1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
15 #include <linux/anon_inodes.h>
16 #include <linux/slab.h>
17 #include <linux/sched/autogroup.h>
18 #include <linux/sched/mm.h>
19 #include <linux/sched/coredump.h>
20 #include <linux/sched/user.h>
21 #include <linux/sched/numa_balancing.h>
22 #include <linux/sched/stat.h>
23 #include <linux/sched/task.h>
24 #include <linux/sched/task_stack.h>
25 #include <linux/sched/cputime.h>
26 #include <linux/seq_file.h>
27 #include <linux/rtmutex.h>
28 #include <linux/init.h>
29 #include <linux/unistd.h>
30 #include <linux/module.h>
31 #include <linux/vmalloc.h>
32 #include <linux/completion.h>
33 #include <linux/personality.h>
34 #include <linux/mempolicy.h>
35 #include <linux/sem.h>
36 #include <linux/file.h>
37 #include <linux/fdtable.h>
38 #include <linux/iocontext.h>
39 #include <linux/key.h>
40 #include <linux/binfmts.h>
41 #include <linux/mman.h>
42 #include <linux/mmu_notifier.h>
45 #include <linux/vmacache.h>
46 #include <linux/nsproxy.h>
47 #include <linux/capability.h>
48 #include <linux/cpu.h>
49 #include <linux/cgroup.h>
50 #include <linux/security.h>
51 #include <linux/hugetlb.h>
52 #include <linux/seccomp.h>
53 #include <linux/swap.h>
54 #include <linux/syscalls.h>
55 #include <linux/jiffies.h>
56 #include <linux/futex.h>
57 #include <linux/compat.h>
58 #include <linux/kthread.h>
59 #include <linux/task_io_accounting_ops.h>
60 #include <linux/rcupdate.h>
61 #include <linux/ptrace.h>
62 #include <linux/mount.h>
63 #include <linux/audit.h>
64 #include <linux/memcontrol.h>
65 #include <linux/ftrace.h>
66 #include <linux/proc_fs.h>
67 #include <linux/profile.h>
68 #include <linux/rmap.h>
69 #include <linux/ksm.h>
70 #include <linux/acct.h>
71 #include <linux/userfaultfd_k.h>
72 #include <linux/tsacct_kern.h>
73 #include <linux/cn_proc.h>
74 #include <linux/freezer.h>
75 #include <linux/delayacct.h>
76 #include <linux/taskstats_kern.h>
77 #include <linux/random.h>
78 #include <linux/tty.h>
79 #include <linux/blkdev.h>
80 #include <linux/fs_struct.h>
81 #include <linux/magic.h>
82 #include <linux/perf_event.h>
83 #include <linux/posix-timers.h>
84 #include <linux/user-return-notifier.h>
85 #include <linux/oom.h>
86 #include <linux/khugepaged.h>
87 #include <linux/signalfd.h>
88 #include <linux/uprobes.h>
89 #include <linux/aio.h>
90 #include <linux/compiler.h>
91 #include <linux/sysctl.h>
92 #include <linux/kcov.h>
93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
98 #include <asm/pgtable.h>
99 #include <asm/pgalloc.h>
100 #include <linux/uaccess.h>
101 #include <asm/mmu_context.h>
102 #include <asm/cacheflush.h>
103 #include <asm/tlbflush.h>
105 #include <trace/events/sched.h>
107 #define CREATE_TRACE_POINTS
108 #include <trace/events/task.h>
111 * Minimum number of threads to boot the kernel
113 #define MIN_THREADS 20
116 * Maximum number of threads
118 #define MAX_THREADS FUTEX_TID_MASK
121 * Protected counters by write_lock_irq(&tasklist_lock)
123 unsigned long total_forks; /* Handle normal Linux uptimes. */
124 int nr_threads; /* The idle threads do not count.. */
126 static int max_threads; /* tunable limit on nr_threads */
128 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
130 static const char * const resident_page_types[] = {
131 NAMED_ARRAY_INDEX(MM_FILEPAGES),
132 NAMED_ARRAY_INDEX(MM_ANONPAGES),
133 NAMED_ARRAY_INDEX(MM_SWAPENTS),
134 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
137 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
139 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
141 #ifdef CONFIG_PROVE_RCU
142 int lockdep_tasklist_lock_is_held(void)
144 return lockdep_is_held(&tasklist_lock);
146 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
147 #endif /* #ifdef CONFIG_PROVE_RCU */
149 int nr_processes(void)
154 for_each_possible_cpu(cpu)
155 total += per_cpu(process_counts, cpu);
160 void __weak arch_release_task_struct(struct task_struct *tsk)
164 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
165 static struct kmem_cache *task_struct_cachep;
167 static inline struct task_struct *alloc_task_struct_node(int node)
169 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
172 static inline void free_task_struct(struct task_struct *tsk)
174 kmem_cache_free(task_struct_cachep, tsk);
178 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
181 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
182 * kmemcache based allocator.
184 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
186 #ifdef CONFIG_VMAP_STACK
188 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
189 * flush. Try to minimize the number of calls by caching stacks.
191 #define NR_CACHED_STACKS 2
192 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
194 static int free_vm_stack_cache(unsigned int cpu)
196 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
199 for (i = 0; i < NR_CACHED_STACKS; i++) {
200 struct vm_struct *vm_stack = cached_vm_stacks[i];
205 vfree(vm_stack->addr);
206 cached_vm_stacks[i] = NULL;
213 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
215 #ifdef CONFIG_VMAP_STACK
219 for (i = 0; i < NR_CACHED_STACKS; i++) {
222 s = this_cpu_xchg(cached_stacks[i], NULL);
227 /* Clear the KASAN shadow of the stack. */
228 kasan_unpoison_shadow(s->addr, THREAD_SIZE);
230 /* Clear stale pointers from reused stack. */
231 memset(s->addr, 0, THREAD_SIZE);
233 tsk->stack_vm_area = s;
234 tsk->stack = s->addr;
239 * Allocated stacks are cached and later reused by new threads,
240 * so memcg accounting is performed manually on assigning/releasing
241 * stacks to tasks. Drop __GFP_ACCOUNT.
243 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
244 VMALLOC_START, VMALLOC_END,
245 THREADINFO_GFP & ~__GFP_ACCOUNT,
247 0, node, __builtin_return_address(0));
250 * We can't call find_vm_area() in interrupt context, and
251 * free_thread_stack() can be called in interrupt context,
252 * so cache the vm_struct.
255 tsk->stack_vm_area = find_vm_area(stack);
260 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
264 tsk->stack = page_address(page);
271 static inline void free_thread_stack(struct task_struct *tsk)
273 #ifdef CONFIG_VMAP_STACK
274 struct vm_struct *vm = task_stack_vm_area(tsk);
279 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
280 mod_memcg_page_state(vm->pages[i],
281 MEMCG_KERNEL_STACK_KB,
282 -(int)(PAGE_SIZE / 1024));
284 memcg_kmem_uncharge_page(vm->pages[i], 0);
287 for (i = 0; i < NR_CACHED_STACKS; i++) {
288 if (this_cpu_cmpxchg(cached_stacks[i],
289 NULL, tsk->stack_vm_area) != NULL)
295 vfree_atomic(tsk->stack);
300 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
303 static struct kmem_cache *thread_stack_cache;
305 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
308 unsigned long *stack;
309 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
314 static void free_thread_stack(struct task_struct *tsk)
316 kmem_cache_free(thread_stack_cache, tsk->stack);
319 void thread_stack_cache_init(void)
321 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
322 THREAD_SIZE, THREAD_SIZE, 0, 0,
324 BUG_ON(thread_stack_cache == NULL);
329 /* SLAB cache for signal_struct structures (tsk->signal) */
330 static struct kmem_cache *signal_cachep;
332 /* SLAB cache for sighand_struct structures (tsk->sighand) */
333 struct kmem_cache *sighand_cachep;
335 /* SLAB cache for files_struct structures (tsk->files) */
336 struct kmem_cache *files_cachep;
338 /* SLAB cache for fs_struct structures (tsk->fs) */
339 struct kmem_cache *fs_cachep;
341 /* SLAB cache for vm_area_struct structures */
342 static struct kmem_cache *vm_area_cachep;
344 /* SLAB cache for mm_struct structures (tsk->mm) */
345 static struct kmem_cache *mm_cachep;
347 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
349 struct vm_area_struct *vma;
351 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
357 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
359 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
363 INIT_LIST_HEAD(&new->anon_vma_chain);
368 void vm_area_free(struct vm_area_struct *vma)
370 kmem_cache_free(vm_area_cachep, vma);
373 static void account_kernel_stack(struct task_struct *tsk, int account)
375 void *stack = task_stack_page(tsk);
376 struct vm_struct *vm = task_stack_vm_area(tsk);
378 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
383 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
385 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
386 mod_zone_page_state(page_zone(vm->pages[i]),
388 PAGE_SIZE / 1024 * account);
392 * All stack pages are in the same zone and belong to the
395 struct page *first_page = virt_to_page(stack);
397 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
398 THREAD_SIZE / 1024 * account);
400 mod_memcg_obj_state(stack, MEMCG_KERNEL_STACK_KB,
401 account * (THREAD_SIZE / 1024));
405 static int memcg_charge_kernel_stack(struct task_struct *tsk)
407 #ifdef CONFIG_VMAP_STACK
408 struct vm_struct *vm = task_stack_vm_area(tsk);
414 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
416 * If memcg_kmem_charge_page() fails, page->mem_cgroup
417 * pointer is NULL, and both memcg_kmem_uncharge_page()
418 * and mod_memcg_page_state() in free_thread_stack()
419 * will ignore this page. So it's safe.
421 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
426 mod_memcg_page_state(vm->pages[i],
427 MEMCG_KERNEL_STACK_KB,
435 static void release_task_stack(struct task_struct *tsk)
437 if (WARN_ON(tsk->state != TASK_DEAD))
438 return; /* Better to leak the stack than to free prematurely */
440 account_kernel_stack(tsk, -1);
441 free_thread_stack(tsk);
443 #ifdef CONFIG_VMAP_STACK
444 tsk->stack_vm_area = NULL;
448 #ifdef CONFIG_THREAD_INFO_IN_TASK
449 void put_task_stack(struct task_struct *tsk)
451 if (refcount_dec_and_test(&tsk->stack_refcount))
452 release_task_stack(tsk);
456 void free_task(struct task_struct *tsk)
458 #ifndef CONFIG_THREAD_INFO_IN_TASK
460 * The task is finally done with both the stack and thread_info,
463 release_task_stack(tsk);
466 * If the task had a separate stack allocation, it should be gone
469 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
471 rt_mutex_debug_task_free(tsk);
472 ftrace_graph_exit_task(tsk);
473 put_seccomp_filter(tsk);
474 arch_release_task_struct(tsk);
475 if (tsk->flags & PF_KTHREAD)
476 free_kthread_struct(tsk);
477 free_task_struct(tsk);
479 EXPORT_SYMBOL(free_task);
482 static __latent_entropy int dup_mmap(struct mm_struct *mm,
483 struct mm_struct *oldmm)
485 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
486 struct rb_node **rb_link, *rb_parent;
488 unsigned long charge;
491 uprobe_start_dup_mmap();
492 if (down_write_killable(&oldmm->mmap_sem)) {
494 goto fail_uprobe_end;
496 flush_cache_dup_mm(oldmm);
497 uprobe_dup_mmap(oldmm, mm);
499 * Not linked in yet - no deadlock potential:
501 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
503 /* No ordering required: file already has been exposed. */
504 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
506 mm->total_vm = oldmm->total_vm;
507 mm->data_vm = oldmm->data_vm;
508 mm->exec_vm = oldmm->exec_vm;
509 mm->stack_vm = oldmm->stack_vm;
511 rb_link = &mm->mm_rb.rb_node;
514 retval = ksm_fork(mm, oldmm);
517 retval = khugepaged_fork(mm, oldmm);
522 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
525 if (mpnt->vm_flags & VM_DONTCOPY) {
526 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
531 * Don't duplicate many vmas if we've been oom-killed (for
534 if (fatal_signal_pending(current)) {
538 if (mpnt->vm_flags & VM_ACCOUNT) {
539 unsigned long len = vma_pages(mpnt);
541 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
545 tmp = vm_area_dup(mpnt);
548 retval = vma_dup_policy(mpnt, tmp);
550 goto fail_nomem_policy;
552 retval = dup_userfaultfd(tmp, &uf);
554 goto fail_nomem_anon_vma_fork;
555 if (tmp->vm_flags & VM_WIPEONFORK) {
556 /* VM_WIPEONFORK gets a clean slate in the child. */
557 tmp->anon_vma = NULL;
558 if (anon_vma_prepare(tmp))
559 goto fail_nomem_anon_vma_fork;
560 } else if (anon_vma_fork(tmp, mpnt))
561 goto fail_nomem_anon_vma_fork;
562 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
563 tmp->vm_next = tmp->vm_prev = NULL;
566 struct inode *inode = file_inode(file);
567 struct address_space *mapping = file->f_mapping;
570 if (tmp->vm_flags & VM_DENYWRITE)
571 atomic_dec(&inode->i_writecount);
572 i_mmap_lock_write(mapping);
573 if (tmp->vm_flags & VM_SHARED)
574 atomic_inc(&mapping->i_mmap_writable);
575 flush_dcache_mmap_lock(mapping);
576 /* insert tmp into the share list, just after mpnt */
577 vma_interval_tree_insert_after(tmp, mpnt,
579 flush_dcache_mmap_unlock(mapping);
580 i_mmap_unlock_write(mapping);
584 * Clear hugetlb-related page reserves for children. This only
585 * affects MAP_PRIVATE mappings. Faults generated by the child
586 * are not guaranteed to succeed, even if read-only
588 if (is_vm_hugetlb_page(tmp))
589 reset_vma_resv_huge_pages(tmp);
592 * Link in the new vma and copy the page table entries.
595 pprev = &tmp->vm_next;
599 __vma_link_rb(mm, tmp, rb_link, rb_parent);
600 rb_link = &tmp->vm_rb.rb_right;
601 rb_parent = &tmp->vm_rb;
604 if (!(tmp->vm_flags & VM_WIPEONFORK))
605 retval = copy_page_range(mm, oldmm, mpnt);
607 if (tmp->vm_ops && tmp->vm_ops->open)
608 tmp->vm_ops->open(tmp);
613 /* a new mm has just been created */
614 retval = arch_dup_mmap(oldmm, mm);
616 up_write(&mm->mmap_sem);
618 up_write(&oldmm->mmap_sem);
619 dup_userfaultfd_complete(&uf);
621 uprobe_end_dup_mmap();
623 fail_nomem_anon_vma_fork:
624 mpol_put(vma_policy(tmp));
629 vm_unacct_memory(charge);
633 static inline int mm_alloc_pgd(struct mm_struct *mm)
635 mm->pgd = pgd_alloc(mm);
636 if (unlikely(!mm->pgd))
641 static inline void mm_free_pgd(struct mm_struct *mm)
643 pgd_free(mm, mm->pgd);
646 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
648 down_write(&oldmm->mmap_sem);
649 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
650 up_write(&oldmm->mmap_sem);
653 #define mm_alloc_pgd(mm) (0)
654 #define mm_free_pgd(mm)
655 #endif /* CONFIG_MMU */
657 static void check_mm(struct mm_struct *mm)
661 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
662 "Please make sure 'struct resident_page_types[]' is updated as well");
664 for (i = 0; i < NR_MM_COUNTERS; i++) {
665 long x = atomic_long_read(&mm->rss_stat.count[i]);
668 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
669 mm, resident_page_types[i], x);
672 if (mm_pgtables_bytes(mm))
673 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
674 mm_pgtables_bytes(mm));
676 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
677 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
681 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
682 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
685 * Called when the last reference to the mm
686 * is dropped: either by a lazy thread or by
687 * mmput. Free the page directory and the mm.
689 void __mmdrop(struct mm_struct *mm)
691 BUG_ON(mm == &init_mm);
692 WARN_ON_ONCE(mm == current->mm);
693 WARN_ON_ONCE(mm == current->active_mm);
696 mmu_notifier_subscriptions_destroy(mm);
698 put_user_ns(mm->user_ns);
701 EXPORT_SYMBOL_GPL(__mmdrop);
703 static void mmdrop_async_fn(struct work_struct *work)
705 struct mm_struct *mm;
707 mm = container_of(work, struct mm_struct, async_put_work);
711 static void mmdrop_async(struct mm_struct *mm)
713 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
714 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
715 schedule_work(&mm->async_put_work);
719 static inline void free_signal_struct(struct signal_struct *sig)
721 taskstats_tgid_free(sig);
722 sched_autogroup_exit(sig);
724 * __mmdrop is not safe to call from softirq context on x86 due to
725 * pgd_dtor so postpone it to the async context
728 mmdrop_async(sig->oom_mm);
729 kmem_cache_free(signal_cachep, sig);
732 static inline void put_signal_struct(struct signal_struct *sig)
734 if (refcount_dec_and_test(&sig->sigcnt))
735 free_signal_struct(sig);
738 void __put_task_struct(struct task_struct *tsk)
740 WARN_ON(!tsk->exit_state);
741 WARN_ON(refcount_read(&tsk->usage));
742 WARN_ON(tsk == current);
745 task_numa_free(tsk, true);
746 security_task_free(tsk);
748 delayacct_tsk_free(tsk);
749 put_signal_struct(tsk->signal);
751 if (!profile_handoff_task(tsk))
754 EXPORT_SYMBOL_GPL(__put_task_struct);
756 void __init __weak arch_task_cache_init(void) { }
761 static void set_max_threads(unsigned int max_threads_suggested)
764 unsigned long nr_pages = totalram_pages();
767 * The number of threads shall be limited such that the thread
768 * structures may only consume a small part of the available memory.
770 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
771 threads = MAX_THREADS;
773 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
774 (u64) THREAD_SIZE * 8UL);
776 if (threads > max_threads_suggested)
777 threads = max_threads_suggested;
779 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
782 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
783 /* Initialized by the architecture: */
784 int arch_task_struct_size __read_mostly;
787 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
788 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
790 /* Fetch thread_struct whitelist for the architecture. */
791 arch_thread_struct_whitelist(offset, size);
794 * Handle zero-sized whitelist or empty thread_struct, otherwise
795 * adjust offset to position of thread_struct in task_struct.
797 if (unlikely(*size == 0))
800 *offset += offsetof(struct task_struct, thread);
802 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
804 void __init fork_init(void)
807 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
808 #ifndef ARCH_MIN_TASKALIGN
809 #define ARCH_MIN_TASKALIGN 0
811 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
812 unsigned long useroffset, usersize;
814 /* create a slab on which task_structs can be allocated */
815 task_struct_whitelist(&useroffset, &usersize);
816 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
817 arch_task_struct_size, align,
818 SLAB_PANIC|SLAB_ACCOUNT,
819 useroffset, usersize, NULL);
822 /* do the arch specific task caches init */
823 arch_task_cache_init();
825 set_max_threads(MAX_THREADS);
827 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
828 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
829 init_task.signal->rlim[RLIMIT_SIGPENDING] =
830 init_task.signal->rlim[RLIMIT_NPROC];
832 for (i = 0; i < UCOUNT_COUNTS; i++) {
833 init_user_ns.ucount_max[i] = max_threads/2;
836 #ifdef CONFIG_VMAP_STACK
837 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
838 NULL, free_vm_stack_cache);
841 lockdep_init_task(&init_task);
845 int __weak arch_dup_task_struct(struct task_struct *dst,
846 struct task_struct *src)
852 void set_task_stack_end_magic(struct task_struct *tsk)
854 unsigned long *stackend;
856 stackend = end_of_stack(tsk);
857 *stackend = STACK_END_MAGIC; /* for overflow detection */
860 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
862 struct task_struct *tsk;
863 unsigned long *stack;
864 struct vm_struct *stack_vm_area __maybe_unused;
867 if (node == NUMA_NO_NODE)
868 node = tsk_fork_get_node(orig);
869 tsk = alloc_task_struct_node(node);
873 stack = alloc_thread_stack_node(tsk, node);
877 if (memcg_charge_kernel_stack(tsk))
880 stack_vm_area = task_stack_vm_area(tsk);
882 err = arch_dup_task_struct(tsk, orig);
885 * arch_dup_task_struct() clobbers the stack-related fields. Make
886 * sure they're properly initialized before using any stack-related
890 #ifdef CONFIG_VMAP_STACK
891 tsk->stack_vm_area = stack_vm_area;
893 #ifdef CONFIG_THREAD_INFO_IN_TASK
894 refcount_set(&tsk->stack_refcount, 1);
900 #ifdef CONFIG_SECCOMP
902 * We must handle setting up seccomp filters once we're under
903 * the sighand lock in case orig has changed between now and
904 * then. Until then, filter must be NULL to avoid messing up
905 * the usage counts on the error path calling free_task.
907 tsk->seccomp.filter = NULL;
910 setup_thread_stack(tsk, orig);
911 clear_user_return_notifier(tsk);
912 clear_tsk_need_resched(tsk);
913 set_task_stack_end_magic(tsk);
915 #ifdef CONFIG_STACKPROTECTOR
916 tsk->stack_canary = get_random_canary();
918 if (orig->cpus_ptr == &orig->cpus_mask)
919 tsk->cpus_ptr = &tsk->cpus_mask;
922 * One for the user space visible state that goes away when reaped.
923 * One for the scheduler.
925 refcount_set(&tsk->rcu_users, 2);
926 /* One for the rcu users */
927 refcount_set(&tsk->usage, 1);
928 #ifdef CONFIG_BLK_DEV_IO_TRACE
931 tsk->splice_pipe = NULL;
932 tsk->task_frag.page = NULL;
933 tsk->wake_q.next = NULL;
935 account_kernel_stack(tsk, 1);
939 #ifdef CONFIG_FAULT_INJECTION
943 #ifdef CONFIG_BLK_CGROUP
944 tsk->throttle_queue = NULL;
945 tsk->use_memdelay = 0;
949 tsk->active_memcg = NULL;
954 free_thread_stack(tsk);
956 free_task_struct(tsk);
960 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
962 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
964 static int __init coredump_filter_setup(char *s)
966 default_dump_filter =
967 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
968 MMF_DUMP_FILTER_MASK;
972 __setup("coredump_filter=", coredump_filter_setup);
974 #include <linux/init_task.h>
976 static void mm_init_aio(struct mm_struct *mm)
979 spin_lock_init(&mm->ioctx_lock);
980 mm->ioctx_table = NULL;
984 static __always_inline void mm_clear_owner(struct mm_struct *mm,
985 struct task_struct *p)
989 WRITE_ONCE(mm->owner, NULL);
993 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1000 static void mm_init_uprobes_state(struct mm_struct *mm)
1002 #ifdef CONFIG_UPROBES
1003 mm->uprobes_state.xol_area = NULL;
1007 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1008 struct user_namespace *user_ns)
1011 mm->mm_rb = RB_ROOT;
1012 mm->vmacache_seqnum = 0;
1013 atomic_set(&mm->mm_users, 1);
1014 atomic_set(&mm->mm_count, 1);
1015 init_rwsem(&mm->mmap_sem);
1016 INIT_LIST_HEAD(&mm->mmlist);
1017 mm->core_state = NULL;
1018 mm_pgtables_bytes_init(mm);
1021 atomic64_set(&mm->pinned_vm, 0);
1022 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1023 spin_lock_init(&mm->page_table_lock);
1024 spin_lock_init(&mm->arg_lock);
1025 mm_init_cpumask(mm);
1027 mm_init_owner(mm, p);
1028 RCU_INIT_POINTER(mm->exe_file, NULL);
1029 mmu_notifier_subscriptions_init(mm);
1030 init_tlb_flush_pending(mm);
1031 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1032 mm->pmd_huge_pte = NULL;
1034 mm_init_uprobes_state(mm);
1037 mm->flags = current->mm->flags & MMF_INIT_MASK;
1038 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1040 mm->flags = default_dump_filter;
1044 if (mm_alloc_pgd(mm))
1047 if (init_new_context(p, mm))
1048 goto fail_nocontext;
1050 mm->user_ns = get_user_ns(user_ns);
1061 * Allocate and initialize an mm_struct.
1063 struct mm_struct *mm_alloc(void)
1065 struct mm_struct *mm;
1071 memset(mm, 0, sizeof(*mm));
1072 return mm_init(mm, current, current_user_ns());
1075 static inline void __mmput(struct mm_struct *mm)
1077 VM_BUG_ON(atomic_read(&mm->mm_users));
1079 uprobe_clear_state(mm);
1082 khugepaged_exit(mm); /* must run before exit_mmap */
1084 mm_put_huge_zero_page(mm);
1085 set_mm_exe_file(mm, NULL);
1086 if (!list_empty(&mm->mmlist)) {
1087 spin_lock(&mmlist_lock);
1088 list_del(&mm->mmlist);
1089 spin_unlock(&mmlist_lock);
1092 module_put(mm->binfmt->module);
1097 * Decrement the use count and release all resources for an mm.
1099 void mmput(struct mm_struct *mm)
1103 if (atomic_dec_and_test(&mm->mm_users))
1106 EXPORT_SYMBOL_GPL(mmput);
1109 static void mmput_async_fn(struct work_struct *work)
1111 struct mm_struct *mm = container_of(work, struct mm_struct,
1117 void mmput_async(struct mm_struct *mm)
1119 if (atomic_dec_and_test(&mm->mm_users)) {
1120 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1121 schedule_work(&mm->async_put_work);
1127 * set_mm_exe_file - change a reference to the mm's executable file
1129 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1131 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1132 * invocations: in mmput() nobody alive left, in execve task is single
1133 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1134 * mm->exe_file, but does so without using set_mm_exe_file() in order
1135 * to do avoid the need for any locks.
1137 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1139 struct file *old_exe_file;
1142 * It is safe to dereference the exe_file without RCU as
1143 * this function is only called if nobody else can access
1144 * this mm -- see comment above for justification.
1146 old_exe_file = rcu_dereference_raw(mm->exe_file);
1149 get_file(new_exe_file);
1150 rcu_assign_pointer(mm->exe_file, new_exe_file);
1156 * get_mm_exe_file - acquire a reference to the mm's executable file
1158 * Returns %NULL if mm has no associated executable file.
1159 * User must release file via fput().
1161 struct file *get_mm_exe_file(struct mm_struct *mm)
1163 struct file *exe_file;
1166 exe_file = rcu_dereference(mm->exe_file);
1167 if (exe_file && !get_file_rcu(exe_file))
1172 EXPORT_SYMBOL(get_mm_exe_file);
1175 * get_task_exe_file - acquire a reference to the task's executable file
1177 * Returns %NULL if task's mm (if any) has no associated executable file or
1178 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1179 * User must release file via fput().
1181 struct file *get_task_exe_file(struct task_struct *task)
1183 struct file *exe_file = NULL;
1184 struct mm_struct *mm;
1189 if (!(task->flags & PF_KTHREAD))
1190 exe_file = get_mm_exe_file(mm);
1195 EXPORT_SYMBOL(get_task_exe_file);
1198 * get_task_mm - acquire a reference to the task's mm
1200 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1201 * this kernel workthread has transiently adopted a user mm with use_mm,
1202 * to do its AIO) is not set and if so returns a reference to it, after
1203 * bumping up the use count. User must release the mm via mmput()
1204 * after use. Typically used by /proc and ptrace.
1206 struct mm_struct *get_task_mm(struct task_struct *task)
1208 struct mm_struct *mm;
1213 if (task->flags & PF_KTHREAD)
1221 EXPORT_SYMBOL_GPL(get_task_mm);
1223 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1225 struct mm_struct *mm;
1228 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1230 return ERR_PTR(err);
1232 mm = get_task_mm(task);
1233 if (mm && mm != current->mm &&
1234 !ptrace_may_access(task, mode)) {
1236 mm = ERR_PTR(-EACCES);
1238 mutex_unlock(&task->signal->cred_guard_mutex);
1243 static void complete_vfork_done(struct task_struct *tsk)
1245 struct completion *vfork;
1248 vfork = tsk->vfork_done;
1249 if (likely(vfork)) {
1250 tsk->vfork_done = NULL;
1256 static int wait_for_vfork_done(struct task_struct *child,
1257 struct completion *vfork)
1261 freezer_do_not_count();
1262 cgroup_enter_frozen();
1263 killed = wait_for_completion_killable(vfork);
1264 cgroup_leave_frozen(false);
1269 child->vfork_done = NULL;
1273 put_task_struct(child);
1277 /* Please note the differences between mmput and mm_release.
1278 * mmput is called whenever we stop holding onto a mm_struct,
1279 * error success whatever.
1281 * mm_release is called after a mm_struct has been removed
1282 * from the current process.
1284 * This difference is important for error handling, when we
1285 * only half set up a mm_struct for a new process and need to restore
1286 * the old one. Because we mmput the new mm_struct before
1287 * restoring the old one. . .
1288 * Eric Biederman 10 January 1998
1290 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1292 uprobe_free_utask(tsk);
1294 /* Get rid of any cached register state */
1295 deactivate_mm(tsk, mm);
1298 * Signal userspace if we're not exiting with a core dump
1299 * because we want to leave the value intact for debugging
1302 if (tsk->clear_child_tid) {
1303 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1304 atomic_read(&mm->mm_users) > 1) {
1306 * We don't check the error code - if userspace has
1307 * not set up a proper pointer then tough luck.
1309 put_user(0, tsk->clear_child_tid);
1310 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1311 1, NULL, NULL, 0, 0);
1313 tsk->clear_child_tid = NULL;
1317 * All done, finally we can wake up parent and return this mm to him.
1318 * Also kthread_stop() uses this completion for synchronization.
1320 if (tsk->vfork_done)
1321 complete_vfork_done(tsk);
1324 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1326 futex_exit_release(tsk);
1327 mm_release(tsk, mm);
1330 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1332 futex_exec_release(tsk);
1333 mm_release(tsk, mm);
1337 * dup_mm() - duplicates an existing mm structure
1338 * @tsk: the task_struct with which the new mm will be associated.
1339 * @oldmm: the mm to duplicate.
1341 * Allocates a new mm structure and duplicates the provided @oldmm structure
1344 * Return: the duplicated mm or NULL on failure.
1346 static struct mm_struct *dup_mm(struct task_struct *tsk,
1347 struct mm_struct *oldmm)
1349 struct mm_struct *mm;
1356 memcpy(mm, oldmm, sizeof(*mm));
1358 if (!mm_init(mm, tsk, mm->user_ns))
1361 err = dup_mmap(mm, oldmm);
1365 mm->hiwater_rss = get_mm_rss(mm);
1366 mm->hiwater_vm = mm->total_vm;
1368 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1374 /* don't put binfmt in mmput, we haven't got module yet */
1376 mm_init_owner(mm, NULL);
1383 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1385 struct mm_struct *mm, *oldmm;
1388 tsk->min_flt = tsk->maj_flt = 0;
1389 tsk->nvcsw = tsk->nivcsw = 0;
1390 #ifdef CONFIG_DETECT_HUNG_TASK
1391 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1392 tsk->last_switch_time = 0;
1396 tsk->active_mm = NULL;
1399 * Are we cloning a kernel thread?
1401 * We need to steal a active VM for that..
1403 oldmm = current->mm;
1407 /* initialize the new vmacache entries */
1408 vmacache_flush(tsk);
1410 if (clone_flags & CLONE_VM) {
1417 mm = dup_mm(tsk, current->mm);
1423 tsk->active_mm = mm;
1430 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1432 struct fs_struct *fs = current->fs;
1433 if (clone_flags & CLONE_FS) {
1434 /* tsk->fs is already what we want */
1435 spin_lock(&fs->lock);
1437 spin_unlock(&fs->lock);
1441 spin_unlock(&fs->lock);
1444 tsk->fs = copy_fs_struct(fs);
1450 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1452 struct files_struct *oldf, *newf;
1456 * A background process may not have any files ...
1458 oldf = current->files;
1462 if (clone_flags & CLONE_FILES) {
1463 atomic_inc(&oldf->count);
1467 newf = dup_fd(oldf, &error);
1477 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1480 struct io_context *ioc = current->io_context;
1481 struct io_context *new_ioc;
1486 * Share io context with parent, if CLONE_IO is set
1488 if (clone_flags & CLONE_IO) {
1490 tsk->io_context = ioc;
1491 } else if (ioprio_valid(ioc->ioprio)) {
1492 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1493 if (unlikely(!new_ioc))
1496 new_ioc->ioprio = ioc->ioprio;
1497 put_io_context(new_ioc);
1503 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1505 struct sighand_struct *sig;
1507 if (clone_flags & CLONE_SIGHAND) {
1508 refcount_inc(¤t->sighand->count);
1511 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1512 RCU_INIT_POINTER(tsk->sighand, sig);
1516 refcount_set(&sig->count, 1);
1517 spin_lock_irq(¤t->sighand->siglock);
1518 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1519 spin_unlock_irq(¤t->sighand->siglock);
1521 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1522 if (clone_flags & CLONE_CLEAR_SIGHAND)
1523 flush_signal_handlers(tsk, 0);
1528 void __cleanup_sighand(struct sighand_struct *sighand)
1530 if (refcount_dec_and_test(&sighand->count)) {
1531 signalfd_cleanup(sighand);
1533 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1534 * without an RCU grace period, see __lock_task_sighand().
1536 kmem_cache_free(sighand_cachep, sighand);
1541 * Initialize POSIX timer handling for a thread group.
1543 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1545 struct posix_cputimers *pct = &sig->posix_cputimers;
1546 unsigned long cpu_limit;
1548 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1549 posix_cputimers_group_init(pct, cpu_limit);
1552 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1554 struct signal_struct *sig;
1556 if (clone_flags & CLONE_THREAD)
1559 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1564 sig->nr_threads = 1;
1565 atomic_set(&sig->live, 1);
1566 refcount_set(&sig->sigcnt, 1);
1568 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1569 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1570 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1572 init_waitqueue_head(&sig->wait_chldexit);
1573 sig->curr_target = tsk;
1574 init_sigpending(&sig->shared_pending);
1575 INIT_HLIST_HEAD(&sig->multiprocess);
1576 seqlock_init(&sig->stats_lock);
1577 prev_cputime_init(&sig->prev_cputime);
1579 #ifdef CONFIG_POSIX_TIMERS
1580 INIT_LIST_HEAD(&sig->posix_timers);
1581 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1582 sig->real_timer.function = it_real_fn;
1585 task_lock(current->group_leader);
1586 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1587 task_unlock(current->group_leader);
1589 posix_cpu_timers_init_group(sig);
1591 tty_audit_fork(sig);
1592 sched_autogroup_fork(sig);
1594 sig->oom_score_adj = current->signal->oom_score_adj;
1595 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1597 mutex_init(&sig->cred_guard_mutex);
1602 static void copy_seccomp(struct task_struct *p)
1604 #ifdef CONFIG_SECCOMP
1606 * Must be called with sighand->lock held, which is common to
1607 * all threads in the group. Holding cred_guard_mutex is not
1608 * needed because this new task is not yet running and cannot
1611 assert_spin_locked(¤t->sighand->siglock);
1613 /* Ref-count the new filter user, and assign it. */
1614 get_seccomp_filter(current);
1615 p->seccomp = current->seccomp;
1618 * Explicitly enable no_new_privs here in case it got set
1619 * between the task_struct being duplicated and holding the
1620 * sighand lock. The seccomp state and nnp must be in sync.
1622 if (task_no_new_privs(current))
1623 task_set_no_new_privs(p);
1626 * If the parent gained a seccomp mode after copying thread
1627 * flags and between before we held the sighand lock, we have
1628 * to manually enable the seccomp thread flag here.
1630 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1631 set_tsk_thread_flag(p, TIF_SECCOMP);
1635 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1637 current->clear_child_tid = tidptr;
1639 return task_pid_vnr(current);
1642 static void rt_mutex_init_task(struct task_struct *p)
1644 raw_spin_lock_init(&p->pi_lock);
1645 #ifdef CONFIG_RT_MUTEXES
1646 p->pi_waiters = RB_ROOT_CACHED;
1647 p->pi_top_task = NULL;
1648 p->pi_blocked_on = NULL;
1652 static inline void init_task_pid_links(struct task_struct *task)
1656 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1657 INIT_HLIST_NODE(&task->pid_links[type]);
1662 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1664 if (type == PIDTYPE_PID)
1665 task->thread_pid = pid;
1667 task->signal->pids[type] = pid;
1670 static inline void rcu_copy_process(struct task_struct *p)
1672 #ifdef CONFIG_PREEMPT_RCU
1673 p->rcu_read_lock_nesting = 0;
1674 p->rcu_read_unlock_special.s = 0;
1675 p->rcu_blocked_node = NULL;
1676 INIT_LIST_HEAD(&p->rcu_node_entry);
1677 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1678 #ifdef CONFIG_TASKS_RCU
1679 p->rcu_tasks_holdout = false;
1680 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1681 p->rcu_tasks_idle_cpu = -1;
1682 #endif /* #ifdef CONFIG_TASKS_RCU */
1685 struct pid *pidfd_pid(const struct file *file)
1687 if (file->f_op == &pidfd_fops)
1688 return file->private_data;
1690 return ERR_PTR(-EBADF);
1693 static int pidfd_release(struct inode *inode, struct file *file)
1695 struct pid *pid = file->private_data;
1697 file->private_data = NULL;
1702 #ifdef CONFIG_PROC_FS
1704 * pidfd_show_fdinfo - print information about a pidfd
1705 * @m: proc fdinfo file
1706 * @f: file referencing a pidfd
1709 * This function will print the pid that a given pidfd refers to in the
1710 * pid namespace of the procfs instance.
1711 * If the pid namespace of the process is not a descendant of the pid
1712 * namespace of the procfs instance 0 will be shown as its pid. This is
1713 * similar to calling getppid() on a process whose parent is outside of
1714 * its pid namespace.
1717 * If pid namespaces are supported then this function will also print
1718 * the pid of a given pidfd refers to for all descendant pid namespaces
1719 * starting from the current pid namespace of the instance, i.e. the
1720 * Pid field and the first entry in the NSpid field will be identical.
1721 * If the pid namespace of the process is not a descendant of the pid
1722 * namespace of the procfs instance 0 will be shown as its first NSpid
1723 * entry and no others will be shown.
1724 * Note that this differs from the Pid and NSpid fields in
1725 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1726 * the pid namespace of the procfs instance. The difference becomes
1727 * obvious when sending around a pidfd between pid namespaces from a
1728 * different branch of the tree, i.e. where no ancestoral relation is
1729 * present between the pid namespaces:
1730 * - create two new pid namespaces ns1 and ns2 in the initial pid
1731 * namespace (also take care to create new mount namespaces in the
1732 * new pid namespace and mount procfs)
1733 * - create a process with a pidfd in ns1
1734 * - send pidfd from ns1 to ns2
1735 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1736 * have exactly one entry, which is 0
1738 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1740 struct pid *pid = f->private_data;
1741 struct pid_namespace *ns;
1744 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1745 ns = proc_pid_ns(file_inode(m->file));
1746 nr = pid_nr_ns(pid, ns);
1749 seq_put_decimal_ll(m, "Pid:\t", nr);
1751 #ifdef CONFIG_PID_NS
1752 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1756 /* If nr is non-zero it means that 'pid' is valid and that
1757 * ns, i.e. the pid namespace associated with the procfs
1758 * instance, is in the pid namespace hierarchy of pid.
1759 * Start at one below the already printed level.
1761 for (i = ns->level + 1; i <= pid->level; i++)
1762 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1770 * Poll support for process exit notification.
1772 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1774 struct task_struct *task;
1775 struct pid *pid = file->private_data;
1776 __poll_t poll_flags = 0;
1778 poll_wait(file, &pid->wait_pidfd, pts);
1781 task = pid_task(pid, PIDTYPE_PID);
1783 * Inform pollers only when the whole thread group exits.
1784 * If the thread group leader exits before all other threads in the
1785 * group, then poll(2) should block, similar to the wait(2) family.
1787 if (!task || (task->exit_state && thread_group_empty(task)))
1788 poll_flags = EPOLLIN | EPOLLRDNORM;
1794 const struct file_operations pidfd_fops = {
1795 .release = pidfd_release,
1797 #ifdef CONFIG_PROC_FS
1798 .show_fdinfo = pidfd_show_fdinfo,
1802 static void __delayed_free_task(struct rcu_head *rhp)
1804 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1809 static __always_inline void delayed_free_task(struct task_struct *tsk)
1811 if (IS_ENABLED(CONFIG_MEMCG))
1812 call_rcu(&tsk->rcu, __delayed_free_task);
1818 * This creates a new process as a copy of the old one,
1819 * but does not actually start it yet.
1821 * It copies the registers, and all the appropriate
1822 * parts of the process environment (as per the clone
1823 * flags). The actual kick-off is left to the caller.
1825 static __latent_entropy struct task_struct *copy_process(
1829 struct kernel_clone_args *args)
1831 int pidfd = -1, retval;
1832 struct task_struct *p;
1833 struct multiprocess_signals delayed;
1834 struct file *pidfile = NULL;
1835 u64 clone_flags = args->flags;
1836 struct nsproxy *nsp = current->nsproxy;
1839 * Don't allow sharing the root directory with processes in a different
1842 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1843 return ERR_PTR(-EINVAL);
1845 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1846 return ERR_PTR(-EINVAL);
1849 * Thread groups must share signals as well, and detached threads
1850 * can only be started up within the thread group.
1852 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1853 return ERR_PTR(-EINVAL);
1856 * Shared signal handlers imply shared VM. By way of the above,
1857 * thread groups also imply shared VM. Blocking this case allows
1858 * for various simplifications in other code.
1860 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1861 return ERR_PTR(-EINVAL);
1864 * Siblings of global init remain as zombies on exit since they are
1865 * not reaped by their parent (swapper). To solve this and to avoid
1866 * multi-rooted process trees, prevent global and container-inits
1867 * from creating siblings.
1869 if ((clone_flags & CLONE_PARENT) &&
1870 current->signal->flags & SIGNAL_UNKILLABLE)
1871 return ERR_PTR(-EINVAL);
1874 * If the new process will be in a different pid or user namespace
1875 * do not allow it to share a thread group with the forking task.
1877 if (clone_flags & CLONE_THREAD) {
1878 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1879 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1880 return ERR_PTR(-EINVAL);
1884 * If the new process will be in a different time namespace
1885 * do not allow it to share VM or a thread group with the forking task.
1887 if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1888 if (nsp->time_ns != nsp->time_ns_for_children)
1889 return ERR_PTR(-EINVAL);
1892 if (clone_flags & CLONE_PIDFD) {
1894 * - CLONE_DETACHED is blocked so that we can potentially
1895 * reuse it later for CLONE_PIDFD.
1896 * - CLONE_THREAD is blocked until someone really needs it.
1898 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1899 return ERR_PTR(-EINVAL);
1903 * Force any signals received before this point to be delivered
1904 * before the fork happens. Collect up signals sent to multiple
1905 * processes that happen during the fork and delay them so that
1906 * they appear to happen after the fork.
1908 sigemptyset(&delayed.signal);
1909 INIT_HLIST_NODE(&delayed.node);
1911 spin_lock_irq(¤t->sighand->siglock);
1912 if (!(clone_flags & CLONE_THREAD))
1913 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1914 recalc_sigpending();
1915 spin_unlock_irq(¤t->sighand->siglock);
1916 retval = -ERESTARTNOINTR;
1917 if (signal_pending(current))
1921 p = dup_task_struct(current, node);
1926 * This _must_ happen before we call free_task(), i.e. before we jump
1927 * to any of the bad_fork_* labels. This is to avoid freeing
1928 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1929 * kernel threads (PF_KTHREAD).
1931 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1933 * Clear TID on mm_release()?
1935 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1937 ftrace_graph_init_task(p);
1939 rt_mutex_init_task(p);
1941 #ifdef CONFIG_PROVE_LOCKING
1942 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1943 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1946 if (atomic_read(&p->real_cred->user->processes) >=
1947 task_rlimit(p, RLIMIT_NPROC)) {
1948 if (p->real_cred->user != INIT_USER &&
1949 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1952 current->flags &= ~PF_NPROC_EXCEEDED;
1954 retval = copy_creds(p, clone_flags);
1959 * If multiple threads are within copy_process(), then this check
1960 * triggers too late. This doesn't hurt, the check is only there
1961 * to stop root fork bombs.
1964 if (nr_threads >= max_threads)
1965 goto bad_fork_cleanup_count;
1967 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1968 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1969 p->flags |= PF_FORKNOEXEC;
1970 INIT_LIST_HEAD(&p->children);
1971 INIT_LIST_HEAD(&p->sibling);
1972 rcu_copy_process(p);
1973 p->vfork_done = NULL;
1974 spin_lock_init(&p->alloc_lock);
1976 init_sigpending(&p->pending);
1978 p->utime = p->stime = p->gtime = 0;
1979 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1980 p->utimescaled = p->stimescaled = 0;
1982 prev_cputime_init(&p->prev_cputime);
1984 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1985 seqcount_init(&p->vtime.seqcount);
1986 p->vtime.starttime = 0;
1987 p->vtime.state = VTIME_INACTIVE;
1990 #if defined(SPLIT_RSS_COUNTING)
1991 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1994 p->default_timer_slack_ns = current->timer_slack_ns;
2000 task_io_accounting_init(&p->ioac);
2001 acct_clear_integrals(p);
2003 posix_cputimers_init(&p->posix_cputimers);
2005 p->io_context = NULL;
2006 audit_set_context(p, NULL);
2009 p->mempolicy = mpol_dup(p->mempolicy);
2010 if (IS_ERR(p->mempolicy)) {
2011 retval = PTR_ERR(p->mempolicy);
2012 p->mempolicy = NULL;
2013 goto bad_fork_cleanup_threadgroup_lock;
2016 #ifdef CONFIG_CPUSETS
2017 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2018 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2019 seqcount_init(&p->mems_allowed_seq);
2021 #ifdef CONFIG_TRACE_IRQFLAGS
2023 p->hardirqs_enabled = 0;
2024 p->hardirq_enable_ip = 0;
2025 p->hardirq_enable_event = 0;
2026 p->hardirq_disable_ip = _THIS_IP_;
2027 p->hardirq_disable_event = 0;
2028 p->softirqs_enabled = 1;
2029 p->softirq_enable_ip = _THIS_IP_;
2030 p->softirq_enable_event = 0;
2031 p->softirq_disable_ip = 0;
2032 p->softirq_disable_event = 0;
2033 p->hardirq_context = 0;
2034 p->softirq_context = 0;
2037 p->pagefault_disabled = 0;
2039 #ifdef CONFIG_LOCKDEP
2040 lockdep_init_task(p);
2043 #ifdef CONFIG_DEBUG_MUTEXES
2044 p->blocked_on = NULL; /* not blocked yet */
2046 #ifdef CONFIG_BCACHE
2047 p->sequential_io = 0;
2048 p->sequential_io_avg = 0;
2051 /* Perform scheduler related setup. Assign this task to a CPU. */
2052 retval = sched_fork(clone_flags, p);
2054 goto bad_fork_cleanup_policy;
2056 retval = perf_event_init_task(p);
2058 goto bad_fork_cleanup_policy;
2059 retval = audit_alloc(p);
2061 goto bad_fork_cleanup_perf;
2062 /* copy all the process information */
2064 retval = security_task_alloc(p, clone_flags);
2066 goto bad_fork_cleanup_audit;
2067 retval = copy_semundo(clone_flags, p);
2069 goto bad_fork_cleanup_security;
2070 retval = copy_files(clone_flags, p);
2072 goto bad_fork_cleanup_semundo;
2073 retval = copy_fs(clone_flags, p);
2075 goto bad_fork_cleanup_files;
2076 retval = copy_sighand(clone_flags, p);
2078 goto bad_fork_cleanup_fs;
2079 retval = copy_signal(clone_flags, p);
2081 goto bad_fork_cleanup_sighand;
2082 retval = copy_mm(clone_flags, p);
2084 goto bad_fork_cleanup_signal;
2085 retval = copy_namespaces(clone_flags, p);
2087 goto bad_fork_cleanup_mm;
2088 retval = copy_io(clone_flags, p);
2090 goto bad_fork_cleanup_namespaces;
2091 retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2094 goto bad_fork_cleanup_io;
2096 stackleak_task_init(p);
2098 if (pid != &init_struct_pid) {
2099 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2100 args->set_tid_size);
2102 retval = PTR_ERR(pid);
2103 goto bad_fork_cleanup_thread;
2108 * This has to happen after we've potentially unshared the file
2109 * descriptor table (so that the pidfd doesn't leak into the child
2110 * if the fd table isn't shared).
2112 if (clone_flags & CLONE_PIDFD) {
2113 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2115 goto bad_fork_free_pid;
2119 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2120 O_RDWR | O_CLOEXEC);
2121 if (IS_ERR(pidfile)) {
2122 put_unused_fd(pidfd);
2123 retval = PTR_ERR(pidfile);
2124 goto bad_fork_free_pid;
2126 get_pid(pid); /* held by pidfile now */
2128 retval = put_user(pidfd, args->pidfd);
2130 goto bad_fork_put_pidfd;
2139 * sigaltstack should be cleared when sharing the same VM
2141 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2145 * Syscall tracing and stepping should be turned off in the
2146 * child regardless of CLONE_PTRACE.
2148 user_disable_single_step(p);
2149 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2150 #ifdef TIF_SYSCALL_EMU
2151 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2153 clear_tsk_latency_tracing(p);
2155 /* ok, now we should be set up.. */
2156 p->pid = pid_nr(pid);
2157 if (clone_flags & CLONE_THREAD) {
2158 p->exit_signal = -1;
2159 p->group_leader = current->group_leader;
2160 p->tgid = current->tgid;
2162 if (clone_flags & CLONE_PARENT)
2163 p->exit_signal = current->group_leader->exit_signal;
2165 p->exit_signal = args->exit_signal;
2166 p->group_leader = p;
2171 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2172 p->dirty_paused_when = 0;
2174 p->pdeath_signal = 0;
2175 INIT_LIST_HEAD(&p->thread_group);
2176 p->task_works = NULL;
2178 cgroup_threadgroup_change_begin(current);
2180 * Ensure that the cgroup subsystem policies allow the new process to be
2181 * forked. It should be noted the the new process's css_set can be changed
2182 * between here and cgroup_post_fork() if an organisation operation is in
2185 retval = cgroup_can_fork(p);
2187 goto bad_fork_cgroup_threadgroup_change_end;
2190 * From this point on we must avoid any synchronous user-space
2191 * communication until we take the tasklist-lock. In particular, we do
2192 * not want user-space to be able to predict the process start-time by
2193 * stalling fork(2) after we recorded the start_time but before it is
2194 * visible to the system.
2197 p->start_time = ktime_get_ns();
2198 p->start_boottime = ktime_get_boottime_ns();
2201 * Make it visible to the rest of the system, but dont wake it up yet.
2202 * Need tasklist lock for parent etc handling!
2204 write_lock_irq(&tasklist_lock);
2206 /* CLONE_PARENT re-uses the old parent */
2207 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2208 p->real_parent = current->real_parent;
2209 p->parent_exec_id = current->parent_exec_id;
2211 p->real_parent = current;
2212 p->parent_exec_id = current->self_exec_id;
2215 klp_copy_process(p);
2217 spin_lock(¤t->sighand->siglock);
2220 * Copy seccomp details explicitly here, in case they were changed
2221 * before holding sighand lock.
2225 rseq_fork(p, clone_flags);
2227 /* Don't start children in a dying pid namespace */
2228 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2230 goto bad_fork_cancel_cgroup;
2233 /* Let kill terminate clone/fork in the middle */
2234 if (fatal_signal_pending(current)) {
2236 goto bad_fork_cancel_cgroup;
2239 /* past the last point of failure */
2241 fd_install(pidfd, pidfile);
2243 init_task_pid_links(p);
2244 if (likely(p->pid)) {
2245 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2247 init_task_pid(p, PIDTYPE_PID, pid);
2248 if (thread_group_leader(p)) {
2249 init_task_pid(p, PIDTYPE_TGID, pid);
2250 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2251 init_task_pid(p, PIDTYPE_SID, task_session(current));
2253 if (is_child_reaper(pid)) {
2254 ns_of_pid(pid)->child_reaper = p;
2255 p->signal->flags |= SIGNAL_UNKILLABLE;
2257 p->signal->shared_pending.signal = delayed.signal;
2258 p->signal->tty = tty_kref_get(current->signal->tty);
2260 * Inherit has_child_subreaper flag under the same
2261 * tasklist_lock with adding child to the process tree
2262 * for propagate_has_child_subreaper optimization.
2264 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2265 p->real_parent->signal->is_child_subreaper;
2266 list_add_tail(&p->sibling, &p->real_parent->children);
2267 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2268 attach_pid(p, PIDTYPE_TGID);
2269 attach_pid(p, PIDTYPE_PGID);
2270 attach_pid(p, PIDTYPE_SID);
2271 __this_cpu_inc(process_counts);
2273 current->signal->nr_threads++;
2274 atomic_inc(¤t->signal->live);
2275 refcount_inc(¤t->signal->sigcnt);
2276 task_join_group_stop(p);
2277 list_add_tail_rcu(&p->thread_group,
2278 &p->group_leader->thread_group);
2279 list_add_tail_rcu(&p->thread_node,
2280 &p->signal->thread_head);
2282 attach_pid(p, PIDTYPE_PID);
2286 hlist_del_init(&delayed.node);
2287 spin_unlock(¤t->sighand->siglock);
2288 syscall_tracepoint_update(p);
2289 write_unlock_irq(&tasklist_lock);
2291 proc_fork_connector(p);
2292 cgroup_post_fork(p);
2293 cgroup_threadgroup_change_end(current);
2296 trace_task_newtask(p, clone_flags);
2297 uprobe_copy_process(p, clone_flags);
2301 bad_fork_cancel_cgroup:
2302 spin_unlock(¤t->sighand->siglock);
2303 write_unlock_irq(&tasklist_lock);
2304 cgroup_cancel_fork(p);
2305 bad_fork_cgroup_threadgroup_change_end:
2306 cgroup_threadgroup_change_end(current);
2308 if (clone_flags & CLONE_PIDFD) {
2310 put_unused_fd(pidfd);
2313 if (pid != &init_struct_pid)
2315 bad_fork_cleanup_thread:
2317 bad_fork_cleanup_io:
2320 bad_fork_cleanup_namespaces:
2321 exit_task_namespaces(p);
2322 bad_fork_cleanup_mm:
2324 mm_clear_owner(p->mm, p);
2327 bad_fork_cleanup_signal:
2328 if (!(clone_flags & CLONE_THREAD))
2329 free_signal_struct(p->signal);
2330 bad_fork_cleanup_sighand:
2331 __cleanup_sighand(p->sighand);
2332 bad_fork_cleanup_fs:
2333 exit_fs(p); /* blocking */
2334 bad_fork_cleanup_files:
2335 exit_files(p); /* blocking */
2336 bad_fork_cleanup_semundo:
2338 bad_fork_cleanup_security:
2339 security_task_free(p);
2340 bad_fork_cleanup_audit:
2342 bad_fork_cleanup_perf:
2343 perf_event_free_task(p);
2344 bad_fork_cleanup_policy:
2345 lockdep_free_task(p);
2347 mpol_put(p->mempolicy);
2348 bad_fork_cleanup_threadgroup_lock:
2350 delayacct_tsk_free(p);
2351 bad_fork_cleanup_count:
2352 atomic_dec(&p->cred->user->processes);
2355 p->state = TASK_DEAD;
2357 delayed_free_task(p);
2359 spin_lock_irq(¤t->sighand->siglock);
2360 hlist_del_init(&delayed.node);
2361 spin_unlock_irq(¤t->sighand->siglock);
2362 return ERR_PTR(retval);
2365 static inline void init_idle_pids(struct task_struct *idle)
2369 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2370 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2371 init_task_pid(idle, type, &init_struct_pid);
2375 struct task_struct *fork_idle(int cpu)
2377 struct task_struct *task;
2378 struct kernel_clone_args args = {
2382 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2383 if (!IS_ERR(task)) {
2384 init_idle_pids(task);
2385 init_idle(task, cpu);
2391 struct mm_struct *copy_init_mm(void)
2393 return dup_mm(NULL, &init_mm);
2397 * Ok, this is the main fork-routine.
2399 * It copies the process, and if successful kick-starts
2400 * it and waits for it to finish using the VM if required.
2402 * args->exit_signal is expected to be checked for sanity by the caller.
2404 long _do_fork(struct kernel_clone_args *args)
2406 u64 clone_flags = args->flags;
2407 struct completion vfork;
2409 struct task_struct *p;
2414 * Determine whether and which event to report to ptracer. When
2415 * called from kernel_thread or CLONE_UNTRACED is explicitly
2416 * requested, no event is reported; otherwise, report if the event
2417 * for the type of forking is enabled.
2419 if (!(clone_flags & CLONE_UNTRACED)) {
2420 if (clone_flags & CLONE_VFORK)
2421 trace = PTRACE_EVENT_VFORK;
2422 else if (args->exit_signal != SIGCHLD)
2423 trace = PTRACE_EVENT_CLONE;
2425 trace = PTRACE_EVENT_FORK;
2427 if (likely(!ptrace_event_enabled(current, trace)))
2431 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2432 add_latent_entropy();
2438 * Do this prior waking up the new thread - the thread pointer
2439 * might get invalid after that point, if the thread exits quickly.
2441 trace_sched_process_fork(current, p);
2443 pid = get_task_pid(p, PIDTYPE_PID);
2446 if (clone_flags & CLONE_PARENT_SETTID)
2447 put_user(nr, args->parent_tid);
2449 if (clone_flags & CLONE_VFORK) {
2450 p->vfork_done = &vfork;
2451 init_completion(&vfork);
2455 wake_up_new_task(p);
2457 /* forking complete and child started to run, tell ptracer */
2458 if (unlikely(trace))
2459 ptrace_event_pid(trace, pid);
2461 if (clone_flags & CLONE_VFORK) {
2462 if (!wait_for_vfork_done(p, &vfork))
2463 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2470 bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
2472 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2473 if ((kargs->flags & CLONE_PIDFD) &&
2474 (kargs->flags & CLONE_PARENT_SETTID))
2480 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2481 /* For compatibility with architectures that call do_fork directly rather than
2482 * using the syscall entry points below. */
2483 long do_fork(unsigned long clone_flags,
2484 unsigned long stack_start,
2485 unsigned long stack_size,
2486 int __user *parent_tidptr,
2487 int __user *child_tidptr)
2489 struct kernel_clone_args args = {
2490 .flags = (clone_flags & ~CSIGNAL),
2491 .pidfd = parent_tidptr,
2492 .child_tid = child_tidptr,
2493 .parent_tid = parent_tidptr,
2494 .exit_signal = (clone_flags & CSIGNAL),
2495 .stack = stack_start,
2496 .stack_size = stack_size,
2499 if (!legacy_clone_args_valid(&args))
2502 return _do_fork(&args);
2507 * Create a kernel thread.
2509 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2511 struct kernel_clone_args args = {
2512 .flags = ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2513 .exit_signal = (flags & CSIGNAL),
2514 .stack = (unsigned long)fn,
2515 .stack_size = (unsigned long)arg,
2518 return _do_fork(&args);
2521 #ifdef __ARCH_WANT_SYS_FORK
2522 SYSCALL_DEFINE0(fork)
2525 struct kernel_clone_args args = {
2526 .exit_signal = SIGCHLD,
2529 return _do_fork(&args);
2531 /* can not support in nommu mode */
2537 #ifdef __ARCH_WANT_SYS_VFORK
2538 SYSCALL_DEFINE0(vfork)
2540 struct kernel_clone_args args = {
2541 .flags = CLONE_VFORK | CLONE_VM,
2542 .exit_signal = SIGCHLD,
2545 return _do_fork(&args);
2549 #ifdef __ARCH_WANT_SYS_CLONE
2550 #ifdef CONFIG_CLONE_BACKWARDS
2551 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2552 int __user *, parent_tidptr,
2554 int __user *, child_tidptr)
2555 #elif defined(CONFIG_CLONE_BACKWARDS2)
2556 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2557 int __user *, parent_tidptr,
2558 int __user *, child_tidptr,
2560 #elif defined(CONFIG_CLONE_BACKWARDS3)
2561 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2563 int __user *, parent_tidptr,
2564 int __user *, child_tidptr,
2567 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2568 int __user *, parent_tidptr,
2569 int __user *, child_tidptr,
2573 struct kernel_clone_args args = {
2574 .flags = (clone_flags & ~CSIGNAL),
2575 .pidfd = parent_tidptr,
2576 .child_tid = child_tidptr,
2577 .parent_tid = parent_tidptr,
2578 .exit_signal = (clone_flags & CSIGNAL),
2583 if (!legacy_clone_args_valid(&args))
2586 return _do_fork(&args);
2590 #ifdef __ARCH_WANT_SYS_CLONE3
2593 * copy_thread implementations handle CLONE_SETTLS by reading the TLS value from
2594 * the registers containing the syscall arguments for clone. This doesn't work
2595 * with clone3 since the TLS value is passed in clone_args instead.
2597 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2598 #error clone3 requires copy_thread_tls support in arch
2601 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2602 struct clone_args __user *uargs,
2606 struct clone_args args;
2607 pid_t *kset_tid = kargs->set_tid;
2609 if (unlikely(usize > PAGE_SIZE))
2611 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2614 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2618 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2621 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2624 if (unlikely(args.set_tid && args.set_tid_size == 0))
2628 * Verify that higher 32bits of exit_signal are unset and that
2629 * it is a valid signal
2631 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2632 !valid_signal(args.exit_signal)))
2635 *kargs = (struct kernel_clone_args){
2636 .flags = args.flags,
2637 .pidfd = u64_to_user_ptr(args.pidfd),
2638 .child_tid = u64_to_user_ptr(args.child_tid),
2639 .parent_tid = u64_to_user_ptr(args.parent_tid),
2640 .exit_signal = args.exit_signal,
2641 .stack = args.stack,
2642 .stack_size = args.stack_size,
2644 .set_tid_size = args.set_tid_size,
2648 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2649 (kargs->set_tid_size * sizeof(pid_t))))
2652 kargs->set_tid = kset_tid;
2658 * clone3_stack_valid - check and prepare stack
2659 * @kargs: kernel clone args
2661 * Verify that the stack arguments userspace gave us are sane.
2662 * In addition, set the stack direction for userspace since it's easy for us to
2665 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2667 if (kargs->stack == 0) {
2668 if (kargs->stack_size > 0)
2671 if (kargs->stack_size == 0)
2674 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2677 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2678 kargs->stack += kargs->stack_size;
2685 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2687 /* Verify that no unknown flags are passed along. */
2688 if (kargs->flags & ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND))
2692 * - make the CLONE_DETACHED bit reuseable for clone3
2693 * - make the CSIGNAL bits reuseable for clone3
2695 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2698 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2699 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2702 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2706 if (!clone3_stack_valid(kargs))
2713 * clone3 - create a new process with specific properties
2714 * @uargs: argument structure
2715 * @size: size of @uargs
2717 * clone3() is the extensible successor to clone()/clone2().
2718 * It takes a struct as argument that is versioned by its size.
2720 * Return: On success, a positive PID for the child process.
2721 * On error, a negative errno number.
2723 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2727 struct kernel_clone_args kargs;
2728 pid_t set_tid[MAX_PID_NS_LEVEL];
2730 kargs.set_tid = set_tid;
2732 err = copy_clone_args_from_user(&kargs, uargs, size);
2736 if (!clone3_args_valid(&kargs))
2739 return _do_fork(&kargs);
2743 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2745 struct task_struct *leader, *parent, *child;
2748 read_lock(&tasklist_lock);
2749 leader = top = top->group_leader;
2751 for_each_thread(leader, parent) {
2752 list_for_each_entry(child, &parent->children, sibling) {
2753 res = visitor(child, data);
2765 if (leader != top) {
2767 parent = child->real_parent;
2768 leader = parent->group_leader;
2772 read_unlock(&tasklist_lock);
2775 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2776 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2779 static void sighand_ctor(void *data)
2781 struct sighand_struct *sighand = data;
2783 spin_lock_init(&sighand->siglock);
2784 init_waitqueue_head(&sighand->signalfd_wqh);
2787 void __init proc_caches_init(void)
2789 unsigned int mm_size;
2791 sighand_cachep = kmem_cache_create("sighand_cache",
2792 sizeof(struct sighand_struct), 0,
2793 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2794 SLAB_ACCOUNT, sighand_ctor);
2795 signal_cachep = kmem_cache_create("signal_cache",
2796 sizeof(struct signal_struct), 0,
2797 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2799 files_cachep = kmem_cache_create("files_cache",
2800 sizeof(struct files_struct), 0,
2801 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2803 fs_cachep = kmem_cache_create("fs_cache",
2804 sizeof(struct fs_struct), 0,
2805 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2809 * The mm_cpumask is located at the end of mm_struct, and is
2810 * dynamically sized based on the maximum CPU number this system
2811 * can have, taking hotplug into account (nr_cpu_ids).
2813 mm_size = sizeof(struct mm_struct) + cpumask_size();
2815 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2816 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2817 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2818 offsetof(struct mm_struct, saved_auxv),
2819 sizeof_field(struct mm_struct, saved_auxv),
2821 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2823 nsproxy_cache_init();
2827 * Check constraints on flags passed to the unshare system call.
2829 static int check_unshare_flags(unsigned long unshare_flags)
2831 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2832 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2833 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2834 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2838 * Not implemented, but pretend it works if there is nothing
2839 * to unshare. Note that unsharing the address space or the
2840 * signal handlers also need to unshare the signal queues (aka
2843 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2844 if (!thread_group_empty(current))
2847 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2848 if (refcount_read(¤t->sighand->count) > 1)
2851 if (unshare_flags & CLONE_VM) {
2852 if (!current_is_single_threaded())
2860 * Unshare the filesystem structure if it is being shared
2862 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2864 struct fs_struct *fs = current->fs;
2866 if (!(unshare_flags & CLONE_FS) || !fs)
2869 /* don't need lock here; in the worst case we'll do useless copy */
2873 *new_fsp = copy_fs_struct(fs);
2881 * Unshare file descriptor table if it is being shared
2883 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2885 struct files_struct *fd = current->files;
2888 if ((unshare_flags & CLONE_FILES) &&
2889 (fd && atomic_read(&fd->count) > 1)) {
2890 *new_fdp = dup_fd(fd, &error);
2899 * unshare allows a process to 'unshare' part of the process
2900 * context which was originally shared using clone. copy_*
2901 * functions used by do_fork() cannot be used here directly
2902 * because they modify an inactive task_struct that is being
2903 * constructed. Here we are modifying the current, active,
2906 int ksys_unshare(unsigned long unshare_flags)
2908 struct fs_struct *fs, *new_fs = NULL;
2909 struct files_struct *fd, *new_fd = NULL;
2910 struct cred *new_cred = NULL;
2911 struct nsproxy *new_nsproxy = NULL;
2916 * If unsharing a user namespace must also unshare the thread group
2917 * and unshare the filesystem root and working directories.
2919 if (unshare_flags & CLONE_NEWUSER)
2920 unshare_flags |= CLONE_THREAD | CLONE_FS;
2922 * If unsharing vm, must also unshare signal handlers.
2924 if (unshare_flags & CLONE_VM)
2925 unshare_flags |= CLONE_SIGHAND;
2927 * If unsharing a signal handlers, must also unshare the signal queues.
2929 if (unshare_flags & CLONE_SIGHAND)
2930 unshare_flags |= CLONE_THREAD;
2932 * If unsharing namespace, must also unshare filesystem information.
2934 if (unshare_flags & CLONE_NEWNS)
2935 unshare_flags |= CLONE_FS;
2937 err = check_unshare_flags(unshare_flags);
2939 goto bad_unshare_out;
2941 * CLONE_NEWIPC must also detach from the undolist: after switching
2942 * to a new ipc namespace, the semaphore arrays from the old
2943 * namespace are unreachable.
2945 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2947 err = unshare_fs(unshare_flags, &new_fs);
2949 goto bad_unshare_out;
2950 err = unshare_fd(unshare_flags, &new_fd);
2952 goto bad_unshare_cleanup_fs;
2953 err = unshare_userns(unshare_flags, &new_cred);
2955 goto bad_unshare_cleanup_fd;
2956 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2959 goto bad_unshare_cleanup_cred;
2961 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2964 * CLONE_SYSVSEM is equivalent to sys_exit().
2968 if (unshare_flags & CLONE_NEWIPC) {
2969 /* Orphan segments in old ns (see sem above). */
2971 shm_init_task(current);
2975 switch_task_namespaces(current, new_nsproxy);
2981 spin_lock(&fs->lock);
2982 current->fs = new_fs;
2987 spin_unlock(&fs->lock);
2991 fd = current->files;
2992 current->files = new_fd;
2996 task_unlock(current);
2999 /* Install the new user namespace */
3000 commit_creds(new_cred);
3005 perf_event_namespaces(current);
3007 bad_unshare_cleanup_cred:
3010 bad_unshare_cleanup_fd:
3012 put_files_struct(new_fd);
3014 bad_unshare_cleanup_fs:
3016 free_fs_struct(new_fs);
3022 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3024 return ksys_unshare(unshare_flags);
3028 * Helper to unshare the files of the current task.
3029 * We don't want to expose copy_files internals to
3030 * the exec layer of the kernel.
3033 int unshare_files(struct files_struct **displaced)
3035 struct task_struct *task = current;
3036 struct files_struct *copy = NULL;
3039 error = unshare_fd(CLONE_FILES, ©);
3040 if (error || !copy) {
3044 *displaced = task->files;
3051 int sysctl_max_threads(struct ctl_table *table, int write,
3052 void __user *buffer, size_t *lenp, loff_t *ppos)
3056 int threads = max_threads;
3058 int max = MAX_THREADS;
3065 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3069 max_threads = threads;