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>
43 #include <linux/hmm.h>
46 #include <linux/vmacache.h>
47 #include <linux/nsproxy.h>
48 #include <linux/capability.h>
49 #include <linux/cpu.h>
50 #include <linux/cgroup.h>
51 #include <linux/security.h>
52 #include <linux/hugetlb.h>
53 #include <linux/seccomp.h>
54 #include <linux/swap.h>
55 #include <linux/syscalls.h>
56 #include <linux/jiffies.h>
57 #include <linux/futex.h>
58 #include <linux/compat.h>
59 #include <linux/kthread.h>
60 #include <linux/task_io_accounting_ops.h>
61 #include <linux/rcupdate.h>
62 #include <linux/ptrace.h>
63 #include <linux/mount.h>
64 #include <linux/audit.h>
65 #include <linux/memcontrol.h>
66 #include <linux/ftrace.h>
67 #include <linux/proc_fs.h>
68 #include <linux/profile.h>
69 #include <linux/rmap.h>
70 #include <linux/ksm.h>
71 #include <linux/acct.h>
72 #include <linux/userfaultfd_k.h>
73 #include <linux/tsacct_kern.h>
74 #include <linux/cn_proc.h>
75 #include <linux/freezer.h>
76 #include <linux/delayacct.h>
77 #include <linux/taskstats_kern.h>
78 #include <linux/random.h>
79 #include <linux/tty.h>
80 #include <linux/blkdev.h>
81 #include <linux/fs_struct.h>
82 #include <linux/magic.h>
83 #include <linux/perf_event.h>
84 #include <linux/posix-timers.h>
85 #include <linux/user-return-notifier.h>
86 #include <linux/oom.h>
87 #include <linux/khugepaged.h>
88 #include <linux/signalfd.h>
89 #include <linux/uprobes.h>
90 #include <linux/aio.h>
91 #include <linux/compiler.h>
92 #include <linux/sysctl.h>
93 #include <linux/kcov.h>
94 #include <linux/livepatch.h>
95 #include <linux/thread_info.h>
96 #include <linux/stackleak.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_PER_CPU(unsigned long, process_counts) = 0;
130 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
132 #ifdef CONFIG_PROVE_RCU
133 int lockdep_tasklist_lock_is_held(void)
135 return lockdep_is_held(&tasklist_lock);
137 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
138 #endif /* #ifdef CONFIG_PROVE_RCU */
140 int nr_processes(void)
145 for_each_possible_cpu(cpu)
146 total += per_cpu(process_counts, cpu);
151 void __weak arch_release_task_struct(struct task_struct *tsk)
155 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
156 static struct kmem_cache *task_struct_cachep;
158 static inline struct task_struct *alloc_task_struct_node(int node)
160 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
163 static inline void free_task_struct(struct task_struct *tsk)
165 kmem_cache_free(task_struct_cachep, tsk);
169 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
172 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
173 * kmemcache based allocator.
175 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
177 #ifdef CONFIG_VMAP_STACK
179 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
180 * flush. Try to minimize the number of calls by caching stacks.
182 #define NR_CACHED_STACKS 2
183 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
185 static int free_vm_stack_cache(unsigned int cpu)
187 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
190 for (i = 0; i < NR_CACHED_STACKS; i++) {
191 struct vm_struct *vm_stack = cached_vm_stacks[i];
196 vfree(vm_stack->addr);
197 cached_vm_stacks[i] = NULL;
204 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
206 #ifdef CONFIG_VMAP_STACK
210 for (i = 0; i < NR_CACHED_STACKS; i++) {
213 s = this_cpu_xchg(cached_stacks[i], NULL);
218 /* Clear stale pointers from reused stack. */
219 memset(s->addr, 0, THREAD_SIZE);
221 tsk->stack_vm_area = s;
222 tsk->stack = s->addr;
227 * Allocated stacks are cached and later reused by new threads,
228 * so memcg accounting is performed manually on assigning/releasing
229 * stacks to tasks. Drop __GFP_ACCOUNT.
231 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
232 VMALLOC_START, VMALLOC_END,
233 THREADINFO_GFP & ~__GFP_ACCOUNT,
235 0, node, __builtin_return_address(0));
238 * We can't call find_vm_area() in interrupt context, and
239 * free_thread_stack() can be called in interrupt context,
240 * so cache the vm_struct.
243 tsk->stack_vm_area = find_vm_area(stack);
248 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
252 tsk->stack = page_address(page);
259 static inline void free_thread_stack(struct task_struct *tsk)
261 #ifdef CONFIG_VMAP_STACK
262 struct vm_struct *vm = task_stack_vm_area(tsk);
267 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
268 mod_memcg_page_state(vm->pages[i],
269 MEMCG_KERNEL_STACK_KB,
270 -(int)(PAGE_SIZE / 1024));
272 memcg_kmem_uncharge(vm->pages[i], 0);
275 for (i = 0; i < NR_CACHED_STACKS; i++) {
276 if (this_cpu_cmpxchg(cached_stacks[i],
277 NULL, tsk->stack_vm_area) != NULL)
283 vfree_atomic(tsk->stack);
288 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
291 static struct kmem_cache *thread_stack_cache;
293 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
296 unsigned long *stack;
297 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
302 static void free_thread_stack(struct task_struct *tsk)
304 kmem_cache_free(thread_stack_cache, tsk->stack);
307 void thread_stack_cache_init(void)
309 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
310 THREAD_SIZE, THREAD_SIZE, 0, 0,
312 BUG_ON(thread_stack_cache == NULL);
317 /* SLAB cache for signal_struct structures (tsk->signal) */
318 static struct kmem_cache *signal_cachep;
320 /* SLAB cache for sighand_struct structures (tsk->sighand) */
321 struct kmem_cache *sighand_cachep;
323 /* SLAB cache for files_struct structures (tsk->files) */
324 struct kmem_cache *files_cachep;
326 /* SLAB cache for fs_struct structures (tsk->fs) */
327 struct kmem_cache *fs_cachep;
329 /* SLAB cache for vm_area_struct structures */
330 static struct kmem_cache *vm_area_cachep;
332 /* SLAB cache for mm_struct structures (tsk->mm) */
333 static struct kmem_cache *mm_cachep;
335 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
337 struct vm_area_struct *vma;
339 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
345 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
347 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
351 INIT_LIST_HEAD(&new->anon_vma_chain);
356 void vm_area_free(struct vm_area_struct *vma)
358 kmem_cache_free(vm_area_cachep, vma);
361 static void account_kernel_stack(struct task_struct *tsk, int account)
363 void *stack = task_stack_page(tsk);
364 struct vm_struct *vm = task_stack_vm_area(tsk);
366 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
371 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
373 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
374 mod_zone_page_state(page_zone(vm->pages[i]),
376 PAGE_SIZE / 1024 * account);
380 * All stack pages are in the same zone and belong to the
383 struct page *first_page = virt_to_page(stack);
385 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
386 THREAD_SIZE / 1024 * account);
388 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
389 account * (THREAD_SIZE / 1024));
393 static int memcg_charge_kernel_stack(struct task_struct *tsk)
395 #ifdef CONFIG_VMAP_STACK
396 struct vm_struct *vm = task_stack_vm_area(tsk);
402 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
404 * If memcg_kmem_charge() fails, page->mem_cgroup
405 * pointer is NULL, and both memcg_kmem_uncharge()
406 * and mod_memcg_page_state() in free_thread_stack()
407 * will ignore this page. So it's safe.
409 ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
413 mod_memcg_page_state(vm->pages[i],
414 MEMCG_KERNEL_STACK_KB,
422 static void release_task_stack(struct task_struct *tsk)
424 if (WARN_ON(tsk->state != TASK_DEAD))
425 return; /* Better to leak the stack than to free prematurely */
427 account_kernel_stack(tsk, -1);
428 free_thread_stack(tsk);
430 #ifdef CONFIG_VMAP_STACK
431 tsk->stack_vm_area = NULL;
435 #ifdef CONFIG_THREAD_INFO_IN_TASK
436 void put_task_stack(struct task_struct *tsk)
438 if (refcount_dec_and_test(&tsk->stack_refcount))
439 release_task_stack(tsk);
443 void free_task(struct task_struct *tsk)
445 #ifndef CONFIG_THREAD_INFO_IN_TASK
447 * The task is finally done with both the stack and thread_info,
450 release_task_stack(tsk);
453 * If the task had a separate stack allocation, it should be gone
456 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
458 rt_mutex_debug_task_free(tsk);
459 ftrace_graph_exit_task(tsk);
460 put_seccomp_filter(tsk);
461 arch_release_task_struct(tsk);
462 if (tsk->flags & PF_KTHREAD)
463 free_kthread_struct(tsk);
464 free_task_struct(tsk);
466 EXPORT_SYMBOL(free_task);
469 static __latent_entropy int dup_mmap(struct mm_struct *mm,
470 struct mm_struct *oldmm)
472 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
473 struct rb_node **rb_link, *rb_parent;
475 unsigned long charge;
478 uprobe_start_dup_mmap();
479 if (down_write_killable(&oldmm->mmap_sem)) {
481 goto fail_uprobe_end;
483 flush_cache_dup_mm(oldmm);
484 uprobe_dup_mmap(oldmm, mm);
486 * Not linked in yet - no deadlock potential:
488 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
490 /* No ordering required: file already has been exposed. */
491 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
493 mm->total_vm = oldmm->total_vm;
494 mm->data_vm = oldmm->data_vm;
495 mm->exec_vm = oldmm->exec_vm;
496 mm->stack_vm = oldmm->stack_vm;
498 rb_link = &mm->mm_rb.rb_node;
501 retval = ksm_fork(mm, oldmm);
504 retval = khugepaged_fork(mm, oldmm);
509 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
512 if (mpnt->vm_flags & VM_DONTCOPY) {
513 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
518 * Don't duplicate many vmas if we've been oom-killed (for
521 if (fatal_signal_pending(current)) {
525 if (mpnt->vm_flags & VM_ACCOUNT) {
526 unsigned long len = vma_pages(mpnt);
528 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
532 tmp = vm_area_dup(mpnt);
535 retval = vma_dup_policy(mpnt, tmp);
537 goto fail_nomem_policy;
539 retval = dup_userfaultfd(tmp, &uf);
541 goto fail_nomem_anon_vma_fork;
542 if (tmp->vm_flags & VM_WIPEONFORK) {
543 /* VM_WIPEONFORK gets a clean slate in the child. */
544 tmp->anon_vma = NULL;
545 if (anon_vma_prepare(tmp))
546 goto fail_nomem_anon_vma_fork;
547 } else if (anon_vma_fork(tmp, mpnt))
548 goto fail_nomem_anon_vma_fork;
549 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
550 tmp->vm_next = tmp->vm_prev = NULL;
553 struct inode *inode = file_inode(file);
554 struct address_space *mapping = file->f_mapping;
557 if (tmp->vm_flags & VM_DENYWRITE)
558 atomic_dec(&inode->i_writecount);
559 i_mmap_lock_write(mapping);
560 if (tmp->vm_flags & VM_SHARED)
561 atomic_inc(&mapping->i_mmap_writable);
562 flush_dcache_mmap_lock(mapping);
563 /* insert tmp into the share list, just after mpnt */
564 vma_interval_tree_insert_after(tmp, mpnt,
566 flush_dcache_mmap_unlock(mapping);
567 i_mmap_unlock_write(mapping);
571 * Clear hugetlb-related page reserves for children. This only
572 * affects MAP_PRIVATE mappings. Faults generated by the child
573 * are not guaranteed to succeed, even if read-only
575 if (is_vm_hugetlb_page(tmp))
576 reset_vma_resv_huge_pages(tmp);
579 * Link in the new vma and copy the page table entries.
582 pprev = &tmp->vm_next;
586 __vma_link_rb(mm, tmp, rb_link, rb_parent);
587 rb_link = &tmp->vm_rb.rb_right;
588 rb_parent = &tmp->vm_rb;
591 if (!(tmp->vm_flags & VM_WIPEONFORK))
592 retval = copy_page_range(mm, oldmm, mpnt);
594 if (tmp->vm_ops && tmp->vm_ops->open)
595 tmp->vm_ops->open(tmp);
600 /* a new mm has just been created */
601 retval = arch_dup_mmap(oldmm, mm);
603 up_write(&mm->mmap_sem);
605 up_write(&oldmm->mmap_sem);
606 dup_userfaultfd_complete(&uf);
608 uprobe_end_dup_mmap();
610 fail_nomem_anon_vma_fork:
611 mpol_put(vma_policy(tmp));
616 vm_unacct_memory(charge);
620 static inline int mm_alloc_pgd(struct mm_struct *mm)
622 mm->pgd = pgd_alloc(mm);
623 if (unlikely(!mm->pgd))
628 static inline void mm_free_pgd(struct mm_struct *mm)
630 pgd_free(mm, mm->pgd);
633 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
635 down_write(&oldmm->mmap_sem);
636 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
637 up_write(&oldmm->mmap_sem);
640 #define mm_alloc_pgd(mm) (0)
641 #define mm_free_pgd(mm)
642 #endif /* CONFIG_MMU */
644 static void check_mm(struct mm_struct *mm)
648 for (i = 0; i < NR_MM_COUNTERS; i++) {
649 long x = atomic_long_read(&mm->rss_stat.count[i]);
652 printk(KERN_ALERT "BUG: Bad rss-counter state "
653 "mm:%p idx:%d val:%ld\n", mm, i, x);
656 if (mm_pgtables_bytes(mm))
657 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
658 mm_pgtables_bytes(mm));
660 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
661 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
665 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
666 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
669 * Called when the last reference to the mm
670 * is dropped: either by a lazy thread or by
671 * mmput. Free the page directory and the mm.
673 void __mmdrop(struct mm_struct *mm)
675 BUG_ON(mm == &init_mm);
676 WARN_ON_ONCE(mm == current->mm);
677 WARN_ON_ONCE(mm == current->active_mm);
680 mmu_notifier_mm_destroy(mm);
682 put_user_ns(mm->user_ns);
685 EXPORT_SYMBOL_GPL(__mmdrop);
687 static void mmdrop_async_fn(struct work_struct *work)
689 struct mm_struct *mm;
691 mm = container_of(work, struct mm_struct, async_put_work);
695 static void mmdrop_async(struct mm_struct *mm)
697 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
698 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
699 schedule_work(&mm->async_put_work);
703 static inline void free_signal_struct(struct signal_struct *sig)
705 taskstats_tgid_free(sig);
706 sched_autogroup_exit(sig);
708 * __mmdrop is not safe to call from softirq context on x86 due to
709 * pgd_dtor so postpone it to the async context
712 mmdrop_async(sig->oom_mm);
713 kmem_cache_free(signal_cachep, sig);
716 static inline void put_signal_struct(struct signal_struct *sig)
718 if (refcount_dec_and_test(&sig->sigcnt))
719 free_signal_struct(sig);
722 void __put_task_struct(struct task_struct *tsk)
724 WARN_ON(!tsk->exit_state);
725 WARN_ON(refcount_read(&tsk->usage));
726 WARN_ON(tsk == current);
729 task_numa_free(tsk, true);
730 security_task_free(tsk);
732 delayacct_tsk_free(tsk);
733 put_signal_struct(tsk->signal);
735 if (!profile_handoff_task(tsk))
738 EXPORT_SYMBOL_GPL(__put_task_struct);
740 void __init __weak arch_task_cache_init(void) { }
745 static void set_max_threads(unsigned int max_threads_suggested)
748 unsigned long nr_pages = totalram_pages();
751 * The number of threads shall be limited such that the thread
752 * structures may only consume a small part of the available memory.
754 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
755 threads = MAX_THREADS;
757 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
758 (u64) THREAD_SIZE * 8UL);
760 if (threads > max_threads_suggested)
761 threads = max_threads_suggested;
763 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
766 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
767 /* Initialized by the architecture: */
768 int arch_task_struct_size __read_mostly;
771 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
772 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
774 /* Fetch thread_struct whitelist for the architecture. */
775 arch_thread_struct_whitelist(offset, size);
778 * Handle zero-sized whitelist or empty thread_struct, otherwise
779 * adjust offset to position of thread_struct in task_struct.
781 if (unlikely(*size == 0))
784 *offset += offsetof(struct task_struct, thread);
786 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
788 void __init fork_init(void)
791 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
792 #ifndef ARCH_MIN_TASKALIGN
793 #define ARCH_MIN_TASKALIGN 0
795 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
796 unsigned long useroffset, usersize;
798 /* create a slab on which task_structs can be allocated */
799 task_struct_whitelist(&useroffset, &usersize);
800 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
801 arch_task_struct_size, align,
802 SLAB_PANIC|SLAB_ACCOUNT,
803 useroffset, usersize, NULL);
806 /* do the arch specific task caches init */
807 arch_task_cache_init();
809 set_max_threads(MAX_THREADS);
811 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
812 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
813 init_task.signal->rlim[RLIMIT_SIGPENDING] =
814 init_task.signal->rlim[RLIMIT_NPROC];
816 for (i = 0; i < UCOUNT_COUNTS; i++) {
817 init_user_ns.ucount_max[i] = max_threads/2;
820 #ifdef CONFIG_VMAP_STACK
821 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
822 NULL, free_vm_stack_cache);
825 lockdep_init_task(&init_task);
829 int __weak arch_dup_task_struct(struct task_struct *dst,
830 struct task_struct *src)
836 void set_task_stack_end_magic(struct task_struct *tsk)
838 unsigned long *stackend;
840 stackend = end_of_stack(tsk);
841 *stackend = STACK_END_MAGIC; /* for overflow detection */
844 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
846 struct task_struct *tsk;
847 unsigned long *stack;
848 struct vm_struct *stack_vm_area __maybe_unused;
851 if (node == NUMA_NO_NODE)
852 node = tsk_fork_get_node(orig);
853 tsk = alloc_task_struct_node(node);
857 stack = alloc_thread_stack_node(tsk, node);
861 if (memcg_charge_kernel_stack(tsk))
864 stack_vm_area = task_stack_vm_area(tsk);
866 err = arch_dup_task_struct(tsk, orig);
869 * arch_dup_task_struct() clobbers the stack-related fields. Make
870 * sure they're properly initialized before using any stack-related
874 #ifdef CONFIG_VMAP_STACK
875 tsk->stack_vm_area = stack_vm_area;
877 #ifdef CONFIG_THREAD_INFO_IN_TASK
878 refcount_set(&tsk->stack_refcount, 1);
884 #ifdef CONFIG_SECCOMP
886 * We must handle setting up seccomp filters once we're under
887 * the sighand lock in case orig has changed between now and
888 * then. Until then, filter must be NULL to avoid messing up
889 * the usage counts on the error path calling free_task.
891 tsk->seccomp.filter = NULL;
894 setup_thread_stack(tsk, orig);
895 clear_user_return_notifier(tsk);
896 clear_tsk_need_resched(tsk);
897 set_task_stack_end_magic(tsk);
899 #ifdef CONFIG_STACKPROTECTOR
900 tsk->stack_canary = get_random_canary();
902 if (orig->cpus_ptr == &orig->cpus_mask)
903 tsk->cpus_ptr = &tsk->cpus_mask;
906 * One for us, one for whoever does the "release_task()" (usually
909 refcount_set(&tsk->usage, 2);
910 #ifdef CONFIG_BLK_DEV_IO_TRACE
913 tsk->splice_pipe = NULL;
914 tsk->task_frag.page = NULL;
915 tsk->wake_q.next = NULL;
917 account_kernel_stack(tsk, 1);
921 #ifdef CONFIG_FAULT_INJECTION
925 #ifdef CONFIG_BLK_CGROUP
926 tsk->throttle_queue = NULL;
927 tsk->use_memdelay = 0;
931 tsk->active_memcg = NULL;
936 free_thread_stack(tsk);
938 free_task_struct(tsk);
942 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
944 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
946 static int __init coredump_filter_setup(char *s)
948 default_dump_filter =
949 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
950 MMF_DUMP_FILTER_MASK;
954 __setup("coredump_filter=", coredump_filter_setup);
956 #include <linux/init_task.h>
958 static void mm_init_aio(struct mm_struct *mm)
961 spin_lock_init(&mm->ioctx_lock);
962 mm->ioctx_table = NULL;
966 static __always_inline void mm_clear_owner(struct mm_struct *mm,
967 struct task_struct *p)
971 WRITE_ONCE(mm->owner, NULL);
975 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
982 static void mm_init_uprobes_state(struct mm_struct *mm)
984 #ifdef CONFIG_UPROBES
985 mm->uprobes_state.xol_area = NULL;
989 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
990 struct user_namespace *user_ns)
994 mm->vmacache_seqnum = 0;
995 atomic_set(&mm->mm_users, 1);
996 atomic_set(&mm->mm_count, 1);
997 init_rwsem(&mm->mmap_sem);
998 INIT_LIST_HEAD(&mm->mmlist);
999 mm->core_state = NULL;
1000 mm_pgtables_bytes_init(mm);
1003 atomic64_set(&mm->pinned_vm, 0);
1004 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1005 spin_lock_init(&mm->page_table_lock);
1006 spin_lock_init(&mm->arg_lock);
1007 mm_init_cpumask(mm);
1009 mm_init_owner(mm, p);
1010 RCU_INIT_POINTER(mm->exe_file, NULL);
1011 mmu_notifier_mm_init(mm);
1013 init_tlb_flush_pending(mm);
1014 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1015 mm->pmd_huge_pte = NULL;
1017 mm_init_uprobes_state(mm);
1020 mm->flags = current->mm->flags & MMF_INIT_MASK;
1021 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1023 mm->flags = default_dump_filter;
1027 if (mm_alloc_pgd(mm))
1030 if (init_new_context(p, mm))
1031 goto fail_nocontext;
1033 mm->user_ns = get_user_ns(user_ns);
1044 * Allocate and initialize an mm_struct.
1046 struct mm_struct *mm_alloc(void)
1048 struct mm_struct *mm;
1054 memset(mm, 0, sizeof(*mm));
1055 return mm_init(mm, current, current_user_ns());
1058 static inline void __mmput(struct mm_struct *mm)
1060 VM_BUG_ON(atomic_read(&mm->mm_users));
1062 uprobe_clear_state(mm);
1065 khugepaged_exit(mm); /* must run before exit_mmap */
1067 mm_put_huge_zero_page(mm);
1068 set_mm_exe_file(mm, NULL);
1069 if (!list_empty(&mm->mmlist)) {
1070 spin_lock(&mmlist_lock);
1071 list_del(&mm->mmlist);
1072 spin_unlock(&mmlist_lock);
1075 module_put(mm->binfmt->module);
1080 * Decrement the use count and release all resources for an mm.
1082 void mmput(struct mm_struct *mm)
1086 if (atomic_dec_and_test(&mm->mm_users))
1089 EXPORT_SYMBOL_GPL(mmput);
1092 static void mmput_async_fn(struct work_struct *work)
1094 struct mm_struct *mm = container_of(work, struct mm_struct,
1100 void mmput_async(struct mm_struct *mm)
1102 if (atomic_dec_and_test(&mm->mm_users)) {
1103 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1104 schedule_work(&mm->async_put_work);
1110 * set_mm_exe_file - change a reference to the mm's executable file
1112 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1114 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1115 * invocations: in mmput() nobody alive left, in execve task is single
1116 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1117 * mm->exe_file, but does so without using set_mm_exe_file() in order
1118 * to do avoid the need for any locks.
1120 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1122 struct file *old_exe_file;
1125 * It is safe to dereference the exe_file without RCU as
1126 * this function is only called if nobody else can access
1127 * this mm -- see comment above for justification.
1129 old_exe_file = rcu_dereference_raw(mm->exe_file);
1132 get_file(new_exe_file);
1133 rcu_assign_pointer(mm->exe_file, new_exe_file);
1139 * get_mm_exe_file - acquire a reference to the mm's executable file
1141 * Returns %NULL if mm has no associated executable file.
1142 * User must release file via fput().
1144 struct file *get_mm_exe_file(struct mm_struct *mm)
1146 struct file *exe_file;
1149 exe_file = rcu_dereference(mm->exe_file);
1150 if (exe_file && !get_file_rcu(exe_file))
1155 EXPORT_SYMBOL(get_mm_exe_file);
1158 * get_task_exe_file - acquire a reference to the task's executable file
1160 * Returns %NULL if task's mm (if any) has no associated executable file or
1161 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1162 * User must release file via fput().
1164 struct file *get_task_exe_file(struct task_struct *task)
1166 struct file *exe_file = NULL;
1167 struct mm_struct *mm;
1172 if (!(task->flags & PF_KTHREAD))
1173 exe_file = get_mm_exe_file(mm);
1178 EXPORT_SYMBOL(get_task_exe_file);
1181 * get_task_mm - acquire a reference to the task's mm
1183 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1184 * this kernel workthread has transiently adopted a user mm with use_mm,
1185 * to do its AIO) is not set and if so returns a reference to it, after
1186 * bumping up the use count. User must release the mm via mmput()
1187 * after use. Typically used by /proc and ptrace.
1189 struct mm_struct *get_task_mm(struct task_struct *task)
1191 struct mm_struct *mm;
1196 if (task->flags & PF_KTHREAD)
1204 EXPORT_SYMBOL_GPL(get_task_mm);
1206 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1208 struct mm_struct *mm;
1211 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1213 return ERR_PTR(err);
1215 mm = get_task_mm(task);
1216 if (mm && mm != current->mm &&
1217 !ptrace_may_access(task, mode)) {
1219 mm = ERR_PTR(-EACCES);
1221 mutex_unlock(&task->signal->cred_guard_mutex);
1226 static void complete_vfork_done(struct task_struct *tsk)
1228 struct completion *vfork;
1231 vfork = tsk->vfork_done;
1232 if (likely(vfork)) {
1233 tsk->vfork_done = NULL;
1239 static int wait_for_vfork_done(struct task_struct *child,
1240 struct completion *vfork)
1244 freezer_do_not_count();
1245 cgroup_enter_frozen();
1246 killed = wait_for_completion_killable(vfork);
1247 cgroup_leave_frozen(false);
1252 child->vfork_done = NULL;
1256 put_task_struct(child);
1260 /* Please note the differences between mmput and mm_release.
1261 * mmput is called whenever we stop holding onto a mm_struct,
1262 * error success whatever.
1264 * mm_release is called after a mm_struct has been removed
1265 * from the current process.
1267 * This difference is important for error handling, when we
1268 * only half set up a mm_struct for a new process and need to restore
1269 * the old one. Because we mmput the new mm_struct before
1270 * restoring the old one. . .
1271 * Eric Biederman 10 January 1998
1273 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1275 /* Get rid of any futexes when releasing the mm */
1277 if (unlikely(tsk->robust_list)) {
1278 exit_robust_list(tsk);
1279 tsk->robust_list = NULL;
1281 #ifdef CONFIG_COMPAT
1282 if (unlikely(tsk->compat_robust_list)) {
1283 compat_exit_robust_list(tsk);
1284 tsk->compat_robust_list = NULL;
1287 if (unlikely(!list_empty(&tsk->pi_state_list)))
1288 exit_pi_state_list(tsk);
1291 uprobe_free_utask(tsk);
1293 /* Get rid of any cached register state */
1294 deactivate_mm(tsk, mm);
1297 * Signal userspace if we're not exiting with a core dump
1298 * because we want to leave the value intact for debugging
1301 if (tsk->clear_child_tid) {
1302 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1303 atomic_read(&mm->mm_users) > 1) {
1305 * We don't check the error code - if userspace has
1306 * not set up a proper pointer then tough luck.
1308 put_user(0, tsk->clear_child_tid);
1309 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1310 1, NULL, NULL, 0, 0);
1312 tsk->clear_child_tid = NULL;
1316 * All done, finally we can wake up parent and return this mm to him.
1317 * Also kthread_stop() uses this completion for synchronization.
1319 if (tsk->vfork_done)
1320 complete_vfork_done(tsk);
1324 * dup_mm() - duplicates an existing mm structure
1325 * @tsk: the task_struct with which the new mm will be associated.
1326 * @oldmm: the mm to duplicate.
1328 * Allocates a new mm structure and duplicates the provided @oldmm structure
1331 * Return: the duplicated mm or NULL on failure.
1333 static struct mm_struct *dup_mm(struct task_struct *tsk,
1334 struct mm_struct *oldmm)
1336 struct mm_struct *mm;
1343 memcpy(mm, oldmm, sizeof(*mm));
1345 if (!mm_init(mm, tsk, mm->user_ns))
1348 err = dup_mmap(mm, oldmm);
1352 mm->hiwater_rss = get_mm_rss(mm);
1353 mm->hiwater_vm = mm->total_vm;
1355 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1361 /* don't put binfmt in mmput, we haven't got module yet */
1363 mm_init_owner(mm, NULL);
1370 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1372 struct mm_struct *mm, *oldmm;
1375 tsk->min_flt = tsk->maj_flt = 0;
1376 tsk->nvcsw = tsk->nivcsw = 0;
1377 #ifdef CONFIG_DETECT_HUNG_TASK
1378 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1379 tsk->last_switch_time = 0;
1383 tsk->active_mm = NULL;
1386 * Are we cloning a kernel thread?
1388 * We need to steal a active VM for that..
1390 oldmm = current->mm;
1394 /* initialize the new vmacache entries */
1395 vmacache_flush(tsk);
1397 if (clone_flags & CLONE_VM) {
1404 mm = dup_mm(tsk, current->mm);
1410 tsk->active_mm = mm;
1417 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1419 struct fs_struct *fs = current->fs;
1420 if (clone_flags & CLONE_FS) {
1421 /* tsk->fs is already what we want */
1422 spin_lock(&fs->lock);
1424 spin_unlock(&fs->lock);
1428 spin_unlock(&fs->lock);
1431 tsk->fs = copy_fs_struct(fs);
1437 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1439 struct files_struct *oldf, *newf;
1443 * A background process may not have any files ...
1445 oldf = current->files;
1449 if (clone_flags & CLONE_FILES) {
1450 atomic_inc(&oldf->count);
1454 newf = dup_fd(oldf, &error);
1464 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1467 struct io_context *ioc = current->io_context;
1468 struct io_context *new_ioc;
1473 * Share io context with parent, if CLONE_IO is set
1475 if (clone_flags & CLONE_IO) {
1477 tsk->io_context = ioc;
1478 } else if (ioprio_valid(ioc->ioprio)) {
1479 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1480 if (unlikely(!new_ioc))
1483 new_ioc->ioprio = ioc->ioprio;
1484 put_io_context(new_ioc);
1490 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1492 struct sighand_struct *sig;
1494 if (clone_flags & CLONE_SIGHAND) {
1495 refcount_inc(¤t->sighand->count);
1498 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1499 rcu_assign_pointer(tsk->sighand, sig);
1503 refcount_set(&sig->count, 1);
1504 spin_lock_irq(¤t->sighand->siglock);
1505 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1506 spin_unlock_irq(¤t->sighand->siglock);
1510 void __cleanup_sighand(struct sighand_struct *sighand)
1512 if (refcount_dec_and_test(&sighand->count)) {
1513 signalfd_cleanup(sighand);
1515 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1516 * without an RCU grace period, see __lock_task_sighand().
1518 kmem_cache_free(sighand_cachep, sighand);
1523 * Initialize POSIX timer handling for a thread group.
1525 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1527 struct posix_cputimers *pct = &sig->posix_cputimers;
1528 unsigned long cpu_limit;
1530 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1531 posix_cputimers_group_init(pct, cpu_limit);
1534 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1536 struct signal_struct *sig;
1538 if (clone_flags & CLONE_THREAD)
1541 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1546 sig->nr_threads = 1;
1547 atomic_set(&sig->live, 1);
1548 refcount_set(&sig->sigcnt, 1);
1550 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1551 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1552 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1554 init_waitqueue_head(&sig->wait_chldexit);
1555 sig->curr_target = tsk;
1556 init_sigpending(&sig->shared_pending);
1557 INIT_HLIST_HEAD(&sig->multiprocess);
1558 seqlock_init(&sig->stats_lock);
1559 prev_cputime_init(&sig->prev_cputime);
1561 #ifdef CONFIG_POSIX_TIMERS
1562 INIT_LIST_HEAD(&sig->posix_timers);
1563 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1564 sig->real_timer.function = it_real_fn;
1567 task_lock(current->group_leader);
1568 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1569 task_unlock(current->group_leader);
1571 posix_cpu_timers_init_group(sig);
1573 tty_audit_fork(sig);
1574 sched_autogroup_fork(sig);
1576 sig->oom_score_adj = current->signal->oom_score_adj;
1577 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1579 mutex_init(&sig->cred_guard_mutex);
1584 static void copy_seccomp(struct task_struct *p)
1586 #ifdef CONFIG_SECCOMP
1588 * Must be called with sighand->lock held, which is common to
1589 * all threads in the group. Holding cred_guard_mutex is not
1590 * needed because this new task is not yet running and cannot
1593 assert_spin_locked(¤t->sighand->siglock);
1595 /* Ref-count the new filter user, and assign it. */
1596 get_seccomp_filter(current);
1597 p->seccomp = current->seccomp;
1600 * Explicitly enable no_new_privs here in case it got set
1601 * between the task_struct being duplicated and holding the
1602 * sighand lock. The seccomp state and nnp must be in sync.
1604 if (task_no_new_privs(current))
1605 task_set_no_new_privs(p);
1608 * If the parent gained a seccomp mode after copying thread
1609 * flags and between before we held the sighand lock, we have
1610 * to manually enable the seccomp thread flag here.
1612 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1613 set_tsk_thread_flag(p, TIF_SECCOMP);
1617 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1619 current->clear_child_tid = tidptr;
1621 return task_pid_vnr(current);
1624 static void rt_mutex_init_task(struct task_struct *p)
1626 raw_spin_lock_init(&p->pi_lock);
1627 #ifdef CONFIG_RT_MUTEXES
1628 p->pi_waiters = RB_ROOT_CACHED;
1629 p->pi_top_task = NULL;
1630 p->pi_blocked_on = NULL;
1634 static inline void init_task_pid_links(struct task_struct *task)
1638 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1639 INIT_HLIST_NODE(&task->pid_links[type]);
1644 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1646 if (type == PIDTYPE_PID)
1647 task->thread_pid = pid;
1649 task->signal->pids[type] = pid;
1652 static inline void rcu_copy_process(struct task_struct *p)
1654 #ifdef CONFIG_PREEMPT_RCU
1655 p->rcu_read_lock_nesting = 0;
1656 p->rcu_read_unlock_special.s = 0;
1657 p->rcu_blocked_node = NULL;
1658 INIT_LIST_HEAD(&p->rcu_node_entry);
1659 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1660 #ifdef CONFIG_TASKS_RCU
1661 p->rcu_tasks_holdout = false;
1662 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1663 p->rcu_tasks_idle_cpu = -1;
1664 #endif /* #ifdef CONFIG_TASKS_RCU */
1667 struct pid *pidfd_pid(const struct file *file)
1669 if (file->f_op == &pidfd_fops)
1670 return file->private_data;
1672 return ERR_PTR(-EBADF);
1675 static int pidfd_release(struct inode *inode, struct file *file)
1677 struct pid *pid = file->private_data;
1679 file->private_data = NULL;
1684 #ifdef CONFIG_PROC_FS
1685 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1687 struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1688 struct pid *pid = f->private_data;
1690 seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1696 * Poll support for process exit notification.
1698 static unsigned int pidfd_poll(struct file *file, struct poll_table_struct *pts)
1700 struct task_struct *task;
1701 struct pid *pid = file->private_data;
1704 poll_wait(file, &pid->wait_pidfd, pts);
1707 task = pid_task(pid, PIDTYPE_PID);
1709 * Inform pollers only when the whole thread group exits.
1710 * If the thread group leader exits before all other threads in the
1711 * group, then poll(2) should block, similar to the wait(2) family.
1713 if (!task || (task->exit_state && thread_group_empty(task)))
1714 poll_flags = POLLIN | POLLRDNORM;
1720 const struct file_operations pidfd_fops = {
1721 .release = pidfd_release,
1723 #ifdef CONFIG_PROC_FS
1724 .show_fdinfo = pidfd_show_fdinfo,
1728 static void __delayed_free_task(struct rcu_head *rhp)
1730 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1735 static __always_inline void delayed_free_task(struct task_struct *tsk)
1737 if (IS_ENABLED(CONFIG_MEMCG))
1738 call_rcu(&tsk->rcu, __delayed_free_task);
1744 * This creates a new process as a copy of the old one,
1745 * but does not actually start it yet.
1747 * It copies the registers, and all the appropriate
1748 * parts of the process environment (as per the clone
1749 * flags). The actual kick-off is left to the caller.
1751 static __latent_entropy struct task_struct *copy_process(
1755 struct kernel_clone_args *args)
1757 int pidfd = -1, retval;
1758 struct task_struct *p;
1759 struct multiprocess_signals delayed;
1760 struct file *pidfile = NULL;
1761 u64 clone_flags = args->flags;
1764 * Don't allow sharing the root directory with processes in a different
1767 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1768 return ERR_PTR(-EINVAL);
1770 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1771 return ERR_PTR(-EINVAL);
1774 * Thread groups must share signals as well, and detached threads
1775 * can only be started up within the thread group.
1777 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1778 return ERR_PTR(-EINVAL);
1781 * Shared signal handlers imply shared VM. By way of the above,
1782 * thread groups also imply shared VM. Blocking this case allows
1783 * for various simplifications in other code.
1785 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1786 return ERR_PTR(-EINVAL);
1789 * Siblings of global init remain as zombies on exit since they are
1790 * not reaped by their parent (swapper). To solve this and to avoid
1791 * multi-rooted process trees, prevent global and container-inits
1792 * from creating siblings.
1794 if ((clone_flags & CLONE_PARENT) &&
1795 current->signal->flags & SIGNAL_UNKILLABLE)
1796 return ERR_PTR(-EINVAL);
1799 * If the new process will be in a different pid or user namespace
1800 * do not allow it to share a thread group with the forking task.
1802 if (clone_flags & CLONE_THREAD) {
1803 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1804 (task_active_pid_ns(current) !=
1805 current->nsproxy->pid_ns_for_children))
1806 return ERR_PTR(-EINVAL);
1809 if (clone_flags & CLONE_PIDFD) {
1811 * - CLONE_DETACHED is blocked so that we can potentially
1812 * reuse it later for CLONE_PIDFD.
1813 * - CLONE_THREAD is blocked until someone really needs it.
1815 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1816 return ERR_PTR(-EINVAL);
1820 * Force any signals received before this point to be delivered
1821 * before the fork happens. Collect up signals sent to multiple
1822 * processes that happen during the fork and delay them so that
1823 * they appear to happen after the fork.
1825 sigemptyset(&delayed.signal);
1826 INIT_HLIST_NODE(&delayed.node);
1828 spin_lock_irq(¤t->sighand->siglock);
1829 if (!(clone_flags & CLONE_THREAD))
1830 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1831 recalc_sigpending();
1832 spin_unlock_irq(¤t->sighand->siglock);
1833 retval = -ERESTARTNOINTR;
1834 if (signal_pending(current))
1838 p = dup_task_struct(current, node);
1843 * This _must_ happen before we call free_task(), i.e. before we jump
1844 * to any of the bad_fork_* labels. This is to avoid freeing
1845 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1846 * kernel threads (PF_KTHREAD).
1848 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1850 * Clear TID on mm_release()?
1852 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1854 ftrace_graph_init_task(p);
1856 rt_mutex_init_task(p);
1858 #ifdef CONFIG_PROVE_LOCKING
1859 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1860 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1863 if (atomic_read(&p->real_cred->user->processes) >=
1864 task_rlimit(p, RLIMIT_NPROC)) {
1865 if (p->real_cred->user != INIT_USER &&
1866 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1869 current->flags &= ~PF_NPROC_EXCEEDED;
1871 retval = copy_creds(p, clone_flags);
1876 * If multiple threads are within copy_process(), then this check
1877 * triggers too late. This doesn't hurt, the check is only there
1878 * to stop root fork bombs.
1881 if (nr_threads >= max_threads)
1882 goto bad_fork_cleanup_count;
1884 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1885 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1886 p->flags |= PF_FORKNOEXEC;
1887 INIT_LIST_HEAD(&p->children);
1888 INIT_LIST_HEAD(&p->sibling);
1889 rcu_copy_process(p);
1890 p->vfork_done = NULL;
1891 spin_lock_init(&p->alloc_lock);
1893 init_sigpending(&p->pending);
1895 p->utime = p->stime = p->gtime = 0;
1896 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1897 p->utimescaled = p->stimescaled = 0;
1899 prev_cputime_init(&p->prev_cputime);
1901 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1902 seqcount_init(&p->vtime.seqcount);
1903 p->vtime.starttime = 0;
1904 p->vtime.state = VTIME_INACTIVE;
1907 #if defined(SPLIT_RSS_COUNTING)
1908 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1911 p->default_timer_slack_ns = current->timer_slack_ns;
1917 task_io_accounting_init(&p->ioac);
1918 acct_clear_integrals(p);
1920 posix_cputimers_init(&p->posix_cputimers);
1922 p->io_context = NULL;
1923 audit_set_context(p, NULL);
1926 p->mempolicy = mpol_dup(p->mempolicy);
1927 if (IS_ERR(p->mempolicy)) {
1928 retval = PTR_ERR(p->mempolicy);
1929 p->mempolicy = NULL;
1930 goto bad_fork_cleanup_threadgroup_lock;
1933 #ifdef CONFIG_CPUSETS
1934 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1935 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1936 seqcount_init(&p->mems_allowed_seq);
1938 #ifdef CONFIG_TRACE_IRQFLAGS
1940 p->hardirqs_enabled = 0;
1941 p->hardirq_enable_ip = 0;
1942 p->hardirq_enable_event = 0;
1943 p->hardirq_disable_ip = _THIS_IP_;
1944 p->hardirq_disable_event = 0;
1945 p->softirqs_enabled = 1;
1946 p->softirq_enable_ip = _THIS_IP_;
1947 p->softirq_enable_event = 0;
1948 p->softirq_disable_ip = 0;
1949 p->softirq_disable_event = 0;
1950 p->hardirq_context = 0;
1951 p->softirq_context = 0;
1954 p->pagefault_disabled = 0;
1956 #ifdef CONFIG_LOCKDEP
1957 lockdep_init_task(p);
1960 #ifdef CONFIG_DEBUG_MUTEXES
1961 p->blocked_on = NULL; /* not blocked yet */
1963 #ifdef CONFIG_BCACHE
1964 p->sequential_io = 0;
1965 p->sequential_io_avg = 0;
1968 /* Perform scheduler related setup. Assign this task to a CPU. */
1969 retval = sched_fork(clone_flags, p);
1971 goto bad_fork_cleanup_policy;
1973 retval = perf_event_init_task(p);
1975 goto bad_fork_cleanup_policy;
1976 retval = audit_alloc(p);
1978 goto bad_fork_cleanup_perf;
1979 /* copy all the process information */
1981 retval = security_task_alloc(p, clone_flags);
1983 goto bad_fork_cleanup_audit;
1984 retval = copy_semundo(clone_flags, p);
1986 goto bad_fork_cleanup_security;
1987 retval = copy_files(clone_flags, p);
1989 goto bad_fork_cleanup_semundo;
1990 retval = copy_fs(clone_flags, p);
1992 goto bad_fork_cleanup_files;
1993 retval = copy_sighand(clone_flags, p);
1995 goto bad_fork_cleanup_fs;
1996 retval = copy_signal(clone_flags, p);
1998 goto bad_fork_cleanup_sighand;
1999 retval = copy_mm(clone_flags, p);
2001 goto bad_fork_cleanup_signal;
2002 retval = copy_namespaces(clone_flags, p);
2004 goto bad_fork_cleanup_mm;
2005 retval = copy_io(clone_flags, p);
2007 goto bad_fork_cleanup_namespaces;
2008 retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2011 goto bad_fork_cleanup_io;
2013 stackleak_task_init(p);
2015 if (pid != &init_struct_pid) {
2016 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
2018 retval = PTR_ERR(pid);
2019 goto bad_fork_cleanup_thread;
2024 * This has to happen after we've potentially unshared the file
2025 * descriptor table (so that the pidfd doesn't leak into the child
2026 * if the fd table isn't shared).
2028 if (clone_flags & CLONE_PIDFD) {
2029 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2031 goto bad_fork_free_pid;
2035 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2036 O_RDWR | O_CLOEXEC);
2037 if (IS_ERR(pidfile)) {
2038 put_unused_fd(pidfd);
2039 retval = PTR_ERR(pidfile);
2040 goto bad_fork_free_pid;
2042 get_pid(pid); /* held by pidfile now */
2044 retval = put_user(pidfd, args->pidfd);
2046 goto bad_fork_put_pidfd;
2053 p->robust_list = NULL;
2054 #ifdef CONFIG_COMPAT
2055 p->compat_robust_list = NULL;
2057 INIT_LIST_HEAD(&p->pi_state_list);
2058 p->pi_state_cache = NULL;
2061 * sigaltstack should be cleared when sharing the same VM
2063 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2067 * Syscall tracing and stepping should be turned off in the
2068 * child regardless of CLONE_PTRACE.
2070 user_disable_single_step(p);
2071 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2072 #ifdef TIF_SYSCALL_EMU
2073 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2075 clear_tsk_latency_tracing(p);
2077 /* ok, now we should be set up.. */
2078 p->pid = pid_nr(pid);
2079 if (clone_flags & CLONE_THREAD) {
2080 p->exit_signal = -1;
2081 p->group_leader = current->group_leader;
2082 p->tgid = current->tgid;
2084 if (clone_flags & CLONE_PARENT)
2085 p->exit_signal = current->group_leader->exit_signal;
2087 p->exit_signal = args->exit_signal;
2088 p->group_leader = p;
2093 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2094 p->dirty_paused_when = 0;
2096 p->pdeath_signal = 0;
2097 INIT_LIST_HEAD(&p->thread_group);
2098 p->task_works = NULL;
2100 cgroup_threadgroup_change_begin(current);
2102 * Ensure that the cgroup subsystem policies allow the new process to be
2103 * forked. It should be noted the the new process's css_set can be changed
2104 * between here and cgroup_post_fork() if an organisation operation is in
2107 retval = cgroup_can_fork(p);
2109 goto bad_fork_cgroup_threadgroup_change_end;
2112 * From this point on we must avoid any synchronous user-space
2113 * communication until we take the tasklist-lock. In particular, we do
2114 * not want user-space to be able to predict the process start-time by
2115 * stalling fork(2) after we recorded the start_time but before it is
2116 * visible to the system.
2119 p->start_time = ktime_get_ns();
2120 p->real_start_time = ktime_get_boottime_ns();
2123 * Make it visible to the rest of the system, but dont wake it up yet.
2124 * Need tasklist lock for parent etc handling!
2126 write_lock_irq(&tasklist_lock);
2128 /* CLONE_PARENT re-uses the old parent */
2129 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2130 p->real_parent = current->real_parent;
2131 p->parent_exec_id = current->parent_exec_id;
2133 p->real_parent = current;
2134 p->parent_exec_id = current->self_exec_id;
2137 klp_copy_process(p);
2139 spin_lock(¤t->sighand->siglock);
2142 * Copy seccomp details explicitly here, in case they were changed
2143 * before holding sighand lock.
2147 rseq_fork(p, clone_flags);
2149 /* Don't start children in a dying pid namespace */
2150 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2152 goto bad_fork_cancel_cgroup;
2155 /* Let kill terminate clone/fork in the middle */
2156 if (fatal_signal_pending(current)) {
2158 goto bad_fork_cancel_cgroup;
2161 /* past the last point of failure */
2163 fd_install(pidfd, pidfile);
2165 init_task_pid_links(p);
2166 if (likely(p->pid)) {
2167 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2169 init_task_pid(p, PIDTYPE_PID, pid);
2170 if (thread_group_leader(p)) {
2171 init_task_pid(p, PIDTYPE_TGID, pid);
2172 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2173 init_task_pid(p, PIDTYPE_SID, task_session(current));
2175 if (is_child_reaper(pid)) {
2176 ns_of_pid(pid)->child_reaper = p;
2177 p->signal->flags |= SIGNAL_UNKILLABLE;
2179 p->signal->shared_pending.signal = delayed.signal;
2180 p->signal->tty = tty_kref_get(current->signal->tty);
2182 * Inherit has_child_subreaper flag under the same
2183 * tasklist_lock with adding child to the process tree
2184 * for propagate_has_child_subreaper optimization.
2186 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2187 p->real_parent->signal->is_child_subreaper;
2188 list_add_tail(&p->sibling, &p->real_parent->children);
2189 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2190 attach_pid(p, PIDTYPE_TGID);
2191 attach_pid(p, PIDTYPE_PGID);
2192 attach_pid(p, PIDTYPE_SID);
2193 __this_cpu_inc(process_counts);
2195 current->signal->nr_threads++;
2196 atomic_inc(¤t->signal->live);
2197 refcount_inc(¤t->signal->sigcnt);
2198 task_join_group_stop(p);
2199 list_add_tail_rcu(&p->thread_group,
2200 &p->group_leader->thread_group);
2201 list_add_tail_rcu(&p->thread_node,
2202 &p->signal->thread_head);
2204 attach_pid(p, PIDTYPE_PID);
2208 hlist_del_init(&delayed.node);
2209 spin_unlock(¤t->sighand->siglock);
2210 syscall_tracepoint_update(p);
2211 write_unlock_irq(&tasklist_lock);
2213 proc_fork_connector(p);
2214 cgroup_post_fork(p);
2215 cgroup_threadgroup_change_end(current);
2218 trace_task_newtask(p, clone_flags);
2219 uprobe_copy_process(p, clone_flags);
2223 bad_fork_cancel_cgroup:
2224 spin_unlock(¤t->sighand->siglock);
2225 write_unlock_irq(&tasklist_lock);
2226 cgroup_cancel_fork(p);
2227 bad_fork_cgroup_threadgroup_change_end:
2228 cgroup_threadgroup_change_end(current);
2230 if (clone_flags & CLONE_PIDFD) {
2232 put_unused_fd(pidfd);
2235 if (pid != &init_struct_pid)
2237 bad_fork_cleanup_thread:
2239 bad_fork_cleanup_io:
2242 bad_fork_cleanup_namespaces:
2243 exit_task_namespaces(p);
2244 bad_fork_cleanup_mm:
2246 mm_clear_owner(p->mm, p);
2249 bad_fork_cleanup_signal:
2250 if (!(clone_flags & CLONE_THREAD))
2251 free_signal_struct(p->signal);
2252 bad_fork_cleanup_sighand:
2253 __cleanup_sighand(p->sighand);
2254 bad_fork_cleanup_fs:
2255 exit_fs(p); /* blocking */
2256 bad_fork_cleanup_files:
2257 exit_files(p); /* blocking */
2258 bad_fork_cleanup_semundo:
2260 bad_fork_cleanup_security:
2261 security_task_free(p);
2262 bad_fork_cleanup_audit:
2264 bad_fork_cleanup_perf:
2265 perf_event_free_task(p);
2266 bad_fork_cleanup_policy:
2267 lockdep_free_task(p);
2269 mpol_put(p->mempolicy);
2270 bad_fork_cleanup_threadgroup_lock:
2272 delayacct_tsk_free(p);
2273 bad_fork_cleanup_count:
2274 atomic_dec(&p->cred->user->processes);
2277 p->state = TASK_DEAD;
2279 delayed_free_task(p);
2281 spin_lock_irq(¤t->sighand->siglock);
2282 hlist_del_init(&delayed.node);
2283 spin_unlock_irq(¤t->sighand->siglock);
2284 return ERR_PTR(retval);
2287 static inline void init_idle_pids(struct task_struct *idle)
2291 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2292 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2293 init_task_pid(idle, type, &init_struct_pid);
2297 struct task_struct *fork_idle(int cpu)
2299 struct task_struct *task;
2300 struct kernel_clone_args args = {
2304 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2305 if (!IS_ERR(task)) {
2306 init_idle_pids(task);
2307 init_idle(task, cpu);
2313 struct mm_struct *copy_init_mm(void)
2315 return dup_mm(NULL, &init_mm);
2319 * Ok, this is the main fork-routine.
2321 * It copies the process, and if successful kick-starts
2322 * it and waits for it to finish using the VM if required.
2324 * args->exit_signal is expected to be checked for sanity by the caller.
2326 long _do_fork(struct kernel_clone_args *args)
2328 u64 clone_flags = args->flags;
2329 struct completion vfork;
2331 struct task_struct *p;
2336 * Determine whether and which event to report to ptracer. When
2337 * called from kernel_thread or CLONE_UNTRACED is explicitly
2338 * requested, no event is reported; otherwise, report if the event
2339 * for the type of forking is enabled.
2341 if (!(clone_flags & CLONE_UNTRACED)) {
2342 if (clone_flags & CLONE_VFORK)
2343 trace = PTRACE_EVENT_VFORK;
2344 else if (args->exit_signal != SIGCHLD)
2345 trace = PTRACE_EVENT_CLONE;
2347 trace = PTRACE_EVENT_FORK;
2349 if (likely(!ptrace_event_enabled(current, trace)))
2353 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2354 add_latent_entropy();
2360 * Do this prior waking up the new thread - the thread pointer
2361 * might get invalid after that point, if the thread exits quickly.
2363 trace_sched_process_fork(current, p);
2365 pid = get_task_pid(p, PIDTYPE_PID);
2368 if (clone_flags & CLONE_PARENT_SETTID)
2369 put_user(nr, args->parent_tid);
2371 if (clone_flags & CLONE_VFORK) {
2372 p->vfork_done = &vfork;
2373 init_completion(&vfork);
2377 wake_up_new_task(p);
2379 /* forking complete and child started to run, tell ptracer */
2380 if (unlikely(trace))
2381 ptrace_event_pid(trace, pid);
2383 if (clone_flags & CLONE_VFORK) {
2384 if (!wait_for_vfork_done(p, &vfork))
2385 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2392 bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
2394 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2395 if ((kargs->flags & CLONE_PIDFD) &&
2396 (kargs->flags & CLONE_PARENT_SETTID))
2402 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2403 /* For compatibility with architectures that call do_fork directly rather than
2404 * using the syscall entry points below. */
2405 long do_fork(unsigned long clone_flags,
2406 unsigned long stack_start,
2407 unsigned long stack_size,
2408 int __user *parent_tidptr,
2409 int __user *child_tidptr)
2411 struct kernel_clone_args args = {
2412 .flags = (clone_flags & ~CSIGNAL),
2413 .pidfd = parent_tidptr,
2414 .child_tid = child_tidptr,
2415 .parent_tid = parent_tidptr,
2416 .exit_signal = (clone_flags & CSIGNAL),
2417 .stack = stack_start,
2418 .stack_size = stack_size,
2421 if (!legacy_clone_args_valid(&args))
2424 return _do_fork(&args);
2429 * Create a kernel thread.
2431 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2433 struct kernel_clone_args args = {
2434 .flags = ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2435 .exit_signal = (flags & CSIGNAL),
2436 .stack = (unsigned long)fn,
2437 .stack_size = (unsigned long)arg,
2440 return _do_fork(&args);
2443 #ifdef __ARCH_WANT_SYS_FORK
2444 SYSCALL_DEFINE0(fork)
2447 struct kernel_clone_args args = {
2448 .exit_signal = SIGCHLD,
2451 return _do_fork(&args);
2453 /* can not support in nommu mode */
2459 #ifdef __ARCH_WANT_SYS_VFORK
2460 SYSCALL_DEFINE0(vfork)
2462 struct kernel_clone_args args = {
2463 .flags = CLONE_VFORK | CLONE_VM,
2464 .exit_signal = SIGCHLD,
2467 return _do_fork(&args);
2471 #ifdef __ARCH_WANT_SYS_CLONE
2472 #ifdef CONFIG_CLONE_BACKWARDS
2473 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2474 int __user *, parent_tidptr,
2476 int __user *, child_tidptr)
2477 #elif defined(CONFIG_CLONE_BACKWARDS2)
2478 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2479 int __user *, parent_tidptr,
2480 int __user *, child_tidptr,
2482 #elif defined(CONFIG_CLONE_BACKWARDS3)
2483 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2485 int __user *, parent_tidptr,
2486 int __user *, child_tidptr,
2489 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2490 int __user *, parent_tidptr,
2491 int __user *, child_tidptr,
2495 struct kernel_clone_args args = {
2496 .flags = (clone_flags & ~CSIGNAL),
2497 .pidfd = parent_tidptr,
2498 .child_tid = child_tidptr,
2499 .parent_tid = parent_tidptr,
2500 .exit_signal = (clone_flags & CSIGNAL),
2505 if (!legacy_clone_args_valid(&args))
2508 return _do_fork(&args);
2512 #ifdef __ARCH_WANT_SYS_CLONE3
2513 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2514 struct clone_args __user *uargs,
2517 struct clone_args args;
2519 if (unlikely(size > PAGE_SIZE))
2522 if (unlikely(size < sizeof(struct clone_args)))
2525 if (unlikely(!access_ok(uargs, size)))
2528 if (size > sizeof(struct clone_args)) {
2529 unsigned char __user *addr;
2530 unsigned char __user *end;
2533 addr = (void __user *)uargs + sizeof(struct clone_args);
2534 end = (void __user *)uargs + size;
2536 for (; addr < end; addr++) {
2537 if (get_user(val, addr))
2543 size = sizeof(struct clone_args);
2546 if (copy_from_user(&args, uargs, size))
2550 * Verify that higher 32bits of exit_signal are unset and that
2551 * it is a valid signal
2553 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2554 !valid_signal(args.exit_signal)))
2557 *kargs = (struct kernel_clone_args){
2558 .flags = args.flags,
2559 .pidfd = u64_to_user_ptr(args.pidfd),
2560 .child_tid = u64_to_user_ptr(args.child_tid),
2561 .parent_tid = u64_to_user_ptr(args.parent_tid),
2562 .exit_signal = args.exit_signal,
2563 .stack = args.stack,
2564 .stack_size = args.stack_size,
2571 static bool clone3_args_valid(const struct kernel_clone_args *kargs)
2574 * All lower bits of the flag word are taken.
2575 * Verify that no other unknown flags are passed along.
2577 if (kargs->flags & ~CLONE_LEGACY_FLAGS)
2581 * - make the CLONE_DETACHED bit reuseable for clone3
2582 * - make the CSIGNAL bits reuseable for clone3
2584 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2587 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2594 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2598 struct kernel_clone_args kargs;
2600 err = copy_clone_args_from_user(&kargs, uargs, size);
2604 if (!clone3_args_valid(&kargs))
2607 return _do_fork(&kargs);
2611 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2613 struct task_struct *leader, *parent, *child;
2616 read_lock(&tasklist_lock);
2617 leader = top = top->group_leader;
2619 for_each_thread(leader, parent) {
2620 list_for_each_entry(child, &parent->children, sibling) {
2621 res = visitor(child, data);
2633 if (leader != top) {
2635 parent = child->real_parent;
2636 leader = parent->group_leader;
2640 read_unlock(&tasklist_lock);
2643 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2644 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2647 static void sighand_ctor(void *data)
2649 struct sighand_struct *sighand = data;
2651 spin_lock_init(&sighand->siglock);
2652 init_waitqueue_head(&sighand->signalfd_wqh);
2655 void __init proc_caches_init(void)
2657 unsigned int mm_size;
2659 sighand_cachep = kmem_cache_create("sighand_cache",
2660 sizeof(struct sighand_struct), 0,
2661 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2662 SLAB_ACCOUNT, sighand_ctor);
2663 signal_cachep = kmem_cache_create("signal_cache",
2664 sizeof(struct signal_struct), 0,
2665 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2667 files_cachep = kmem_cache_create("files_cache",
2668 sizeof(struct files_struct), 0,
2669 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2671 fs_cachep = kmem_cache_create("fs_cache",
2672 sizeof(struct fs_struct), 0,
2673 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2677 * The mm_cpumask is located at the end of mm_struct, and is
2678 * dynamically sized based on the maximum CPU number this system
2679 * can have, taking hotplug into account (nr_cpu_ids).
2681 mm_size = sizeof(struct mm_struct) + cpumask_size();
2683 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2684 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2685 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2686 offsetof(struct mm_struct, saved_auxv),
2687 sizeof_field(struct mm_struct, saved_auxv),
2689 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2691 nsproxy_cache_init();
2695 * Check constraints on flags passed to the unshare system call.
2697 static int check_unshare_flags(unsigned long unshare_flags)
2699 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2700 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2701 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2702 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2705 * Not implemented, but pretend it works if there is nothing
2706 * to unshare. Note that unsharing the address space or the
2707 * signal handlers also need to unshare the signal queues (aka
2710 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2711 if (!thread_group_empty(current))
2714 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2715 if (refcount_read(¤t->sighand->count) > 1)
2718 if (unshare_flags & CLONE_VM) {
2719 if (!current_is_single_threaded())
2727 * Unshare the filesystem structure if it is being shared
2729 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2731 struct fs_struct *fs = current->fs;
2733 if (!(unshare_flags & CLONE_FS) || !fs)
2736 /* don't need lock here; in the worst case we'll do useless copy */
2740 *new_fsp = copy_fs_struct(fs);
2748 * Unshare file descriptor table if it is being shared
2750 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2752 struct files_struct *fd = current->files;
2755 if ((unshare_flags & CLONE_FILES) &&
2756 (fd && atomic_read(&fd->count) > 1)) {
2757 *new_fdp = dup_fd(fd, &error);
2766 * unshare allows a process to 'unshare' part of the process
2767 * context which was originally shared using clone. copy_*
2768 * functions used by do_fork() cannot be used here directly
2769 * because they modify an inactive task_struct that is being
2770 * constructed. Here we are modifying the current, active,
2773 int ksys_unshare(unsigned long unshare_flags)
2775 struct fs_struct *fs, *new_fs = NULL;
2776 struct files_struct *fd, *new_fd = NULL;
2777 struct cred *new_cred = NULL;
2778 struct nsproxy *new_nsproxy = NULL;
2783 * If unsharing a user namespace must also unshare the thread group
2784 * and unshare the filesystem root and working directories.
2786 if (unshare_flags & CLONE_NEWUSER)
2787 unshare_flags |= CLONE_THREAD | CLONE_FS;
2789 * If unsharing vm, must also unshare signal handlers.
2791 if (unshare_flags & CLONE_VM)
2792 unshare_flags |= CLONE_SIGHAND;
2794 * If unsharing a signal handlers, must also unshare the signal queues.
2796 if (unshare_flags & CLONE_SIGHAND)
2797 unshare_flags |= CLONE_THREAD;
2799 * If unsharing namespace, must also unshare filesystem information.
2801 if (unshare_flags & CLONE_NEWNS)
2802 unshare_flags |= CLONE_FS;
2804 err = check_unshare_flags(unshare_flags);
2806 goto bad_unshare_out;
2808 * CLONE_NEWIPC must also detach from the undolist: after switching
2809 * to a new ipc namespace, the semaphore arrays from the old
2810 * namespace are unreachable.
2812 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2814 err = unshare_fs(unshare_flags, &new_fs);
2816 goto bad_unshare_out;
2817 err = unshare_fd(unshare_flags, &new_fd);
2819 goto bad_unshare_cleanup_fs;
2820 err = unshare_userns(unshare_flags, &new_cred);
2822 goto bad_unshare_cleanup_fd;
2823 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2826 goto bad_unshare_cleanup_cred;
2828 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2831 * CLONE_SYSVSEM is equivalent to sys_exit().
2835 if (unshare_flags & CLONE_NEWIPC) {
2836 /* Orphan segments in old ns (see sem above). */
2838 shm_init_task(current);
2842 switch_task_namespaces(current, new_nsproxy);
2848 spin_lock(&fs->lock);
2849 current->fs = new_fs;
2854 spin_unlock(&fs->lock);
2858 fd = current->files;
2859 current->files = new_fd;
2863 task_unlock(current);
2866 /* Install the new user namespace */
2867 commit_creds(new_cred);
2872 perf_event_namespaces(current);
2874 bad_unshare_cleanup_cred:
2877 bad_unshare_cleanup_fd:
2879 put_files_struct(new_fd);
2881 bad_unshare_cleanup_fs:
2883 free_fs_struct(new_fs);
2889 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2891 return ksys_unshare(unshare_flags);
2895 * Helper to unshare the files of the current task.
2896 * We don't want to expose copy_files internals to
2897 * the exec layer of the kernel.
2900 int unshare_files(struct files_struct **displaced)
2902 struct task_struct *task = current;
2903 struct files_struct *copy = NULL;
2906 error = unshare_fd(CLONE_FILES, ©);
2907 if (error || !copy) {
2911 *displaced = task->files;
2918 int sysctl_max_threads(struct ctl_table *table, int write,
2919 void __user *buffer, size_t *lenp, loff_t *ppos)
2923 int threads = max_threads;
2924 int min = MIN_THREADS;
2925 int max = MAX_THREADS;
2932 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2936 set_max_threads(threads);