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>
99 #include <asm/pgalloc.h>
100 #include <linux/uaccess.h>
101 #include <asm/mmu_context.h>
102 #include <asm/cacheflush.h>
103 #include <asm/tlbflush.h>
105 #include <trace/events/sched.h>
107 #define CREATE_TRACE_POINTS
108 #include <trace/events/task.h>
111 * Minimum number of threads to boot the kernel
113 #define MIN_THREADS 20
116 * Maximum number of threads
118 #define MAX_THREADS FUTEX_TID_MASK
121 * Protected counters by write_lock_irq(&tasklist_lock)
123 unsigned long total_forks; /* Handle normal Linux uptimes. */
124 int nr_threads; /* The idle threads do not count.. */
126 static int max_threads; /* tunable limit on nr_threads */
128 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
130 static const char * const resident_page_types[] = {
131 NAMED_ARRAY_INDEX(MM_FILEPAGES),
132 NAMED_ARRAY_INDEX(MM_ANONPAGES),
133 NAMED_ARRAY_INDEX(MM_SWAPENTS),
134 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
137 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
139 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
141 #ifdef CONFIG_PROVE_RCU
142 int lockdep_tasklist_lock_is_held(void)
144 return lockdep_is_held(&tasklist_lock);
146 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
147 #endif /* #ifdef CONFIG_PROVE_RCU */
149 int nr_processes(void)
154 for_each_possible_cpu(cpu)
155 total += per_cpu(process_counts, cpu);
160 void __weak arch_release_task_struct(struct task_struct *tsk)
164 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
165 static struct kmem_cache *task_struct_cachep;
167 static inline struct task_struct *alloc_task_struct_node(int node)
169 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
172 static inline void free_task_struct(struct task_struct *tsk)
174 kmem_cache_free(task_struct_cachep, tsk);
178 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
181 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
182 * kmemcache based allocator.
184 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
186 #ifdef CONFIG_VMAP_STACK
188 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
189 * flush. Try to minimize the number of calls by caching stacks.
191 #define NR_CACHED_STACKS 2
192 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
194 static int free_vm_stack_cache(unsigned int cpu)
196 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
199 for (i = 0; i < NR_CACHED_STACKS; i++) {
200 struct vm_struct *vm_stack = cached_vm_stacks[i];
205 vfree(vm_stack->addr);
206 cached_vm_stacks[i] = NULL;
213 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
215 #ifdef CONFIG_VMAP_STACK
219 for (i = 0; i < NR_CACHED_STACKS; i++) {
222 s = this_cpu_xchg(cached_stacks[i], NULL);
227 /* Clear the KASAN shadow of the stack. */
228 kasan_unpoison_shadow(s->addr, THREAD_SIZE);
230 /* Clear stale pointers from reused stack. */
231 memset(s->addr, 0, THREAD_SIZE);
233 tsk->stack_vm_area = s;
234 tsk->stack = s->addr;
239 * Allocated stacks are cached and later reused by new threads,
240 * so memcg accounting is performed manually on assigning/releasing
241 * stacks to tasks. Drop __GFP_ACCOUNT.
243 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
244 VMALLOC_START, VMALLOC_END,
245 THREADINFO_GFP & ~__GFP_ACCOUNT,
247 0, node, __builtin_return_address(0));
250 * We can't call find_vm_area() in interrupt context, and
251 * free_thread_stack() can be called in interrupt context,
252 * so cache the vm_struct.
255 tsk->stack_vm_area = find_vm_area(stack);
260 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
264 tsk->stack = page_address(page);
271 static inline void free_thread_stack(struct task_struct *tsk)
273 #ifdef CONFIG_VMAP_STACK
274 struct vm_struct *vm = task_stack_vm_area(tsk);
279 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
280 memcg_kmem_uncharge_page(vm->pages[i], 0);
282 for (i = 0; i < NR_CACHED_STACKS; i++) {
283 if (this_cpu_cmpxchg(cached_stacks[i],
284 NULL, tsk->stack_vm_area) != NULL)
290 vfree_atomic(tsk->stack);
295 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
298 static struct kmem_cache *thread_stack_cache;
300 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
303 unsigned long *stack;
304 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
309 static void free_thread_stack(struct task_struct *tsk)
311 kmem_cache_free(thread_stack_cache, tsk->stack);
314 void thread_stack_cache_init(void)
316 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
317 THREAD_SIZE, THREAD_SIZE, 0, 0,
319 BUG_ON(thread_stack_cache == NULL);
324 /* SLAB cache for signal_struct structures (tsk->signal) */
325 static struct kmem_cache *signal_cachep;
327 /* SLAB cache for sighand_struct structures (tsk->sighand) */
328 struct kmem_cache *sighand_cachep;
330 /* SLAB cache for files_struct structures (tsk->files) */
331 struct kmem_cache *files_cachep;
333 /* SLAB cache for fs_struct structures (tsk->fs) */
334 struct kmem_cache *fs_cachep;
336 /* SLAB cache for vm_area_struct structures */
337 static struct kmem_cache *vm_area_cachep;
339 /* SLAB cache for mm_struct structures (tsk->mm) */
340 static struct kmem_cache *mm_cachep;
342 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
344 struct vm_area_struct *vma;
346 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
352 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
354 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
357 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
358 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
360 * orig->shared.rb may be modified concurrently, but the clone
361 * will be reinitialized.
363 *new = data_race(*orig);
364 INIT_LIST_HEAD(&new->anon_vma_chain);
365 new->vm_next = new->vm_prev = NULL;
370 void vm_area_free(struct vm_area_struct *vma)
372 kmem_cache_free(vm_area_cachep, vma);
375 static void account_kernel_stack(struct task_struct *tsk, int account)
377 void *stack = task_stack_page(tsk);
378 struct vm_struct *vm = task_stack_vm_area(tsk);
381 /* All stack pages are in the same node. */
383 mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
384 account * (THREAD_SIZE / 1024));
386 mod_lruvec_slab_state(stack, NR_KERNEL_STACK_KB,
387 account * (THREAD_SIZE / 1024));
390 static int memcg_charge_kernel_stack(struct task_struct *tsk)
392 #ifdef CONFIG_VMAP_STACK
393 struct vm_struct *vm = task_stack_vm_area(tsk);
396 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
401 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
403 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
405 * If memcg_kmem_charge_page() fails, page->mem_cgroup
406 * pointer is NULL, and memcg_kmem_uncharge_page() in
407 * free_thread_stack() will ignore this page.
409 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
419 static void release_task_stack(struct task_struct *tsk)
421 if (WARN_ON(tsk->state != TASK_DEAD))
422 return; /* Better to leak the stack than to free prematurely */
424 account_kernel_stack(tsk, -1);
425 free_thread_stack(tsk);
427 #ifdef CONFIG_VMAP_STACK
428 tsk->stack_vm_area = NULL;
432 #ifdef CONFIG_THREAD_INFO_IN_TASK
433 void put_task_stack(struct task_struct *tsk)
435 if (refcount_dec_and_test(&tsk->stack_refcount))
436 release_task_stack(tsk);
440 void free_task(struct task_struct *tsk)
444 #ifndef CONFIG_THREAD_INFO_IN_TASK
446 * The task is finally done with both the stack and thread_info,
449 release_task_stack(tsk);
452 * If the task had a separate stack allocation, it should be gone
455 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
457 rt_mutex_debug_task_free(tsk);
458 ftrace_graph_exit_task(tsk);
459 arch_release_task_struct(tsk);
460 if (tsk->flags & PF_KTHREAD)
461 free_kthread_struct(tsk);
462 free_task_struct(tsk);
464 EXPORT_SYMBOL(free_task);
467 static __latent_entropy int dup_mmap(struct mm_struct *mm,
468 struct mm_struct *oldmm)
470 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
471 struct rb_node **rb_link, *rb_parent;
473 unsigned long charge;
476 uprobe_start_dup_mmap();
477 if (mmap_write_lock_killable(oldmm)) {
479 goto fail_uprobe_end;
481 flush_cache_dup_mm(oldmm);
482 uprobe_dup_mmap(oldmm, mm);
484 * Not linked in yet - no deadlock potential:
486 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
488 /* No ordering required: file already has been exposed. */
489 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
491 mm->total_vm = oldmm->total_vm;
492 mm->data_vm = oldmm->data_vm;
493 mm->exec_vm = oldmm->exec_vm;
494 mm->stack_vm = oldmm->stack_vm;
496 rb_link = &mm->mm_rb.rb_node;
499 retval = ksm_fork(mm, oldmm);
502 retval = khugepaged_fork(mm, oldmm);
507 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
510 if (mpnt->vm_flags & VM_DONTCOPY) {
511 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
516 * Don't duplicate many vmas if we've been oom-killed (for
519 if (fatal_signal_pending(current)) {
523 if (mpnt->vm_flags & VM_ACCOUNT) {
524 unsigned long len = vma_pages(mpnt);
526 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
530 tmp = vm_area_dup(mpnt);
533 retval = vma_dup_policy(mpnt, tmp);
535 goto fail_nomem_policy;
537 retval = dup_userfaultfd(tmp, &uf);
539 goto fail_nomem_anon_vma_fork;
540 if (tmp->vm_flags & VM_WIPEONFORK) {
542 * VM_WIPEONFORK gets a clean slate in the child.
543 * Don't prepare anon_vma until fault since we don't
544 * copy page for current vma.
546 tmp->anon_vma = NULL;
547 } else if (anon_vma_fork(tmp, mpnt))
548 goto fail_nomem_anon_vma_fork;
549 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
552 struct inode *inode = file_inode(file);
553 struct address_space *mapping = file->f_mapping;
556 if (tmp->vm_flags & VM_DENYWRITE)
557 atomic_dec(&inode->i_writecount);
558 i_mmap_lock_write(mapping);
559 if (tmp->vm_flags & VM_SHARED)
560 atomic_inc(&mapping->i_mmap_writable);
561 flush_dcache_mmap_lock(mapping);
562 /* insert tmp into the share list, just after mpnt */
563 vma_interval_tree_insert_after(tmp, mpnt,
565 flush_dcache_mmap_unlock(mapping);
566 i_mmap_unlock_write(mapping);
570 * Clear hugetlb-related page reserves for children. This only
571 * affects MAP_PRIVATE mappings. Faults generated by the child
572 * are not guaranteed to succeed, even if read-only
574 if (is_vm_hugetlb_page(tmp))
575 reset_vma_resv_huge_pages(tmp);
578 * Link in the new vma and copy the page table entries.
581 pprev = &tmp->vm_next;
585 __vma_link_rb(mm, tmp, rb_link, rb_parent);
586 rb_link = &tmp->vm_rb.rb_right;
587 rb_parent = &tmp->vm_rb;
590 if (!(tmp->vm_flags & VM_WIPEONFORK))
591 retval = copy_page_range(mm, oldmm, mpnt);
593 if (tmp->vm_ops && tmp->vm_ops->open)
594 tmp->vm_ops->open(tmp);
599 /* a new mm has just been created */
600 retval = arch_dup_mmap(oldmm, mm);
602 mmap_write_unlock(mm);
604 mmap_write_unlock(oldmm);
605 dup_userfaultfd_complete(&uf);
607 uprobe_end_dup_mmap();
609 fail_nomem_anon_vma_fork:
610 mpol_put(vma_policy(tmp));
615 vm_unacct_memory(charge);
619 static inline int mm_alloc_pgd(struct mm_struct *mm)
621 mm->pgd = pgd_alloc(mm);
622 if (unlikely(!mm->pgd))
627 static inline void mm_free_pgd(struct mm_struct *mm)
629 pgd_free(mm, mm->pgd);
632 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
634 mmap_write_lock(oldmm);
635 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
636 mmap_write_unlock(oldmm);
639 #define mm_alloc_pgd(mm) (0)
640 #define mm_free_pgd(mm)
641 #endif /* CONFIG_MMU */
643 static void check_mm(struct mm_struct *mm)
647 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
648 "Please make sure 'struct resident_page_types[]' is updated as well");
650 for (i = 0; i < NR_MM_COUNTERS; i++) {
651 long x = atomic_long_read(&mm->rss_stat.count[i]);
654 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
655 mm, resident_page_types[i], x);
658 if (mm_pgtables_bytes(mm))
659 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
660 mm_pgtables_bytes(mm));
662 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
663 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
667 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
668 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
671 * Called when the last reference to the mm
672 * is dropped: either by a lazy thread or by
673 * mmput. Free the page directory and the mm.
675 void __mmdrop(struct mm_struct *mm)
677 BUG_ON(mm == &init_mm);
678 WARN_ON_ONCE(mm == current->mm);
679 WARN_ON_ONCE(mm == current->active_mm);
682 mmu_notifier_subscriptions_destroy(mm);
684 put_user_ns(mm->user_ns);
687 EXPORT_SYMBOL_GPL(__mmdrop);
689 static void mmdrop_async_fn(struct work_struct *work)
691 struct mm_struct *mm;
693 mm = container_of(work, struct mm_struct, async_put_work);
697 static void mmdrop_async(struct mm_struct *mm)
699 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
700 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
701 schedule_work(&mm->async_put_work);
705 static inline void free_signal_struct(struct signal_struct *sig)
707 taskstats_tgid_free(sig);
708 sched_autogroup_exit(sig);
710 * __mmdrop is not safe to call from softirq context on x86 due to
711 * pgd_dtor so postpone it to the async context
714 mmdrop_async(sig->oom_mm);
715 kmem_cache_free(signal_cachep, sig);
718 static inline void put_signal_struct(struct signal_struct *sig)
720 if (refcount_dec_and_test(&sig->sigcnt))
721 free_signal_struct(sig);
724 void __put_task_struct(struct task_struct *tsk)
726 WARN_ON(!tsk->exit_state);
727 WARN_ON(refcount_read(&tsk->usage));
728 WARN_ON(tsk == current);
731 task_numa_free(tsk, true);
732 security_task_free(tsk);
734 delayacct_tsk_free(tsk);
735 put_signal_struct(tsk->signal);
737 if (!profile_handoff_task(tsk))
740 EXPORT_SYMBOL_GPL(__put_task_struct);
742 void __init __weak arch_task_cache_init(void) { }
747 static void set_max_threads(unsigned int max_threads_suggested)
750 unsigned long nr_pages = totalram_pages();
753 * The number of threads shall be limited such that the thread
754 * structures may only consume a small part of the available memory.
756 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
757 threads = MAX_THREADS;
759 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
760 (u64) THREAD_SIZE * 8UL);
762 if (threads > max_threads_suggested)
763 threads = max_threads_suggested;
765 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
768 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
769 /* Initialized by the architecture: */
770 int arch_task_struct_size __read_mostly;
773 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
774 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
776 /* Fetch thread_struct whitelist for the architecture. */
777 arch_thread_struct_whitelist(offset, size);
780 * Handle zero-sized whitelist or empty thread_struct, otherwise
781 * adjust offset to position of thread_struct in task_struct.
783 if (unlikely(*size == 0))
786 *offset += offsetof(struct task_struct, thread);
788 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
790 void __init fork_init(void)
793 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
794 #ifndef ARCH_MIN_TASKALIGN
795 #define ARCH_MIN_TASKALIGN 0
797 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
798 unsigned long useroffset, usersize;
800 /* create a slab on which task_structs can be allocated */
801 task_struct_whitelist(&useroffset, &usersize);
802 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
803 arch_task_struct_size, align,
804 SLAB_PANIC|SLAB_ACCOUNT,
805 useroffset, usersize, NULL);
808 /* do the arch specific task caches init */
809 arch_task_cache_init();
811 set_max_threads(MAX_THREADS);
813 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
814 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
815 init_task.signal->rlim[RLIMIT_SIGPENDING] =
816 init_task.signal->rlim[RLIMIT_NPROC];
818 for (i = 0; i < UCOUNT_COUNTS; i++) {
819 init_user_ns.ucount_max[i] = max_threads/2;
822 #ifdef CONFIG_VMAP_STACK
823 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
824 NULL, free_vm_stack_cache);
829 lockdep_init_task(&init_task);
833 int __weak arch_dup_task_struct(struct task_struct *dst,
834 struct task_struct *src)
840 void set_task_stack_end_magic(struct task_struct *tsk)
842 unsigned long *stackend;
844 stackend = end_of_stack(tsk);
845 *stackend = STACK_END_MAGIC; /* for overflow detection */
848 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
850 struct task_struct *tsk;
851 unsigned long *stack;
852 struct vm_struct *stack_vm_area __maybe_unused;
855 if (node == NUMA_NO_NODE)
856 node = tsk_fork_get_node(orig);
857 tsk = alloc_task_struct_node(node);
861 stack = alloc_thread_stack_node(tsk, node);
865 if (memcg_charge_kernel_stack(tsk))
868 stack_vm_area = task_stack_vm_area(tsk);
870 err = arch_dup_task_struct(tsk, orig);
873 * arch_dup_task_struct() clobbers the stack-related fields. Make
874 * sure they're properly initialized before using any stack-related
878 #ifdef CONFIG_VMAP_STACK
879 tsk->stack_vm_area = stack_vm_area;
881 #ifdef CONFIG_THREAD_INFO_IN_TASK
882 refcount_set(&tsk->stack_refcount, 1);
888 err = scs_prepare(tsk, node);
892 #ifdef CONFIG_SECCOMP
894 * We must handle setting up seccomp filters once we're under
895 * the sighand lock in case orig has changed between now and
896 * then. Until then, filter must be NULL to avoid messing up
897 * the usage counts on the error path calling free_task.
899 tsk->seccomp.filter = NULL;
902 setup_thread_stack(tsk, orig);
903 clear_user_return_notifier(tsk);
904 clear_tsk_need_resched(tsk);
905 set_task_stack_end_magic(tsk);
907 #ifdef CONFIG_STACKPROTECTOR
908 tsk->stack_canary = get_random_canary();
910 if (orig->cpus_ptr == &orig->cpus_mask)
911 tsk->cpus_ptr = &tsk->cpus_mask;
914 * One for the user space visible state that goes away when reaped.
915 * One for the scheduler.
917 refcount_set(&tsk->rcu_users, 2);
918 /* One for the rcu users */
919 refcount_set(&tsk->usage, 1);
920 #ifdef CONFIG_BLK_DEV_IO_TRACE
923 tsk->splice_pipe = NULL;
924 tsk->task_frag.page = NULL;
925 tsk->wake_q.next = NULL;
927 account_kernel_stack(tsk, 1);
931 #ifdef CONFIG_FAULT_INJECTION
935 #ifdef CONFIG_BLK_CGROUP
936 tsk->throttle_queue = NULL;
937 tsk->use_memdelay = 0;
941 tsk->active_memcg = NULL;
946 free_thread_stack(tsk);
948 free_task_struct(tsk);
952 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
954 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
956 static int __init coredump_filter_setup(char *s)
958 default_dump_filter =
959 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
960 MMF_DUMP_FILTER_MASK;
964 __setup("coredump_filter=", coredump_filter_setup);
966 #include <linux/init_task.h>
968 static void mm_init_aio(struct mm_struct *mm)
971 spin_lock_init(&mm->ioctx_lock);
972 mm->ioctx_table = NULL;
976 static __always_inline void mm_clear_owner(struct mm_struct *mm,
977 struct task_struct *p)
981 WRITE_ONCE(mm->owner, NULL);
985 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
992 static void mm_init_uprobes_state(struct mm_struct *mm)
994 #ifdef CONFIG_UPROBES
995 mm->uprobes_state.xol_area = NULL;
999 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1000 struct user_namespace *user_ns)
1003 mm->mm_rb = RB_ROOT;
1004 mm->vmacache_seqnum = 0;
1005 atomic_set(&mm->mm_users, 1);
1006 atomic_set(&mm->mm_count, 1);
1008 INIT_LIST_HEAD(&mm->mmlist);
1009 mm->core_state = NULL;
1010 mm_pgtables_bytes_init(mm);
1013 atomic64_set(&mm->pinned_vm, 0);
1014 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1015 spin_lock_init(&mm->page_table_lock);
1016 spin_lock_init(&mm->arg_lock);
1017 mm_init_cpumask(mm);
1019 mm_init_owner(mm, p);
1020 RCU_INIT_POINTER(mm->exe_file, NULL);
1021 mmu_notifier_subscriptions_init(mm);
1022 init_tlb_flush_pending(mm);
1023 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1024 mm->pmd_huge_pte = NULL;
1026 mm_init_uprobes_state(mm);
1029 mm->flags = current->mm->flags & MMF_INIT_MASK;
1030 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1032 mm->flags = default_dump_filter;
1036 if (mm_alloc_pgd(mm))
1039 if (init_new_context(p, mm))
1040 goto fail_nocontext;
1042 mm->user_ns = get_user_ns(user_ns);
1053 * Allocate and initialize an mm_struct.
1055 struct mm_struct *mm_alloc(void)
1057 struct mm_struct *mm;
1063 memset(mm, 0, sizeof(*mm));
1064 return mm_init(mm, current, current_user_ns());
1067 static inline void __mmput(struct mm_struct *mm)
1069 VM_BUG_ON(atomic_read(&mm->mm_users));
1071 uprobe_clear_state(mm);
1074 khugepaged_exit(mm); /* must run before exit_mmap */
1076 mm_put_huge_zero_page(mm);
1077 set_mm_exe_file(mm, NULL);
1078 if (!list_empty(&mm->mmlist)) {
1079 spin_lock(&mmlist_lock);
1080 list_del(&mm->mmlist);
1081 spin_unlock(&mmlist_lock);
1084 module_put(mm->binfmt->module);
1089 * Decrement the use count and release all resources for an mm.
1091 void mmput(struct mm_struct *mm)
1095 if (atomic_dec_and_test(&mm->mm_users))
1098 EXPORT_SYMBOL_GPL(mmput);
1101 static void mmput_async_fn(struct work_struct *work)
1103 struct mm_struct *mm = container_of(work, struct mm_struct,
1109 void mmput_async(struct mm_struct *mm)
1111 if (atomic_dec_and_test(&mm->mm_users)) {
1112 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1113 schedule_work(&mm->async_put_work);
1119 * set_mm_exe_file - change a reference to the mm's executable file
1121 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1123 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1124 * invocations: in mmput() nobody alive left, in execve task is single
1125 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1126 * mm->exe_file, but does so without using set_mm_exe_file() in order
1127 * to do avoid the need for any locks.
1129 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1131 struct file *old_exe_file;
1134 * It is safe to dereference the exe_file without RCU as
1135 * this function is only called if nobody else can access
1136 * this mm -- see comment above for justification.
1138 old_exe_file = rcu_dereference_raw(mm->exe_file);
1141 get_file(new_exe_file);
1142 rcu_assign_pointer(mm->exe_file, new_exe_file);
1148 * get_mm_exe_file - acquire a reference to the mm's executable file
1150 * Returns %NULL if mm has no associated executable file.
1151 * User must release file via fput().
1153 struct file *get_mm_exe_file(struct mm_struct *mm)
1155 struct file *exe_file;
1158 exe_file = rcu_dereference(mm->exe_file);
1159 if (exe_file && !get_file_rcu(exe_file))
1164 EXPORT_SYMBOL(get_mm_exe_file);
1167 * get_task_exe_file - acquire a reference to the task's executable file
1169 * Returns %NULL if task's mm (if any) has no associated executable file or
1170 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1171 * User must release file via fput().
1173 struct file *get_task_exe_file(struct task_struct *task)
1175 struct file *exe_file = NULL;
1176 struct mm_struct *mm;
1181 if (!(task->flags & PF_KTHREAD))
1182 exe_file = get_mm_exe_file(mm);
1187 EXPORT_SYMBOL(get_task_exe_file);
1190 * get_task_mm - acquire a reference to the task's mm
1192 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1193 * this kernel workthread has transiently adopted a user mm with use_mm,
1194 * to do its AIO) is not set and if so returns a reference to it, after
1195 * bumping up the use count. User must release the mm via mmput()
1196 * after use. Typically used by /proc and ptrace.
1198 struct mm_struct *get_task_mm(struct task_struct *task)
1200 struct mm_struct *mm;
1205 if (task->flags & PF_KTHREAD)
1213 EXPORT_SYMBOL_GPL(get_task_mm);
1215 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1217 struct mm_struct *mm;
1220 err = mutex_lock_killable(&task->signal->exec_update_mutex);
1222 return ERR_PTR(err);
1224 mm = get_task_mm(task);
1225 if (mm && mm != current->mm &&
1226 !ptrace_may_access(task, mode)) {
1228 mm = ERR_PTR(-EACCES);
1230 mutex_unlock(&task->signal->exec_update_mutex);
1235 static void complete_vfork_done(struct task_struct *tsk)
1237 struct completion *vfork;
1240 vfork = tsk->vfork_done;
1241 if (likely(vfork)) {
1242 tsk->vfork_done = NULL;
1248 static int wait_for_vfork_done(struct task_struct *child,
1249 struct completion *vfork)
1253 freezer_do_not_count();
1254 cgroup_enter_frozen();
1255 killed = wait_for_completion_killable(vfork);
1256 cgroup_leave_frozen(false);
1261 child->vfork_done = NULL;
1265 put_task_struct(child);
1269 /* Please note the differences between mmput and mm_release.
1270 * mmput is called whenever we stop holding onto a mm_struct,
1271 * error success whatever.
1273 * mm_release is called after a mm_struct has been removed
1274 * from the current process.
1276 * This difference is important for error handling, when we
1277 * only half set up a mm_struct for a new process and need to restore
1278 * the old one. Because we mmput the new mm_struct before
1279 * restoring the old one. . .
1280 * Eric Biederman 10 January 1998
1282 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1284 uprobe_free_utask(tsk);
1286 /* Get rid of any cached register state */
1287 deactivate_mm(tsk, mm);
1290 * Signal userspace if we're not exiting with a core dump
1291 * because we want to leave the value intact for debugging
1294 if (tsk->clear_child_tid) {
1295 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1296 atomic_read(&mm->mm_users) > 1) {
1298 * We don't check the error code - if userspace has
1299 * not set up a proper pointer then tough luck.
1301 put_user(0, tsk->clear_child_tid);
1302 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1303 1, NULL, NULL, 0, 0);
1305 tsk->clear_child_tid = NULL;
1309 * All done, finally we can wake up parent and return this mm to him.
1310 * Also kthread_stop() uses this completion for synchronization.
1312 if (tsk->vfork_done)
1313 complete_vfork_done(tsk);
1316 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1318 futex_exit_release(tsk);
1319 mm_release(tsk, mm);
1322 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1324 futex_exec_release(tsk);
1325 mm_release(tsk, mm);
1329 * dup_mm() - duplicates an existing mm structure
1330 * @tsk: the task_struct with which the new mm will be associated.
1331 * @oldmm: the mm to duplicate.
1333 * Allocates a new mm structure and duplicates the provided @oldmm structure
1336 * Return: the duplicated mm or NULL on failure.
1338 static struct mm_struct *dup_mm(struct task_struct *tsk,
1339 struct mm_struct *oldmm)
1341 struct mm_struct *mm;
1348 memcpy(mm, oldmm, sizeof(*mm));
1350 if (!mm_init(mm, tsk, mm->user_ns))
1353 err = dup_mmap(mm, oldmm);
1357 mm->hiwater_rss = get_mm_rss(mm);
1358 mm->hiwater_vm = mm->total_vm;
1360 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1366 /* don't put binfmt in mmput, we haven't got module yet */
1368 mm_init_owner(mm, NULL);
1375 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1377 struct mm_struct *mm, *oldmm;
1380 tsk->min_flt = tsk->maj_flt = 0;
1381 tsk->nvcsw = tsk->nivcsw = 0;
1382 #ifdef CONFIG_DETECT_HUNG_TASK
1383 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1384 tsk->last_switch_time = 0;
1388 tsk->active_mm = NULL;
1391 * Are we cloning a kernel thread?
1393 * We need to steal a active VM for that..
1395 oldmm = current->mm;
1399 /* initialize the new vmacache entries */
1400 vmacache_flush(tsk);
1402 if (clone_flags & CLONE_VM) {
1409 mm = dup_mm(tsk, current->mm);
1415 tsk->active_mm = mm;
1422 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1424 struct fs_struct *fs = current->fs;
1425 if (clone_flags & CLONE_FS) {
1426 /* tsk->fs is already what we want */
1427 spin_lock(&fs->lock);
1429 spin_unlock(&fs->lock);
1433 spin_unlock(&fs->lock);
1436 tsk->fs = copy_fs_struct(fs);
1442 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1444 struct files_struct *oldf, *newf;
1448 * A background process may not have any files ...
1450 oldf = current->files;
1454 if (clone_flags & CLONE_FILES) {
1455 atomic_inc(&oldf->count);
1459 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1469 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1472 struct io_context *ioc = current->io_context;
1473 struct io_context *new_ioc;
1478 * Share io context with parent, if CLONE_IO is set
1480 if (clone_flags & CLONE_IO) {
1482 tsk->io_context = ioc;
1483 } else if (ioprio_valid(ioc->ioprio)) {
1484 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1485 if (unlikely(!new_ioc))
1488 new_ioc->ioprio = ioc->ioprio;
1489 put_io_context(new_ioc);
1495 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1497 struct sighand_struct *sig;
1499 if (clone_flags & CLONE_SIGHAND) {
1500 refcount_inc(¤t->sighand->count);
1503 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1504 RCU_INIT_POINTER(tsk->sighand, sig);
1508 refcount_set(&sig->count, 1);
1509 spin_lock_irq(¤t->sighand->siglock);
1510 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1511 spin_unlock_irq(¤t->sighand->siglock);
1513 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1514 if (clone_flags & CLONE_CLEAR_SIGHAND)
1515 flush_signal_handlers(tsk, 0);
1520 void __cleanup_sighand(struct sighand_struct *sighand)
1522 if (refcount_dec_and_test(&sighand->count)) {
1523 signalfd_cleanup(sighand);
1525 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1526 * without an RCU grace period, see __lock_task_sighand().
1528 kmem_cache_free(sighand_cachep, sighand);
1533 * Initialize POSIX timer handling for a thread group.
1535 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1537 struct posix_cputimers *pct = &sig->posix_cputimers;
1538 unsigned long cpu_limit;
1540 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1541 posix_cputimers_group_init(pct, cpu_limit);
1544 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1546 struct signal_struct *sig;
1548 if (clone_flags & CLONE_THREAD)
1551 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1556 sig->nr_threads = 1;
1557 atomic_set(&sig->live, 1);
1558 refcount_set(&sig->sigcnt, 1);
1560 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1561 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1562 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1564 init_waitqueue_head(&sig->wait_chldexit);
1565 sig->curr_target = tsk;
1566 init_sigpending(&sig->shared_pending);
1567 INIT_HLIST_HEAD(&sig->multiprocess);
1568 seqlock_init(&sig->stats_lock);
1569 prev_cputime_init(&sig->prev_cputime);
1571 #ifdef CONFIG_POSIX_TIMERS
1572 INIT_LIST_HEAD(&sig->posix_timers);
1573 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1574 sig->real_timer.function = it_real_fn;
1577 task_lock(current->group_leader);
1578 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1579 task_unlock(current->group_leader);
1581 posix_cpu_timers_init_group(sig);
1583 tty_audit_fork(sig);
1584 sched_autogroup_fork(sig);
1586 sig->oom_score_adj = current->signal->oom_score_adj;
1587 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1589 mutex_init(&sig->cred_guard_mutex);
1590 mutex_init(&sig->exec_update_mutex);
1595 static void copy_seccomp(struct task_struct *p)
1597 #ifdef CONFIG_SECCOMP
1599 * Must be called with sighand->lock held, which is common to
1600 * all threads in the group. Holding cred_guard_mutex is not
1601 * needed because this new task is not yet running and cannot
1604 assert_spin_locked(¤t->sighand->siglock);
1606 /* Ref-count the new filter user, and assign it. */
1607 get_seccomp_filter(current);
1608 p->seccomp = current->seccomp;
1611 * Explicitly enable no_new_privs here in case it got set
1612 * between the task_struct being duplicated and holding the
1613 * sighand lock. The seccomp state and nnp must be in sync.
1615 if (task_no_new_privs(current))
1616 task_set_no_new_privs(p);
1619 * If the parent gained a seccomp mode after copying thread
1620 * flags and between before we held the sighand lock, we have
1621 * to manually enable the seccomp thread flag here.
1623 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1624 set_tsk_thread_flag(p, TIF_SECCOMP);
1628 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1630 current->clear_child_tid = tidptr;
1632 return task_pid_vnr(current);
1635 static void rt_mutex_init_task(struct task_struct *p)
1637 raw_spin_lock_init(&p->pi_lock);
1638 #ifdef CONFIG_RT_MUTEXES
1639 p->pi_waiters = RB_ROOT_CACHED;
1640 p->pi_top_task = NULL;
1641 p->pi_blocked_on = NULL;
1645 static inline void init_task_pid_links(struct task_struct *task)
1649 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1650 INIT_HLIST_NODE(&task->pid_links[type]);
1655 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1657 if (type == PIDTYPE_PID)
1658 task->thread_pid = pid;
1660 task->signal->pids[type] = pid;
1663 static inline void rcu_copy_process(struct task_struct *p)
1665 #ifdef CONFIG_PREEMPT_RCU
1666 p->rcu_read_lock_nesting = 0;
1667 p->rcu_read_unlock_special.s = 0;
1668 p->rcu_blocked_node = NULL;
1669 INIT_LIST_HEAD(&p->rcu_node_entry);
1670 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1671 #ifdef CONFIG_TASKS_RCU
1672 p->rcu_tasks_holdout = false;
1673 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1674 p->rcu_tasks_idle_cpu = -1;
1675 #endif /* #ifdef CONFIG_TASKS_RCU */
1676 #ifdef CONFIG_TASKS_TRACE_RCU
1677 p->trc_reader_nesting = 0;
1678 p->trc_reader_special.s = 0;
1679 INIT_LIST_HEAD(&p->trc_holdout_list);
1680 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1683 struct pid *pidfd_pid(const struct file *file)
1685 if (file->f_op == &pidfd_fops)
1686 return file->private_data;
1688 return ERR_PTR(-EBADF);
1691 static int pidfd_release(struct inode *inode, struct file *file)
1693 struct pid *pid = file->private_data;
1695 file->private_data = NULL;
1700 #ifdef CONFIG_PROC_FS
1702 * pidfd_show_fdinfo - print information about a pidfd
1703 * @m: proc fdinfo file
1704 * @f: file referencing a pidfd
1707 * This function will print the pid that a given pidfd refers to in the
1708 * pid namespace of the procfs instance.
1709 * If the pid namespace of the process is not a descendant of the pid
1710 * namespace of the procfs instance 0 will be shown as its pid. This is
1711 * similar to calling getppid() on a process whose parent is outside of
1712 * its pid namespace.
1715 * If pid namespaces are supported then this function will also print
1716 * the pid of a given pidfd refers to for all descendant pid namespaces
1717 * starting from the current pid namespace of the instance, i.e. the
1718 * Pid field and the first entry in the NSpid field will be identical.
1719 * If the pid namespace of the process is not a descendant of the pid
1720 * namespace of the procfs instance 0 will be shown as its first NSpid
1721 * entry and no others will be shown.
1722 * Note that this differs from the Pid and NSpid fields in
1723 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1724 * the pid namespace of the procfs instance. The difference becomes
1725 * obvious when sending around a pidfd between pid namespaces from a
1726 * different branch of the tree, i.e. where no ancestoral relation is
1727 * present between the pid namespaces:
1728 * - create two new pid namespaces ns1 and ns2 in the initial pid
1729 * namespace (also take care to create new mount namespaces in the
1730 * new pid namespace and mount procfs)
1731 * - create a process with a pidfd in ns1
1732 * - send pidfd from ns1 to ns2
1733 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1734 * have exactly one entry, which is 0
1736 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1738 struct pid *pid = f->private_data;
1739 struct pid_namespace *ns;
1742 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1743 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1744 nr = pid_nr_ns(pid, ns);
1747 seq_put_decimal_ll(m, "Pid:\t", nr);
1749 #ifdef CONFIG_PID_NS
1750 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1754 /* If nr is non-zero it means that 'pid' is valid and that
1755 * ns, i.e. the pid namespace associated with the procfs
1756 * instance, is in the pid namespace hierarchy of pid.
1757 * Start at one below the already printed level.
1759 for (i = ns->level + 1; i <= pid->level; i++)
1760 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1768 * Poll support for process exit notification.
1770 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1772 struct pid *pid = file->private_data;
1773 __poll_t poll_flags = 0;
1775 poll_wait(file, &pid->wait_pidfd, pts);
1778 * Inform pollers only when the whole thread group exits.
1779 * If the thread group leader exits before all other threads in the
1780 * group, then poll(2) should block, similar to the wait(2) family.
1782 if (thread_group_exited(pid))
1783 poll_flags = EPOLLIN | EPOLLRDNORM;
1788 const struct file_operations pidfd_fops = {
1789 .release = pidfd_release,
1791 #ifdef CONFIG_PROC_FS
1792 .show_fdinfo = pidfd_show_fdinfo,
1796 static void __delayed_free_task(struct rcu_head *rhp)
1798 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1803 static __always_inline void delayed_free_task(struct task_struct *tsk)
1805 if (IS_ENABLED(CONFIG_MEMCG))
1806 call_rcu(&tsk->rcu, __delayed_free_task);
1812 * This creates a new process as a copy of the old one,
1813 * but does not actually start it yet.
1815 * It copies the registers, and all the appropriate
1816 * parts of the process environment (as per the clone
1817 * flags). The actual kick-off is left to the caller.
1819 static __latent_entropy struct task_struct *copy_process(
1823 struct kernel_clone_args *args)
1825 int pidfd = -1, retval;
1826 struct task_struct *p;
1827 struct multiprocess_signals delayed;
1828 struct file *pidfile = NULL;
1829 u64 clone_flags = args->flags;
1830 struct nsproxy *nsp = current->nsproxy;
1833 * Don't allow sharing the root directory with processes in a different
1836 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1837 return ERR_PTR(-EINVAL);
1839 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1840 return ERR_PTR(-EINVAL);
1843 * Thread groups must share signals as well, and detached threads
1844 * can only be started up within the thread group.
1846 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1847 return ERR_PTR(-EINVAL);
1850 * Shared signal handlers imply shared VM. By way of the above,
1851 * thread groups also imply shared VM. Blocking this case allows
1852 * for various simplifications in other code.
1854 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1855 return ERR_PTR(-EINVAL);
1858 * Siblings of global init remain as zombies on exit since they are
1859 * not reaped by their parent (swapper). To solve this and to avoid
1860 * multi-rooted process trees, prevent global and container-inits
1861 * from creating siblings.
1863 if ((clone_flags & CLONE_PARENT) &&
1864 current->signal->flags & SIGNAL_UNKILLABLE)
1865 return ERR_PTR(-EINVAL);
1868 * If the new process will be in a different pid or user namespace
1869 * do not allow it to share a thread group with the forking task.
1871 if (clone_flags & CLONE_THREAD) {
1872 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1873 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1874 return ERR_PTR(-EINVAL);
1878 * If the new process will be in a different time namespace
1879 * do not allow it to share VM or a thread group with the forking task.
1881 if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1882 if (nsp->time_ns != nsp->time_ns_for_children)
1883 return ERR_PTR(-EINVAL);
1886 if (clone_flags & CLONE_PIDFD) {
1888 * - CLONE_DETACHED is blocked so that we can potentially
1889 * reuse it later for CLONE_PIDFD.
1890 * - CLONE_THREAD is blocked until someone really needs it.
1892 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1893 return ERR_PTR(-EINVAL);
1897 * Force any signals received before this point to be delivered
1898 * before the fork happens. Collect up signals sent to multiple
1899 * processes that happen during the fork and delay them so that
1900 * they appear to happen after the fork.
1902 sigemptyset(&delayed.signal);
1903 INIT_HLIST_NODE(&delayed.node);
1905 spin_lock_irq(¤t->sighand->siglock);
1906 if (!(clone_flags & CLONE_THREAD))
1907 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1908 recalc_sigpending();
1909 spin_unlock_irq(¤t->sighand->siglock);
1910 retval = -ERESTARTNOINTR;
1911 if (signal_pending(current))
1915 p = dup_task_struct(current, node);
1920 * This _must_ happen before we call free_task(), i.e. before we jump
1921 * to any of the bad_fork_* labels. This is to avoid freeing
1922 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1923 * kernel threads (PF_KTHREAD).
1925 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1927 * Clear TID on mm_release()?
1929 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1931 ftrace_graph_init_task(p);
1933 rt_mutex_init_task(p);
1935 lockdep_assert_irqs_enabled();
1936 #ifdef CONFIG_PROVE_LOCKING
1937 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1940 if (atomic_read(&p->real_cred->user->processes) >=
1941 task_rlimit(p, RLIMIT_NPROC)) {
1942 if (p->real_cred->user != INIT_USER &&
1943 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1946 current->flags &= ~PF_NPROC_EXCEEDED;
1948 retval = copy_creds(p, clone_flags);
1953 * If multiple threads are within copy_process(), then this check
1954 * triggers too late. This doesn't hurt, the check is only there
1955 * to stop root fork bombs.
1958 if (data_race(nr_threads >= max_threads))
1959 goto bad_fork_cleanup_count;
1961 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1962 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1963 p->flags |= PF_FORKNOEXEC;
1964 INIT_LIST_HEAD(&p->children);
1965 INIT_LIST_HEAD(&p->sibling);
1966 rcu_copy_process(p);
1967 p->vfork_done = NULL;
1968 spin_lock_init(&p->alloc_lock);
1970 init_sigpending(&p->pending);
1972 p->utime = p->stime = p->gtime = 0;
1973 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1974 p->utimescaled = p->stimescaled = 0;
1976 prev_cputime_init(&p->prev_cputime);
1978 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1979 seqcount_init(&p->vtime.seqcount);
1980 p->vtime.starttime = 0;
1981 p->vtime.state = VTIME_INACTIVE;
1984 #if defined(SPLIT_RSS_COUNTING)
1985 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1988 p->default_timer_slack_ns = current->timer_slack_ns;
1994 task_io_accounting_init(&p->ioac);
1995 acct_clear_integrals(p);
1997 posix_cputimers_init(&p->posix_cputimers);
1999 p->io_context = NULL;
2000 audit_set_context(p, NULL);
2003 p->mempolicy = mpol_dup(p->mempolicy);
2004 if (IS_ERR(p->mempolicy)) {
2005 retval = PTR_ERR(p->mempolicy);
2006 p->mempolicy = NULL;
2007 goto bad_fork_cleanup_threadgroup_lock;
2010 #ifdef CONFIG_CPUSETS
2011 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2012 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2013 seqcount_init(&p->mems_allowed_seq);
2015 #ifdef CONFIG_TRACE_IRQFLAGS
2016 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2017 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2018 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2019 p->softirqs_enabled = 1;
2020 p->softirq_context = 0;
2023 p->pagefault_disabled = 0;
2025 #ifdef CONFIG_LOCKDEP
2026 lockdep_init_task(p);
2029 #ifdef CONFIG_DEBUG_MUTEXES
2030 p->blocked_on = NULL; /* not blocked yet */
2032 #ifdef CONFIG_BCACHE
2033 p->sequential_io = 0;
2034 p->sequential_io_avg = 0;
2037 /* Perform scheduler related setup. Assign this task to a CPU. */
2038 retval = sched_fork(clone_flags, p);
2040 goto bad_fork_cleanup_policy;
2042 retval = perf_event_init_task(p);
2044 goto bad_fork_cleanup_policy;
2045 retval = audit_alloc(p);
2047 goto bad_fork_cleanup_perf;
2048 /* copy all the process information */
2050 retval = security_task_alloc(p, clone_flags);
2052 goto bad_fork_cleanup_audit;
2053 retval = copy_semundo(clone_flags, p);
2055 goto bad_fork_cleanup_security;
2056 retval = copy_files(clone_flags, p);
2058 goto bad_fork_cleanup_semundo;
2059 retval = copy_fs(clone_flags, p);
2061 goto bad_fork_cleanup_files;
2062 retval = copy_sighand(clone_flags, p);
2064 goto bad_fork_cleanup_fs;
2065 retval = copy_signal(clone_flags, p);
2067 goto bad_fork_cleanup_sighand;
2068 retval = copy_mm(clone_flags, p);
2070 goto bad_fork_cleanup_signal;
2071 retval = copy_namespaces(clone_flags, p);
2073 goto bad_fork_cleanup_mm;
2074 retval = copy_io(clone_flags, p);
2076 goto bad_fork_cleanup_namespaces;
2077 retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2079 goto bad_fork_cleanup_io;
2081 stackleak_task_init(p);
2083 if (pid != &init_struct_pid) {
2084 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2085 args->set_tid_size);
2087 retval = PTR_ERR(pid);
2088 goto bad_fork_cleanup_thread;
2093 * This has to happen after we've potentially unshared the file
2094 * descriptor table (so that the pidfd doesn't leak into the child
2095 * if the fd table isn't shared).
2097 if (clone_flags & CLONE_PIDFD) {
2098 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2100 goto bad_fork_free_pid;
2104 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2105 O_RDWR | O_CLOEXEC);
2106 if (IS_ERR(pidfile)) {
2107 put_unused_fd(pidfd);
2108 retval = PTR_ERR(pidfile);
2109 goto bad_fork_free_pid;
2111 get_pid(pid); /* held by pidfile now */
2113 retval = put_user(pidfd, args->pidfd);
2115 goto bad_fork_put_pidfd;
2124 * sigaltstack should be cleared when sharing the same VM
2126 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2130 * Syscall tracing and stepping should be turned off in the
2131 * child regardless of CLONE_PTRACE.
2133 user_disable_single_step(p);
2134 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2135 #ifdef TIF_SYSCALL_EMU
2136 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2138 clear_tsk_latency_tracing(p);
2140 /* ok, now we should be set up.. */
2141 p->pid = pid_nr(pid);
2142 if (clone_flags & CLONE_THREAD) {
2143 p->exit_signal = -1;
2144 p->group_leader = current->group_leader;
2145 p->tgid = current->tgid;
2147 if (clone_flags & CLONE_PARENT)
2148 p->exit_signal = current->group_leader->exit_signal;
2150 p->exit_signal = args->exit_signal;
2151 p->group_leader = p;
2156 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2157 p->dirty_paused_when = 0;
2159 p->pdeath_signal = 0;
2160 INIT_LIST_HEAD(&p->thread_group);
2161 p->task_works = NULL;
2164 * Ensure that the cgroup subsystem policies allow the new process to be
2165 * forked. It should be noted the the new process's css_set can be changed
2166 * between here and cgroup_post_fork() if an organisation operation is in
2169 retval = cgroup_can_fork(p, args);
2171 goto bad_fork_put_pidfd;
2174 * From this point on we must avoid any synchronous user-space
2175 * communication until we take the tasklist-lock. In particular, we do
2176 * not want user-space to be able to predict the process start-time by
2177 * stalling fork(2) after we recorded the start_time but before it is
2178 * visible to the system.
2181 p->start_time = ktime_get_ns();
2182 p->start_boottime = ktime_get_boottime_ns();
2185 * Make it visible to the rest of the system, but dont wake it up yet.
2186 * Need tasklist lock for parent etc handling!
2188 write_lock_irq(&tasklist_lock);
2190 /* CLONE_PARENT re-uses the old parent */
2191 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2192 p->real_parent = current->real_parent;
2193 p->parent_exec_id = current->parent_exec_id;
2195 p->real_parent = current;
2196 p->parent_exec_id = current->self_exec_id;
2199 klp_copy_process(p);
2201 spin_lock(¤t->sighand->siglock);
2204 * Copy seccomp details explicitly here, in case they were changed
2205 * before holding sighand lock.
2209 rseq_fork(p, clone_flags);
2211 /* Don't start children in a dying pid namespace */
2212 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2214 goto bad_fork_cancel_cgroup;
2217 /* Let kill terminate clone/fork in the middle */
2218 if (fatal_signal_pending(current)) {
2220 goto bad_fork_cancel_cgroup;
2223 /* past the last point of failure */
2225 fd_install(pidfd, pidfile);
2227 init_task_pid_links(p);
2228 if (likely(p->pid)) {
2229 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2231 init_task_pid(p, PIDTYPE_PID, pid);
2232 if (thread_group_leader(p)) {
2233 init_task_pid(p, PIDTYPE_TGID, pid);
2234 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2235 init_task_pid(p, PIDTYPE_SID, task_session(current));
2237 if (is_child_reaper(pid)) {
2238 ns_of_pid(pid)->child_reaper = p;
2239 p->signal->flags |= SIGNAL_UNKILLABLE;
2241 p->signal->shared_pending.signal = delayed.signal;
2242 p->signal->tty = tty_kref_get(current->signal->tty);
2244 * Inherit has_child_subreaper flag under the same
2245 * tasklist_lock with adding child to the process tree
2246 * for propagate_has_child_subreaper optimization.
2248 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2249 p->real_parent->signal->is_child_subreaper;
2250 list_add_tail(&p->sibling, &p->real_parent->children);
2251 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2252 attach_pid(p, PIDTYPE_TGID);
2253 attach_pid(p, PIDTYPE_PGID);
2254 attach_pid(p, PIDTYPE_SID);
2255 __this_cpu_inc(process_counts);
2257 current->signal->nr_threads++;
2258 atomic_inc(¤t->signal->live);
2259 refcount_inc(¤t->signal->sigcnt);
2260 task_join_group_stop(p);
2261 list_add_tail_rcu(&p->thread_group,
2262 &p->group_leader->thread_group);
2263 list_add_tail_rcu(&p->thread_node,
2264 &p->signal->thread_head);
2266 attach_pid(p, PIDTYPE_PID);
2270 hlist_del_init(&delayed.node);
2271 spin_unlock(¤t->sighand->siglock);
2272 syscall_tracepoint_update(p);
2273 write_unlock_irq(&tasklist_lock);
2275 proc_fork_connector(p);
2277 cgroup_post_fork(p, args);
2280 trace_task_newtask(p, clone_flags);
2281 uprobe_copy_process(p, clone_flags);
2285 bad_fork_cancel_cgroup:
2286 spin_unlock(¤t->sighand->siglock);
2287 write_unlock_irq(&tasklist_lock);
2288 cgroup_cancel_fork(p, args);
2290 if (clone_flags & CLONE_PIDFD) {
2292 put_unused_fd(pidfd);
2295 if (pid != &init_struct_pid)
2297 bad_fork_cleanup_thread:
2299 bad_fork_cleanup_io:
2302 bad_fork_cleanup_namespaces:
2303 exit_task_namespaces(p);
2304 bad_fork_cleanup_mm:
2306 mm_clear_owner(p->mm, p);
2309 bad_fork_cleanup_signal:
2310 if (!(clone_flags & CLONE_THREAD))
2311 free_signal_struct(p->signal);
2312 bad_fork_cleanup_sighand:
2313 __cleanup_sighand(p->sighand);
2314 bad_fork_cleanup_fs:
2315 exit_fs(p); /* blocking */
2316 bad_fork_cleanup_files:
2317 exit_files(p); /* blocking */
2318 bad_fork_cleanup_semundo:
2320 bad_fork_cleanup_security:
2321 security_task_free(p);
2322 bad_fork_cleanup_audit:
2324 bad_fork_cleanup_perf:
2325 perf_event_free_task(p);
2326 bad_fork_cleanup_policy:
2327 lockdep_free_task(p);
2329 mpol_put(p->mempolicy);
2330 bad_fork_cleanup_threadgroup_lock:
2332 delayacct_tsk_free(p);
2333 bad_fork_cleanup_count:
2334 atomic_dec(&p->cred->user->processes);
2337 p->state = TASK_DEAD;
2339 delayed_free_task(p);
2341 spin_lock_irq(¤t->sighand->siglock);
2342 hlist_del_init(&delayed.node);
2343 spin_unlock_irq(¤t->sighand->siglock);
2344 return ERR_PTR(retval);
2347 static inline void init_idle_pids(struct task_struct *idle)
2351 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2352 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2353 init_task_pid(idle, type, &init_struct_pid);
2357 struct task_struct *fork_idle(int cpu)
2359 struct task_struct *task;
2360 struct kernel_clone_args args = {
2364 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2365 if (!IS_ERR(task)) {
2366 init_idle_pids(task);
2367 init_idle(task, cpu);
2373 struct mm_struct *copy_init_mm(void)
2375 return dup_mm(NULL, &init_mm);
2379 * Ok, this is the main fork-routine.
2381 * It copies the process, and if successful kick-starts
2382 * it and waits for it to finish using the VM if required.
2384 * args->exit_signal is expected to be checked for sanity by the caller.
2386 long _do_fork(struct kernel_clone_args *args)
2388 u64 clone_flags = args->flags;
2389 struct completion vfork;
2391 struct task_struct *p;
2396 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2397 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2398 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2399 * field in struct clone_args and it still doesn't make sense to have
2400 * them both point at the same memory location. Performing this check
2401 * here has the advantage that we don't need to have a separate helper
2402 * to check for legacy clone().
2404 if ((args->flags & CLONE_PIDFD) &&
2405 (args->flags & CLONE_PARENT_SETTID) &&
2406 (args->pidfd == args->parent_tid))
2410 * Determine whether and which event to report to ptracer. When
2411 * called from kernel_thread or CLONE_UNTRACED is explicitly
2412 * requested, no event is reported; otherwise, report if the event
2413 * for the type of forking is enabled.
2415 if (!(clone_flags & CLONE_UNTRACED)) {
2416 if (clone_flags & CLONE_VFORK)
2417 trace = PTRACE_EVENT_VFORK;
2418 else if (args->exit_signal != SIGCHLD)
2419 trace = PTRACE_EVENT_CLONE;
2421 trace = PTRACE_EVENT_FORK;
2423 if (likely(!ptrace_event_enabled(current, trace)))
2427 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2428 add_latent_entropy();
2434 * Do this prior waking up the new thread - the thread pointer
2435 * might get invalid after that point, if the thread exits quickly.
2437 trace_sched_process_fork(current, p);
2439 pid = get_task_pid(p, PIDTYPE_PID);
2442 if (clone_flags & CLONE_PARENT_SETTID)
2443 put_user(nr, args->parent_tid);
2445 if (clone_flags & CLONE_VFORK) {
2446 p->vfork_done = &vfork;
2447 init_completion(&vfork);
2451 wake_up_new_task(p);
2453 /* forking complete and child started to run, tell ptracer */
2454 if (unlikely(trace))
2455 ptrace_event_pid(trace, pid);
2457 if (clone_flags & CLONE_VFORK) {
2458 if (!wait_for_vfork_done(p, &vfork))
2459 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2467 * Create a kernel thread.
2469 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2471 struct kernel_clone_args args = {
2472 .flags = ((lower_32_bits(flags) | CLONE_VM |
2473 CLONE_UNTRACED) & ~CSIGNAL),
2474 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2475 .stack = (unsigned long)fn,
2476 .stack_size = (unsigned long)arg,
2479 return _do_fork(&args);
2482 #ifdef __ARCH_WANT_SYS_FORK
2483 SYSCALL_DEFINE0(fork)
2486 struct kernel_clone_args args = {
2487 .exit_signal = SIGCHLD,
2490 return _do_fork(&args);
2492 /* can not support in nommu mode */
2498 #ifdef __ARCH_WANT_SYS_VFORK
2499 SYSCALL_DEFINE0(vfork)
2501 struct kernel_clone_args args = {
2502 .flags = CLONE_VFORK | CLONE_VM,
2503 .exit_signal = SIGCHLD,
2506 return _do_fork(&args);
2510 #ifdef __ARCH_WANT_SYS_CLONE
2511 #ifdef CONFIG_CLONE_BACKWARDS
2512 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2513 int __user *, parent_tidptr,
2515 int __user *, child_tidptr)
2516 #elif defined(CONFIG_CLONE_BACKWARDS2)
2517 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2518 int __user *, parent_tidptr,
2519 int __user *, child_tidptr,
2521 #elif defined(CONFIG_CLONE_BACKWARDS3)
2522 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2524 int __user *, parent_tidptr,
2525 int __user *, child_tidptr,
2528 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2529 int __user *, parent_tidptr,
2530 int __user *, child_tidptr,
2534 struct kernel_clone_args args = {
2535 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2536 .pidfd = parent_tidptr,
2537 .child_tid = child_tidptr,
2538 .parent_tid = parent_tidptr,
2539 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2544 return _do_fork(&args);
2548 #ifdef __ARCH_WANT_SYS_CLONE3
2550 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2551 struct clone_args __user *uargs,
2555 struct clone_args args;
2556 pid_t *kset_tid = kargs->set_tid;
2558 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2559 CLONE_ARGS_SIZE_VER0);
2560 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2561 CLONE_ARGS_SIZE_VER1);
2562 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2563 CLONE_ARGS_SIZE_VER2);
2564 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2566 if (unlikely(usize > PAGE_SIZE))
2568 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2571 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2575 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2578 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2581 if (unlikely(args.set_tid && args.set_tid_size == 0))
2585 * Verify that higher 32bits of exit_signal are unset and that
2586 * it is a valid signal
2588 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2589 !valid_signal(args.exit_signal)))
2592 if ((args.flags & CLONE_INTO_CGROUP) &&
2593 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2596 *kargs = (struct kernel_clone_args){
2597 .flags = args.flags,
2598 .pidfd = u64_to_user_ptr(args.pidfd),
2599 .child_tid = u64_to_user_ptr(args.child_tid),
2600 .parent_tid = u64_to_user_ptr(args.parent_tid),
2601 .exit_signal = args.exit_signal,
2602 .stack = args.stack,
2603 .stack_size = args.stack_size,
2605 .set_tid_size = args.set_tid_size,
2606 .cgroup = args.cgroup,
2610 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2611 (kargs->set_tid_size * sizeof(pid_t))))
2614 kargs->set_tid = kset_tid;
2620 * clone3_stack_valid - check and prepare stack
2621 * @kargs: kernel clone args
2623 * Verify that the stack arguments userspace gave us are sane.
2624 * In addition, set the stack direction for userspace since it's easy for us to
2627 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2629 if (kargs->stack == 0) {
2630 if (kargs->stack_size > 0)
2633 if (kargs->stack_size == 0)
2636 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2639 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2640 kargs->stack += kargs->stack_size;
2647 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2649 /* Verify that no unknown flags are passed along. */
2651 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2655 * - make the CLONE_DETACHED bit reuseable for clone3
2656 * - make the CSIGNAL bits reuseable for clone3
2658 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2661 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2662 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2665 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2669 if (!clone3_stack_valid(kargs))
2676 * clone3 - create a new process with specific properties
2677 * @uargs: argument structure
2678 * @size: size of @uargs
2680 * clone3() is the extensible successor to clone()/clone2().
2681 * It takes a struct as argument that is versioned by its size.
2683 * Return: On success, a positive PID for the child process.
2684 * On error, a negative errno number.
2686 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2690 struct kernel_clone_args kargs;
2691 pid_t set_tid[MAX_PID_NS_LEVEL];
2693 kargs.set_tid = set_tid;
2695 err = copy_clone_args_from_user(&kargs, uargs, size);
2699 if (!clone3_args_valid(&kargs))
2702 return _do_fork(&kargs);
2706 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2708 struct task_struct *leader, *parent, *child;
2711 read_lock(&tasklist_lock);
2712 leader = top = top->group_leader;
2714 for_each_thread(leader, parent) {
2715 list_for_each_entry(child, &parent->children, sibling) {
2716 res = visitor(child, data);
2728 if (leader != top) {
2730 parent = child->real_parent;
2731 leader = parent->group_leader;
2735 read_unlock(&tasklist_lock);
2738 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2739 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2742 static void sighand_ctor(void *data)
2744 struct sighand_struct *sighand = data;
2746 spin_lock_init(&sighand->siglock);
2747 init_waitqueue_head(&sighand->signalfd_wqh);
2750 void __init proc_caches_init(void)
2752 unsigned int mm_size;
2754 sighand_cachep = kmem_cache_create("sighand_cache",
2755 sizeof(struct sighand_struct), 0,
2756 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2757 SLAB_ACCOUNT, sighand_ctor);
2758 signal_cachep = kmem_cache_create("signal_cache",
2759 sizeof(struct signal_struct), 0,
2760 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2762 files_cachep = kmem_cache_create("files_cache",
2763 sizeof(struct files_struct), 0,
2764 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2766 fs_cachep = kmem_cache_create("fs_cache",
2767 sizeof(struct fs_struct), 0,
2768 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2772 * The mm_cpumask is located at the end of mm_struct, and is
2773 * dynamically sized based on the maximum CPU number this system
2774 * can have, taking hotplug into account (nr_cpu_ids).
2776 mm_size = sizeof(struct mm_struct) + cpumask_size();
2778 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2779 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2780 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2781 offsetof(struct mm_struct, saved_auxv),
2782 sizeof_field(struct mm_struct, saved_auxv),
2784 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2786 nsproxy_cache_init();
2790 * Check constraints on flags passed to the unshare system call.
2792 static int check_unshare_flags(unsigned long unshare_flags)
2794 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2795 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2796 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2797 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2801 * Not implemented, but pretend it works if there is nothing
2802 * to unshare. Note that unsharing the address space or the
2803 * signal handlers also need to unshare the signal queues (aka
2806 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2807 if (!thread_group_empty(current))
2810 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2811 if (refcount_read(¤t->sighand->count) > 1)
2814 if (unshare_flags & CLONE_VM) {
2815 if (!current_is_single_threaded())
2823 * Unshare the filesystem structure if it is being shared
2825 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2827 struct fs_struct *fs = current->fs;
2829 if (!(unshare_flags & CLONE_FS) || !fs)
2832 /* don't need lock here; in the worst case we'll do useless copy */
2836 *new_fsp = copy_fs_struct(fs);
2844 * Unshare file descriptor table if it is being shared
2846 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2847 struct files_struct **new_fdp)
2849 struct files_struct *fd = current->files;
2852 if ((unshare_flags & CLONE_FILES) &&
2853 (fd && atomic_read(&fd->count) > 1)) {
2854 *new_fdp = dup_fd(fd, max_fds, &error);
2863 * unshare allows a process to 'unshare' part of the process
2864 * context which was originally shared using clone. copy_*
2865 * functions used by _do_fork() cannot be used here directly
2866 * because they modify an inactive task_struct that is being
2867 * constructed. Here we are modifying the current, active,
2870 int ksys_unshare(unsigned long unshare_flags)
2872 struct fs_struct *fs, *new_fs = NULL;
2873 struct files_struct *fd, *new_fd = NULL;
2874 struct cred *new_cred = NULL;
2875 struct nsproxy *new_nsproxy = NULL;
2880 * If unsharing a user namespace must also unshare the thread group
2881 * and unshare the filesystem root and working directories.
2883 if (unshare_flags & CLONE_NEWUSER)
2884 unshare_flags |= CLONE_THREAD | CLONE_FS;
2886 * If unsharing vm, must also unshare signal handlers.
2888 if (unshare_flags & CLONE_VM)
2889 unshare_flags |= CLONE_SIGHAND;
2891 * If unsharing a signal handlers, must also unshare the signal queues.
2893 if (unshare_flags & CLONE_SIGHAND)
2894 unshare_flags |= CLONE_THREAD;
2896 * If unsharing namespace, must also unshare filesystem information.
2898 if (unshare_flags & CLONE_NEWNS)
2899 unshare_flags |= CLONE_FS;
2901 err = check_unshare_flags(unshare_flags);
2903 goto bad_unshare_out;
2905 * CLONE_NEWIPC must also detach from the undolist: after switching
2906 * to a new ipc namespace, the semaphore arrays from the old
2907 * namespace are unreachable.
2909 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2911 err = unshare_fs(unshare_flags, &new_fs);
2913 goto bad_unshare_out;
2914 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2916 goto bad_unshare_cleanup_fs;
2917 err = unshare_userns(unshare_flags, &new_cred);
2919 goto bad_unshare_cleanup_fd;
2920 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2923 goto bad_unshare_cleanup_cred;
2925 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2928 * CLONE_SYSVSEM is equivalent to sys_exit().
2932 if (unshare_flags & CLONE_NEWIPC) {
2933 /* Orphan segments in old ns (see sem above). */
2935 shm_init_task(current);
2939 switch_task_namespaces(current, new_nsproxy);
2945 spin_lock(&fs->lock);
2946 current->fs = new_fs;
2951 spin_unlock(&fs->lock);
2955 fd = current->files;
2956 current->files = new_fd;
2960 task_unlock(current);
2963 /* Install the new user namespace */
2964 commit_creds(new_cred);
2969 perf_event_namespaces(current);
2971 bad_unshare_cleanup_cred:
2974 bad_unshare_cleanup_fd:
2976 put_files_struct(new_fd);
2978 bad_unshare_cleanup_fs:
2980 free_fs_struct(new_fs);
2986 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2988 return ksys_unshare(unshare_flags);
2992 * Helper to unshare the files of the current task.
2993 * We don't want to expose copy_files internals to
2994 * the exec layer of the kernel.
2997 int unshare_files(struct files_struct **displaced)
2999 struct task_struct *task = current;
3000 struct files_struct *copy = NULL;
3003 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3004 if (error || !copy) {
3008 *displaced = task->files;
3015 int sysctl_max_threads(struct ctl_table *table, int write,
3016 void __user *buffer, size_t *lenp, loff_t *ppos)
3020 int threads = max_threads;
3022 int max = MAX_THREADS;
3029 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3033 max_threads = threads;