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>
97 #include <linux/scs.h>
98 #include <linux/io_uring.h>
99 #include <linux/bpf.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]);
196 static int free_vm_stack_cache(unsigned int cpu)
198 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
201 for (i = 0; i < NR_CACHED_STACKS; i++) {
202 struct vm_struct *vm_stack = cached_vm_stacks[i];
207 vfree(vm_stack->addr);
208 cached_vm_stacks[i] = NULL;
215 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
217 #ifdef CONFIG_VMAP_STACK
221 for (i = 0; i < NR_CACHED_STACKS; i++) {
224 s = this_cpu_xchg(cached_stacks[i], NULL);
229 /* Mark stack accessible for KASAN. */
230 kasan_unpoison_range(s->addr, THREAD_SIZE);
232 /* Clear stale pointers from reused stack. */
233 memset(s->addr, 0, THREAD_SIZE);
235 tsk->stack_vm_area = s;
236 tsk->stack = s->addr;
241 * Allocated stacks are cached and later reused by new threads,
242 * so memcg accounting is performed manually on assigning/releasing
243 * stacks to tasks. Drop __GFP_ACCOUNT.
245 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
246 VMALLOC_START, VMALLOC_END,
247 THREADINFO_GFP & ~__GFP_ACCOUNT,
249 0, node, __builtin_return_address(0));
252 * We can't call find_vm_area() in interrupt context, and
253 * free_thread_stack() can be called in interrupt context,
254 * so cache the vm_struct.
257 tsk->stack_vm_area = find_vm_area(stack);
262 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
266 tsk->stack = kasan_reset_tag(page_address(page));
273 static inline void free_thread_stack(struct task_struct *tsk)
275 #ifdef CONFIG_VMAP_STACK
276 struct vm_struct *vm = task_stack_vm_area(tsk);
281 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
282 memcg_kmem_uncharge_page(vm->pages[i], 0);
284 for (i = 0; i < NR_CACHED_STACKS; i++) {
285 if (this_cpu_cmpxchg(cached_stacks[i],
286 NULL, tsk->stack_vm_area) != NULL)
292 vfree_atomic(tsk->stack);
297 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
300 static struct kmem_cache *thread_stack_cache;
302 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
305 unsigned long *stack;
306 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
307 stack = kasan_reset_tag(stack);
312 static void free_thread_stack(struct task_struct *tsk)
314 kmem_cache_free(thread_stack_cache, tsk->stack);
317 void thread_stack_cache_init(void)
319 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
320 THREAD_SIZE, THREAD_SIZE, 0, 0,
322 BUG_ON(thread_stack_cache == NULL);
327 /* SLAB cache for signal_struct structures (tsk->signal) */
328 static struct kmem_cache *signal_cachep;
330 /* SLAB cache for sighand_struct structures (tsk->sighand) */
331 struct kmem_cache *sighand_cachep;
333 /* SLAB cache for files_struct structures (tsk->files) */
334 struct kmem_cache *files_cachep;
336 /* SLAB cache for fs_struct structures (tsk->fs) */
337 struct kmem_cache *fs_cachep;
339 /* SLAB cache for vm_area_struct structures */
340 static struct kmem_cache *vm_area_cachep;
342 /* SLAB cache for mm_struct structures (tsk->mm) */
343 static struct kmem_cache *mm_cachep;
345 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
347 struct vm_area_struct *vma;
349 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
355 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
357 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
360 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
361 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
363 * orig->shared.rb may be modified concurrently, but the clone
364 * will be reinitialized.
366 *new = data_race(*orig);
367 INIT_LIST_HEAD(&new->anon_vma_chain);
368 new->vm_next = new->vm_prev = NULL;
373 void vm_area_free(struct vm_area_struct *vma)
375 kmem_cache_free(vm_area_cachep, vma);
378 static void account_kernel_stack(struct task_struct *tsk, int account)
380 void *stack = task_stack_page(tsk);
381 struct vm_struct *vm = task_stack_vm_area(tsk);
384 /* All stack pages are in the same node. */
386 mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
387 account * (THREAD_SIZE / 1024));
389 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
390 account * (THREAD_SIZE / 1024));
393 static int memcg_charge_kernel_stack(struct task_struct *tsk)
395 #ifdef CONFIG_VMAP_STACK
396 struct vm_struct *vm = task_stack_vm_area(tsk);
399 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
404 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
406 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
408 * If memcg_kmem_charge_page() fails, page's
409 * memory cgroup pointer is NULL, and
410 * memcg_kmem_uncharge_page() in free_thread_stack()
411 * will ignore this page.
413 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
423 static void release_task_stack(struct task_struct *tsk)
425 if (WARN_ON(tsk->state != TASK_DEAD))
426 return; /* Better to leak the stack than to free prematurely */
428 account_kernel_stack(tsk, -1);
429 free_thread_stack(tsk);
431 #ifdef CONFIG_VMAP_STACK
432 tsk->stack_vm_area = NULL;
436 #ifdef CONFIG_THREAD_INFO_IN_TASK
437 void put_task_stack(struct task_struct *tsk)
439 if (refcount_dec_and_test(&tsk->stack_refcount))
440 release_task_stack(tsk);
444 void free_task(struct task_struct *tsk)
448 #ifndef CONFIG_THREAD_INFO_IN_TASK
450 * The task is finally done with both the stack and thread_info,
453 release_task_stack(tsk);
456 * If the task had a separate stack allocation, it should be gone
459 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
461 rt_mutex_debug_task_free(tsk);
462 ftrace_graph_exit_task(tsk);
463 arch_release_task_struct(tsk);
464 if (tsk->flags & PF_KTHREAD)
465 free_kthread_struct(tsk);
466 free_task_struct(tsk);
468 EXPORT_SYMBOL(free_task);
471 static __latent_entropy int dup_mmap(struct mm_struct *mm,
472 struct mm_struct *oldmm)
474 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
475 struct rb_node **rb_link, *rb_parent;
477 unsigned long charge;
480 uprobe_start_dup_mmap();
481 if (mmap_write_lock_killable(oldmm)) {
483 goto fail_uprobe_end;
485 flush_cache_dup_mm(oldmm);
486 uprobe_dup_mmap(oldmm, mm);
488 * Not linked in yet - no deadlock potential:
490 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
492 /* No ordering required: file already has been exposed. */
493 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
495 mm->total_vm = oldmm->total_vm;
496 mm->data_vm = oldmm->data_vm;
497 mm->exec_vm = oldmm->exec_vm;
498 mm->stack_vm = oldmm->stack_vm;
500 rb_link = &mm->mm_rb.rb_node;
503 retval = ksm_fork(mm, oldmm);
506 retval = khugepaged_fork(mm, oldmm);
511 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
514 if (mpnt->vm_flags & VM_DONTCOPY) {
515 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
520 * Don't duplicate many vmas if we've been oom-killed (for
523 if (fatal_signal_pending(current)) {
527 if (mpnt->vm_flags & VM_ACCOUNT) {
528 unsigned long len = vma_pages(mpnt);
530 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
534 tmp = vm_area_dup(mpnt);
537 retval = vma_dup_policy(mpnt, tmp);
539 goto fail_nomem_policy;
541 retval = dup_userfaultfd(tmp, &uf);
543 goto fail_nomem_anon_vma_fork;
544 if (tmp->vm_flags & VM_WIPEONFORK) {
546 * VM_WIPEONFORK gets a clean slate in the child.
547 * Don't prepare anon_vma until fault since we don't
548 * copy page for current vma.
550 tmp->anon_vma = NULL;
551 } else if (anon_vma_fork(tmp, mpnt))
552 goto fail_nomem_anon_vma_fork;
553 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
556 struct inode *inode = file_inode(file);
557 struct address_space *mapping = file->f_mapping;
560 if (tmp->vm_flags & VM_DENYWRITE)
561 put_write_access(inode);
562 i_mmap_lock_write(mapping);
563 if (tmp->vm_flags & VM_SHARED)
564 mapping_allow_writable(mapping);
565 flush_dcache_mmap_lock(mapping);
566 /* insert tmp into the share list, just after mpnt */
567 vma_interval_tree_insert_after(tmp, mpnt,
569 flush_dcache_mmap_unlock(mapping);
570 i_mmap_unlock_write(mapping);
574 * Clear hugetlb-related page reserves for children. This only
575 * affects MAP_PRIVATE mappings. Faults generated by the child
576 * are not guaranteed to succeed, even if read-only
578 if (is_vm_hugetlb_page(tmp))
579 reset_vma_resv_huge_pages(tmp);
582 * Link in the new vma and copy the page table entries.
585 pprev = &tmp->vm_next;
589 __vma_link_rb(mm, tmp, rb_link, rb_parent);
590 rb_link = &tmp->vm_rb.rb_right;
591 rb_parent = &tmp->vm_rb;
594 if (!(tmp->vm_flags & VM_WIPEONFORK))
595 retval = copy_page_range(tmp, mpnt);
597 if (tmp->vm_ops && tmp->vm_ops->open)
598 tmp->vm_ops->open(tmp);
603 /* a new mm has just been created */
604 retval = arch_dup_mmap(oldmm, mm);
606 mmap_write_unlock(mm);
608 mmap_write_unlock(oldmm);
609 dup_userfaultfd_complete(&uf);
611 uprobe_end_dup_mmap();
613 fail_nomem_anon_vma_fork:
614 mpol_put(vma_policy(tmp));
619 vm_unacct_memory(charge);
623 static inline int mm_alloc_pgd(struct mm_struct *mm)
625 mm->pgd = pgd_alloc(mm);
626 if (unlikely(!mm->pgd))
631 static inline void mm_free_pgd(struct mm_struct *mm)
633 pgd_free(mm, mm->pgd);
636 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
638 mmap_write_lock(oldmm);
639 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
640 mmap_write_unlock(oldmm);
643 #define mm_alloc_pgd(mm) (0)
644 #define mm_free_pgd(mm)
645 #endif /* CONFIG_MMU */
647 static void check_mm(struct mm_struct *mm)
651 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
652 "Please make sure 'struct resident_page_types[]' is updated as well");
654 for (i = 0; i < NR_MM_COUNTERS; i++) {
655 long x = atomic_long_read(&mm->rss_stat.count[i]);
658 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
659 mm, resident_page_types[i], x);
662 if (mm_pgtables_bytes(mm))
663 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
664 mm_pgtables_bytes(mm));
666 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
667 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
671 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
672 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
675 * Called when the last reference to the mm
676 * is dropped: either by a lazy thread or by
677 * mmput. Free the page directory and the mm.
679 void __mmdrop(struct mm_struct *mm)
681 BUG_ON(mm == &init_mm);
682 WARN_ON_ONCE(mm == current->mm);
683 WARN_ON_ONCE(mm == current->active_mm);
686 mmu_notifier_subscriptions_destroy(mm);
688 put_user_ns(mm->user_ns);
691 EXPORT_SYMBOL_GPL(__mmdrop);
693 static void mmdrop_async_fn(struct work_struct *work)
695 struct mm_struct *mm;
697 mm = container_of(work, struct mm_struct, async_put_work);
701 static void mmdrop_async(struct mm_struct *mm)
703 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
704 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
705 schedule_work(&mm->async_put_work);
709 static inline void free_signal_struct(struct signal_struct *sig)
711 taskstats_tgid_free(sig);
712 sched_autogroup_exit(sig);
714 * __mmdrop is not safe to call from softirq context on x86 due to
715 * pgd_dtor so postpone it to the async context
718 mmdrop_async(sig->oom_mm);
719 kmem_cache_free(signal_cachep, sig);
722 static inline void put_signal_struct(struct signal_struct *sig)
724 if (refcount_dec_and_test(&sig->sigcnt))
725 free_signal_struct(sig);
728 void __put_task_struct(struct task_struct *tsk)
730 WARN_ON(!tsk->exit_state);
731 WARN_ON(refcount_read(&tsk->usage));
732 WARN_ON(tsk == current);
736 task_numa_free(tsk, true);
737 security_task_free(tsk);
738 bpf_task_storage_free(tsk);
740 delayacct_tsk_free(tsk);
741 put_signal_struct(tsk->signal);
743 if (!profile_handoff_task(tsk))
746 EXPORT_SYMBOL_GPL(__put_task_struct);
748 void __init __weak arch_task_cache_init(void) { }
753 static void set_max_threads(unsigned int max_threads_suggested)
756 unsigned long nr_pages = totalram_pages();
759 * The number of threads shall be limited such that the thread
760 * structures may only consume a small part of the available memory.
762 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
763 threads = MAX_THREADS;
765 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
766 (u64) THREAD_SIZE * 8UL);
768 if (threads > max_threads_suggested)
769 threads = max_threads_suggested;
771 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
774 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
775 /* Initialized by the architecture: */
776 int arch_task_struct_size __read_mostly;
779 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
780 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
782 /* Fetch thread_struct whitelist for the architecture. */
783 arch_thread_struct_whitelist(offset, size);
786 * Handle zero-sized whitelist or empty thread_struct, otherwise
787 * adjust offset to position of thread_struct in task_struct.
789 if (unlikely(*size == 0))
792 *offset += offsetof(struct task_struct, thread);
794 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
796 void __init fork_init(void)
799 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
800 #ifndef ARCH_MIN_TASKALIGN
801 #define ARCH_MIN_TASKALIGN 0
803 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
804 unsigned long useroffset, usersize;
806 /* create a slab on which task_structs can be allocated */
807 task_struct_whitelist(&useroffset, &usersize);
808 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
809 arch_task_struct_size, align,
810 SLAB_PANIC|SLAB_ACCOUNT,
811 useroffset, usersize, NULL);
814 /* do the arch specific task caches init */
815 arch_task_cache_init();
817 set_max_threads(MAX_THREADS);
819 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
820 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
821 init_task.signal->rlim[RLIMIT_SIGPENDING] =
822 init_task.signal->rlim[RLIMIT_NPROC];
824 for (i = 0; i < UCOUNT_COUNTS; i++)
825 init_user_ns.ucount_max[i] = max_threads/2;
827 #ifdef CONFIG_VMAP_STACK
828 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
829 NULL, free_vm_stack_cache);
834 lockdep_init_task(&init_task);
838 int __weak arch_dup_task_struct(struct task_struct *dst,
839 struct task_struct *src)
845 void set_task_stack_end_magic(struct task_struct *tsk)
847 unsigned long *stackend;
849 stackend = end_of_stack(tsk);
850 *stackend = STACK_END_MAGIC; /* for overflow detection */
853 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
855 struct task_struct *tsk;
856 unsigned long *stack;
857 struct vm_struct *stack_vm_area __maybe_unused;
860 if (node == NUMA_NO_NODE)
861 node = tsk_fork_get_node(orig);
862 tsk = alloc_task_struct_node(node);
866 stack = alloc_thread_stack_node(tsk, node);
870 if (memcg_charge_kernel_stack(tsk))
873 stack_vm_area = task_stack_vm_area(tsk);
875 err = arch_dup_task_struct(tsk, orig);
878 * arch_dup_task_struct() clobbers the stack-related fields. Make
879 * sure they're properly initialized before using any stack-related
883 #ifdef CONFIG_VMAP_STACK
884 tsk->stack_vm_area = stack_vm_area;
886 #ifdef CONFIG_THREAD_INFO_IN_TASK
887 refcount_set(&tsk->stack_refcount, 1);
893 err = scs_prepare(tsk, node);
897 #ifdef CONFIG_SECCOMP
899 * We must handle setting up seccomp filters once we're under
900 * the sighand lock in case orig has changed between now and
901 * then. Until then, filter must be NULL to avoid messing up
902 * the usage counts on the error path calling free_task.
904 tsk->seccomp.filter = NULL;
907 setup_thread_stack(tsk, orig);
908 clear_user_return_notifier(tsk);
909 clear_tsk_need_resched(tsk);
910 set_task_stack_end_magic(tsk);
911 clear_syscall_work_syscall_user_dispatch(tsk);
913 #ifdef CONFIG_STACKPROTECTOR
914 tsk->stack_canary = get_random_canary();
916 if (orig->cpus_ptr == &orig->cpus_mask)
917 tsk->cpus_ptr = &tsk->cpus_mask;
920 * One for the user space visible state that goes away when reaped.
921 * One for the scheduler.
923 refcount_set(&tsk->rcu_users, 2);
924 /* One for the rcu users */
925 refcount_set(&tsk->usage, 1);
926 #ifdef CONFIG_BLK_DEV_IO_TRACE
929 tsk->splice_pipe = NULL;
930 tsk->task_frag.page = NULL;
931 tsk->wake_q.next = NULL;
932 tsk->pf_io_worker = NULL;
934 account_kernel_stack(tsk, 1);
937 kmap_local_fork(tsk);
939 #ifdef CONFIG_FAULT_INJECTION
943 #ifdef CONFIG_BLK_CGROUP
944 tsk->throttle_queue = NULL;
945 tsk->use_memdelay = 0;
949 tsk->active_memcg = NULL;
954 free_thread_stack(tsk);
956 free_task_struct(tsk);
960 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
962 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
964 static int __init coredump_filter_setup(char *s)
966 default_dump_filter =
967 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
968 MMF_DUMP_FILTER_MASK;
972 __setup("coredump_filter=", coredump_filter_setup);
974 #include <linux/init_task.h>
976 static void mm_init_aio(struct mm_struct *mm)
979 spin_lock_init(&mm->ioctx_lock);
980 mm->ioctx_table = NULL;
984 static __always_inline void mm_clear_owner(struct mm_struct *mm,
985 struct task_struct *p)
989 WRITE_ONCE(mm->owner, NULL);
993 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1000 static void mm_init_pasid(struct mm_struct *mm)
1002 #ifdef CONFIG_IOMMU_SUPPORT
1003 mm->pasid = INIT_PASID;
1007 static void mm_init_uprobes_state(struct mm_struct *mm)
1009 #ifdef CONFIG_UPROBES
1010 mm->uprobes_state.xol_area = NULL;
1014 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1015 struct user_namespace *user_ns)
1018 mm->mm_rb = RB_ROOT;
1019 mm->vmacache_seqnum = 0;
1020 atomic_set(&mm->mm_users, 1);
1021 atomic_set(&mm->mm_count, 1);
1022 seqcount_init(&mm->write_protect_seq);
1024 INIT_LIST_HEAD(&mm->mmlist);
1025 mm->core_state = NULL;
1026 mm_pgtables_bytes_init(mm);
1029 atomic_set(&mm->has_pinned, 0);
1030 atomic64_set(&mm->pinned_vm, 0);
1031 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1032 spin_lock_init(&mm->page_table_lock);
1033 spin_lock_init(&mm->arg_lock);
1034 mm_init_cpumask(mm);
1036 mm_init_owner(mm, p);
1038 RCU_INIT_POINTER(mm->exe_file, NULL);
1039 mmu_notifier_subscriptions_init(mm);
1040 init_tlb_flush_pending(mm);
1041 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1042 mm->pmd_huge_pte = NULL;
1044 mm_init_uprobes_state(mm);
1047 mm->flags = current->mm->flags & MMF_INIT_MASK;
1048 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1050 mm->flags = default_dump_filter;
1054 if (mm_alloc_pgd(mm))
1057 if (init_new_context(p, mm))
1058 goto fail_nocontext;
1060 mm->user_ns = get_user_ns(user_ns);
1071 * Allocate and initialize an mm_struct.
1073 struct mm_struct *mm_alloc(void)
1075 struct mm_struct *mm;
1081 memset(mm, 0, sizeof(*mm));
1082 return mm_init(mm, current, current_user_ns());
1085 static inline void __mmput(struct mm_struct *mm)
1087 VM_BUG_ON(atomic_read(&mm->mm_users));
1089 uprobe_clear_state(mm);
1092 khugepaged_exit(mm); /* must run before exit_mmap */
1094 mm_put_huge_zero_page(mm);
1095 set_mm_exe_file(mm, NULL);
1096 if (!list_empty(&mm->mmlist)) {
1097 spin_lock(&mmlist_lock);
1098 list_del(&mm->mmlist);
1099 spin_unlock(&mmlist_lock);
1102 module_put(mm->binfmt->module);
1107 * Decrement the use count and release all resources for an mm.
1109 void mmput(struct mm_struct *mm)
1113 if (atomic_dec_and_test(&mm->mm_users))
1116 EXPORT_SYMBOL_GPL(mmput);
1119 static void mmput_async_fn(struct work_struct *work)
1121 struct mm_struct *mm = container_of(work, struct mm_struct,
1127 void mmput_async(struct mm_struct *mm)
1129 if (atomic_dec_and_test(&mm->mm_users)) {
1130 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1131 schedule_work(&mm->async_put_work);
1137 * set_mm_exe_file - change a reference to the mm's executable file
1139 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1141 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1142 * invocations: in mmput() nobody alive left, in execve task is single
1143 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1144 * mm->exe_file, but does so without using set_mm_exe_file() in order
1145 * to do avoid the need for any locks.
1147 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1149 struct file *old_exe_file;
1152 * It is safe to dereference the exe_file without RCU as
1153 * this function is only called if nobody else can access
1154 * this mm -- see comment above for justification.
1156 old_exe_file = rcu_dereference_raw(mm->exe_file);
1159 get_file(new_exe_file);
1160 rcu_assign_pointer(mm->exe_file, new_exe_file);
1166 * get_mm_exe_file - acquire a reference to the mm's executable file
1168 * Returns %NULL if mm has no associated executable file.
1169 * User must release file via fput().
1171 struct file *get_mm_exe_file(struct mm_struct *mm)
1173 struct file *exe_file;
1176 exe_file = rcu_dereference(mm->exe_file);
1177 if (exe_file && !get_file_rcu(exe_file))
1182 EXPORT_SYMBOL(get_mm_exe_file);
1185 * get_task_exe_file - acquire a reference to the task's executable file
1187 * Returns %NULL if task's mm (if any) has no associated executable file or
1188 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1189 * User must release file via fput().
1191 struct file *get_task_exe_file(struct task_struct *task)
1193 struct file *exe_file = NULL;
1194 struct mm_struct *mm;
1199 if (!(task->flags & PF_KTHREAD))
1200 exe_file = get_mm_exe_file(mm);
1205 EXPORT_SYMBOL(get_task_exe_file);
1208 * get_task_mm - acquire a reference to the task's mm
1210 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1211 * this kernel workthread has transiently adopted a user mm with use_mm,
1212 * to do its AIO) is not set and if so returns a reference to it, after
1213 * bumping up the use count. User must release the mm via mmput()
1214 * after use. Typically used by /proc and ptrace.
1216 struct mm_struct *get_task_mm(struct task_struct *task)
1218 struct mm_struct *mm;
1223 if (task->flags & PF_KTHREAD)
1231 EXPORT_SYMBOL_GPL(get_task_mm);
1233 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1235 struct mm_struct *mm;
1238 err = down_read_killable(&task->signal->exec_update_lock);
1240 return ERR_PTR(err);
1242 mm = get_task_mm(task);
1243 if (mm && mm != current->mm &&
1244 !ptrace_may_access(task, mode)) {
1246 mm = ERR_PTR(-EACCES);
1248 up_read(&task->signal->exec_update_lock);
1253 static void complete_vfork_done(struct task_struct *tsk)
1255 struct completion *vfork;
1258 vfork = tsk->vfork_done;
1259 if (likely(vfork)) {
1260 tsk->vfork_done = NULL;
1266 static int wait_for_vfork_done(struct task_struct *child,
1267 struct completion *vfork)
1271 freezer_do_not_count();
1272 cgroup_enter_frozen();
1273 killed = wait_for_completion_killable(vfork);
1274 cgroup_leave_frozen(false);
1279 child->vfork_done = NULL;
1283 put_task_struct(child);
1287 /* Please note the differences between mmput and mm_release.
1288 * mmput is called whenever we stop holding onto a mm_struct,
1289 * error success whatever.
1291 * mm_release is called after a mm_struct has been removed
1292 * from the current process.
1294 * This difference is important for error handling, when we
1295 * only half set up a mm_struct for a new process and need to restore
1296 * the old one. Because we mmput the new mm_struct before
1297 * restoring the old one. . .
1298 * Eric Biederman 10 January 1998
1300 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1302 uprobe_free_utask(tsk);
1304 /* Get rid of any cached register state */
1305 deactivate_mm(tsk, mm);
1308 * Signal userspace if we're not exiting with a core dump
1309 * because we want to leave the value intact for debugging
1312 if (tsk->clear_child_tid) {
1313 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1314 atomic_read(&mm->mm_users) > 1) {
1316 * We don't check the error code - if userspace has
1317 * not set up a proper pointer then tough luck.
1319 put_user(0, tsk->clear_child_tid);
1320 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1321 1, NULL, NULL, 0, 0);
1323 tsk->clear_child_tid = NULL;
1327 * All done, finally we can wake up parent and return this mm to him.
1328 * Also kthread_stop() uses this completion for synchronization.
1330 if (tsk->vfork_done)
1331 complete_vfork_done(tsk);
1334 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1336 futex_exit_release(tsk);
1337 mm_release(tsk, mm);
1340 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1342 futex_exec_release(tsk);
1343 mm_release(tsk, mm);
1347 * dup_mm() - duplicates an existing mm structure
1348 * @tsk: the task_struct with which the new mm will be associated.
1349 * @oldmm: the mm to duplicate.
1351 * Allocates a new mm structure and duplicates the provided @oldmm structure
1354 * Return: the duplicated mm or NULL on failure.
1356 static struct mm_struct *dup_mm(struct task_struct *tsk,
1357 struct mm_struct *oldmm)
1359 struct mm_struct *mm;
1366 memcpy(mm, oldmm, sizeof(*mm));
1368 if (!mm_init(mm, tsk, mm->user_ns))
1371 err = dup_mmap(mm, oldmm);
1375 mm->hiwater_rss = get_mm_rss(mm);
1376 mm->hiwater_vm = mm->total_vm;
1378 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1384 /* don't put binfmt in mmput, we haven't got module yet */
1386 mm_init_owner(mm, NULL);
1393 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1395 struct mm_struct *mm, *oldmm;
1398 tsk->min_flt = tsk->maj_flt = 0;
1399 tsk->nvcsw = tsk->nivcsw = 0;
1400 #ifdef CONFIG_DETECT_HUNG_TASK
1401 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1402 tsk->last_switch_time = 0;
1406 tsk->active_mm = NULL;
1409 * Are we cloning a kernel thread?
1411 * We need to steal a active VM for that..
1413 oldmm = current->mm;
1417 /* initialize the new vmacache entries */
1418 vmacache_flush(tsk);
1420 if (clone_flags & CLONE_VM) {
1427 mm = dup_mm(tsk, current->mm);
1433 tsk->active_mm = mm;
1440 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1442 struct fs_struct *fs = current->fs;
1443 if (clone_flags & CLONE_FS) {
1444 /* tsk->fs is already what we want */
1445 spin_lock(&fs->lock);
1447 spin_unlock(&fs->lock);
1451 spin_unlock(&fs->lock);
1454 tsk->fs = copy_fs_struct(fs);
1460 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1462 struct files_struct *oldf, *newf;
1466 * A background process may not have any files ...
1468 oldf = current->files;
1472 if (clone_flags & CLONE_FILES) {
1473 atomic_inc(&oldf->count);
1477 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1487 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1490 struct io_context *ioc = current->io_context;
1491 struct io_context *new_ioc;
1496 * Share io context with parent, if CLONE_IO is set
1498 if (clone_flags & CLONE_IO) {
1500 tsk->io_context = ioc;
1501 } else if (ioprio_valid(ioc->ioprio)) {
1502 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1503 if (unlikely(!new_ioc))
1506 new_ioc->ioprio = ioc->ioprio;
1507 put_io_context(new_ioc);
1513 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1515 struct sighand_struct *sig;
1517 if (clone_flags & CLONE_SIGHAND) {
1518 refcount_inc(¤t->sighand->count);
1521 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1522 RCU_INIT_POINTER(tsk->sighand, sig);
1526 refcount_set(&sig->count, 1);
1527 spin_lock_irq(¤t->sighand->siglock);
1528 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1529 spin_unlock_irq(¤t->sighand->siglock);
1531 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1532 if (clone_flags & CLONE_CLEAR_SIGHAND)
1533 flush_signal_handlers(tsk, 0);
1538 void __cleanup_sighand(struct sighand_struct *sighand)
1540 if (refcount_dec_and_test(&sighand->count)) {
1541 signalfd_cleanup(sighand);
1543 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1544 * without an RCU grace period, see __lock_task_sighand().
1546 kmem_cache_free(sighand_cachep, sighand);
1551 * Initialize POSIX timer handling for a thread group.
1553 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1555 struct posix_cputimers *pct = &sig->posix_cputimers;
1556 unsigned long cpu_limit;
1558 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1559 posix_cputimers_group_init(pct, cpu_limit);
1562 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1564 struct signal_struct *sig;
1566 if (clone_flags & CLONE_THREAD)
1569 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1574 sig->nr_threads = 1;
1575 atomic_set(&sig->live, 1);
1576 refcount_set(&sig->sigcnt, 1);
1578 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1579 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1580 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1582 init_waitqueue_head(&sig->wait_chldexit);
1583 sig->curr_target = tsk;
1584 init_sigpending(&sig->shared_pending);
1585 INIT_HLIST_HEAD(&sig->multiprocess);
1586 seqlock_init(&sig->stats_lock);
1587 prev_cputime_init(&sig->prev_cputime);
1589 #ifdef CONFIG_POSIX_TIMERS
1590 INIT_LIST_HEAD(&sig->posix_timers);
1591 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1592 sig->real_timer.function = it_real_fn;
1595 task_lock(current->group_leader);
1596 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1597 task_unlock(current->group_leader);
1599 posix_cpu_timers_init_group(sig);
1601 tty_audit_fork(sig);
1602 sched_autogroup_fork(sig);
1604 sig->oom_score_adj = current->signal->oom_score_adj;
1605 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1607 mutex_init(&sig->cred_guard_mutex);
1608 init_rwsem(&sig->exec_update_lock);
1613 static void copy_seccomp(struct task_struct *p)
1615 #ifdef CONFIG_SECCOMP
1617 * Must be called with sighand->lock held, which is common to
1618 * all threads in the group. Holding cred_guard_mutex is not
1619 * needed because this new task is not yet running and cannot
1622 assert_spin_locked(¤t->sighand->siglock);
1624 /* Ref-count the new filter user, and assign it. */
1625 get_seccomp_filter(current);
1626 p->seccomp = current->seccomp;
1629 * Explicitly enable no_new_privs here in case it got set
1630 * between the task_struct being duplicated and holding the
1631 * sighand lock. The seccomp state and nnp must be in sync.
1633 if (task_no_new_privs(current))
1634 task_set_no_new_privs(p);
1637 * If the parent gained a seccomp mode after copying thread
1638 * flags and between before we held the sighand lock, we have
1639 * to manually enable the seccomp thread flag here.
1641 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1642 set_task_syscall_work(p, SECCOMP);
1646 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1648 current->clear_child_tid = tidptr;
1650 return task_pid_vnr(current);
1653 static void rt_mutex_init_task(struct task_struct *p)
1655 raw_spin_lock_init(&p->pi_lock);
1656 #ifdef CONFIG_RT_MUTEXES
1657 p->pi_waiters = RB_ROOT_CACHED;
1658 p->pi_top_task = NULL;
1659 p->pi_blocked_on = NULL;
1663 static inline void init_task_pid_links(struct task_struct *task)
1667 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1668 INIT_HLIST_NODE(&task->pid_links[type]);
1672 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1674 if (type == PIDTYPE_PID)
1675 task->thread_pid = pid;
1677 task->signal->pids[type] = pid;
1680 static inline void rcu_copy_process(struct task_struct *p)
1682 #ifdef CONFIG_PREEMPT_RCU
1683 p->rcu_read_lock_nesting = 0;
1684 p->rcu_read_unlock_special.s = 0;
1685 p->rcu_blocked_node = NULL;
1686 INIT_LIST_HEAD(&p->rcu_node_entry);
1687 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1688 #ifdef CONFIG_TASKS_RCU
1689 p->rcu_tasks_holdout = false;
1690 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1691 p->rcu_tasks_idle_cpu = -1;
1692 #endif /* #ifdef CONFIG_TASKS_RCU */
1693 #ifdef CONFIG_TASKS_TRACE_RCU
1694 p->trc_reader_nesting = 0;
1695 p->trc_reader_special.s = 0;
1696 INIT_LIST_HEAD(&p->trc_holdout_list);
1697 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1700 struct pid *pidfd_pid(const struct file *file)
1702 if (file->f_op == &pidfd_fops)
1703 return file->private_data;
1705 return ERR_PTR(-EBADF);
1708 static int pidfd_release(struct inode *inode, struct file *file)
1710 struct pid *pid = file->private_data;
1712 file->private_data = NULL;
1717 #ifdef CONFIG_PROC_FS
1719 * pidfd_show_fdinfo - print information about a pidfd
1720 * @m: proc fdinfo file
1721 * @f: file referencing a pidfd
1724 * This function will print the pid that a given pidfd refers to in the
1725 * pid namespace of the procfs instance.
1726 * If the pid namespace of the process is not a descendant of the pid
1727 * namespace of the procfs instance 0 will be shown as its pid. This is
1728 * similar to calling getppid() on a process whose parent is outside of
1729 * its pid namespace.
1732 * If pid namespaces are supported then this function will also print
1733 * the pid of a given pidfd refers to for all descendant pid namespaces
1734 * starting from the current pid namespace of the instance, i.e. the
1735 * Pid field and the first entry in the NSpid field will be identical.
1736 * If the pid namespace of the process is not a descendant of the pid
1737 * namespace of the procfs instance 0 will be shown as its first NSpid
1738 * entry and no others will be shown.
1739 * Note that this differs from the Pid and NSpid fields in
1740 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1741 * the pid namespace of the procfs instance. The difference becomes
1742 * obvious when sending around a pidfd between pid namespaces from a
1743 * different branch of the tree, i.e. where no ancestoral relation is
1744 * present between the pid namespaces:
1745 * - create two new pid namespaces ns1 and ns2 in the initial pid
1746 * namespace (also take care to create new mount namespaces in the
1747 * new pid namespace and mount procfs)
1748 * - create a process with a pidfd in ns1
1749 * - send pidfd from ns1 to ns2
1750 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1751 * have exactly one entry, which is 0
1753 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1755 struct pid *pid = f->private_data;
1756 struct pid_namespace *ns;
1759 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1760 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1761 nr = pid_nr_ns(pid, ns);
1764 seq_put_decimal_ll(m, "Pid:\t", nr);
1766 #ifdef CONFIG_PID_NS
1767 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1771 /* If nr is non-zero it means that 'pid' is valid and that
1772 * ns, i.e. the pid namespace associated with the procfs
1773 * instance, is in the pid namespace hierarchy of pid.
1774 * Start at one below the already printed level.
1776 for (i = ns->level + 1; i <= pid->level; i++)
1777 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1785 * Poll support for process exit notification.
1787 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1789 struct pid *pid = file->private_data;
1790 __poll_t poll_flags = 0;
1792 poll_wait(file, &pid->wait_pidfd, pts);
1795 * Inform pollers only when the whole thread group exits.
1796 * If the thread group leader exits before all other threads in the
1797 * group, then poll(2) should block, similar to the wait(2) family.
1799 if (thread_group_exited(pid))
1800 poll_flags = EPOLLIN | EPOLLRDNORM;
1805 const struct file_operations pidfd_fops = {
1806 .release = pidfd_release,
1808 #ifdef CONFIG_PROC_FS
1809 .show_fdinfo = pidfd_show_fdinfo,
1813 static void __delayed_free_task(struct rcu_head *rhp)
1815 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1820 static __always_inline void delayed_free_task(struct task_struct *tsk)
1822 if (IS_ENABLED(CONFIG_MEMCG))
1823 call_rcu(&tsk->rcu, __delayed_free_task);
1828 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1830 /* Skip if kernel thread */
1834 /* Skip if spawning a thread or using vfork */
1835 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1838 /* We need to synchronize with __set_oom_adj */
1839 mutex_lock(&oom_adj_mutex);
1840 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1841 /* Update the values in case they were changed after copy_signal */
1842 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1843 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1844 mutex_unlock(&oom_adj_mutex);
1848 * This creates a new process as a copy of the old one,
1849 * but does not actually start it yet.
1851 * It copies the registers, and all the appropriate
1852 * parts of the process environment (as per the clone
1853 * flags). The actual kick-off is left to the caller.
1855 static __latent_entropy struct task_struct *copy_process(
1859 struct kernel_clone_args *args)
1861 int pidfd = -1, retval;
1862 struct task_struct *p;
1863 struct multiprocess_signals delayed;
1864 struct file *pidfile = NULL;
1865 u64 clone_flags = args->flags;
1866 struct nsproxy *nsp = current->nsproxy;
1869 * Don't allow sharing the root directory with processes in a different
1872 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1873 return ERR_PTR(-EINVAL);
1875 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1876 return ERR_PTR(-EINVAL);
1879 * Thread groups must share signals as well, and detached threads
1880 * can only be started up within the thread group.
1882 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1883 return ERR_PTR(-EINVAL);
1886 * Shared signal handlers imply shared VM. By way of the above,
1887 * thread groups also imply shared VM. Blocking this case allows
1888 * for various simplifications in other code.
1890 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1891 return ERR_PTR(-EINVAL);
1894 * Siblings of global init remain as zombies on exit since they are
1895 * not reaped by their parent (swapper). To solve this and to avoid
1896 * multi-rooted process trees, prevent global and container-inits
1897 * from creating siblings.
1899 if ((clone_flags & CLONE_PARENT) &&
1900 current->signal->flags & SIGNAL_UNKILLABLE)
1901 return ERR_PTR(-EINVAL);
1904 * If the new process will be in a different pid or user namespace
1905 * do not allow it to share a thread group with the forking task.
1907 if (clone_flags & CLONE_THREAD) {
1908 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1909 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1910 return ERR_PTR(-EINVAL);
1914 * If the new process will be in a different time namespace
1915 * do not allow it to share VM or a thread group with the forking task.
1917 if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1918 if (nsp->time_ns != nsp->time_ns_for_children)
1919 return ERR_PTR(-EINVAL);
1922 if (clone_flags & CLONE_PIDFD) {
1924 * - CLONE_DETACHED is blocked so that we can potentially
1925 * reuse it later for CLONE_PIDFD.
1926 * - CLONE_THREAD is blocked until someone really needs it.
1928 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1929 return ERR_PTR(-EINVAL);
1933 * Force any signals received before this point to be delivered
1934 * before the fork happens. Collect up signals sent to multiple
1935 * processes that happen during the fork and delay them so that
1936 * they appear to happen after the fork.
1938 sigemptyset(&delayed.signal);
1939 INIT_HLIST_NODE(&delayed.node);
1941 spin_lock_irq(¤t->sighand->siglock);
1942 if (!(clone_flags & CLONE_THREAD))
1943 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1944 recalc_sigpending();
1945 spin_unlock_irq(¤t->sighand->siglock);
1946 retval = -ERESTARTNOINTR;
1947 if (task_sigpending(current))
1951 p = dup_task_struct(current, node);
1954 if (args->io_thread) {
1956 * Mark us an IO worker, and block any signal that isn't
1959 p->flags |= PF_IO_WORKER;
1960 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
1964 * This _must_ happen before we call free_task(), i.e. before we jump
1965 * to any of the bad_fork_* labels. This is to avoid freeing
1966 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1967 * kernel threads (PF_KTHREAD).
1969 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1971 * Clear TID on mm_release()?
1973 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1975 ftrace_graph_init_task(p);
1977 rt_mutex_init_task(p);
1979 lockdep_assert_irqs_enabled();
1980 #ifdef CONFIG_PROVE_LOCKING
1981 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1984 if (atomic_read(&p->real_cred->user->processes) >=
1985 task_rlimit(p, RLIMIT_NPROC)) {
1986 if (p->real_cred->user != INIT_USER &&
1987 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1990 current->flags &= ~PF_NPROC_EXCEEDED;
1992 retval = copy_creds(p, clone_flags);
1997 * If multiple threads are within copy_process(), then this check
1998 * triggers too late. This doesn't hurt, the check is only there
1999 * to stop root fork bombs.
2002 if (data_race(nr_threads >= max_threads))
2003 goto bad_fork_cleanup_count;
2005 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2006 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
2007 p->flags |= PF_FORKNOEXEC;
2008 INIT_LIST_HEAD(&p->children);
2009 INIT_LIST_HEAD(&p->sibling);
2010 rcu_copy_process(p);
2011 p->vfork_done = NULL;
2012 spin_lock_init(&p->alloc_lock);
2014 init_sigpending(&p->pending);
2015 p->sigqueue_cache = NULL;
2017 p->utime = p->stime = p->gtime = 0;
2018 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2019 p->utimescaled = p->stimescaled = 0;
2021 prev_cputime_init(&p->prev_cputime);
2023 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2024 seqcount_init(&p->vtime.seqcount);
2025 p->vtime.starttime = 0;
2026 p->vtime.state = VTIME_INACTIVE;
2029 #ifdef CONFIG_IO_URING
2033 #if defined(SPLIT_RSS_COUNTING)
2034 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2037 p->default_timer_slack_ns = current->timer_slack_ns;
2043 task_io_accounting_init(&p->ioac);
2044 acct_clear_integrals(p);
2046 posix_cputimers_init(&p->posix_cputimers);
2048 p->io_context = NULL;
2049 audit_set_context(p, NULL);
2052 p->mempolicy = mpol_dup(p->mempolicy);
2053 if (IS_ERR(p->mempolicy)) {
2054 retval = PTR_ERR(p->mempolicy);
2055 p->mempolicy = NULL;
2056 goto bad_fork_cleanup_threadgroup_lock;
2059 #ifdef CONFIG_CPUSETS
2060 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2061 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2062 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2064 #ifdef CONFIG_TRACE_IRQFLAGS
2065 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2066 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2067 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2068 p->softirqs_enabled = 1;
2069 p->softirq_context = 0;
2072 p->pagefault_disabled = 0;
2074 #ifdef CONFIG_LOCKDEP
2075 lockdep_init_task(p);
2078 #ifdef CONFIG_DEBUG_MUTEXES
2079 p->blocked_on = NULL; /* not blocked yet */
2081 #ifdef CONFIG_BCACHE
2082 p->sequential_io = 0;
2083 p->sequential_io_avg = 0;
2085 #ifdef CONFIG_BPF_SYSCALL
2086 RCU_INIT_POINTER(p->bpf_storage, NULL);
2089 /* Perform scheduler related setup. Assign this task to a CPU. */
2090 retval = sched_fork(clone_flags, p);
2092 goto bad_fork_cleanup_policy;
2094 retval = perf_event_init_task(p, clone_flags);
2096 goto bad_fork_cleanup_policy;
2097 retval = audit_alloc(p);
2099 goto bad_fork_cleanup_perf;
2100 /* copy all the process information */
2102 retval = security_task_alloc(p, clone_flags);
2104 goto bad_fork_cleanup_audit;
2105 retval = copy_semundo(clone_flags, p);
2107 goto bad_fork_cleanup_security;
2108 retval = copy_files(clone_flags, p);
2110 goto bad_fork_cleanup_semundo;
2111 retval = copy_fs(clone_flags, p);
2113 goto bad_fork_cleanup_files;
2114 retval = copy_sighand(clone_flags, p);
2116 goto bad_fork_cleanup_fs;
2117 retval = copy_signal(clone_flags, p);
2119 goto bad_fork_cleanup_sighand;
2120 retval = copy_mm(clone_flags, p);
2122 goto bad_fork_cleanup_signal;
2123 retval = copy_namespaces(clone_flags, p);
2125 goto bad_fork_cleanup_mm;
2126 retval = copy_io(clone_flags, p);
2128 goto bad_fork_cleanup_namespaces;
2129 retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2131 goto bad_fork_cleanup_io;
2133 stackleak_task_init(p);
2135 if (pid != &init_struct_pid) {
2136 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2137 args->set_tid_size);
2139 retval = PTR_ERR(pid);
2140 goto bad_fork_cleanup_thread;
2145 * This has to happen after we've potentially unshared the file
2146 * descriptor table (so that the pidfd doesn't leak into the child
2147 * if the fd table isn't shared).
2149 if (clone_flags & CLONE_PIDFD) {
2150 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2152 goto bad_fork_free_pid;
2156 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2157 O_RDWR | O_CLOEXEC);
2158 if (IS_ERR(pidfile)) {
2159 put_unused_fd(pidfd);
2160 retval = PTR_ERR(pidfile);
2161 goto bad_fork_free_pid;
2163 get_pid(pid); /* held by pidfile now */
2165 retval = put_user(pidfd, args->pidfd);
2167 goto bad_fork_put_pidfd;
2176 * sigaltstack should be cleared when sharing the same VM
2178 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2182 * Syscall tracing and stepping should be turned off in the
2183 * child regardless of CLONE_PTRACE.
2185 user_disable_single_step(p);
2186 clear_task_syscall_work(p, SYSCALL_TRACE);
2187 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2188 clear_task_syscall_work(p, SYSCALL_EMU);
2190 clear_tsk_latency_tracing(p);
2192 /* ok, now we should be set up.. */
2193 p->pid = pid_nr(pid);
2194 if (clone_flags & CLONE_THREAD) {
2195 p->group_leader = current->group_leader;
2196 p->tgid = current->tgid;
2198 p->group_leader = p;
2203 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2204 p->dirty_paused_when = 0;
2206 p->pdeath_signal = 0;
2207 INIT_LIST_HEAD(&p->thread_group);
2208 p->task_works = NULL;
2210 #ifdef CONFIG_KRETPROBES
2211 p->kretprobe_instances.first = NULL;
2215 * Ensure that the cgroup subsystem policies allow the new process to be
2216 * forked. It should be noted that the new process's css_set can be changed
2217 * between here and cgroup_post_fork() if an organisation operation is in
2220 retval = cgroup_can_fork(p, args);
2222 goto bad_fork_put_pidfd;
2225 * From this point on we must avoid any synchronous user-space
2226 * communication until we take the tasklist-lock. In particular, we do
2227 * not want user-space to be able to predict the process start-time by
2228 * stalling fork(2) after we recorded the start_time but before it is
2229 * visible to the system.
2232 p->start_time = ktime_get_ns();
2233 p->start_boottime = ktime_get_boottime_ns();
2236 * Make it visible to the rest of the system, but dont wake it up yet.
2237 * Need tasklist lock for parent etc handling!
2239 write_lock_irq(&tasklist_lock);
2241 /* CLONE_PARENT re-uses the old parent */
2242 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2243 p->real_parent = current->real_parent;
2244 p->parent_exec_id = current->parent_exec_id;
2245 if (clone_flags & CLONE_THREAD)
2246 p->exit_signal = -1;
2248 p->exit_signal = current->group_leader->exit_signal;
2250 p->real_parent = current;
2251 p->parent_exec_id = current->self_exec_id;
2252 p->exit_signal = args->exit_signal;
2255 klp_copy_process(p);
2257 spin_lock(¤t->sighand->siglock);
2260 * Copy seccomp details explicitly here, in case they were changed
2261 * before holding sighand lock.
2265 rseq_fork(p, clone_flags);
2267 /* Don't start children in a dying pid namespace */
2268 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2270 goto bad_fork_cancel_cgroup;
2273 /* Let kill terminate clone/fork in the middle */
2274 if (fatal_signal_pending(current)) {
2276 goto bad_fork_cancel_cgroup;
2279 /* past the last point of failure */
2281 fd_install(pidfd, pidfile);
2283 init_task_pid_links(p);
2284 if (likely(p->pid)) {
2285 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2287 init_task_pid(p, PIDTYPE_PID, pid);
2288 if (thread_group_leader(p)) {
2289 init_task_pid(p, PIDTYPE_TGID, pid);
2290 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2291 init_task_pid(p, PIDTYPE_SID, task_session(current));
2293 if (is_child_reaper(pid)) {
2294 ns_of_pid(pid)->child_reaper = p;
2295 p->signal->flags |= SIGNAL_UNKILLABLE;
2297 p->signal->shared_pending.signal = delayed.signal;
2298 p->signal->tty = tty_kref_get(current->signal->tty);
2300 * Inherit has_child_subreaper flag under the same
2301 * tasklist_lock with adding child to the process tree
2302 * for propagate_has_child_subreaper optimization.
2304 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2305 p->real_parent->signal->is_child_subreaper;
2306 list_add_tail(&p->sibling, &p->real_parent->children);
2307 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2308 attach_pid(p, PIDTYPE_TGID);
2309 attach_pid(p, PIDTYPE_PGID);
2310 attach_pid(p, PIDTYPE_SID);
2311 __this_cpu_inc(process_counts);
2313 current->signal->nr_threads++;
2314 atomic_inc(¤t->signal->live);
2315 refcount_inc(¤t->signal->sigcnt);
2316 task_join_group_stop(p);
2317 list_add_tail_rcu(&p->thread_group,
2318 &p->group_leader->thread_group);
2319 list_add_tail_rcu(&p->thread_node,
2320 &p->signal->thread_head);
2322 attach_pid(p, PIDTYPE_PID);
2326 hlist_del_init(&delayed.node);
2327 spin_unlock(¤t->sighand->siglock);
2328 syscall_tracepoint_update(p);
2329 write_unlock_irq(&tasklist_lock);
2331 proc_fork_connector(p);
2333 cgroup_post_fork(p, args);
2336 trace_task_newtask(p, clone_flags);
2337 uprobe_copy_process(p, clone_flags);
2339 copy_oom_score_adj(clone_flags, p);
2343 bad_fork_cancel_cgroup:
2344 spin_unlock(¤t->sighand->siglock);
2345 write_unlock_irq(&tasklist_lock);
2346 cgroup_cancel_fork(p, args);
2348 if (clone_flags & CLONE_PIDFD) {
2350 put_unused_fd(pidfd);
2353 if (pid != &init_struct_pid)
2355 bad_fork_cleanup_thread:
2357 bad_fork_cleanup_io:
2360 bad_fork_cleanup_namespaces:
2361 exit_task_namespaces(p);
2362 bad_fork_cleanup_mm:
2364 mm_clear_owner(p->mm, p);
2367 bad_fork_cleanup_signal:
2368 if (!(clone_flags & CLONE_THREAD))
2369 free_signal_struct(p->signal);
2370 bad_fork_cleanup_sighand:
2371 __cleanup_sighand(p->sighand);
2372 bad_fork_cleanup_fs:
2373 exit_fs(p); /* blocking */
2374 bad_fork_cleanup_files:
2375 exit_files(p); /* blocking */
2376 bad_fork_cleanup_semundo:
2378 bad_fork_cleanup_security:
2379 security_task_free(p);
2380 bad_fork_cleanup_audit:
2382 bad_fork_cleanup_perf:
2383 perf_event_free_task(p);
2384 bad_fork_cleanup_policy:
2385 lockdep_free_task(p);
2387 mpol_put(p->mempolicy);
2388 bad_fork_cleanup_threadgroup_lock:
2390 delayacct_tsk_free(p);
2391 bad_fork_cleanup_count:
2392 atomic_dec(&p->cred->user->processes);
2395 p->state = TASK_DEAD;
2397 delayed_free_task(p);
2399 spin_lock_irq(¤t->sighand->siglock);
2400 hlist_del_init(&delayed.node);
2401 spin_unlock_irq(¤t->sighand->siglock);
2402 return ERR_PTR(retval);
2405 static inline void init_idle_pids(struct task_struct *idle)
2409 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2410 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2411 init_task_pid(idle, type, &init_struct_pid);
2415 struct task_struct *fork_idle(int cpu)
2417 struct task_struct *task;
2418 struct kernel_clone_args args = {
2422 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2423 if (!IS_ERR(task)) {
2424 init_idle_pids(task);
2425 init_idle(task, cpu);
2431 struct mm_struct *copy_init_mm(void)
2433 return dup_mm(NULL, &init_mm);
2437 * This is like kernel_clone(), but shaved down and tailored to just
2438 * creating io_uring workers. It returns a created task, or an error pointer.
2439 * The returned task is inactive, and the caller must fire it up through
2440 * wake_up_new_task(p). All signals are blocked in the created task.
2442 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2444 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2446 struct kernel_clone_args args = {
2447 .flags = ((lower_32_bits(flags) | CLONE_VM |
2448 CLONE_UNTRACED) & ~CSIGNAL),
2449 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2450 .stack = (unsigned long)fn,
2451 .stack_size = (unsigned long)arg,
2455 return copy_process(NULL, 0, node, &args);
2459 * Ok, this is the main fork-routine.
2461 * It copies the process, and if successful kick-starts
2462 * it and waits for it to finish using the VM if required.
2464 * args->exit_signal is expected to be checked for sanity by the caller.
2466 pid_t kernel_clone(struct kernel_clone_args *args)
2468 u64 clone_flags = args->flags;
2469 struct completion vfork;
2471 struct task_struct *p;
2476 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2477 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2478 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2479 * field in struct clone_args and it still doesn't make sense to have
2480 * them both point at the same memory location. Performing this check
2481 * here has the advantage that we don't need to have a separate helper
2482 * to check for legacy clone().
2484 if ((args->flags & CLONE_PIDFD) &&
2485 (args->flags & CLONE_PARENT_SETTID) &&
2486 (args->pidfd == args->parent_tid))
2490 * Determine whether and which event to report to ptracer. When
2491 * called from kernel_thread or CLONE_UNTRACED is explicitly
2492 * requested, no event is reported; otherwise, report if the event
2493 * for the type of forking is enabled.
2495 if (!(clone_flags & CLONE_UNTRACED)) {
2496 if (clone_flags & CLONE_VFORK)
2497 trace = PTRACE_EVENT_VFORK;
2498 else if (args->exit_signal != SIGCHLD)
2499 trace = PTRACE_EVENT_CLONE;
2501 trace = PTRACE_EVENT_FORK;
2503 if (likely(!ptrace_event_enabled(current, trace)))
2507 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2508 add_latent_entropy();
2514 * Do this prior waking up the new thread - the thread pointer
2515 * might get invalid after that point, if the thread exits quickly.
2517 trace_sched_process_fork(current, p);
2519 pid = get_task_pid(p, PIDTYPE_PID);
2522 if (clone_flags & CLONE_PARENT_SETTID)
2523 put_user(nr, args->parent_tid);
2525 if (clone_flags & CLONE_VFORK) {
2526 p->vfork_done = &vfork;
2527 init_completion(&vfork);
2531 wake_up_new_task(p);
2533 /* forking complete and child started to run, tell ptracer */
2534 if (unlikely(trace))
2535 ptrace_event_pid(trace, pid);
2537 if (clone_flags & CLONE_VFORK) {
2538 if (!wait_for_vfork_done(p, &vfork))
2539 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2547 * Create a kernel thread.
2549 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2551 struct kernel_clone_args args = {
2552 .flags = ((lower_32_bits(flags) | CLONE_VM |
2553 CLONE_UNTRACED) & ~CSIGNAL),
2554 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2555 .stack = (unsigned long)fn,
2556 .stack_size = (unsigned long)arg,
2559 return kernel_clone(&args);
2562 #ifdef __ARCH_WANT_SYS_FORK
2563 SYSCALL_DEFINE0(fork)
2566 struct kernel_clone_args args = {
2567 .exit_signal = SIGCHLD,
2570 return kernel_clone(&args);
2572 /* can not support in nommu mode */
2578 #ifdef __ARCH_WANT_SYS_VFORK
2579 SYSCALL_DEFINE0(vfork)
2581 struct kernel_clone_args args = {
2582 .flags = CLONE_VFORK | CLONE_VM,
2583 .exit_signal = SIGCHLD,
2586 return kernel_clone(&args);
2590 #ifdef __ARCH_WANT_SYS_CLONE
2591 #ifdef CONFIG_CLONE_BACKWARDS
2592 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2593 int __user *, parent_tidptr,
2595 int __user *, child_tidptr)
2596 #elif defined(CONFIG_CLONE_BACKWARDS2)
2597 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2598 int __user *, parent_tidptr,
2599 int __user *, child_tidptr,
2601 #elif defined(CONFIG_CLONE_BACKWARDS3)
2602 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2604 int __user *, parent_tidptr,
2605 int __user *, child_tidptr,
2608 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2609 int __user *, parent_tidptr,
2610 int __user *, child_tidptr,
2614 struct kernel_clone_args args = {
2615 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2616 .pidfd = parent_tidptr,
2617 .child_tid = child_tidptr,
2618 .parent_tid = parent_tidptr,
2619 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2624 return kernel_clone(&args);
2628 #ifdef __ARCH_WANT_SYS_CLONE3
2630 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2631 struct clone_args __user *uargs,
2635 struct clone_args args;
2636 pid_t *kset_tid = kargs->set_tid;
2638 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2639 CLONE_ARGS_SIZE_VER0);
2640 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2641 CLONE_ARGS_SIZE_VER1);
2642 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2643 CLONE_ARGS_SIZE_VER2);
2644 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2646 if (unlikely(usize > PAGE_SIZE))
2648 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2651 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2655 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2658 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2661 if (unlikely(args.set_tid && args.set_tid_size == 0))
2665 * Verify that higher 32bits of exit_signal are unset and that
2666 * it is a valid signal
2668 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2669 !valid_signal(args.exit_signal)))
2672 if ((args.flags & CLONE_INTO_CGROUP) &&
2673 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2676 *kargs = (struct kernel_clone_args){
2677 .flags = args.flags,
2678 .pidfd = u64_to_user_ptr(args.pidfd),
2679 .child_tid = u64_to_user_ptr(args.child_tid),
2680 .parent_tid = u64_to_user_ptr(args.parent_tid),
2681 .exit_signal = args.exit_signal,
2682 .stack = args.stack,
2683 .stack_size = args.stack_size,
2685 .set_tid_size = args.set_tid_size,
2686 .cgroup = args.cgroup,
2690 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2691 (kargs->set_tid_size * sizeof(pid_t))))
2694 kargs->set_tid = kset_tid;
2700 * clone3_stack_valid - check and prepare stack
2701 * @kargs: kernel clone args
2703 * Verify that the stack arguments userspace gave us are sane.
2704 * In addition, set the stack direction for userspace since it's easy for us to
2707 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2709 if (kargs->stack == 0) {
2710 if (kargs->stack_size > 0)
2713 if (kargs->stack_size == 0)
2716 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2719 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2720 kargs->stack += kargs->stack_size;
2727 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2729 /* Verify that no unknown flags are passed along. */
2731 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2735 * - make the CLONE_DETACHED bit reuseable for clone3
2736 * - make the CSIGNAL bits reuseable for clone3
2738 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2741 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2742 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2745 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2749 if (!clone3_stack_valid(kargs))
2756 * clone3 - create a new process with specific properties
2757 * @uargs: argument structure
2758 * @size: size of @uargs
2760 * clone3() is the extensible successor to clone()/clone2().
2761 * It takes a struct as argument that is versioned by its size.
2763 * Return: On success, a positive PID for the child process.
2764 * On error, a negative errno number.
2766 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2770 struct kernel_clone_args kargs;
2771 pid_t set_tid[MAX_PID_NS_LEVEL];
2773 kargs.set_tid = set_tid;
2775 err = copy_clone_args_from_user(&kargs, uargs, size);
2779 if (!clone3_args_valid(&kargs))
2782 return kernel_clone(&kargs);
2786 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2788 struct task_struct *leader, *parent, *child;
2791 read_lock(&tasklist_lock);
2792 leader = top = top->group_leader;
2794 for_each_thread(leader, parent) {
2795 list_for_each_entry(child, &parent->children, sibling) {
2796 res = visitor(child, data);
2808 if (leader != top) {
2810 parent = child->real_parent;
2811 leader = parent->group_leader;
2815 read_unlock(&tasklist_lock);
2818 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2819 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2822 static void sighand_ctor(void *data)
2824 struct sighand_struct *sighand = data;
2826 spin_lock_init(&sighand->siglock);
2827 init_waitqueue_head(&sighand->signalfd_wqh);
2830 void __init proc_caches_init(void)
2832 unsigned int mm_size;
2834 sighand_cachep = kmem_cache_create("sighand_cache",
2835 sizeof(struct sighand_struct), 0,
2836 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2837 SLAB_ACCOUNT, sighand_ctor);
2838 signal_cachep = kmem_cache_create("signal_cache",
2839 sizeof(struct signal_struct), 0,
2840 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2842 files_cachep = kmem_cache_create("files_cache",
2843 sizeof(struct files_struct), 0,
2844 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2846 fs_cachep = kmem_cache_create("fs_cache",
2847 sizeof(struct fs_struct), 0,
2848 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2852 * The mm_cpumask is located at the end of mm_struct, and is
2853 * dynamically sized based on the maximum CPU number this system
2854 * can have, taking hotplug into account (nr_cpu_ids).
2856 mm_size = sizeof(struct mm_struct) + cpumask_size();
2858 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2859 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2860 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2861 offsetof(struct mm_struct, saved_auxv),
2862 sizeof_field(struct mm_struct, saved_auxv),
2864 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2866 nsproxy_cache_init();
2870 * Check constraints on flags passed to the unshare system call.
2872 static int check_unshare_flags(unsigned long unshare_flags)
2874 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2875 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2876 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2877 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2881 * Not implemented, but pretend it works if there is nothing
2882 * to unshare. Note that unsharing the address space or the
2883 * signal handlers also need to unshare the signal queues (aka
2886 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2887 if (!thread_group_empty(current))
2890 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2891 if (refcount_read(¤t->sighand->count) > 1)
2894 if (unshare_flags & CLONE_VM) {
2895 if (!current_is_single_threaded())
2903 * Unshare the filesystem structure if it is being shared
2905 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2907 struct fs_struct *fs = current->fs;
2909 if (!(unshare_flags & CLONE_FS) || !fs)
2912 /* don't need lock here; in the worst case we'll do useless copy */
2916 *new_fsp = copy_fs_struct(fs);
2924 * Unshare file descriptor table if it is being shared
2926 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2927 struct files_struct **new_fdp)
2929 struct files_struct *fd = current->files;
2932 if ((unshare_flags & CLONE_FILES) &&
2933 (fd && atomic_read(&fd->count) > 1)) {
2934 *new_fdp = dup_fd(fd, max_fds, &error);
2943 * unshare allows a process to 'unshare' part of the process
2944 * context which was originally shared using clone. copy_*
2945 * functions used by kernel_clone() cannot be used here directly
2946 * because they modify an inactive task_struct that is being
2947 * constructed. Here we are modifying the current, active,
2950 int ksys_unshare(unsigned long unshare_flags)
2952 struct fs_struct *fs, *new_fs = NULL;
2953 struct files_struct *fd, *new_fd = NULL;
2954 struct cred *new_cred = NULL;
2955 struct nsproxy *new_nsproxy = NULL;
2960 * If unsharing a user namespace must also unshare the thread group
2961 * and unshare the filesystem root and working directories.
2963 if (unshare_flags & CLONE_NEWUSER)
2964 unshare_flags |= CLONE_THREAD | CLONE_FS;
2966 * If unsharing vm, must also unshare signal handlers.
2968 if (unshare_flags & CLONE_VM)
2969 unshare_flags |= CLONE_SIGHAND;
2971 * If unsharing a signal handlers, must also unshare the signal queues.
2973 if (unshare_flags & CLONE_SIGHAND)
2974 unshare_flags |= CLONE_THREAD;
2976 * If unsharing namespace, must also unshare filesystem information.
2978 if (unshare_flags & CLONE_NEWNS)
2979 unshare_flags |= CLONE_FS;
2981 err = check_unshare_flags(unshare_flags);
2983 goto bad_unshare_out;
2985 * CLONE_NEWIPC must also detach from the undolist: after switching
2986 * to a new ipc namespace, the semaphore arrays from the old
2987 * namespace are unreachable.
2989 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2991 err = unshare_fs(unshare_flags, &new_fs);
2993 goto bad_unshare_out;
2994 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2996 goto bad_unshare_cleanup_fs;
2997 err = unshare_userns(unshare_flags, &new_cred);
2999 goto bad_unshare_cleanup_fd;
3000 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3003 goto bad_unshare_cleanup_cred;
3005 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3008 * CLONE_SYSVSEM is equivalent to sys_exit().
3012 if (unshare_flags & CLONE_NEWIPC) {
3013 /* Orphan segments in old ns (see sem above). */
3015 shm_init_task(current);
3019 switch_task_namespaces(current, new_nsproxy);
3025 spin_lock(&fs->lock);
3026 current->fs = new_fs;
3031 spin_unlock(&fs->lock);
3035 fd = current->files;
3036 current->files = new_fd;
3040 task_unlock(current);
3043 /* Install the new user namespace */
3044 commit_creds(new_cred);
3049 perf_event_namespaces(current);
3051 bad_unshare_cleanup_cred:
3054 bad_unshare_cleanup_fd:
3056 put_files_struct(new_fd);
3058 bad_unshare_cleanup_fs:
3060 free_fs_struct(new_fs);
3066 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3068 return ksys_unshare(unshare_flags);
3072 * Helper to unshare the files of the current task.
3073 * We don't want to expose copy_files internals to
3074 * the exec layer of the kernel.
3077 int unshare_files(void)
3079 struct task_struct *task = current;
3080 struct files_struct *old, *copy = NULL;
3083 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3091 put_files_struct(old);
3095 int sysctl_max_threads(struct ctl_table *table, int write,
3096 void *buffer, size_t *lenp, loff_t *ppos)
3100 int threads = max_threads;
3102 int max = MAX_THREADS;
3109 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3113 max_threads = threads;