREAD_SIZE, THREAD_SIZE, 0, 0,
413 BUG_ON(thread_stack_cache == NULL);
416 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
417 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
419 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
421 unsigned long *stack;
423 stack = arch_alloc_thread_stack_node(tsk, node);
425 return stack ? 0 : -ENOMEM;
428 static void free_thread_stack(struct task_struct *tsk)
430 arch_free_thread_stack(tsk);
434 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
436 /* SLAB cache for signal_struct structures (tsk->signal) */
437 static struct kmem_cache *signal_cachep;
439 /* SLAB cache for sighand_struct structures (tsk->sighand) */
440 struct kmem_cache *sighand_cachep;
442 /* SLAB cache for files_struct structures (tsk->files) */
443 struct kmem_cache *files_cachep;
445 /* SLAB cache for fs_struct structures (tsk->fs) */
446 struct kmem_cache *fs_cachep;
448 /* SLAB cache for vm_area_struct structures */
449 static struct kmem_cache *vm_area_cachep;
451 /* SLAB cache for mm_struct structures (tsk->mm) */
452 static struct kmem_cache *mm_cachep;
454 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
456 struct vm_area_struct *vma;
458 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
464 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
466 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
469 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
470 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
472 * orig->shared.rb may be modified concurrently, but the clone
473 * will be reinitialized.
475 data_race(memcpy(new, orig, sizeof(*new)));
476 INIT_LIST_HEAD(&new->anon_vma_chain);
477 dup_anon_vma_name(orig, new);
482 void vm_area_free(struct vm_area_struct *vma)
484 free_anon_vma_name(vma);
485 kmem_cache_free(vm_area_cachep, vma);
488 static void account_kernel_stack(struct task_struct *tsk, int account)
490 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
491 struct vm_struct *vm = task_stack_vm_area(tsk);
494 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
495 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
496 account * (PAGE_SIZE / 1024));
498 void *stack = task_stack_page(tsk);
500 /* All stack pages are in the same node. */
501 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
502 account * (THREAD_SIZE / 1024));
506 void exit_task_stack_account(struct task_struct *tsk)
508 account_kernel_stack(tsk, -1);
510 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
511 struct vm_struct *vm;
514 vm = task_stack_vm_area(tsk);
515 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
516 memcg_kmem_uncharge_page(vm->pages[i], 0);
520 static void release_task_stack(struct task_struct *tsk)
522 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
523 return; /* Better to leak the stack than to free prematurely */
525 free_thread_stack(tsk);
528 #ifdef CONFIG_THREAD_INFO_IN_TASK
529 void put_task_stack(struct task_struct *tsk)
531 if (refcount_dec_and_test(&tsk->stack_refcount))
532 release_task_stack(tsk);
536 void free_task(struct task_struct *tsk)
538 #ifdef CONFIG_SECCOMP
539 WARN_ON_ONCE(tsk->seccomp.filter);
541 release_user_cpus_ptr(tsk);
544 #ifndef CONFIG_THREAD_INFO_IN_TASK
546 * The task is finally done with both the stack and thread_info,
549 release_task_stack(tsk);
552 * If the task had a separate stack allocation, it should be gone
555 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
557 rt_mutex_debug_task_free(tsk);
558 ftrace_graph_exit_task(tsk);
559 arch_release_task_struct(tsk);
560 if (tsk->flags & PF_KTHREAD)
561 free_kthread_struct(tsk);
562 free_task_struct(tsk);
564 EXPORT_SYMBOL(free_task);
566 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
568 struct file *exe_file;
570 exe_file = get_mm_exe_file(oldmm);
571 RCU_INIT_POINTER(mm->exe_file, exe_file);
573 * We depend on the oldmm having properly denied write access to the
576 if (exe_file && deny_write_access(exe_file))
577 pr_warn_once("deny_write_access() failed in %s\n", __func__);
581 static __latent_entropy int dup_mmap(struct mm_struct *mm,
582 struct mm_struct *oldmm)
584 struct vm_area_struct *mpnt, *tmp;
586 unsigned long charge = 0;
588 VMA_ITERATOR(old_vmi, oldmm, 0);
589 VMA_ITERATOR(vmi, mm, 0);
591 uprobe_start_dup_mmap();
592 if (mmap_write_lock_killable(oldmm)) {
594 goto fail_uprobe_end;
596 flush_cache_dup_mm(oldmm);
597 uprobe_dup_mmap(oldmm, mm);
599 * Not linked in yet - no deadlock potential:
601 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
603 /* No ordering required: file already has been exposed. */
604 dup_mm_exe_file(mm, oldmm);
606 mm->total_vm = oldmm->total_vm;
607 mm->data_vm = oldmm->data_vm;
608 mm->exec_vm = oldmm->exec_vm;
609 mm->stack_vm = oldmm->stack_vm;
611 retval = ksm_fork(mm, oldmm);
614 khugepaged_fork(mm, oldmm);
616 retval = vma_iter_bulk_alloc(&vmi, oldmm->map_count);
620 for_each_vma(old_vmi, mpnt) {
623 if (mpnt->vm_flags & VM_DONTCOPY) {
624 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
629 * Don't duplicate many vmas if we've been oom-killed (for
632 if (fatal_signal_pending(current)) {
636 if (mpnt->vm_flags & VM_ACCOUNT) {
637 unsigned long len = vma_pages(mpnt);
639 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
643 tmp = vm_area_dup(mpnt);
646 retval = vma_dup_policy(mpnt, tmp);
648 goto fail_nomem_policy;
650 retval = dup_userfaultfd(tmp, &uf);
652 goto fail_nomem_anon_vma_fork;
653 if (tmp->vm_flags & VM_WIPEONFORK) {
655 * VM_WIPEONFORK gets a clean slate in the child.
656 * Don't prepare anon_vma until fault since we don't
657 * copy page for current vma.
659 tmp->anon_vma = NULL;
660 } else if (anon_vma_fork(tmp, mpnt))
661 goto fail_nomem_anon_vma_fork;
662 vm_flags_clear(tmp, VM_LOCKED_MASK);
665 struct address_space *mapping = file->f_mapping;
668 i_mmap_lock_write(mapping);
669 if (tmp->vm_flags & VM_SHARED)
670 mapping_allow_writable(mapping);
671 flush_dcache_mmap_lock(mapping);
672 /* insert tmp into the share list, just after mpnt */
673 vma_interval_tree_insert_after(tmp, mpnt,
675 flush_dcache_mmap_unlock(mapping);
676 i_mmap_unlock_write(mapping);
680 * Copy/update hugetlb private vma information.
682 if (is_vm_hugetlb_page(tmp))
683 hugetlb_dup_vma_private(tmp);
685 /* Link the vma into the MT */
686 if (vma_iter_bulk_store(&vmi, tmp))
687 goto fail_nomem_vmi_store;
690 if (!(tmp->vm_flags & VM_WIPEONFORK))
691 retval = copy_page_range(tmp, mpnt);
693 if (tmp->vm_ops && tmp->vm_ops->open)
694 tmp->vm_ops->open(tmp);
699 /* a new mm has just been created */
700 retval = arch_dup_mmap(oldmm, mm);
704 mmap_write_unlock(mm);
706 mmap_write_unlock(oldmm);
707 dup_userfaultfd_complete(&uf);
709 uprobe_end_dup_mmap();
712 fail_nomem_vmi_store:
713 unlink_anon_vmas(tmp);
714 fail_nomem_anon_vma_fork:
715 mpol_put(vma_policy(tmp));
720 vm_unacct_memory(charge);
724 static inline int mm_alloc_pgd(struct mm_struct *mm)
726 mm->pgd = pgd_alloc(mm);
727 if (unlikely(!mm->pgd))
732 static inline void mm_free_pgd(struct mm_struct *mm)
734 pgd_free(mm, mm->pgd);
737 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
739 mmap_write_lock(oldmm);
740 dup_mm_exe_file(mm, oldmm);
741 mmap_write_unlock(oldmm);
744 #define mm_alloc_pgd(mm) (0)
745 #define mm_free_pgd(mm)
746 #endif /* CONFIG_MMU */
748 static void check_mm(struct mm_struct *mm)
752 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
753 "Please make sure 'struct resident_page_types[]' is updated as well");
755 for (i = 0; i < NR_MM_COUNTERS; i++) {
756 long x = percpu_counter_sum(&mm->rss_stat[i]);
759 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
760 mm, resident_page_types[i], x);
763 if (mm_pgtables_bytes(mm))
764 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
765 mm_pgtables_bytes(mm));
767 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
768 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
772 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
773 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
776 * Called when the last reference to the mm
777 * is dropped: either by a lazy thread or by
778 * mmput. Free the page directory and the mm.
780 void __mmdrop(struct mm_struct *mm)
784 BUG_ON(mm == &init_mm);
785 WARN_ON_ONCE(mm == current->mm);
786 WARN_ON_ONCE(mm == current->active_mm);
789 mmu_notifier_subscriptions_destroy(mm);
791 put_user_ns(mm->user_ns);
794 for (i = 0; i < NR_MM_COUNTERS; i++)
795 percpu_counter_destroy(&mm->rss_stat[i]);
798 EXPORT_SYMBOL_GPL(__mmdrop);
800 static void mmdrop_async_fn(struct work_struct *work)
802 struct mm_struct *mm;
804 mm = container_of(work, struct mm_struct, async_put_work);
808 static void mmdrop_async(struct mm_struct *mm)
810 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
811 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
812 schedule_work(&mm->async_put_work);
816 static inline void free_signal_struct(struct signal_struct *sig)
818 taskstats_tgid_free(sig);
819 sched_autogroup_exit(sig);
821 * __mmdrop is not safe to call from softirq context on x86 due to
822 * pgd_dtor so postpone it to the async context
825 mmdrop_async(sig->oom_mm);
826 kmem_cache_free(signal_cachep, sig);
829 static inline void put_signal_struct(struct signal_struct *sig)
831 if (refcount_dec_and_test(&sig->sigcnt))
832 free_signal_struct(sig);
835 void __put_task_struct(struct task_struct *tsk)
837 WARN_ON(!tsk->exit_state);
838 WARN_ON(refcount_read(&tsk->usage));
839 WARN_ON(tsk == current);
843 task_numa_free(tsk, true);
844 security_task_free(tsk);
845 bpf_task_storage_free(tsk);
847 delayacct_tsk_free(tsk);
848 put_signal_struct(tsk->signal);
849 sched_core_free(tsk);
852 EXPORT_SYMBOL_GPL(__put_task_struct);
854 void __init __weak arch_task_cache_init(void) { }
859 static void set_max_threads(unsigned int max_threads_suggested)
862 unsigned long nr_pages = totalram_pages();
865 * The number of threads shall be limited such that the thread
866 * structures may only consume a small part of the available memory.
868 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
869 threads = MAX_THREADS;
871 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
872 (u64) THREAD_SIZE * 8UL);
874 if (threads > max_threads_suggested)
875 threads = max_threads_suggested;
877 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
880 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
881 /* Initialized by the architecture: */
882 int arch_task_struct_size __read_mostly;
885 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
886 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
888 /* Fetch thread_struct whitelist for the architecture. */
889 arch_thread_struct_whitelist(offset, size);
892 * Handle zero-sized whitelist or empty thread_struct, otherwise
893 * adjust offset to position of thread_struct in task_struct.
895 if (unlikely(*size == 0))
898 *offset += offsetof(struct task_struct, thread);
900 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
902 void __init fork_init(void)
905 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
906 #ifndef ARCH_MIN_TASKALIGN
907 #define ARCH_MIN_TASKALIGN 0
909 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
910 unsigned long useroffset, usersize;
912 /* create a slab on which task_structs can be allocated */
913 task_struct_whitelist(&useroffset, &usersize);
914 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
915 arch_task_struct_size, align,
916 SLAB_PANIC|SLAB_ACCOUNT,
917 useroffset, usersize, NULL);
920 /* do the arch specific task caches init */
921 arch_task_cache_init();
923 set_max_threads(MAX_THREADS);
925 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
926 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
927 init_task.signal->rlim[RLIMIT_SIGPENDING] =
928 init_task.signal->rlim[RLIMIT_NPROC];
930 for (i = 0; i < UCOUNT_COUNTS; i++)
931 init_user_ns.ucount_max[i] = max_threads/2;
933 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
934 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
935 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
936 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
938 #ifdef CONFIG_VMAP_STACK
939 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
940 NULL, free_vm_stack_cache);
945 lockdep_init_task(&init_task);
949 int __weak arch_dup_task_struct(struct task_struct *dst,
950 struct task_struct *src)
956 void set_task_stack_end_magic(struct task_struct *tsk)
958 unsigned long *stackend;
960 stackend = end_of_stack(tsk);
961 *stackend = STACK_END_MAGIC; /* for overflow detection */
964 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
966 struct task_struct *tsk;
969 if (node == NUMA_NO_NODE)
970 node = tsk_fork_get_node(orig);
971 tsk = alloc_task_struct_node(node);
975 err = arch_dup_task_struct(tsk, orig);
979 err = alloc_thread_stack_node(tsk, node);
983 #ifdef CONFIG_THREAD_INFO_IN_TASK
984 refcount_set(&tsk->stack_refcount, 1);
986 account_kernel_stack(tsk, 1);
988 err = scs_prepare(tsk, node);
992 #ifdef CONFIG_SECCOMP
994 * We must handle setting up seccomp filters once we're under
995 * the sighand lock in case orig has changed between now and
996 * then. Until then, filter must be NULL to avoid messing up
997 * the usage counts on the error path calling free_task.
999 tsk->seccomp.filter = NULL;
1002 setup_thread_stack(tsk, orig);
1003 clear_user_return_notifier(tsk);
1004 clear_tsk_need_resched(tsk);
1005 set_task_stack_end_magic(tsk);
1006 clear_syscall_work_syscall_user_dispatch(tsk);
1008 #ifdef CONFIG_STACKPROTECTOR
1009 tsk->stack_canary = get_random_canary();
1011 if (orig->cpus_ptr == &orig->cpus_mask)
1012 tsk->cpus_ptr = &tsk->cpus_mask;
1013 dup_user_cpus_ptr(tsk, orig, node);
1016 * One for the user space visible state that goes away when reaped.
1017 * One for the scheduler.
1019 refcount_set(&tsk->rcu_users, 2);
1020 /* One for the rcu users */
1021 refcount_set(&tsk->usage, 1);
1022 #ifdef CONFIG_BLK_DEV_IO_TRACE
1023 tsk->btrace_seq = 0;
1025 tsk->splice_pipe = NULL;
1026 tsk->task_frag.page = NULL;
1027 tsk->wake_q.next = NULL;
1028 tsk->worker_private = NULL;
1030 kcov_task_init(tsk);
1031 kmsan_task_create(tsk);
1032 kmap_local_fork(tsk);
1034 #ifdef CONFIG_FAULT_INJECTION
1038 #ifdef CONFIG_BLK_CGROUP
1039 tsk->throttle_disk = NULL;
1040 tsk->use_memdelay = 0;
1043 #ifdef CONFIG_IOMMU_SVA
1044 tsk->pasid_activated = 0;
1048 tsk->active_memcg = NULL;
1051 #ifdef CONFIG_CPU_SUP_INTEL
1052 tsk->reported_split_lock = 0;
1055 #ifdef CONFIG_SCHED_MM_CID
1057 tsk->mm_cid_active = 0;
1062 exit_task_stack_account(tsk);
1063 free_thread_stack(tsk);
1065 free_task_struct(tsk);
1069 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1071 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1073 static int __init coredump_filter_setup(char *s)
1075 default_dump_filter =
1076 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1077 MMF_DUMP_FILTER_MASK;
1081 __setup("coredump_filter=", coredump_filter_setup);
1083 #include <linux/init_task.h>
1085 static void mm_init_aio(struct mm_struct *mm)
1088 spin_lock_init(&mm->ioctx_lock);
1089 mm->ioctx_table = NULL;
1093 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1094 struct task_struct *p)
1098 WRITE_ONCE(mm->owner, NULL);
1102 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1109 static void mm_init_uprobes_state(struct mm_struct *mm)
1111 #ifdef CONFIG_UPROBES
1112 mm->uprobes_state.xol_area = NULL;
1116 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1117 struct user_namespace *user_ns)
1121 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1122 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1123 atomic_set(&mm->mm_users, 1);
1124 atomic_set(&mm->mm_count, 1);
1125 seqcount_init(&mm->write_protect_seq);
1127 INIT_LIST_HEAD(&mm->mmlist);
1128 mm_pgtables_bytes_init(mm);
1131 atomic64_set(&mm->pinned_vm, 0);
1132 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1133 spin_lock_init(&mm->page_table_lock);
1134 spin_lock_init(&mm->arg_lock);
1135 mm_init_cpumask(mm);
1137 mm_init_owner(mm, p);
1139 RCU_INIT_POINTER(mm->exe_file, NULL);
1140 mmu_notifier_subscriptions_init(mm);
1141 init_tlb_flush_pending(mm);
1142 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1143 mm->pmd_huge_pte = NULL;
1145 mm_init_uprobes_state(mm);
1146 hugetlb_count_init(mm);
1149 mm->flags = current->mm->flags & MMF_INIT_MASK;
1150 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1152 mm->flags = default_dump_filter;
1156 if (mm_alloc_pgd(mm))
1159 if (init_new_context(p, mm))
1160 goto fail_nocontext;
1162 for (i = 0; i < NR_MM_COUNTERS; i++)
1163 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1166 mm->user_ns = get_user_ns(user_ns);
1167 lru_gen_init_mm(mm);
1173 percpu_counter_destroy(&mm->rss_stat[--i]);
1182 * Allocate and initialize an mm_struct.
1184 struct mm_struct *mm_alloc(void)
1186 struct mm_struct *mm;
1192 memset(mm, 0, sizeof(*mm));
1193 return mm_init(mm, current, current_user_ns());
1196 static inline void __mmput(struct mm_struct *mm)
1198 VM_BUG_ON(atomic_read(&mm->mm_users));
1200 uprobe_clear_state(mm);
1203 khugepaged_exit(mm); /* must run before exit_mmap */
1205 mm_put_huge_zero_page(mm);
1206 set_mm_exe_file(mm, NULL);
1207 if (!list_empty(&mm->mmlist)) {
1208 spin_lock(&mmlist_lock);
1209 list_del(&mm->mmlist);
1210 spin_unlock(&mmlist_lock);
1213 module_put(mm->binfmt->module);
1219 * Decrement the use count and release all resources for an mm.
1221 void mmput(struct mm_struct *mm)
1225 if (atomic_dec_and_test(&mm->mm_users))
1228 EXPORT_SYMBOL_GPL(mmput);
1231 static void mmput_async_fn(struct work_struct *work)
1233 struct mm_struct *mm = container_of(work, struct mm_struct,
1239 void mmput_async(struct mm_struct *mm)
1241 if (atomic_dec_and_test(&mm->mm_users)) {
1242 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1243 schedule_work(&mm->async_put_work);
1246 EXPORT_SYMBOL_GPL(mmput_async);
1250 * set_mm_exe_file - change a reference to the mm's executable file
1252 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1254 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1255 * invocations: in mmput() nobody alive left, in execve task is single
1258 * Can only fail if new_exe_file != NULL.
1260 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1262 struct file *old_exe_file;
1265 * It is safe to dereference the exe_file without RCU as
1266 * this function is only called if nobody else can access
1267 * this mm -- see comment above for justification.
1269 old_exe_file = rcu_dereference_raw(mm->exe_file);
1273 * We expect the caller (i.e., sys_execve) to already denied
1274 * write access, so this is unlikely to fail.
1276 if (unlikely(deny_write_access(new_exe_file)))
1278 get_file(new_exe_file);
1280 rcu_assign_pointer(mm->exe_file, new_exe_file);
1282 allow_write_access(old_exe_file);
1289 * replace_mm_exe_file - replace a reference to the mm's executable file
1291 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1292 * dealing with concurrent invocation and without grabbing the mmap lock in
1295 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1297 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1299 struct vm_area_struct *vma;
1300 struct file *old_exe_file;
1303 /* Forbid mm->exe_file change if old file still mapped. */
1304 old_exe_file = get_mm_exe_file(mm);
1306 VMA_ITERATOR(vmi, mm, 0);
1308 for_each_vma(vmi, vma) {
1311 if (path_equal(&vma->vm_file->f_path,
1312 &old_exe_file->f_path)) {
1317 mmap_read_unlock(mm);
1323 /* set the new file, lockless */
1324 ret = deny_write_access(new_exe_file);
1327 get_file(new_exe_file);
1329 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1332 * Don't race with dup_mmap() getting the file and disallowing
1333 * write access while someone might open the file writable.
1336 allow_write_access(old_exe_file);
1338 mmap_read_unlock(mm);
1344 * get_mm_exe_file - acquire a reference to the mm's executable file
1346 * Returns %NULL if mm has no associated executable file.
1347 * User must release file via fput().
1349 struct file *get_mm_exe_file(struct mm_struct *mm)
1351 struct file *exe_file;
1354 exe_file = rcu_dereference(mm->exe_file);
1355 if (exe_file && !get_file_rcu(exe_file))
1362 * get_task_exe_file - acquire a reference to the task's executable file
1364 * Returns %NULL if task's mm (if any) has no associated executable file or
1365 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1366 * User must release file via fput().
1368 struct file *get_task_exe_file(struct task_struct *task)
1370 struct file *exe_file = NULL;
1371 struct mm_struct *mm;
1376 if (!(task->flags & PF_KTHREAD))
1377 exe_file = get_mm_exe_file(mm);
1384 * get_task_mm - acquire a reference to the task's mm
1386 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1387 * this kernel workthread has transiently adopted a user mm with use_mm,
1388 * to do its AIO) is not set and if so returns a reference to it, after
1389 * bumping up the use count. User must release the mm via mmput()
1390 * after use. Typically used by /proc and ptrace.
1392 struct mm_struct *get_task_mm(struct task_struct *task)
1394 struct mm_struct *mm;
1399 if (task->flags & PF_KTHREAD)
1407 EXPORT_SYMBOL_GPL(get_task_mm);
1409 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1411 struct mm_struct *mm;
1414 err = down_read_killable(&task->signal->exec_update_lock);
1416 return ERR_PTR(err);
1418 mm = get_task_mm(task);
1419 if (mm && mm != current->mm &&
1420 !ptrace_may_access(task, mode)) {
1422 mm = ERR_PTR(-EACCES);
1424 up_read(&task->signal->exec_update_lock);
1429 static void complete_vfork_done(struct task_struct *tsk)
1431 struct completion *vfork;
1434 vfork = tsk->vfork_done;
1435 if (likely(vfork)) {
1436 tsk->vfork_done = NULL;
1442 static int wait_for_vfork_done(struct task_struct *child,
1443 struct completion *vfork)
1445 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1448 cgroup_enter_frozen();
1449 killed = wait_for_completion_state(vfork, state);
1450 cgroup_leave_frozen(false);
1454 child->vfork_done = NULL;
1458 put_task_struct(child);
1462 /* Please note the differences between mmput and mm_release.
1463 * mmput is called whenever we stop holding onto a mm_struct,
1464 * error success whatever.
1466 * mm_release is called after a mm_struct has been removed
1467 * from the current process.
1469 * This difference is important for error handling, when we
1470 * only half set up a mm_struct for a new process and need to restore
1471 * the old one. Because we mmput the new mm_struct before
1472 * restoring the old one. . .
1473 * Eric Biederman 10 January 1998
1475 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1477 uprobe_free_utask(tsk);
1479 /* Get rid of any cached register state */
1480 deactivate_mm(tsk, mm);
1483 * Signal userspace if we're not exiting with a core dump
1484 * because we want to leave the value intact for debugging
1487 if (tsk->clear_child_tid) {
1488 if (atomic_read(&mm->mm_users) > 1) {
1490 * We don't check the error code - if userspace has
1491 * not set up a proper pointer then tough luck.
1493 put_user(0, tsk->clear_child_tid);
1494 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1495 1, NULL, NULL, 0, 0);
1497 tsk->clear_child_tid = NULL;
1501 * All done, finally we can wake up parent and return this mm to him.
1502 * Also kthread_stop() uses this completion for synchronization.
1504 if (tsk->vfork_done)
1505 complete_vfork_done(tsk);
1508 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1510 futex_exit_release(tsk);
1511 mm_release(tsk, mm);
1514 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1516 futex_exec_release(tsk);
1517 mm_release(tsk, mm);
1521 * dup_mm() - duplicates an existing mm structure
1522 * @tsk: the task_struct with which the new mm will be associated.
1523 * @oldmm: the mm to duplicate.
1525 * Allocates a new mm structure and duplicates the provided @oldmm structure
1528 * Return: the duplicated mm or NULL on failure.
1530 static struct mm_struct *dup_mm(struct task_struct *tsk,
1531 struct mm_struct *oldmm)
1533 struct mm_struct *mm;
1540 memcpy(mm, oldmm, sizeof(*mm));
1542 if (!mm_init(mm, tsk, mm->user_ns))
1545 err = dup_mmap(mm, oldmm);
1549 mm->hiwater_rss = get_mm_rss(mm);
1550 mm->hiwater_vm = mm->total_vm;
1552 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1558 /* don't put binfmt in mmput, we haven't got module yet */
1560 mm_init_owner(mm, NULL);
1567 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1569 struct mm_struct *mm, *oldmm;
1571 tsk->min_flt = tsk->maj_flt = 0;
1572 tsk->nvcsw = tsk->nivcsw = 0;
1573 #ifdef CONFIG_DETECT_HUNG_TASK
1574 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1575 tsk->last_switch_time = 0;
1579 tsk->active_mm = NULL;
1582 * Are we cloning a kernel thread?
1584 * We need to steal a active VM for that..
1586 oldmm = current->mm;
1590 if (clone_flags & CLONE_VM) {
1594 mm = dup_mm(tsk, current->mm);
1600 tsk->active_mm = mm;
1601 sched_mm_cid_fork(tsk);
1605 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1607 struct fs_struct *fs = current->fs;
1608 if (clone_flags & CLONE_FS) {
1609 /* tsk->fs is already what we want */
1610 spin_lock(&fs->lock);
1612 spin_unlock(&fs->lock);
1616 spin_unlock(&fs->lock);
1619 tsk->fs = copy_fs_struct(fs);
1625 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1627 struct files_struct *oldf, *newf;
1631 * A background process may not have any files ...
1633 oldf = current->files;
1637 if (clone_flags & CLONE_FILES) {
1638 atomic_inc(&oldf->count);
1642 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1652 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1654 struct sighand_struct *sig;
1656 if (clone_flags & CLONE_SIGHAND) {
1657 refcount_inc(¤t->sighand->count);
1660 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1661 RCU_INIT_POINTER(tsk->sighand, sig);
1665 refcount_set(&sig->count, 1);
1666 spin_lock_irq(¤t->sighand->siglock);
1667 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1668 spin_unlock_irq(¤t->sighand->siglock);
1670 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1671 if (clone_flags & CLONE_CLEAR_SIGHAND)
1672 flush_signal_handlers(tsk, 0);
1677 void __cleanup_sighand(struct sighand_struct *sighand)
1679 if (refcount_dec_and_test(&sighand->count)) {
1680 signalfd_cleanup(sighand);
1682 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1683 * without an RCU grace period, see __lock_task_sighand().
1685 kmem_cache_free(sighand_cachep, sighand);
1690 * Initialize POSIX timer handling for a thread group.
1692 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1694 struct posix_cputimers *pct = &sig->posix_cputimers;
1695 unsigned long cpu_limit;
1697 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1698 posix_cputimers_group_init(pct, cpu_limit);
1701 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1703 struct signal_struct *sig;
1705 if (clone_flags & CLONE_THREAD)
1708 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1713 sig->nr_threads = 1;
1714 sig->quick_threads = 1;
1715 atomic_set(&sig->live, 1);
1716 refcount_set(&sig->sigcnt, 1);
1718 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1719 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1720 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1722 init_waitqueue_head(&sig->wait_chldexit);
1723 sig->curr_target = tsk;
1724 init_sigpending(&sig->shared_pending);
1725 INIT_HLIST_HEAD(&sig->multiprocess);
1726 seqlock_init(&sig->stats_lock);
1727 prev_cputime_init(&sig->prev_cputime);
1729 #ifdef CONFIG_POSIX_TIMERS
1730 INIT_LIST_HEAD(&sig->posix_timers);
1731 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1732 sig->real_timer.function = it_real_fn;
1735 task_lock(current->group_leader);
1736 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1737 task_unlock(current->group_leader);
1739 posix_cpu_timers_init_group(sig);
1741 tty_audit_fork(sig);
1742 sched_autogroup_fork(sig);
1744 sig->oom_score_adj = current->signal->oom_score_adj;
1745 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1747 mutex_init(&sig->cred_guard_mutex);
1748 init_rwsem(&sig->exec_update_lock);
1753 static void copy_seccomp(struct task_struct *p)
1755 #ifdef CONFIG_SECCOMP
1757 * Must be called with sighand->lock held, which is common to
1758 * all threads in the group. Holding cred_guard_mutex is not
1759 * needed because this new task is not yet running and cannot
1762 assert_spin_locked(¤t->sighand->siglock);
1764 /* Ref-count the new filter user, and assign it. */
1765 get_seccomp_filter(current);
1766 p->seccomp = current->seccomp;
1769 * Explicitly enable no_new_privs here in case it got set
1770 * between the task_struct being duplicated and holding the
1771 * sighand lock. The seccomp state and nnp must be in sync.
1773 if (task_no_new_privs(current))
1774 task_set_no_new_privs(p);
1777 * If the parent gained a seccomp mode after copying thread
1778 * flags and between before we held the sighand lock, we have
1779 * to manually enable the seccomp thread flag here.
1781 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1782 set_task_syscall_work(p, SECCOMP);
1786 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1788 current->clear_child_tid = tidptr;
1790 return task_pid_vnr(current);
1793 static void rt_mutex_init_task(struct task_struct *p)
1795 raw_spin_lock_init(&p->pi_lock);
1796 #ifdef CONFIG_RT_MUTEXES
1797 p->pi_waiters = RB_ROOT_CACHED;
1798 p->pi_top_task = NULL;
1799 p->pi_blocked_on = NULL;
1803 static inline void init_task_pid_links(struct task_struct *task)
1807 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1808 INIT_HLIST_NODE(&task->pid_links[type]);
1812 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1814 if (type == PIDTYPE_PID)
1815 task->thread_pid = pid;
1817 task->signal->pids[type] = pid;
1820 static inline void rcu_copy_process(struct task_struct *p)
1822 #ifdef CONFIG_PREEMPT_RCU
1823 p->rcu_read_lock_nesting = 0;
1824 p->rcu_read_unlock_special.s = 0;
1825 p->rcu_blocked_node = NULL;
1826 INIT_LIST_HEAD(&p->rcu_node_entry);
1827 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1828 #ifdef CONFIG_TASKS_RCU
1829 p->rcu_tasks_holdout = false;
1830 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1831 p->rcu_tasks_idle_cpu = -1;
1832 #endif /* #ifdef CONFIG_TASKS_RCU */
1833 #ifdef CONFIG_TASKS_TRACE_RCU
1834 p->trc_reader_nesting = 0;
1835 p->trc_reader_special.s = 0;
1836 INIT_LIST_HEAD(&p->trc_holdout_list);
1837 INIT_LIST_HEAD(&p->trc_blkd_node);
1838 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1841 struct pid *pidfd_pid(const struct file *file)
1843 if (file->f_op == &pidfd_fops)
1844 return file->private_data;
1846 return ERR_PTR(-EBADF);
1849 static int pidfd_release(struct inode *inode, struct file *file)
1851 struct pid *pid = file->private_data;
1853 file->private_data = NULL;
1858 #ifdef CONFIG_PROC_FS
1860 * pidfd_show_fdinfo - print information about a pidfd
1861 * @m: proc fdinfo file
1862 * @f: file referencing a pidfd
1865 * This function will print the pid that a given pidfd refers to in the
1866 * pid namespace of the procfs instance.
1867 * If the pid namespace of the process is not a descendant of the pid
1868 * namespace of the procfs instance 0 will be shown as its pid. This is
1869 * similar to calling getppid() on a process whose parent is outside of
1870 * its pid namespace.
1873 * If pid namespaces are supported then this function will also print
1874 * the pid of a given pidfd refers to for all descendant pid namespaces
1875 * starting from the current pid namespace of the instance, i.e. the
1876 * Pid field and the first entry in the NSpid field will be identical.
1877 * If the pid namespace of the process is not a descendant of the pid
1878 * namespace of the procfs instance 0 will be shown as its first NSpid
1879 * entry and no others will be shown.
1880 * Note that this differs from the Pid and NSpid fields in
1881 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1882 * the pid namespace of the procfs instance. The difference becomes
1883 * obvious when sending around a pidfd between pid namespaces from a
1884 * different branch of the tree, i.e. where no ancestral relation is
1885 * present between the pid namespaces:
1886 * - create two new pid namespaces ns1 and ns2 in the initial pid
1887 * namespace (also take care to create new mount namespaces in the
1888 * new pid namespace and mount procfs)
1889 * - create a process with a pidfd in ns1
1890 * - send pidfd from ns1 to ns2
1891 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1892 * have exactly one entry, which is 0
1894 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1896 struct pid *pid = f->private_data;
1897 struct pid_namespace *ns;
1900 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1901 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1902 nr = pid_nr_ns(pid, ns);
1905 seq_put_decimal_ll(m, "Pid:\t", nr);
1907 #ifdef CONFIG_PID_NS
1908 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1912 /* If nr is non-zero it means that 'pid' is valid and that
1913 * ns, i.e. the pid namespace associated with the procfs
1914 * instance, is in the pid namespace hierarchy of pid.
1915 * Start at one below the already printed level.
1917 for (i = ns->level + 1; i <= pid->level; i++)
1918 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1926 * Poll support for process exit notification.
1928 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1930 struct pid *pid = file->private_data;
1931 __poll_t poll_flags = 0;
1933 poll_wait(file, &pid->wait_pidfd, pts);
1936 * Inform pollers only when the whole thread group exits.
1937 * If the thread group leader exits before all other threads in the
1938 * group, then poll(2) should block, similar to the wait(2) family.
1940 if (thread_group_exited(pid))
1941 poll_flags = EPOLLIN | EPOLLRDNORM;
1946 const struct file_operations pidfd_fops = {
1947 .release = pidfd_release,
1949 #ifdef CONFIG_PROC_FS
1950 .show_fdinfo = pidfd_show_fdinfo,
1954 static void __delayed_free_task(struct rcu_head *rhp)
1956 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1961 static __always_inline void delayed_free_task(struct task_struct *tsk)
1963 if (IS_ENABLED(CONFIG_MEMCG))
1964 call_rcu(&tsk->rcu, __delayed_free_task);
1969 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1971 /* Skip if kernel thread */
1975 /* Skip if spawning a thread or using vfork */
1976 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1979 /* We need to synchronize with __set_oom_adj */
1980 mutex_lock(&oom_adj_mutex);
1981 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1982 /* Update the values in case they were changed after copy_signal */
1983 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1984 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1985 mutex_unlock(&oom_adj_mutex);
1989 static void rv_task_fork(struct task_struct *p)
1993 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
1994 p->rv[i].da_mon.monitoring = false;
1997 #define rv_task_fork(p) do {} while (0)
2001 * This creates a new process as a copy of the old one,
2002 * but does not actually start it yet.
2004 * It copies the registers, and all the appropriate
2005 * parts of the process environment (as per the clone
2006 * flags). The actual kick-off is left to the caller.
2008 static __latent_entropy struct task_struct *copy_process(
2012 struct kernel_clone_args *args)
2014 int pidfd = -1, retval;
2015 struct task_struct *p;
2016 struct multiprocess_signals delayed;
2017 struct file *pidfile = NULL;
2018 const u64 clone_flags = args->flags;
2019 struct nsproxy *nsp = current->nsproxy;
2022 * Don't allow sharing the root directory with processes in a different
2025 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2026 return ERR_PTR(-EINVAL);
2028 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2029 return ERR_PTR(-EINVAL);
2032 * Thread groups must share signals as well, and detached threads
2033 * can only be started up within the thread group.
2035 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2036 return ERR_PTR(-EINVAL);
2039 * Shared signal handlers imply shared VM. By way of the above,
2040 * thread groups also imply shared VM. Blocking this case allows
2041 * for various simplifications in other code.
2043 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2044 return ERR_PTR(-EINVAL);
2047 * Siblings of global init remain as zombies on exit since they are
2048 * not reaped by their parent (swapper). To solve this and to avoid
2049 * multi-rooted process trees, prevent global and container-inits
2050 * from creating siblings.
2052 if ((clone_flags & CLONE_PARENT) &&
2053 current->signal->flags & SIGNAL_UNKILLABLE)
2054 return ERR_PTR(-EINVAL);
2057 * If the new process will be in a different pid or user namespace
2058 * do not allow it to share a thread group with the forking task.
2060 if (clone_flags & CLONE_THREAD) {
2061 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2062 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2063 return ERR_PTR(-EINVAL);
2066 if (clone_flags & CLONE_PIDFD) {
2068 * - CLONE_DETACHED is blocked so that we can potentially
2069 * reuse it later for CLONE_PIDFD.
2070 * - CLONE_THREAD is blocked until someone really needs it.
2072 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2073 return ERR_PTR(-EINVAL);
2077 * Force any signals received before this point to be delivered
2078 * before the fork happens. Collect up signals sent to multiple
2079 * processes that happen during the fork and delay them so that
2080 * they appear to happen after the fork.
2082 sigemptyset(&delayed.signal);
2083 INIT_HLIST_NODE(&delayed.node);
2085 spin_lock_irq(¤t->sighand->siglock);
2086 if (!(clone_flags & CLONE_THREAD))
2087 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2088 recalc_sigpending();
2089 spin_unlock_irq(¤t->sighand->siglock);
2090 retval = -ERESTARTNOINTR;
2091 if (task_sigpending(current))
2095 p = dup_task_struct(current, node);
2098 p->flags &= ~PF_KTHREAD;
2100 p->flags |= PF_KTHREAD;
2101 if (args->io_thread) {
2103 * Mark us an IO worker, and block any signal that isn't
2106 p->flags |= PF_IO_WORKER;
2107 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2110 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2112 * Clear TID on mm_release()?
2114 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2116 ftrace_graph_init_task(p);
2118 rt_mutex_init_task(p);
2120 lockdep_assert_irqs_enabled();
2121 #ifdef CONFIG_PROVE_LOCKING
2122 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2124 retval = copy_creds(p, clone_flags);
2129 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2130 if (p->real_cred->user != INIT_USER &&
2131 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2132 goto bad_fork_cleanup_count;
2134 current->flags &= ~PF_NPROC_EXCEEDED;
2137 * If multiple threads are within copy_process(), then this check
2138 * triggers too late. This doesn't hurt, the check is only there
2139 * to stop root fork bombs.
2142 if (data_race(nr_threads >= max_threads))
2143 goto bad_fork_cleanup_count;
2145 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2146 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2147 p->flags |= PF_FORKNOEXEC;
2148 INIT_LIST_HEAD(&p->children);
2149 INIT_LIST_HEAD(&p->sibling);
2150 rcu_copy_process(p);
2151 p->vfork_done = NULL;
2152 spin_lock_init(&p->alloc_lock);
2154 init_sigpending(&p->pending);
2156 p->utime = p->stime = p->gtime = 0;
2157 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2158 p->utimescaled = p->stimescaled = 0;
2160 prev_cputime_init(&p->prev_cputime);
2162 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2163 seqcount_init(&p->vtime.seqcount);
2164 p->vtime.starttime = 0;
2165 p->vtime.state = VTIME_INACTIVE;
2168 #ifdef CONFIG_IO_URING
2172 #if defined(SPLIT_RSS_COUNTING)
2173 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2176 p->default_timer_slack_ns = current->timer_slack_ns;
2182 task_io_accounting_init(&p->ioac);
2183 acct_clear_integrals(p);
2185 posix_cputimers_init(&p->posix_cputimers);
2187 p->io_context = NULL;
2188 audit_set_context(p, NULL);
2190 if (args->kthread) {
2191 if (!set_kthread_struct(p))
2192 goto bad_fork_cleanup_delayacct;
2195 p->mempolicy = mpol_dup(p->mempolicy);
2196 if (IS_ERR(p->mempolicy)) {
2197 retval = PTR_ERR(p->mempolicy);
2198 p->mempolicy = NULL;
2199 goto bad_fork_cleanup_delayacct;
2202 #ifdef CONFIG_CPUSETS
2203 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2204 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2205 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2207 #ifdef CONFIG_TRACE_IRQFLAGS
2208 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2209 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2210 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2211 p->softirqs_enabled = 1;
2212 p->softirq_context = 0;
2215 p->pagefault_disabled = 0;
2217 #ifdef CONFIG_LOCKDEP
2218 lockdep_init_task(p);
2221 #ifdef CONFIG_DEBUG_MUTEXES
2222 p->blocked_on = NULL; /* not blocked yet */
2224 #ifdef CONFIG_BCACHE
2225 p->sequential_io = 0;
2226 p->sequential_io_avg = 0;
2228 #ifdef CONFIG_BPF_SYSCALL
2229 RCU_INIT_POINTER(p->bpf_storage, NULL);
2233 /* Perform scheduler related setup. Assign this task to a CPU. */
2234 retval = sched_fork(clone_flags, p);
2236 goto bad_fork_cleanup_policy;
2238 retval = perf_event_init_task(p, clone_flags);
2240 goto bad_fork_cleanup_policy;
2241 retval = audit_alloc(p);
2243 goto bad_fork_cleanup_perf;
2244 /* copy all the process information */
2246 retval = security_task_alloc(p, clone_flags);
2248 goto bad_fork_cleanup_audit;
2249 retval = copy_semundo(clone_flags, p);
2251 goto bad_fork_cleanup_security;
2252 retval = copy_files(clone_flags, p);
2254 goto bad_fork_cleanup_semundo;
2255 retval = copy_fs(clone_flags, p);
2257 goto bad_fork_cleanup_files;
2258 retval = copy_sighand(clone_flags, p);
2260 goto bad_fork_cleanup_fs;
2261 retval = copy_signal(clone_flags, p);
2263 goto bad_fork_cleanup_sighand;
2264 retval = copy_mm(clone_flags, p);
2266 goto bad_fork_cleanup_signal;
2267 retval = copy_namespaces(clone_flags, p);
2269 goto bad_fork_cleanup_mm;
2270 retval = copy_io(clone_flags, p);
2272 goto bad_fork_cleanup_namespaces;
2273 retval = copy_thread(p, args);
2275 goto bad_fork_cleanup_io;
2277 stackleak_task_init(p);
2279 if (pid != &init_struct_pid) {
2280 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2281 args->set_tid_size);
2283 retval = PTR_ERR(pid);
2284 goto bad_fork_cleanup_thread;
2289 * This has to happen after we've potentially unshared the file
2290 * descriptor table (so that the pidfd doesn't leak into the child
2291 * if the fd table isn't shared).
2293 if (clone_flags & CLONE_PIDFD) {
2294 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2296 goto bad_fork_free_pid;
2300 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2301 O_RDWR | O_CLOEXEC);
2302 if (IS_ERR(pidfile)) {
2303 put_unused_fd(pidfd);
2304 retval = PTR_ERR(pidfile);
2305 goto bad_fork_free_pid;
2307 get_pid(pid); /* held by pidfile now */
2309 retval = put_user(pidfd, args->pidfd);
2311 goto bad_fork_put_pidfd;
2320 * sigaltstack should be cleared when sharing the same VM
2322 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2326 * Syscall tracing and stepping should be turned off in the
2327 * child regardless of CLONE_PTRACE.
2329 user_disable_single_step(p);
2330 clear_task_syscall_work(p, SYSCALL_TRACE);
2331 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2332 clear_task_syscall_work(p, SYSCALL_EMU);
2334 clear_tsk_latency_tracing(p);
2336 /* ok, now we should be set up.. */
2337 p->pid = pid_nr(pid);
2338 if (clone_flags & CLONE_THREAD) {
2339 p->group_leader = current->group_leader;
2340 p->tgid = current->tgid;
2342 p->group_leader = p;
2347 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2348 p->dirty_paused_when = 0;
2350 p->pdeath_signal = 0;
2351 INIT_LIST_HEAD(&p->thread_group);
2352 p->task_works = NULL;
2353 clear_posix_cputimers_work(p);
2355 #ifdef CONFIG_KRETPROBES
2356 p->kretprobe_instances.first = NULL;
2358 #ifdef CONFIG_RETHOOK
2359 p->rethooks.first = NULL;
2363 * Ensure that the cgroup subsystem policies allow the new process to be
2364 * forked. It should be noted that the new process's css_set can be changed
2365 * between here and cgroup_post_fork() if an organisation operation is in
2368 retval = cgroup_can_fork(p, args);
2370 goto bad_fork_put_pidfd;
2373 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2374 * the new task on the correct runqueue. All this *before* the task
2377 * This isn't part of ->can_fork() because while the re-cloning is
2378 * cgroup specific, it unconditionally needs to place the task on a
2381 sched_cgroup_fork(p, args);
2384 * From this point on we must avoid any synchronous user-space
2385 * communication until we take the tasklist-lock. In particular, we do
2386 * not want user-space to be able to predict the process start-time by
2387 * stalling fork(2) after we recorded the start_time but before it is
2388 * visible to the system.
2391 p->start_time = ktime_get_ns();
2392 p->start_boottime = ktime_get_boottime_ns();
2395 * Make it visible to the rest of the system, but dont wake it up yet.
2396 * Need tasklist lock for parent etc handling!
2398 write_lock_irq(&tasklist_lock);
2400 /* CLONE_PARENT re-uses the old parent */
2401 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2402 p->real_parent = current->real_parent;
2403 p->parent_exec_id = current->parent_exec_id;
2404 if (clone_flags & CLONE_THREAD)
2405 p->exit_signal = -1;
2407 p->exit_signal = current->group_leader->exit_signal;
2409 p->real_parent = current;
2410 p->parent_exec_id = current->self_exec_id;
2411 p->exit_signal = args->exit_signal;
2414 klp_copy_process(p);
2418 spin_lock(¤t->sighand->siglock);
2422 rseq_fork(p, clone_flags);
2424 /* Don't start children in a dying pid namespace */
2425 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2427 goto bad_fork_cancel_cgroup;
2430 /* Let kill terminate clone/fork in the middle */
2431 if (fatal_signal_pending(current)) {
2433 goto bad_fork_cancel_cgroup;
2436 /* No more failure paths after this point. */
2439 * Copy seccomp details explicitly here, in case they were changed
2440 * before holding sighand lock.
2444 init_task_pid_links(p);
2445 if (likely(p->pid)) {
2446 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2448 init_task_pid(p, PIDTYPE_PID, pid);
2449 if (thread_group_leader(p)) {
2450 init_task_pid(p, PIDTYPE_TGID, pid);
2451 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2452 init_task_pid(p, PIDTYPE_SID, task_session(current));
2454 if (is_child_reaper(pid)) {
2455 ns_of_pid(pid)->child_reaper = p;
2456 p->signal->flags |= SIGNAL_UNKILLABLE;
2458 p->signal->shared_pending.signal = delayed.signal;
2459 p->signal->tty = tty_kref_get(current->signal->tty);
2461 * Inherit has_child_subreaper flag under the same
2462 * tasklist_lock with adding child to the process tree
2463 * for propagate_has_child_subreaper optimization.
2465 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2466 p->real_parent->signal->is_child_subreaper;
2467 list_add_tail(&p->sibling, &p->real_parent->children);
2468 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2469 attach_pid(p, PIDTYPE_TGID);
2470 attach_pid(p, PIDTYPE_PGID);
2471 attach_pid(p, PIDTYPE_SID);
2472 __this_cpu_inc(process_counts);
2474 current->signal->nr_threads++;
2475 current->signal->quick_threads++;
2476 atomic_inc(¤t->signal->live);
2477 refcount_inc(¤t->signal->sigcnt);
2478 task_join_group_stop(p);
2479 list_add_tail_rcu(&p->thread_group,
2480 &p->group_leader->thread_group);
2481 list_add_tail_rcu(&p->thread_node,
2482 &p->signal->thread_head);
2484 attach_pid(p, PIDTYPE_PID);
2488 hlist_del_init(&delayed.node);
2489 spin_unlock(¤t->sighand->siglock);
2490 syscall_tracepoint_update(p);
2491 write_unlock_irq(&tasklist_lock);
2494 fd_install(pidfd, pidfile);
2496 proc_fork_connector(p);
2498 cgroup_post_fork(p, args);
2501 trace_task_newtask(p, clone_flags);
2502 uprobe_copy_process(p, clone_flags);
2504 copy_oom_score_adj(clone_flags, p);
2508 bad_fork_cancel_cgroup:
2510 spin_unlock(¤t->sighand->siglock);
2511 write_unlock_irq(&tasklist_lock);
2512 cgroup_cancel_fork(p, args);
2514 if (clone_flags & CLONE_PIDFD) {
2516 put_unused_fd(pidfd);
2519 if (pid != &init_struct_pid)
2521 bad_fork_cleanup_thread:
2523 bad_fork_cleanup_io:
2526 bad_fork_cleanup_namespaces:
2527 exit_task_namespaces(p);
2528 bad_fork_cleanup_mm:
2530 mm_clear_owner(p->mm, p);
2533 bad_fork_cleanup_signal:
2534 if (!(clone_flags & CLONE_THREAD))
2535 free_signal_struct(p->signal);
2536 bad_fork_cleanup_sighand:
2537 __cleanup_sighand(p->sighand);
2538 bad_fork_cleanup_fs:
2539 exit_fs(p); /* blocking */
2540 bad_fork_cleanup_files:
2541 exit_files(p); /* blocking */
2542 bad_fork_cleanup_semundo:
2544 bad_fork_cleanup_security:
2545 security_task_free(p);
2546 bad_fork_cleanup_audit:
2548 bad_fork_cleanup_perf:
2549 perf_event_free_task(p);
2550 bad_fork_cleanup_policy:
2551 lockdep_free_task(p);
2553 mpol_put(p->mempolicy);
2555 bad_fork_cleanup_delayacct:
2556 delayacct_tsk_free(p);
2557 bad_fork_cleanup_count:
2558 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2561 WRITE_ONCE(p->__state, TASK_DEAD);
2562 exit_task_stack_account(p);
2564 delayed_free_task(p);
2566 spin_lock_irq(¤t->sighand->siglock);
2567 hlist_del_init(&delayed.node);
2568 spin_unlock_irq(¤t->sighand->siglock);
2569 return ERR_PTR(retval);
2572 static inline void init_idle_pids(struct task_struct *idle)
2576 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2577 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2578 init_task_pid(idle, type, &init_struct_pid);
2582 static int idle_dummy(void *dummy)
2584 /* This function is never called */
2588 struct task_struct * __init fork_idle(int cpu)
2590 struct task_struct *task;
2591 struct kernel_clone_args args = {
2599 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2600 if (!IS_ERR(task)) {
2601 init_idle_pids(task);
2602 init_idle(task, cpu);
2609 * This is like kernel_clone(), but shaved down and tailored to just
2610 * creating io_uring workers. It returns a created task, or an error pointer.
2611 * The returned task is inactive, and the caller must fire it up through
2612 * wake_up_new_task(p). All signals are blocked in the created task.
2614 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2616 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2618 struct kernel_clone_args args = {
2619 .flags = ((lower_32_bits(flags) | CLONE_VM |
2620 CLONE_UNTRACED) & ~CSIGNAL),
2621 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2627 return copy_process(NULL, 0, node, &args);
2631 * Ok, this is the main fork-routine.
2633 * It copies the process, and if successful kick-starts
2634 * it and waits for it to finish using the VM if required.
2636 * args->exit_signal is expected to be checked for sanity by the caller.
2638 pid_t kernel_clone(struct kernel_clone_args *args)
2640 u64 clone_flags = args->flags;
2641 struct completion vfork;
2643 struct task_struct *p;
2648 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2649 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2650 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2651 * field in struct clone_args and it still doesn't make sense to have
2652 * them both point at the same memory location. Performing this check
2653 * here has the advantage that we don't need to have a separate helper
2654 * to check for legacy clone().
2656 if ((args->flags & CLONE_PIDFD) &&
2657 (args->flags & CLONE_PARENT_SETTID) &&
2658 (args->pidfd == args->parent_tid))
2662 * Determine whether and which event to report to ptracer. When
2663 * called from kernel_thread or CLONE_UNTRACED is explicitly
2664 * requested, no event is reported; otherwise, report if the event
2665 * for the type of forking is enabled.
2667 if (!(clone_flags & CLONE_UNTRACED)) {
2668 if (clone_flags & CLONE_VFORK)
2669 trace = PTRACE_EVENT_VFORK;
2670 else if (args->exit_signal != SIGCHLD)
2671 trace = PTRACE_EVENT_CLONE;
2673 trace = PTRACE_EVENT_FORK;
2675 if (likely(!ptrace_event_enabled(current, trace)))
2679 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2680 add_latent_entropy();
2686 * Do this prior waking up the new thread - the thread pointer
2687 * might get invalid after that point, if the thread exits quickly.
2689 trace_sched_process_fork(current, p);
2691 pid = get_task_pid(p, PIDTYPE_PID);
2694 if (clone_flags & CLONE_PARENT_SETTID)
2695 put_user(nr, args->parent_tid);
2697 if (clone_flags & CLONE_VFORK) {
2698 p->vfork_done = &vfork;
2699 init_completion(&vfork);
2703 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2704 /* lock the task to synchronize with memcg migration */
2706 lru_gen_add_mm(p->mm);
2710 wake_up_new_task(p);
2712 /* forking complete and child started to run, tell ptracer */
2713 if (unlikely(trace))
2714 ptrace_event_pid(trace, pid);
2716 if (clone_flags & CLONE_VFORK) {
2717 if (!wait_for_vfork_done(p, &vfork))
2718 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2726 * Create a kernel thread.
2728 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2730 struct kernel_clone_args args = {
2731 .flags = ((lower_32_bits(flags) | CLONE_VM |
2732 CLONE_UNTRACED) & ~CSIGNAL),
2733 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2739 return kernel_clone(&args);
2743 * Create a user mode thread.
2745 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2747 struct kernel_clone_args args = {
2748 .flags = ((lower_32_bits(flags) | CLONE_VM |
2749 CLONE_UNTRACED) & ~CSIGNAL),
2750 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2755 return kernel_clone(&args);
2758 #ifdef __ARCH_WANT_SYS_FORK
2759 SYSCALL_DEFINE0(fork)
2762 struct kernel_clone_args args = {
2763 .exit_signal = SIGCHLD,
2766 return kernel_clone(&args);
2768 /* can not support in nommu mode */
2774 #ifdef __ARCH_WANT_SYS_VFORK
2775 SYSCALL_DEFINE0(vfork)
2777 struct kernel_clone_args args = {
2778 .flags = CLONE_VFORK | CLONE_VM,
2779 .exit_signal = SIGCHLD,
2782 return kernel_clone(&args);
2786 #ifdef __ARCH_WANT_SYS_CLONE
2787 #ifdef CONFIG_CLONE_BACKWARDS
2788 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2789 int __user *, parent_tidptr,
2791 int __user *, child_tidptr)
2792 #elif defined(CONFIG_CLONE_BACKWARDS2)
2793 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2794 int __user *, parent_tidptr,
2795 int __user *, child_tidptr,
2797 #elif defined(CONFIG_CLONE_BACKWARDS3)
2798 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2800 int __user *, parent_tidptr,
2801 int __user *, child_tidptr,
2804 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2805 int __user *, parent_tidptr,
2806 int __user *, child_tidptr,
2810 struct kernel_clone_args args = {
2811 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2812 .pidfd = parent_tidptr,
2813 .child_tid = child_tidptr,
2814 .parent_tid = parent_tidptr,
2815 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2820 return kernel_clone(&args);
2824 #ifdef __ARCH_WANT_SYS_CLONE3
2826 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2827 struct clone_args __user *uargs,
2831 struct clone_args args;
2832 pid_t *kset_tid = kargs->set_tid;
2834 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2835 CLONE_ARGS_SIZE_VER0);
2836 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2837 CLONE_ARGS_SIZE_VER1);
2838 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2839 CLONE_ARGS_SIZE_VER2);
2840 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2842 if (unlikely(usize > PAGE_SIZE))
2844 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2847 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2851 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2854 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2857 if (unlikely(args.set_tid && args.set_tid_size == 0))
2861 * Verify that higher 32bits of exit_signal are unset and that
2862 * it is a valid signal
2864 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2865 !valid_signal(args.exit_signal)))
2868 if ((args.flags & CLONE_INTO_CGROUP) &&
2869 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2872 *kargs = (struct kernel_clone_args){
2873 .flags = args.flags,
2874 .pidfd = u64_to_user_ptr(args.pidfd),
2875 .child_tid = u64_to_user_ptr(args.child_tid),
2876 .parent_tid = u64_to_user_ptr(args.parent_tid),
2877 .exit_signal = args.exit_signal,
2878 .stack = args.stack,
2879 .stack_size = args.stack_size,
2881 .set_tid_size = args.set_tid_size,
2882 .cgroup = args.cgroup,
2886 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2887 (kargs->set_tid_size * sizeof(pid_t))))
2890 kargs->set_tid = kset_tid;
2896 * clone3_stack_valid - check and prepare stack
2897 * @kargs: kernel clone args
2899 * Verify that the stack arguments userspace gave us are sane.
2900 * In addition, set the stack direction for userspace since it's easy for us to
2903 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2905 if (kargs->stack == 0) {
2906 if (kargs->stack_size > 0)
2909 if (kargs->stack_size == 0)
2912 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2915 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2916 kargs->stack += kargs->stack_size;
2923 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2925 /* Verify that no unknown flags are passed along. */
2927 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2931 * - make the CLONE_DETACHED bit reusable for clone3
2932 * - make the CSIGNAL bits reusable for clone3
2934 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
2937 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2938 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2941 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2945 if (!clone3_stack_valid(kargs))
2952 * clone3 - create a new process with specific properties
2953 * @uargs: argument structure
2954 * @size: size of @uargs
2956 * clone3() is the extensible successor to clone()/clone2().
2957 * It takes a struct as argument that is versioned by its size.
2959 * Return: On success, a positive PID for the child process.
2960 * On error, a negative errno number.
2962 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2966 struct kernel_clone_args kargs;
2967 pid_t set_tid[MAX_PID_NS_LEVEL];
2969 kargs.set_tid = set_tid;
2971 err = copy_clone_args_from_user(&kargs, uargs, size);
2975 if (!clone3_args_valid(&kargs))
2978 return kernel_clone(&kargs);
2982 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2984 struct task_struct *leader, *parent, *child;
2987 read_lock(&tasklist_lock);
2988 leader = top = top->group_leader;
2990 for_each_thread(leader, parent) {
2991 list_for_each_entry(child, &parent->children, sibling) {
2992 res = visitor(child, data);
3004 if (leader != top) {
3006 parent = child->real_parent;
3007 leader = parent->group_leader;
3011 read_unlock(&tasklist_lock);
3014 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3015 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3018 static void sighand_ctor(void *data)
3020 struct sighand_struct *sighand = data;
3022 spin_lock_init(&sighand->siglock);
3023 init_waitqueue_head(&sighand->signalfd_wqh);
3026 void __init mm_cache_init(void)
3028 unsigned int mm_size;
3031 * The mm_cpumask is located at the end of mm_struct, and is
3032 * dynamically sized based on the maximum CPU number this system
3033 * can have, taking hotplug into account (nr_cpu_ids).
3035 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3037 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3038 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3039 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3040 offsetof(struct mm_struct, saved_auxv),
3041 sizeof_field(struct mm_struct, saved_auxv),
3045 void __init proc_caches_init(void)
3047 sighand_cachep = kmem_cache_create("sighand_cache",
3048 sizeof(struct sighand_struct), 0,
3049 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3050 SLAB_ACCOUNT, sighand_ctor);
3051 signal_cachep = kmem_cache_create("signal_cache",
3052 sizeof(struct signal_struct), 0,
3053 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3055 files_cachep = kmem_cache_create("files_cache",
3056 sizeof(struct files_struct), 0,
3057 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3059 fs_cachep = kmem_cache_create("fs_cache",
3060 sizeof(struct fs_struct), 0,
3061 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3064 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3066 nsproxy_cache_init();
3070 * Check constraints on flags passed to the unshare system call.
3072 static int check_unshare_flags(unsigned long unshare_flags)
3074 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3075 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3076 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3077 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3081 * Not implemented, but pretend it works if there is nothing
3082 * to unshare. Note that unsharing the address space or the
3083 * signal handlers also need to unshare the signal queues (aka
3086 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3087 if (!thread_group_empty(current))
3090 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3091 if (refcount_read(¤t->sighand->count) > 1)
3094 if (unshare_flags & CLONE_VM) {
3095 if (!current_is_single_threaded())
3103 * Unshare the filesystem structure if it is being shared
3105 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3107 struct fs_struct *fs = current->fs;
3109 if (!(unshare_flags & CLONE_FS) || !fs)
3112 /* don't need lock here; in the worst case we'll do useless copy */
3116 *new_fsp = copy_fs_struct(fs);
3124 * Unshare file descriptor table if it is being shared
3126 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3127 struct files_struct **new_fdp)
3129 struct files_struct *fd = current->files;
3132 if ((unshare_flags & CLONE_FILES) &&
3133 (fd && atomic_read(&fd->count) > 1)) {
3134 *new_fdp = dup_fd(fd, max_fds, &error);
3143 * unshare allows a process to 'unshare' part of the process
3144 * context which was originally shared using clone. copy_*
3145 * functions used by kernel_clone() cannot be used here directly
3146 * because they modify an inactive task_struct that is being
3147 * constructed. Here we are modifying the current, active,
3150 int ksys_unshare(unsigned long unshare_flags)
3152 struct fs_struct *fs, *new_fs = NULL;
3153 struct files_struct *new_fd = NULL;
3154 struct cred *new_cred = NULL;
3155 struct nsproxy *new_nsproxy = NULL;
3160 * If unsharing a user namespace must also unshare the thread group
3161 * and unshare the filesystem root and working directories.
3163 if (unshare_flags & CLONE_NEWUSER)
3164 unshare_flags |= CLONE_THREAD | CLONE_FS;
3166 * If unsharing vm, must also unshare signal handlers.
3168 if (unshare_flags & CLONE_VM)
3169 unshare_flags |= CLONE_SIGHAND;
3171 * If unsharing a signal handlers, must also unshare the signal queues.
3173 if (unshare_flags & CLONE_SIGHAND)
3174 unshare_flags |= CLONE_THREAD;
3176 * If unsharing namespace, must also unshare filesystem information.
3178 if (unshare_flags & CLONE_NEWNS)
3179 unshare_flags |= CLONE_FS;
3181 err = check_unshare_flags(unshare_flags);
3183 goto bad_unshare_out;
3185 * CLONE_NEWIPC must also detach from the undolist: after switching
3186 * to a new ipc namespace, the semaphore arrays from the old
3187 * namespace are unreachable.
3189 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3191 err = unshare_fs(unshare_flags, &new_fs);
3193 goto bad_unshare_out;
3194 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3196 goto bad_unshare_cleanup_fs;
3197 err = unshare_userns(unshare_flags, &new_cred);
3199 goto bad_unshare_cleanup_fd;
3200 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3203 goto bad_unshare_cleanup_cred;
3206 err = set_cred_ucounts(new_cred);
3208 goto bad_unshare_cleanup_cred;
3211 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3214 * CLONE_SYSVSEM is equivalent to sys_exit().
3218 if (unshare_flags & CLONE_NEWIPC) {
3219 /* Orphan segments in old ns (see sem above). */
3221 shm_init_task(current);
3225 switch_task_namespaces(current, new_nsproxy);
3231 spin_lock(&fs->lock);
3232 current->fs = new_fs;
3237 spin_unlock(&fs->lock);
3241 swap(current->files, new_fd);
3243 task_unlock(current);
3246 /* Install the new user namespace */
3247 commit_creds(new_cred);
3252 perf_event_namespaces(current);
3254 bad_unshare_cleanup_cred:
3257 bad_unshare_cleanup_fd:
3259 put_files_struct(new_fd);
3261 bad_unshare_cleanup_fs:
3263 free_fs_struct(new_fs);
3269 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3271 return ksys_unshare(unshare_flags);
3275 * Helper to unshare the files of the current task.
3276 * We don't want to expose copy_files internals to
3277 * the exec layer of the kernel.
3280 int unshare_files(void)
3282 struct task_struct *task = current;
3283 struct files_struct *old, *copy = NULL;
3286 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3294 put_files_struct(old);
3298 int sysctl_max_threads(struct ctl_table *table, int write,
3299 void *buffer, size_t *lenp, loff_t *ppos)
3303 int threads = max_threads;
3305 int max = MAX_THREADS;
3312 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3316 max_threads = threads;