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);
386 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
387 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
388 account * (PAGE_SIZE / 1024));
390 /* All stack pages are in the same node. */
391 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
392 account * (THREAD_SIZE / 1024));
396 static int memcg_charge_kernel_stack(struct task_struct *tsk)
398 #ifdef CONFIG_VMAP_STACK
399 struct vm_struct *vm = task_stack_vm_area(tsk);
402 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
407 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
409 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
411 * If memcg_kmem_charge_page() fails, page's
412 * memory cgroup pointer is NULL, and
413 * memcg_kmem_uncharge_page() in free_thread_stack()
414 * will ignore this page.
416 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
426 static void release_task_stack(struct task_struct *tsk)
428 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
429 return; /* Better to leak the stack than to free prematurely */
431 account_kernel_stack(tsk, -1);
432 free_thread_stack(tsk);
434 #ifdef CONFIG_VMAP_STACK
435 tsk->stack_vm_area = NULL;
439 #ifdef CONFIG_THREAD_INFO_IN_TASK
440 void put_task_stack(struct task_struct *tsk)
442 if (refcount_dec_and_test(&tsk->stack_refcount))
443 release_task_stack(tsk);
447 void free_task(struct task_struct *tsk)
451 #ifndef CONFIG_THREAD_INFO_IN_TASK
453 * The task is finally done with both the stack and thread_info,
456 release_task_stack(tsk);
459 * If the task had a separate stack allocation, it should be gone
462 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
464 rt_mutex_debug_task_free(tsk);
465 ftrace_graph_exit_task(tsk);
466 arch_release_task_struct(tsk);
467 if (tsk->flags & PF_KTHREAD)
468 free_kthread_struct(tsk);
469 free_task_struct(tsk);
471 EXPORT_SYMBOL(free_task);
474 static __latent_entropy int dup_mmap(struct mm_struct *mm,
475 struct mm_struct *oldmm)
477 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
478 struct rb_node **rb_link, *rb_parent;
480 unsigned long charge;
483 uprobe_start_dup_mmap();
484 if (mmap_write_lock_killable(oldmm)) {
486 goto fail_uprobe_end;
488 flush_cache_dup_mm(oldmm);
489 uprobe_dup_mmap(oldmm, mm);
491 * Not linked in yet - no deadlock potential:
493 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
495 /* No ordering required: file already has been exposed. */
496 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
498 mm->total_vm = oldmm->total_vm;
499 mm->data_vm = oldmm->data_vm;
500 mm->exec_vm = oldmm->exec_vm;
501 mm->stack_vm = oldmm->stack_vm;
503 rb_link = &mm->mm_rb.rb_node;
506 retval = ksm_fork(mm, oldmm);
509 retval = khugepaged_fork(mm, oldmm);
514 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
517 if (mpnt->vm_flags & VM_DONTCOPY) {
518 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
523 * Don't duplicate many vmas if we've been oom-killed (for
526 if (fatal_signal_pending(current)) {
530 if (mpnt->vm_flags & VM_ACCOUNT) {
531 unsigned long len = vma_pages(mpnt);
533 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
537 tmp = vm_area_dup(mpnt);
540 retval = vma_dup_policy(mpnt, tmp);
542 goto fail_nomem_policy;
544 retval = dup_userfaultfd(tmp, &uf);
546 goto fail_nomem_anon_vma_fork;
547 if (tmp->vm_flags & VM_WIPEONFORK) {
549 * VM_WIPEONFORK gets a clean slate in the child.
550 * Don't prepare anon_vma until fault since we don't
551 * copy page for current vma.
553 tmp->anon_vma = NULL;
554 } else if (anon_vma_fork(tmp, mpnt))
555 goto fail_nomem_anon_vma_fork;
556 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
559 struct inode *inode = file_inode(file);
560 struct address_space *mapping = file->f_mapping;
563 if (tmp->vm_flags & VM_DENYWRITE)
564 put_write_access(inode);
565 i_mmap_lock_write(mapping);
566 if (tmp->vm_flags & VM_SHARED)
567 mapping_allow_writable(mapping);
568 flush_dcache_mmap_lock(mapping);
569 /* insert tmp into the share list, just after mpnt */
570 vma_interval_tree_insert_after(tmp, mpnt,
572 flush_dcache_mmap_unlock(mapping);
573 i_mmap_unlock_write(mapping);
577 * Clear hugetlb-related page reserves for children. This only
578 * affects MAP_PRIVATE mappings. Faults generated by the child
579 * are not guaranteed to succeed, even if read-only
581 if (is_vm_hugetlb_page(tmp))
582 reset_vma_resv_huge_pages(tmp);
585 * Link in the new vma and copy the page table entries.
588 pprev = &tmp->vm_next;
592 __vma_link_rb(mm, tmp, rb_link, rb_parent);
593 rb_link = &tmp->vm_rb.rb_right;
594 rb_parent = &tmp->vm_rb;
597 if (!(tmp->vm_flags & VM_WIPEONFORK))
598 retval = copy_page_range(tmp, mpnt);
600 if (tmp->vm_ops && tmp->vm_ops->open)
601 tmp->vm_ops->open(tmp);
606 /* a new mm has just been created */
607 retval = arch_dup_mmap(oldmm, mm);
609 mmap_write_unlock(mm);
611 mmap_write_unlock(oldmm);
612 dup_userfaultfd_complete(&uf);
614 uprobe_end_dup_mmap();
616 fail_nomem_anon_vma_fork:
617 mpol_put(vma_policy(tmp));
622 vm_unacct_memory(charge);
626 static inline int mm_alloc_pgd(struct mm_struct *mm)
628 mm->pgd = pgd_alloc(mm);
629 if (unlikely(!mm->pgd))
634 static inline void mm_free_pgd(struct mm_struct *mm)
636 pgd_free(mm, mm->pgd);
639 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
641 mmap_write_lock(oldmm);
642 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
643 mmap_write_unlock(oldmm);
646 #define mm_alloc_pgd(mm) (0)
647 #define mm_free_pgd(mm)
648 #endif /* CONFIG_MMU */
650 static void check_mm(struct mm_struct *mm)
654 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
655 "Please make sure 'struct resident_page_types[]' is updated as well");
657 for (i = 0; i < NR_MM_COUNTERS; i++) {
658 long x = atomic_long_read(&mm->rss_stat.count[i]);
661 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
662 mm, resident_page_types[i], x);
665 if (mm_pgtables_bytes(mm))
666 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
667 mm_pgtables_bytes(mm));
669 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
670 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
674 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
675 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
678 * Called when the last reference to the mm
679 * is dropped: either by a lazy thread or by
680 * mmput. Free the page directory and the mm.
682 void __mmdrop(struct mm_struct *mm)
684 BUG_ON(mm == &init_mm);
685 WARN_ON_ONCE(mm == current->mm);
686 WARN_ON_ONCE(mm == current->active_mm);
689 mmu_notifier_subscriptions_destroy(mm);
691 put_user_ns(mm->user_ns);
694 EXPORT_SYMBOL_GPL(__mmdrop);
696 static void mmdrop_async_fn(struct work_struct *work)
698 struct mm_struct *mm;
700 mm = container_of(work, struct mm_struct, async_put_work);
704 static void mmdrop_async(struct mm_struct *mm)
706 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
707 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
708 schedule_work(&mm->async_put_work);
712 static inline void free_signal_struct(struct signal_struct *sig)
714 taskstats_tgid_free(sig);
715 sched_autogroup_exit(sig);
717 * __mmdrop is not safe to call from softirq context on x86 due to
718 * pgd_dtor so postpone it to the async context
721 mmdrop_async(sig->oom_mm);
722 kmem_cache_free(signal_cachep, sig);
725 static inline void put_signal_struct(struct signal_struct *sig)
727 if (refcount_dec_and_test(&sig->sigcnt))
728 free_signal_struct(sig);
731 void __put_task_struct(struct task_struct *tsk)
733 WARN_ON(!tsk->exit_state);
734 WARN_ON(refcount_read(&tsk->usage));
735 WARN_ON(tsk == current);
739 task_numa_free(tsk, true);
740 security_task_free(tsk);
741 bpf_task_storage_free(tsk);
743 delayacct_tsk_free(tsk);
744 put_signal_struct(tsk->signal);
745 sched_core_free(tsk);
747 if (!profile_handoff_task(tsk))
750 EXPORT_SYMBOL_GPL(__put_task_struct);
752 void __init __weak arch_task_cache_init(void) { }
757 static void set_max_threads(unsigned int max_threads_suggested)
760 unsigned long nr_pages = totalram_pages();
763 * The number of threads shall be limited such that the thread
764 * structures may only consume a small part of the available memory.
766 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
767 threads = MAX_THREADS;
769 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
770 (u64) THREAD_SIZE * 8UL);
772 if (threads > max_threads_suggested)
773 threads = max_threads_suggested;
775 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
778 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
779 /* Initialized by the architecture: */
780 int arch_task_struct_size __read_mostly;
783 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
784 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
786 /* Fetch thread_struct whitelist for the architecture. */
787 arch_thread_struct_whitelist(offset, size);
790 * Handle zero-sized whitelist or empty thread_struct, otherwise
791 * adjust offset to position of thread_struct in task_struct.
793 if (unlikely(*size == 0))
796 *offset += offsetof(struct task_struct, thread);
798 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
800 void __init fork_init(void)
803 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
804 #ifndef ARCH_MIN_TASKALIGN
805 #define ARCH_MIN_TASKALIGN 0
807 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
808 unsigned long useroffset, usersize;
810 /* create a slab on which task_structs can be allocated */
811 task_struct_whitelist(&useroffset, &usersize);
812 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
813 arch_task_struct_size, align,
814 SLAB_PANIC|SLAB_ACCOUNT,
815 useroffset, usersize, NULL);
818 /* do the arch specific task caches init */
819 arch_task_cache_init();
821 set_max_threads(MAX_THREADS);
823 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
824 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
825 init_task.signal->rlim[RLIMIT_SIGPENDING] =
826 init_task.signal->rlim[RLIMIT_NPROC];
828 for (i = 0; i < UCOUNT_COUNTS; i++)
829 init_user_ns.ucount_max[i] = max_threads/2;
831 #ifdef CONFIG_VMAP_STACK
832 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
833 NULL, free_vm_stack_cache);
838 lockdep_init_task(&init_task);
842 int __weak arch_dup_task_struct(struct task_struct *dst,
843 struct task_struct *src)
849 void set_task_stack_end_magic(struct task_struct *tsk)
851 unsigned long *stackend;
853 stackend = end_of_stack(tsk);
854 *stackend = STACK_END_MAGIC; /* for overflow detection */
857 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
859 struct task_struct *tsk;
860 unsigned long *stack;
861 struct vm_struct *stack_vm_area __maybe_unused;
864 if (node == NUMA_NO_NODE)
865 node = tsk_fork_get_node(orig);
866 tsk = alloc_task_struct_node(node);
870 stack = alloc_thread_stack_node(tsk, node);
874 if (memcg_charge_kernel_stack(tsk))
877 stack_vm_area = task_stack_vm_area(tsk);
879 err = arch_dup_task_struct(tsk, orig);
882 * arch_dup_task_struct() clobbers the stack-related fields. Make
883 * sure they're properly initialized before using any stack-related
887 #ifdef CONFIG_VMAP_STACK
888 tsk->stack_vm_area = stack_vm_area;
890 #ifdef CONFIG_THREAD_INFO_IN_TASK
891 refcount_set(&tsk->stack_refcount, 1);
897 err = scs_prepare(tsk, node);
901 #ifdef CONFIG_SECCOMP
903 * We must handle setting up seccomp filters once we're under
904 * the sighand lock in case orig has changed between now and
905 * then. Until then, filter must be NULL to avoid messing up
906 * the usage counts on the error path calling free_task.
908 tsk->seccomp.filter = NULL;
911 setup_thread_stack(tsk, orig);
912 clear_user_return_notifier(tsk);
913 clear_tsk_need_resched(tsk);
914 set_task_stack_end_magic(tsk);
915 clear_syscall_work_syscall_user_dispatch(tsk);
917 #ifdef CONFIG_STACKPROTECTOR
918 tsk->stack_canary = get_random_canary();
920 if (orig->cpus_ptr == &orig->cpus_mask)
921 tsk->cpus_ptr = &tsk->cpus_mask;
924 * One for the user space visible state that goes away when reaped.
925 * One for the scheduler.
927 refcount_set(&tsk->rcu_users, 2);
928 /* One for the rcu users */
929 refcount_set(&tsk->usage, 1);
930 #ifdef CONFIG_BLK_DEV_IO_TRACE
933 tsk->splice_pipe = NULL;
934 tsk->task_frag.page = NULL;
935 tsk->wake_q.next = NULL;
936 tsk->pf_io_worker = NULL;
938 account_kernel_stack(tsk, 1);
941 kmap_local_fork(tsk);
943 #ifdef CONFIG_FAULT_INJECTION
947 #ifdef CONFIG_BLK_CGROUP
948 tsk->throttle_queue = NULL;
949 tsk->use_memdelay = 0;
953 tsk->active_memcg = NULL;
958 free_thread_stack(tsk);
960 free_task_struct(tsk);
964 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
966 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
968 static int __init coredump_filter_setup(char *s)
970 default_dump_filter =
971 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
972 MMF_DUMP_FILTER_MASK;
976 __setup("coredump_filter=", coredump_filter_setup);
978 #include <linux/init_task.h>
980 static void mm_init_aio(struct mm_struct *mm)
983 spin_lock_init(&mm->ioctx_lock);
984 mm->ioctx_table = NULL;
988 static __always_inline void mm_clear_owner(struct mm_struct *mm,
989 struct task_struct *p)
993 WRITE_ONCE(mm->owner, NULL);
997 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1004 static void mm_init_pasid(struct mm_struct *mm)
1006 #ifdef CONFIG_IOMMU_SUPPORT
1007 mm->pasid = INIT_PASID;
1011 static void mm_init_uprobes_state(struct mm_struct *mm)
1013 #ifdef CONFIG_UPROBES
1014 mm->uprobes_state.xol_area = NULL;
1018 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1019 struct user_namespace *user_ns)
1022 mm->mm_rb = RB_ROOT;
1023 mm->vmacache_seqnum = 0;
1024 atomic_set(&mm->mm_users, 1);
1025 atomic_set(&mm->mm_count, 1);
1026 seqcount_init(&mm->write_protect_seq);
1028 INIT_LIST_HEAD(&mm->mmlist);
1029 mm->core_state = NULL;
1030 mm_pgtables_bytes_init(mm);
1033 atomic_set(&mm->has_pinned, 0);
1034 atomic64_set(&mm->pinned_vm, 0);
1035 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1036 spin_lock_init(&mm->page_table_lock);
1037 spin_lock_init(&mm->arg_lock);
1038 mm_init_cpumask(mm);
1040 mm_init_owner(mm, p);
1042 RCU_INIT_POINTER(mm->exe_file, NULL);
1043 mmu_notifier_subscriptions_init(mm);
1044 init_tlb_flush_pending(mm);
1045 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1046 mm->pmd_huge_pte = NULL;
1048 mm_init_uprobes_state(mm);
1051 mm->flags = current->mm->flags & MMF_INIT_MASK;
1052 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1054 mm->flags = default_dump_filter;
1058 if (mm_alloc_pgd(mm))
1061 if (init_new_context(p, mm))
1062 goto fail_nocontext;
1064 mm->user_ns = get_user_ns(user_ns);
1075 * Allocate and initialize an mm_struct.
1077 struct mm_struct *mm_alloc(void)
1079 struct mm_struct *mm;
1085 memset(mm, 0, sizeof(*mm));
1086 return mm_init(mm, current, current_user_ns());
1089 static inline void __mmput(struct mm_struct *mm)
1091 VM_BUG_ON(atomic_read(&mm->mm_users));
1093 uprobe_clear_state(mm);
1096 khugepaged_exit(mm); /* must run before exit_mmap */
1098 mm_put_huge_zero_page(mm);
1099 set_mm_exe_file(mm, NULL);
1100 if (!list_empty(&mm->mmlist)) {
1101 spin_lock(&mmlist_lock);
1102 list_del(&mm->mmlist);
1103 spin_unlock(&mmlist_lock);
1106 module_put(mm->binfmt->module);
1111 * Decrement the use count and release all resources for an mm.
1113 void mmput(struct mm_struct *mm)
1117 if (atomic_dec_and_test(&mm->mm_users))
1120 EXPORT_SYMBOL_GPL(mmput);
1123 static void mmput_async_fn(struct work_struct *work)
1125 struct mm_struct *mm = container_of(work, struct mm_struct,
1131 void mmput_async(struct mm_struct *mm)
1133 if (atomic_dec_and_test(&mm->mm_users)) {
1134 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1135 schedule_work(&mm->async_put_work);
1141 * set_mm_exe_file - change a reference to the mm's executable file
1143 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1145 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1146 * invocations: in mmput() nobody alive left, in execve task is single
1147 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1148 * mm->exe_file, but does so without using set_mm_exe_file() in order
1149 * to avoid the need for any locks.
1151 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1153 struct file *old_exe_file;
1156 * It is safe to dereference the exe_file without RCU as
1157 * this function is only called if nobody else can access
1158 * this mm -- see comment above for justification.
1160 old_exe_file = rcu_dereference_raw(mm->exe_file);
1163 get_file(new_exe_file);
1164 rcu_assign_pointer(mm->exe_file, new_exe_file);
1170 * get_mm_exe_file - acquire a reference to the mm's executable file
1172 * Returns %NULL if mm has no associated executable file.
1173 * User must release file via fput().
1175 struct file *get_mm_exe_file(struct mm_struct *mm)
1177 struct file *exe_file;
1180 exe_file = rcu_dereference(mm->exe_file);
1181 if (exe_file && !get_file_rcu(exe_file))
1186 EXPORT_SYMBOL(get_mm_exe_file);
1189 * get_task_exe_file - acquire a reference to the task's executable file
1191 * Returns %NULL if task's mm (if any) has no associated executable file or
1192 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1193 * User must release file via fput().
1195 struct file *get_task_exe_file(struct task_struct *task)
1197 struct file *exe_file = NULL;
1198 struct mm_struct *mm;
1203 if (!(task->flags & PF_KTHREAD))
1204 exe_file = get_mm_exe_file(mm);
1209 EXPORT_SYMBOL(get_task_exe_file);
1212 * get_task_mm - acquire a reference to the task's mm
1214 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1215 * this kernel workthread has transiently adopted a user mm with use_mm,
1216 * to do its AIO) is not set and if so returns a reference to it, after
1217 * bumping up the use count. User must release the mm via mmput()
1218 * after use. Typically used by /proc and ptrace.
1220 struct mm_struct *get_task_mm(struct task_struct *task)
1222 struct mm_struct *mm;
1227 if (task->flags & PF_KTHREAD)
1235 EXPORT_SYMBOL_GPL(get_task_mm);
1237 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1239 struct mm_struct *mm;
1242 err = down_read_killable(&task->signal->exec_update_lock);
1244 return ERR_PTR(err);
1246 mm = get_task_mm(task);
1247 if (mm && mm != current->mm &&
1248 !ptrace_may_access(task, mode)) {
1250 mm = ERR_PTR(-EACCES);
1252 up_read(&task->signal->exec_update_lock);
1257 static void complete_vfork_done(struct task_struct *tsk)
1259 struct completion *vfork;
1262 vfork = tsk->vfork_done;
1263 if (likely(vfork)) {
1264 tsk->vfork_done = NULL;
1270 static int wait_for_vfork_done(struct task_struct *child,
1271 struct completion *vfork)
1275 freezer_do_not_count();
1276 cgroup_enter_frozen();
1277 killed = wait_for_completion_killable(vfork);
1278 cgroup_leave_frozen(false);
1283 child->vfork_done = NULL;
1287 put_task_struct(child);
1291 /* Please note the differences between mmput and mm_release.
1292 * mmput is called whenever we stop holding onto a mm_struct,
1293 * error success whatever.
1295 * mm_release is called after a mm_struct has been removed
1296 * from the current process.
1298 * This difference is important for error handling, when we
1299 * only half set up a mm_struct for a new process and need to restore
1300 * the old one. Because we mmput the new mm_struct before
1301 * restoring the old one. . .
1302 * Eric Biederman 10 January 1998
1304 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1306 uprobe_free_utask(tsk);
1308 /* Get rid of any cached register state */
1309 deactivate_mm(tsk, mm);
1312 * Signal userspace if we're not exiting with a core dump
1313 * because we want to leave the value intact for debugging
1316 if (tsk->clear_child_tid) {
1317 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1318 atomic_read(&mm->mm_users) > 1) {
1320 * We don't check the error code - if userspace has
1321 * not set up a proper pointer then tough luck.
1323 put_user(0, tsk->clear_child_tid);
1324 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1325 1, NULL, NULL, 0, 0);
1327 tsk->clear_child_tid = NULL;
1331 * All done, finally we can wake up parent and return this mm to him.
1332 * Also kthread_stop() uses this completion for synchronization.
1334 if (tsk->vfork_done)
1335 complete_vfork_done(tsk);
1338 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1340 futex_exit_release(tsk);
1341 mm_release(tsk, mm);
1344 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1346 futex_exec_release(tsk);
1347 mm_release(tsk, mm);
1351 * dup_mm() - duplicates an existing mm structure
1352 * @tsk: the task_struct with which the new mm will be associated.
1353 * @oldmm: the mm to duplicate.
1355 * Allocates a new mm structure and duplicates the provided @oldmm structure
1358 * Return: the duplicated mm or NULL on failure.
1360 static struct mm_struct *dup_mm(struct task_struct *tsk,
1361 struct mm_struct *oldmm)
1363 struct mm_struct *mm;
1370 memcpy(mm, oldmm, sizeof(*mm));
1372 if (!mm_init(mm, tsk, mm->user_ns))
1375 err = dup_mmap(mm, oldmm);
1379 mm->hiwater_rss = get_mm_rss(mm);
1380 mm->hiwater_vm = mm->total_vm;
1382 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1388 /* don't put binfmt in mmput, we haven't got module yet */
1390 mm_init_owner(mm, NULL);
1397 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1399 struct mm_struct *mm, *oldmm;
1401 tsk->min_flt = tsk->maj_flt = 0;
1402 tsk->nvcsw = tsk->nivcsw = 0;
1403 #ifdef CONFIG_DETECT_HUNG_TASK
1404 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1405 tsk->last_switch_time = 0;
1409 tsk->active_mm = NULL;
1412 * Are we cloning a kernel thread?
1414 * We need to steal a active VM for that..
1416 oldmm = current->mm;
1420 /* initialize the new vmacache entries */
1421 vmacache_flush(tsk);
1423 if (clone_flags & CLONE_VM) {
1427 mm = dup_mm(tsk, current->mm);
1433 tsk->active_mm = mm;
1437 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1439 struct fs_struct *fs = current->fs;
1440 if (clone_flags & CLONE_FS) {
1441 /* tsk->fs is already what we want */
1442 spin_lock(&fs->lock);
1444 spin_unlock(&fs->lock);
1448 spin_unlock(&fs->lock);
1451 tsk->fs = copy_fs_struct(fs);
1457 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1459 struct files_struct *oldf, *newf;
1463 * A background process may not have any files ...
1465 oldf = current->files;
1469 if (clone_flags & CLONE_FILES) {
1470 atomic_inc(&oldf->count);
1474 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1484 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1487 struct io_context *ioc = current->io_context;
1488 struct io_context *new_ioc;
1493 * Share io context with parent, if CLONE_IO is set
1495 if (clone_flags & CLONE_IO) {
1497 tsk->io_context = ioc;
1498 } else if (ioprio_valid(ioc->ioprio)) {
1499 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1500 if (unlikely(!new_ioc))
1503 new_ioc->ioprio = ioc->ioprio;
1504 put_io_context(new_ioc);
1510 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1512 struct sighand_struct *sig;
1514 if (clone_flags & CLONE_SIGHAND) {
1515 refcount_inc(¤t->sighand->count);
1518 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1519 RCU_INIT_POINTER(tsk->sighand, sig);
1523 refcount_set(&sig->count, 1);
1524 spin_lock_irq(¤t->sighand->siglock);
1525 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1526 spin_unlock_irq(¤t->sighand->siglock);
1528 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1529 if (clone_flags & CLONE_CLEAR_SIGHAND)
1530 flush_signal_handlers(tsk, 0);
1535 void __cleanup_sighand(struct sighand_struct *sighand)
1537 if (refcount_dec_and_test(&sighand->count)) {
1538 signalfd_cleanup(sighand);
1540 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1541 * without an RCU grace period, see __lock_task_sighand().
1543 kmem_cache_free(sighand_cachep, sighand);
1548 * Initialize POSIX timer handling for a thread group.
1550 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1552 struct posix_cputimers *pct = &sig->posix_cputimers;
1553 unsigned long cpu_limit;
1555 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1556 posix_cputimers_group_init(pct, cpu_limit);
1559 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1561 struct signal_struct *sig;
1563 if (clone_flags & CLONE_THREAD)
1566 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1571 sig->nr_threads = 1;
1572 atomic_set(&sig->live, 1);
1573 refcount_set(&sig->sigcnt, 1);
1575 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1576 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1577 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1579 init_waitqueue_head(&sig->wait_chldexit);
1580 sig->curr_target = tsk;
1581 init_sigpending(&sig->shared_pending);
1582 INIT_HLIST_HEAD(&sig->multiprocess);
1583 seqlock_init(&sig->stats_lock);
1584 prev_cputime_init(&sig->prev_cputime);
1586 #ifdef CONFIG_POSIX_TIMERS
1587 INIT_LIST_HEAD(&sig->posix_timers);
1588 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1589 sig->real_timer.function = it_real_fn;
1592 task_lock(current->group_leader);
1593 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1594 task_unlock(current->group_leader);
1596 posix_cpu_timers_init_group(sig);
1598 tty_audit_fork(sig);
1599 sched_autogroup_fork(sig);
1601 sig->oom_score_adj = current->signal->oom_score_adj;
1602 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1604 mutex_init(&sig->cred_guard_mutex);
1605 init_rwsem(&sig->exec_update_lock);
1610 static void copy_seccomp(struct task_struct *p)
1612 #ifdef CONFIG_SECCOMP
1614 * Must be called with sighand->lock held, which is common to
1615 * all threads in the group. Holding cred_guard_mutex is not
1616 * needed because this new task is not yet running and cannot
1619 assert_spin_locked(¤t->sighand->siglock);
1621 /* Ref-count the new filter user, and assign it. */
1622 get_seccomp_filter(current);
1623 p->seccomp = current->seccomp;
1626 * Explicitly enable no_new_privs here in case it got set
1627 * between the task_struct being duplicated and holding the
1628 * sighand lock. The seccomp state and nnp must be in sync.
1630 if (task_no_new_privs(current))
1631 task_set_no_new_privs(p);
1634 * If the parent gained a seccomp mode after copying thread
1635 * flags and between before we held the sighand lock, we have
1636 * to manually enable the seccomp thread flag here.
1638 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1639 set_task_syscall_work(p, SECCOMP);
1643 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1645 current->clear_child_tid = tidptr;
1647 return task_pid_vnr(current);
1650 static void rt_mutex_init_task(struct task_struct *p)
1652 raw_spin_lock_init(&p->pi_lock);
1653 #ifdef CONFIG_RT_MUTEXES
1654 p->pi_waiters = RB_ROOT_CACHED;
1655 p->pi_top_task = NULL;
1656 p->pi_blocked_on = NULL;
1660 static inline void init_task_pid_links(struct task_struct *task)
1664 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1665 INIT_HLIST_NODE(&task->pid_links[type]);
1669 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1671 if (type == PIDTYPE_PID)
1672 task->thread_pid = pid;
1674 task->signal->pids[type] = pid;
1677 static inline void rcu_copy_process(struct task_struct *p)
1679 #ifdef CONFIG_PREEMPT_RCU
1680 p->rcu_read_lock_nesting = 0;
1681 p->rcu_read_unlock_special.s = 0;
1682 p->rcu_blocked_node = NULL;
1683 INIT_LIST_HEAD(&p->rcu_node_entry);
1684 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1685 #ifdef CONFIG_TASKS_RCU
1686 p->rcu_tasks_holdout = false;
1687 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1688 p->rcu_tasks_idle_cpu = -1;
1689 #endif /* #ifdef CONFIG_TASKS_RCU */
1690 #ifdef CONFIG_TASKS_TRACE_RCU
1691 p->trc_reader_nesting = 0;
1692 p->trc_reader_special.s = 0;
1693 INIT_LIST_HEAD(&p->trc_holdout_list);
1694 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1697 struct pid *pidfd_pid(const struct file *file)
1699 if (file->f_op == &pidfd_fops)
1700 return file->private_data;
1702 return ERR_PTR(-EBADF);
1705 static int pidfd_release(struct inode *inode, struct file *file)
1707 struct pid *pid = file->private_data;
1709 file->private_data = NULL;
1714 #ifdef CONFIG_PROC_FS
1716 * pidfd_show_fdinfo - print information about a pidfd
1717 * @m: proc fdinfo file
1718 * @f: file referencing a pidfd
1721 * This function will print the pid that a given pidfd refers to in the
1722 * pid namespace of the procfs instance.
1723 * If the pid namespace of the process is not a descendant of the pid
1724 * namespace of the procfs instance 0 will be shown as its pid. This is
1725 * similar to calling getppid() on a process whose parent is outside of
1726 * its pid namespace.
1729 * If pid namespaces are supported then this function will also print
1730 * the pid of a given pidfd refers to for all descendant pid namespaces
1731 * starting from the current pid namespace of the instance, i.e. the
1732 * Pid field and the first entry in the NSpid field will be identical.
1733 * If the pid namespace of the process is not a descendant of the pid
1734 * namespace of the procfs instance 0 will be shown as its first NSpid
1735 * entry and no others will be shown.
1736 * Note that this differs from the Pid and NSpid fields in
1737 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1738 * the pid namespace of the procfs instance. The difference becomes
1739 * obvious when sending around a pidfd between pid namespaces from a
1740 * different branch of the tree, i.e. where no ancestral relation is
1741 * present between the pid namespaces:
1742 * - create two new pid namespaces ns1 and ns2 in the initial pid
1743 * namespace (also take care to create new mount namespaces in the
1744 * new pid namespace and mount procfs)
1745 * - create a process with a pidfd in ns1
1746 * - send pidfd from ns1 to ns2
1747 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1748 * have exactly one entry, which is 0
1750 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1752 struct pid *pid = f->private_data;
1753 struct pid_namespace *ns;
1756 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1757 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1758 nr = pid_nr_ns(pid, ns);
1761 seq_put_decimal_ll(m, "Pid:\t", nr);
1763 #ifdef CONFIG_PID_NS
1764 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1768 /* If nr is non-zero it means that 'pid' is valid and that
1769 * ns, i.e. the pid namespace associated with the procfs
1770 * instance, is in the pid namespace hierarchy of pid.
1771 * Start at one below the already printed level.
1773 for (i = ns->level + 1; i <= pid->level; i++)
1774 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1782 * Poll support for process exit notification.
1784 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1786 struct pid *pid = file->private_data;
1787 __poll_t poll_flags = 0;
1789 poll_wait(file, &pid->wait_pidfd, pts);
1792 * Inform pollers only when the whole thread group exits.
1793 * If the thread group leader exits before all other threads in the
1794 * group, then poll(2) should block, similar to the wait(2) family.
1796 if (thread_group_exited(pid))
1797 poll_flags = EPOLLIN | EPOLLRDNORM;
1802 const struct file_operations pidfd_fops = {
1803 .release = pidfd_release,
1805 #ifdef CONFIG_PROC_FS
1806 .show_fdinfo = pidfd_show_fdinfo,
1810 static void __delayed_free_task(struct rcu_head *rhp)
1812 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1817 static __always_inline void delayed_free_task(struct task_struct *tsk)
1819 if (IS_ENABLED(CONFIG_MEMCG))
1820 call_rcu(&tsk->rcu, __delayed_free_task);
1825 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1827 /* Skip if kernel thread */
1831 /* Skip if spawning a thread or using vfork */
1832 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1835 /* We need to synchronize with __set_oom_adj */
1836 mutex_lock(&oom_adj_mutex);
1837 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1838 /* Update the values in case they were changed after copy_signal */
1839 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1840 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1841 mutex_unlock(&oom_adj_mutex);
1845 * This creates a new process as a copy of the old one,
1846 * but does not actually start it yet.
1848 * It copies the registers, and all the appropriate
1849 * parts of the process environment (as per the clone
1850 * flags). The actual kick-off is left to the caller.
1852 static __latent_entropy struct task_struct *copy_process(
1856 struct kernel_clone_args *args)
1858 int pidfd = -1, retval;
1859 struct task_struct *p;
1860 struct multiprocess_signals delayed;
1861 struct file *pidfile = NULL;
1862 u64 clone_flags = args->flags;
1863 struct nsproxy *nsp = current->nsproxy;
1866 * Don't allow sharing the root directory with processes in a different
1869 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1870 return ERR_PTR(-EINVAL);
1872 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1873 return ERR_PTR(-EINVAL);
1876 * Thread groups must share signals as well, and detached threads
1877 * can only be started up within the thread group.
1879 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1880 return ERR_PTR(-EINVAL);
1883 * Shared signal handlers imply shared VM. By way of the above,
1884 * thread groups also imply shared VM. Blocking this case allows
1885 * for various simplifications in other code.
1887 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1888 return ERR_PTR(-EINVAL);
1891 * Siblings of global init remain as zombies on exit since they are
1892 * not reaped by their parent (swapper). To solve this and to avoid
1893 * multi-rooted process trees, prevent global and container-inits
1894 * from creating siblings.
1896 if ((clone_flags & CLONE_PARENT) &&
1897 current->signal->flags & SIGNAL_UNKILLABLE)
1898 return ERR_PTR(-EINVAL);
1901 * If the new process will be in a different pid or user namespace
1902 * do not allow it to share a thread group with the forking task.
1904 if (clone_flags & CLONE_THREAD) {
1905 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1906 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1907 return ERR_PTR(-EINVAL);
1911 * If the new process will be in a different time namespace
1912 * do not allow it to share VM or a thread group with the forking task.
1914 if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1915 if (nsp->time_ns != nsp->time_ns_for_children)
1916 return ERR_PTR(-EINVAL);
1919 if (clone_flags & CLONE_PIDFD) {
1921 * - CLONE_DETACHED is blocked so that we can potentially
1922 * reuse it later for CLONE_PIDFD.
1923 * - CLONE_THREAD is blocked until someone really needs it.
1925 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1926 return ERR_PTR(-EINVAL);
1930 * Force any signals received before this point to be delivered
1931 * before the fork happens. Collect up signals sent to multiple
1932 * processes that happen during the fork and delay them so that
1933 * they appear to happen after the fork.
1935 sigemptyset(&delayed.signal);
1936 INIT_HLIST_NODE(&delayed.node);
1938 spin_lock_irq(¤t->sighand->siglock);
1939 if (!(clone_flags & CLONE_THREAD))
1940 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1941 recalc_sigpending();
1942 spin_unlock_irq(¤t->sighand->siglock);
1943 retval = -ERESTARTNOINTR;
1944 if (task_sigpending(current))
1948 p = dup_task_struct(current, node);
1951 if (args->io_thread) {
1953 * Mark us an IO worker, and block any signal that isn't
1956 p->flags |= PF_IO_WORKER;
1957 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
1961 * This _must_ happen before we call free_task(), i.e. before we jump
1962 * to any of the bad_fork_* labels. This is to avoid freeing
1963 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1964 * kernel threads (PF_KTHREAD).
1966 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1968 * Clear TID on mm_release()?
1970 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1972 ftrace_graph_init_task(p);
1974 rt_mutex_init_task(p);
1976 lockdep_assert_irqs_enabled();
1977 #ifdef CONFIG_PROVE_LOCKING
1978 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1981 if (atomic_read(&p->real_cred->user->processes) >=
1982 task_rlimit(p, RLIMIT_NPROC)) {
1983 if (p->real_cred->user != INIT_USER &&
1984 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1987 current->flags &= ~PF_NPROC_EXCEEDED;
1989 retval = copy_creds(p, clone_flags);
1994 * If multiple threads are within copy_process(), then this check
1995 * triggers too late. This doesn't hurt, the check is only there
1996 * to stop root fork bombs.
1999 if (data_race(nr_threads >= max_threads))
2000 goto bad_fork_cleanup_count;
2002 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2003 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2004 p->flags |= PF_FORKNOEXEC;
2005 INIT_LIST_HEAD(&p->children);
2006 INIT_LIST_HEAD(&p->sibling);
2007 rcu_copy_process(p);
2008 p->vfork_done = NULL;
2009 spin_lock_init(&p->alloc_lock);
2011 init_sigpending(&p->pending);
2013 p->utime = p->stime = p->gtime = 0;
2014 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2015 p->utimescaled = p->stimescaled = 0;
2017 prev_cputime_init(&p->prev_cputime);
2019 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2020 seqcount_init(&p->vtime.seqcount);
2021 p->vtime.starttime = 0;
2022 p->vtime.state = VTIME_INACTIVE;
2025 #ifdef CONFIG_IO_URING
2029 #if defined(SPLIT_RSS_COUNTING)
2030 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2033 p->default_timer_slack_ns = current->timer_slack_ns;
2039 task_io_accounting_init(&p->ioac);
2040 acct_clear_integrals(p);
2042 posix_cputimers_init(&p->posix_cputimers);
2044 p->io_context = NULL;
2045 audit_set_context(p, NULL);
2048 p->mempolicy = mpol_dup(p->mempolicy);
2049 if (IS_ERR(p->mempolicy)) {
2050 retval = PTR_ERR(p->mempolicy);
2051 p->mempolicy = NULL;
2052 goto bad_fork_cleanup_threadgroup_lock;
2055 #ifdef CONFIG_CPUSETS
2056 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2057 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2058 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2060 #ifdef CONFIG_TRACE_IRQFLAGS
2061 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2062 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2063 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2064 p->softirqs_enabled = 1;
2065 p->softirq_context = 0;
2068 p->pagefault_disabled = 0;
2070 #ifdef CONFIG_LOCKDEP
2071 lockdep_init_task(p);
2074 #ifdef CONFIG_DEBUG_MUTEXES
2075 p->blocked_on = NULL; /* not blocked yet */
2077 #ifdef CONFIG_BCACHE
2078 p->sequential_io = 0;
2079 p->sequential_io_avg = 0;
2081 #ifdef CONFIG_BPF_SYSCALL
2082 RCU_INIT_POINTER(p->bpf_storage, NULL);
2085 /* Perform scheduler related setup. Assign this task to a CPU. */
2086 retval = sched_fork(clone_flags, p);
2088 goto bad_fork_cleanup_policy;
2090 retval = perf_event_init_task(p, clone_flags);
2092 goto bad_fork_cleanup_policy;
2093 retval = audit_alloc(p);
2095 goto bad_fork_cleanup_perf;
2096 /* copy all the process information */
2098 retval = security_task_alloc(p, clone_flags);
2100 goto bad_fork_cleanup_audit;
2101 retval = copy_semundo(clone_flags, p);
2103 goto bad_fork_cleanup_security;
2104 retval = copy_files(clone_flags, p);
2106 goto bad_fork_cleanup_semundo;
2107 retval = copy_fs(clone_flags, p);
2109 goto bad_fork_cleanup_files;
2110 retval = copy_sighand(clone_flags, p);
2112 goto bad_fork_cleanup_fs;
2113 retval = copy_signal(clone_flags, p);
2115 goto bad_fork_cleanup_sighand;
2116 retval = copy_mm(clone_flags, p);
2118 goto bad_fork_cleanup_signal;
2119 retval = copy_namespaces(clone_flags, p);
2121 goto bad_fork_cleanup_mm;
2122 retval = copy_io(clone_flags, p);
2124 goto bad_fork_cleanup_namespaces;
2125 retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2127 goto bad_fork_cleanup_io;
2129 stackleak_task_init(p);
2131 if (pid != &init_struct_pid) {
2132 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2133 args->set_tid_size);
2135 retval = PTR_ERR(pid);
2136 goto bad_fork_cleanup_thread;
2141 * This has to happen after we've potentially unshared the file
2142 * descriptor table (so that the pidfd doesn't leak into the child
2143 * if the fd table isn't shared).
2145 if (clone_flags & CLONE_PIDFD) {
2146 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2148 goto bad_fork_free_pid;
2152 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2153 O_RDWR | O_CLOEXEC);
2154 if (IS_ERR(pidfile)) {
2155 put_unused_fd(pidfd);
2156 retval = PTR_ERR(pidfile);
2157 goto bad_fork_free_pid;
2159 get_pid(pid); /* held by pidfile now */
2161 retval = put_user(pidfd, args->pidfd);
2163 goto bad_fork_put_pidfd;
2172 * sigaltstack should be cleared when sharing the same VM
2174 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2178 * Syscall tracing and stepping should be turned off in the
2179 * child regardless of CLONE_PTRACE.
2181 user_disable_single_step(p);
2182 clear_task_syscall_work(p, SYSCALL_TRACE);
2183 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2184 clear_task_syscall_work(p, SYSCALL_EMU);
2186 clear_tsk_latency_tracing(p);
2188 /* ok, now we should be set up.. */
2189 p->pid = pid_nr(pid);
2190 if (clone_flags & CLONE_THREAD) {
2191 p->group_leader = current->group_leader;
2192 p->tgid = current->tgid;
2194 p->group_leader = p;
2199 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2200 p->dirty_paused_when = 0;
2202 p->pdeath_signal = 0;
2203 INIT_LIST_HEAD(&p->thread_group);
2204 p->task_works = NULL;
2206 #ifdef CONFIG_KRETPROBES
2207 p->kretprobe_instances.first = NULL;
2211 * Ensure that the cgroup subsystem policies allow the new process to be
2212 * forked. It should be noted that the new process's css_set can be changed
2213 * between here and cgroup_post_fork() if an organisation operation is in
2216 retval = cgroup_can_fork(p, args);
2218 goto bad_fork_put_pidfd;
2221 * From this point on we must avoid any synchronous user-space
2222 * communication until we take the tasklist-lock. In particular, we do
2223 * not want user-space to be able to predict the process start-time by
2224 * stalling fork(2) after we recorded the start_time but before it is
2225 * visible to the system.
2228 p->start_time = ktime_get_ns();
2229 p->start_boottime = ktime_get_boottime_ns();
2232 * Make it visible to the rest of the system, but dont wake it up yet.
2233 * Need tasklist lock for parent etc handling!
2235 write_lock_irq(&tasklist_lock);
2237 /* CLONE_PARENT re-uses the old parent */
2238 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2239 p->real_parent = current->real_parent;
2240 p->parent_exec_id = current->parent_exec_id;
2241 if (clone_flags & CLONE_THREAD)
2242 p->exit_signal = -1;
2244 p->exit_signal = current->group_leader->exit_signal;
2246 p->real_parent = current;
2247 p->parent_exec_id = current->self_exec_id;
2248 p->exit_signal = args->exit_signal;
2251 klp_copy_process(p);
2255 spin_lock(¤t->sighand->siglock);
2258 * Copy seccomp details explicitly here, in case they were changed
2259 * before holding sighand lock.
2263 rseq_fork(p, clone_flags);
2265 /* Don't start children in a dying pid namespace */
2266 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2268 goto bad_fork_cancel_cgroup;
2271 /* Let kill terminate clone/fork in the middle */
2272 if (fatal_signal_pending(current)) {
2274 goto bad_fork_cancel_cgroup;
2277 /* past the last point of failure */
2279 fd_install(pidfd, pidfile);
2281 init_task_pid_links(p);
2282 if (likely(p->pid)) {
2283 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2285 init_task_pid(p, PIDTYPE_PID, pid);
2286 if (thread_group_leader(p)) {
2287 init_task_pid(p, PIDTYPE_TGID, pid);
2288 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2289 init_task_pid(p, PIDTYPE_SID, task_session(current));
2291 if (is_child_reaper(pid)) {
2292 ns_of_pid(pid)->child_reaper = p;
2293 p->signal->flags |= SIGNAL_UNKILLABLE;
2295 p->signal->shared_pending.signal = delayed.signal;
2296 p->signal->tty = tty_kref_get(current->signal->tty);
2298 * Inherit has_child_subreaper flag under the same
2299 * tasklist_lock with adding child to the process tree
2300 * for propagate_has_child_subreaper optimization.
2302 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2303 p->real_parent->signal->is_child_subreaper;
2304 list_add_tail(&p->sibling, &p->real_parent->children);
2305 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2306 attach_pid(p, PIDTYPE_TGID);
2307 attach_pid(p, PIDTYPE_PGID);
2308 attach_pid(p, PIDTYPE_SID);
2309 __this_cpu_inc(process_counts);
2311 current->signal->nr_threads++;
2312 atomic_inc(¤t->signal->live);
2313 refcount_inc(¤t->signal->sigcnt);
2314 task_join_group_stop(p);
2315 list_add_tail_rcu(&p->thread_group,
2316 &p->group_leader->thread_group);
2317 list_add_tail_rcu(&p->thread_node,
2318 &p->signal->thread_head);
2320 attach_pid(p, PIDTYPE_PID);
2324 hlist_del_init(&delayed.node);
2325 spin_unlock(¤t->sighand->siglock);
2326 syscall_tracepoint_update(p);
2327 write_unlock_irq(&tasklist_lock);
2329 proc_fork_connector(p);
2331 cgroup_post_fork(p, args);
2334 trace_task_newtask(p, clone_flags);
2335 uprobe_copy_process(p, clone_flags);
2337 copy_oom_score_adj(clone_flags, p);
2341 bad_fork_cancel_cgroup:
2343 spin_unlock(¤t->sighand->siglock);
2344 write_unlock_irq(&tasklist_lock);
2345 cgroup_cancel_fork(p, args);
2347 if (clone_flags & CLONE_PIDFD) {
2349 put_unused_fd(pidfd);
2352 if (pid != &init_struct_pid)
2354 bad_fork_cleanup_thread:
2356 bad_fork_cleanup_io:
2359 bad_fork_cleanup_namespaces:
2360 exit_task_namespaces(p);
2361 bad_fork_cleanup_mm:
2363 mm_clear_owner(p->mm, p);
2366 bad_fork_cleanup_signal:
2367 if (!(clone_flags & CLONE_THREAD))
2368 free_signal_struct(p->signal);
2369 bad_fork_cleanup_sighand:
2370 __cleanup_sighand(p->sighand);
2371 bad_fork_cleanup_fs:
2372 exit_fs(p); /* blocking */
2373 bad_fork_cleanup_files:
2374 exit_files(p); /* blocking */
2375 bad_fork_cleanup_semundo:
2377 bad_fork_cleanup_security:
2378 security_task_free(p);
2379 bad_fork_cleanup_audit:
2381 bad_fork_cleanup_perf:
2382 perf_event_free_task(p);
2383 bad_fork_cleanup_policy:
2384 lockdep_free_task(p);
2386 mpol_put(p->mempolicy);
2387 bad_fork_cleanup_threadgroup_lock:
2389 delayacct_tsk_free(p);
2390 bad_fork_cleanup_count:
2391 atomic_dec(&p->cred->user->processes);
2394 WRITE_ONCE(p->__state, TASK_DEAD);
2396 delayed_free_task(p);
2398 spin_lock_irq(¤t->sighand->siglock);
2399 hlist_del_init(&delayed.node);
2400 spin_unlock_irq(¤t->sighand->siglock);
2401 return ERR_PTR(retval);
2404 static inline void init_idle_pids(struct task_struct *idle)
2408 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2409 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2410 init_task_pid(idle, type, &init_struct_pid);
2414 struct task_struct * __init fork_idle(int cpu)
2416 struct task_struct *task;
2417 struct kernel_clone_args args = {
2421 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2422 if (!IS_ERR(task)) {
2423 init_idle_pids(task);
2424 init_idle(task, cpu);
2430 struct mm_struct *copy_init_mm(void)
2432 return dup_mm(NULL, &init_mm);
2436 * This is like kernel_clone(), but shaved down and tailored to just
2437 * creating io_uring workers. It returns a created task, or an error pointer.
2438 * The returned task is inactive, and the caller must fire it up through
2439 * wake_up_new_task(p). All signals are blocked in the created task.
2441 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2443 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2445 struct kernel_clone_args args = {
2446 .flags = ((lower_32_bits(flags) | CLONE_VM |
2447 CLONE_UNTRACED) & ~CSIGNAL),
2448 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2449 .stack = (unsigned long)fn,
2450 .stack_size = (unsigned long)arg,
2454 return copy_process(NULL, 0, node, &args);
2458 * Ok, this is the main fork-routine.
2460 * It copies the process, and if successful kick-starts
2461 * it and waits for it to finish using the VM if required.
2463 * args->exit_signal is expected to be checked for sanity by the caller.
2465 pid_t kernel_clone(struct kernel_clone_args *args)
2467 u64 clone_flags = args->flags;
2468 struct completion vfork;
2470 struct task_struct *p;
2475 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2476 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2477 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2478 * field in struct clone_args and it still doesn't make sense to have
2479 * them both point at the same memory location. Performing this check
2480 * here has the advantage that we don't need to have a separate helper
2481 * to check for legacy clone().
2483 if ((args->flags & CLONE_PIDFD) &&
2484 (args->flags & CLONE_PARENT_SETTID) &&
2485 (args->pidfd == args->parent_tid))
2489 * Determine whether and which event to report to ptracer. When
2490 * called from kernel_thread or CLONE_UNTRACED is explicitly
2491 * requested, no event is reported; otherwise, report if the event
2492 * for the type of forking is enabled.
2494 if (!(clone_flags & CLONE_UNTRACED)) {
2495 if (clone_flags & CLONE_VFORK)
2496 trace = PTRACE_EVENT_VFORK;
2497 else if (args->exit_signal != SIGCHLD)
2498 trace = PTRACE_EVENT_CLONE;
2500 trace = PTRACE_EVENT_FORK;
2502 if (likely(!ptrace_event_enabled(current, trace)))
2506 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2507 add_latent_entropy();
2513 * Do this prior waking up the new thread - the thread pointer
2514 * might get invalid after that point, if the thread exits quickly.
2516 trace_sched_process_fork(current, p);
2518 pid = get_task_pid(p, PIDTYPE_PID);
2521 if (clone_flags & CLONE_PARENT_SETTID)
2522 put_user(nr, args->parent_tid);
2524 if (clone_flags & CLONE_VFORK) {
2525 p->vfork_done = &vfork;
2526 init_completion(&vfork);
2530 wake_up_new_task(p);
2532 /* forking complete and child started to run, tell ptracer */
2533 if (unlikely(trace))
2534 ptrace_event_pid(trace, pid);
2536 if (clone_flags & CLONE_VFORK) {
2537 if (!wait_for_vfork_done(p, &vfork))
2538 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2546 * Create a kernel thread.
2548 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2550 struct kernel_clone_args args = {
2551 .flags = ((lower_32_bits(flags) | CLONE_VM |
2552 CLONE_UNTRACED) & ~CSIGNAL),
2553 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2554 .stack = (unsigned long)fn,
2555 .stack_size = (unsigned long)arg,
2558 return kernel_clone(&args);
2561 #ifdef __ARCH_WANT_SYS_FORK
2562 SYSCALL_DEFINE0(fork)
2565 struct kernel_clone_args args = {
2566 .exit_signal = SIGCHLD,
2569 return kernel_clone(&args);
2571 /* can not support in nommu mode */
2577 #ifdef __ARCH_WANT_SYS_VFORK
2578 SYSCALL_DEFINE0(vfork)
2580 struct kernel_clone_args args = {
2581 .flags = CLONE_VFORK | CLONE_VM,
2582 .exit_signal = SIGCHLD,
2585 return kernel_clone(&args);
2589 #ifdef __ARCH_WANT_SYS_CLONE
2590 #ifdef CONFIG_CLONE_BACKWARDS
2591 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2592 int __user *, parent_tidptr,
2594 int __user *, child_tidptr)
2595 #elif defined(CONFIG_CLONE_BACKWARDS2)
2596 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2597 int __user *, parent_tidptr,
2598 int __user *, child_tidptr,
2600 #elif defined(CONFIG_CLONE_BACKWARDS3)
2601 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2603 int __user *, parent_tidptr,
2604 int __user *, child_tidptr,
2607 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2608 int __user *, parent_tidptr,
2609 int __user *, child_tidptr,
2613 struct kernel_clone_args args = {
2614 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2615 .pidfd = parent_tidptr,
2616 .child_tid = child_tidptr,
2617 .parent_tid = parent_tidptr,
2618 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2623 return kernel_clone(&args);
2627 #ifdef __ARCH_WANT_SYS_CLONE3
2629 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2630 struct clone_args __user *uargs,
2634 struct clone_args args;
2635 pid_t *kset_tid = kargs->set_tid;
2637 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2638 CLONE_ARGS_SIZE_VER0);
2639 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2640 CLONE_ARGS_SIZE_VER1);
2641 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2642 CLONE_ARGS_SIZE_VER2);
2643 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2645 if (unlikely(usize > PAGE_SIZE))
2647 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2650 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2654 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2657 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2660 if (unlikely(args.set_tid && args.set_tid_size == 0))
2664 * Verify that higher 32bits of exit_signal are unset and that
2665 * it is a valid signal
2667 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2668 !valid_signal(args.exit_signal)))
2671 if ((args.flags & CLONE_INTO_CGROUP) &&
2672 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2675 *kargs = (struct kernel_clone_args){
2676 .flags = args.flags,
2677 .pidfd = u64_to_user_ptr(args.pidfd),
2678 .child_tid = u64_to_user_ptr(args.child_tid),
2679 .parent_tid = u64_to_user_ptr(args.parent_tid),
2680 .exit_signal = args.exit_signal,
2681 .stack = args.stack,
2682 .stack_size = args.stack_size,
2684 .set_tid_size = args.set_tid_size,
2685 .cgroup = args.cgroup,
2689 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2690 (kargs->set_tid_size * sizeof(pid_t))))
2693 kargs->set_tid = kset_tid;
2699 * clone3_stack_valid - check and prepare stack
2700 * @kargs: kernel clone args
2702 * Verify that the stack arguments userspace gave us are sane.
2703 * In addition, set the stack direction for userspace since it's easy for us to
2706 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2708 if (kargs->stack == 0) {
2709 if (kargs->stack_size > 0)
2712 if (kargs->stack_size == 0)
2715 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2718 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2719 kargs->stack += kargs->stack_size;
2726 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2728 /* Verify that no unknown flags are passed along. */
2730 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2734 * - make the CLONE_DETACHED bit reusable for clone3
2735 * - make the CSIGNAL bits reusable for clone3
2737 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2740 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2741 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2744 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2748 if (!clone3_stack_valid(kargs))
2755 * clone3 - create a new process with specific properties
2756 * @uargs: argument structure
2757 * @size: size of @uargs
2759 * clone3() is the extensible successor to clone()/clone2().
2760 * It takes a struct as argument that is versioned by its size.
2762 * Return: On success, a positive PID for the child process.
2763 * On error, a negative errno number.
2765 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2769 struct kernel_clone_args kargs;
2770 pid_t set_tid[MAX_PID_NS_LEVEL];
2772 kargs.set_tid = set_tid;
2774 err = copy_clone_args_from_user(&kargs, uargs, size);
2778 if (!clone3_args_valid(&kargs))
2781 return kernel_clone(&kargs);
2785 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2787 struct task_struct *leader, *parent, *child;
2790 read_lock(&tasklist_lock);
2791 leader = top = top->group_leader;
2793 for_each_thread(leader, parent) {
2794 list_for_each_entry(child, &parent->children, sibling) {
2795 res = visitor(child, data);
2807 if (leader != top) {
2809 parent = child->real_parent;
2810 leader = parent->group_leader;
2814 read_unlock(&tasklist_lock);
2817 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2818 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2821 static void sighand_ctor(void *data)
2823 struct sighand_struct *sighand = data;
2825 spin_lock_init(&sighand->siglock);
2826 init_waitqueue_head(&sighand->signalfd_wqh);
2829 void __init proc_caches_init(void)
2831 unsigned int mm_size;
2833 sighand_cachep = kmem_cache_create("sighand_cache",
2834 sizeof(struct sighand_struct), 0,
2835 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2836 SLAB_ACCOUNT, sighand_ctor);
2837 signal_cachep = kmem_cache_create("signal_cache",
2838 sizeof(struct signal_struct), 0,
2839 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2841 files_cachep = kmem_cache_create("files_cache",
2842 sizeof(struct files_struct), 0,
2843 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2845 fs_cachep = kmem_cache_create("fs_cache",
2846 sizeof(struct fs_struct), 0,
2847 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2851 * The mm_cpumask is located at the end of mm_struct, and is
2852 * dynamically sized based on the maximum CPU number this system
2853 * can have, taking hotplug into account (nr_cpu_ids).
2855 mm_size = sizeof(struct mm_struct) + cpumask_size();
2857 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2858 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2859 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2860 offsetof(struct mm_struct, saved_auxv),
2861 sizeof_field(struct mm_struct, saved_auxv),
2863 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2865 nsproxy_cache_init();
2869 * Check constraints on flags passed to the unshare system call.
2871 static int check_unshare_flags(unsigned long unshare_flags)
2873 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2874 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2875 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2876 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2880 * Not implemented, but pretend it works if there is nothing
2881 * to unshare. Note that unsharing the address space or the
2882 * signal handlers also need to unshare the signal queues (aka
2885 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2886 if (!thread_group_empty(current))
2889 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2890 if (refcount_read(¤t->sighand->count) > 1)
2893 if (unshare_flags & CLONE_VM) {
2894 if (!current_is_single_threaded())
2902 * Unshare the filesystem structure if it is being shared
2904 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2906 struct fs_struct *fs = current->fs;
2908 if (!(unshare_flags & CLONE_FS) || !fs)
2911 /* don't need lock here; in the worst case we'll do useless copy */
2915 *new_fsp = copy_fs_struct(fs);
2923 * Unshare file descriptor table if it is being shared
2925 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2926 struct files_struct **new_fdp)
2928 struct files_struct *fd = current->files;
2931 if ((unshare_flags & CLONE_FILES) &&
2932 (fd && atomic_read(&fd->count) > 1)) {
2933 *new_fdp = dup_fd(fd, max_fds, &error);
2942 * unshare allows a process to 'unshare' part of the process
2943 * context which was originally shared using clone. copy_*
2944 * functions used by kernel_clone() cannot be used here directly
2945 * because they modify an inactive task_struct that is being
2946 * constructed. Here we are modifying the current, active,
2949 int ksys_unshare(unsigned long unshare_flags)
2951 struct fs_struct *fs, *new_fs = NULL;
2952 struct files_struct *fd, *new_fd = NULL;
2953 struct cred *new_cred = NULL;
2954 struct nsproxy *new_nsproxy = NULL;
2959 * If unsharing a user namespace must also unshare the thread group
2960 * and unshare the filesystem root and working directories.
2962 if (unshare_flags & CLONE_NEWUSER)
2963 unshare_flags |= CLONE_THREAD | CLONE_FS;
2965 * If unsharing vm, must also unshare signal handlers.
2967 if (unshare_flags & CLONE_VM)
2968 unshare_flags |= CLONE_SIGHAND;
2970 * If unsharing a signal handlers, must also unshare the signal queues.
2972 if (unshare_flags & CLONE_SIGHAND)
2973 unshare_flags |= CLONE_THREAD;
2975 * If unsharing namespace, must also unshare filesystem information.
2977 if (unshare_flags & CLONE_NEWNS)
2978 unshare_flags |= CLONE_FS;
2980 err = check_unshare_flags(unshare_flags);
2982 goto bad_unshare_out;
2984 * CLONE_NEWIPC must also detach from the undolist: after switching
2985 * to a new ipc namespace, the semaphore arrays from the old
2986 * namespace are unreachable.
2988 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2990 err = unshare_fs(unshare_flags, &new_fs);
2992 goto bad_unshare_out;
2993 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2995 goto bad_unshare_cleanup_fs;
2996 err = unshare_userns(unshare_flags, &new_cred);
2998 goto bad_unshare_cleanup_fd;
2999 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3002 goto bad_unshare_cleanup_cred;
3004 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3007 * CLONE_SYSVSEM is equivalent to sys_exit().
3011 if (unshare_flags & CLONE_NEWIPC) {
3012 /* Orphan segments in old ns (see sem above). */
3014 shm_init_task(current);
3018 switch_task_namespaces(current, new_nsproxy);
3024 spin_lock(&fs->lock);
3025 current->fs = new_fs;
3030 spin_unlock(&fs->lock);
3034 fd = current->files;
3035 current->files = new_fd;
3039 task_unlock(current);
3042 /* Install the new user namespace */
3043 commit_creds(new_cred);
3048 perf_event_namespaces(current);
3050 bad_unshare_cleanup_cred:
3053 bad_unshare_cleanup_fd:
3055 put_files_struct(new_fd);
3057 bad_unshare_cleanup_fs:
3059 free_fs_struct(new_fs);
3065 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3067 return ksys_unshare(unshare_flags);
3071 * Helper to unshare the files of the current task.
3072 * We don't want to expose copy_files internals to
3073 * the exec layer of the kernel.
3076 int unshare_files(void)
3078 struct task_struct *task = current;
3079 struct files_struct *old, *copy = NULL;
3082 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3090 put_files_struct(old);
3094 int sysctl_max_threads(struct ctl_table *table, int write,
3095 void *buffer, size_t *lenp, loff_t *ppos)
3099 int threads = max_threads;
3101 int max = MAX_THREADS;
3108 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3112 max_threads = threads;