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 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
773 /* Fetch thread_struct whitelist for the architecture. */
774 arch_thread_struct_whitelist(offset, size);
777 * Handle zero-sized whitelist or empty thread_struct, otherwise
778 * adjust offset to position of thread_struct in task_struct.
780 if (unlikely(*size == 0))
783 *offset += offsetof(struct task_struct, thread);
786 void __init fork_init(void)
789 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
790 #ifndef ARCH_MIN_TASKALIGN
791 #define ARCH_MIN_TASKALIGN 0
793 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
794 unsigned long useroffset, usersize;
796 /* create a slab on which task_structs can be allocated */
797 task_struct_whitelist(&useroffset, &usersize);
798 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
799 arch_task_struct_size, align,
800 SLAB_PANIC|SLAB_ACCOUNT,
801 useroffset, usersize, NULL);
804 /* do the arch specific task caches init */
805 arch_task_cache_init();
807 set_max_threads(MAX_THREADS);
809 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
810 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
811 init_task.signal->rlim[RLIMIT_SIGPENDING] =
812 init_task.signal->rlim[RLIMIT_NPROC];
814 for (i = 0; i < UCOUNT_COUNTS; i++) {
815 init_user_ns.ucount_max[i] = max_threads/2;
818 #ifdef CONFIG_VMAP_STACK
819 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
820 NULL, free_vm_stack_cache);
823 lockdep_init_task(&init_task);
827 int __weak arch_dup_task_struct(struct task_struct *dst,
828 struct task_struct *src)
834 void set_task_stack_end_magic(struct task_struct *tsk)
836 unsigned long *stackend;
838 stackend = end_of_stack(tsk);
839 *stackend = STACK_END_MAGIC; /* for overflow detection */
842 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
844 struct task_struct *tsk;
845 unsigned long *stack;
846 struct vm_struct *stack_vm_area __maybe_unused;
849 if (node == NUMA_NO_NODE)
850 node = tsk_fork_get_node(orig);
851 tsk = alloc_task_struct_node(node);
855 stack = alloc_thread_stack_node(tsk, node);
859 if (memcg_charge_kernel_stack(tsk))
862 stack_vm_area = task_stack_vm_area(tsk);
864 err = arch_dup_task_struct(tsk, orig);
867 * arch_dup_task_struct() clobbers the stack-related fields. Make
868 * sure they're properly initialized before using any stack-related
872 #ifdef CONFIG_VMAP_STACK
873 tsk->stack_vm_area = stack_vm_area;
875 #ifdef CONFIG_THREAD_INFO_IN_TASK
876 refcount_set(&tsk->stack_refcount, 1);
882 #ifdef CONFIG_SECCOMP
884 * We must handle setting up seccomp filters once we're under
885 * the sighand lock in case orig has changed between now and
886 * then. Until then, filter must be NULL to avoid messing up
887 * the usage counts on the error path calling free_task.
889 tsk->seccomp.filter = NULL;
892 setup_thread_stack(tsk, orig);
893 clear_user_return_notifier(tsk);
894 clear_tsk_need_resched(tsk);
895 set_task_stack_end_magic(tsk);
897 #ifdef CONFIG_STACKPROTECTOR
898 tsk->stack_canary = get_random_canary();
900 if (orig->cpus_ptr == &orig->cpus_mask)
901 tsk->cpus_ptr = &tsk->cpus_mask;
904 * One for us, one for whoever does the "release_task()" (usually
907 refcount_set(&tsk->usage, 2);
908 #ifdef CONFIG_BLK_DEV_IO_TRACE
911 tsk->splice_pipe = NULL;
912 tsk->task_frag.page = NULL;
913 tsk->wake_q.next = NULL;
915 account_kernel_stack(tsk, 1);
919 #ifdef CONFIG_FAULT_INJECTION
923 #ifdef CONFIG_BLK_CGROUP
924 tsk->throttle_queue = NULL;
925 tsk->use_memdelay = 0;
929 tsk->active_memcg = NULL;
934 free_thread_stack(tsk);
936 free_task_struct(tsk);
940 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
942 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
944 static int __init coredump_filter_setup(char *s)
946 default_dump_filter =
947 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
948 MMF_DUMP_FILTER_MASK;
952 __setup("coredump_filter=", coredump_filter_setup);
954 #include <linux/init_task.h>
956 static void mm_init_aio(struct mm_struct *mm)
959 spin_lock_init(&mm->ioctx_lock);
960 mm->ioctx_table = NULL;
964 static __always_inline void mm_clear_owner(struct mm_struct *mm,
965 struct task_struct *p)
969 WRITE_ONCE(mm->owner, NULL);
973 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
980 static void mm_init_uprobes_state(struct mm_struct *mm)
982 #ifdef CONFIG_UPROBES
983 mm->uprobes_state.xol_area = NULL;
987 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
988 struct user_namespace *user_ns)
992 mm->vmacache_seqnum = 0;
993 atomic_set(&mm->mm_users, 1);
994 atomic_set(&mm->mm_count, 1);
995 init_rwsem(&mm->mmap_sem);
996 INIT_LIST_HEAD(&mm->mmlist);
997 mm->core_state = NULL;
998 mm_pgtables_bytes_init(mm);
1001 atomic64_set(&mm->pinned_vm, 0);
1002 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1003 spin_lock_init(&mm->page_table_lock);
1004 spin_lock_init(&mm->arg_lock);
1005 mm_init_cpumask(mm);
1007 mm_init_owner(mm, p);
1008 RCU_INIT_POINTER(mm->exe_file, NULL);
1009 mmu_notifier_mm_init(mm);
1011 init_tlb_flush_pending(mm);
1012 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1013 mm->pmd_huge_pte = NULL;
1015 mm_init_uprobes_state(mm);
1018 mm->flags = current->mm->flags & MMF_INIT_MASK;
1019 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1021 mm->flags = default_dump_filter;
1025 if (mm_alloc_pgd(mm))
1028 if (init_new_context(p, mm))
1029 goto fail_nocontext;
1031 mm->user_ns = get_user_ns(user_ns);
1042 * Allocate and initialize an mm_struct.
1044 struct mm_struct *mm_alloc(void)
1046 struct mm_struct *mm;
1052 memset(mm, 0, sizeof(*mm));
1053 return mm_init(mm, current, current_user_ns());
1056 static inline void __mmput(struct mm_struct *mm)
1058 VM_BUG_ON(atomic_read(&mm->mm_users));
1060 uprobe_clear_state(mm);
1063 khugepaged_exit(mm); /* must run before exit_mmap */
1065 mm_put_huge_zero_page(mm);
1066 set_mm_exe_file(mm, NULL);
1067 if (!list_empty(&mm->mmlist)) {
1068 spin_lock(&mmlist_lock);
1069 list_del(&mm->mmlist);
1070 spin_unlock(&mmlist_lock);
1073 module_put(mm->binfmt->module);
1078 * Decrement the use count and release all resources for an mm.
1080 void mmput(struct mm_struct *mm)
1084 if (atomic_dec_and_test(&mm->mm_users))
1087 EXPORT_SYMBOL_GPL(mmput);
1090 static void mmput_async_fn(struct work_struct *work)
1092 struct mm_struct *mm = container_of(work, struct mm_struct,
1098 void mmput_async(struct mm_struct *mm)
1100 if (atomic_dec_and_test(&mm->mm_users)) {
1101 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1102 schedule_work(&mm->async_put_work);
1108 * set_mm_exe_file - change a reference to the mm's executable file
1110 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1112 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1113 * invocations: in mmput() nobody alive left, in execve task is single
1114 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1115 * mm->exe_file, but does so without using set_mm_exe_file() in order
1116 * to do avoid the need for any locks.
1118 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1120 struct file *old_exe_file;
1123 * It is safe to dereference the exe_file without RCU as
1124 * this function is only called if nobody else can access
1125 * this mm -- see comment above for justification.
1127 old_exe_file = rcu_dereference_raw(mm->exe_file);
1130 get_file(new_exe_file);
1131 rcu_assign_pointer(mm->exe_file, new_exe_file);
1137 * get_mm_exe_file - acquire a reference to the mm's executable file
1139 * Returns %NULL if mm has no associated executable file.
1140 * User must release file via fput().
1142 struct file *get_mm_exe_file(struct mm_struct *mm)
1144 struct file *exe_file;
1147 exe_file = rcu_dereference(mm->exe_file);
1148 if (exe_file && !get_file_rcu(exe_file))
1153 EXPORT_SYMBOL(get_mm_exe_file);
1156 * get_task_exe_file - acquire a reference to the task's executable file
1158 * Returns %NULL if task's mm (if any) has no associated executable file or
1159 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1160 * User must release file via fput().
1162 struct file *get_task_exe_file(struct task_struct *task)
1164 struct file *exe_file = NULL;
1165 struct mm_struct *mm;
1170 if (!(task->flags & PF_KTHREAD))
1171 exe_file = get_mm_exe_file(mm);
1176 EXPORT_SYMBOL(get_task_exe_file);
1179 * get_task_mm - acquire a reference to the task's mm
1181 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1182 * this kernel workthread has transiently adopted a user mm with use_mm,
1183 * to do its AIO) is not set and if so returns a reference to it, after
1184 * bumping up the use count. User must release the mm via mmput()
1185 * after use. Typically used by /proc and ptrace.
1187 struct mm_struct *get_task_mm(struct task_struct *task)
1189 struct mm_struct *mm;
1194 if (task->flags & PF_KTHREAD)
1202 EXPORT_SYMBOL_GPL(get_task_mm);
1204 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1206 struct mm_struct *mm;
1209 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1211 return ERR_PTR(err);
1213 mm = get_task_mm(task);
1214 if (mm && mm != current->mm &&
1215 !ptrace_may_access(task, mode)) {
1217 mm = ERR_PTR(-EACCES);
1219 mutex_unlock(&task->signal->cred_guard_mutex);
1224 static void complete_vfork_done(struct task_struct *tsk)
1226 struct completion *vfork;
1229 vfork = tsk->vfork_done;
1230 if (likely(vfork)) {
1231 tsk->vfork_done = NULL;
1237 static int wait_for_vfork_done(struct task_struct *child,
1238 struct completion *vfork)
1242 freezer_do_not_count();
1243 cgroup_enter_frozen();
1244 killed = wait_for_completion_killable(vfork);
1245 cgroup_leave_frozen(false);
1250 child->vfork_done = NULL;
1254 put_task_struct(child);
1258 /* Please note the differences between mmput and mm_release.
1259 * mmput is called whenever we stop holding onto a mm_struct,
1260 * error success whatever.
1262 * mm_release is called after a mm_struct has been removed
1263 * from the current process.
1265 * This difference is important for error handling, when we
1266 * only half set up a mm_struct for a new process and need to restore
1267 * the old one. Because we mmput the new mm_struct before
1268 * restoring the old one. . .
1269 * Eric Biederman 10 January 1998
1271 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1273 /* Get rid of any futexes when releasing the mm */
1275 if (unlikely(tsk->robust_list)) {
1276 exit_robust_list(tsk);
1277 tsk->robust_list = NULL;
1279 #ifdef CONFIG_COMPAT
1280 if (unlikely(tsk->compat_robust_list)) {
1281 compat_exit_robust_list(tsk);
1282 tsk->compat_robust_list = NULL;
1285 if (unlikely(!list_empty(&tsk->pi_state_list)))
1286 exit_pi_state_list(tsk);
1289 uprobe_free_utask(tsk);
1291 /* Get rid of any cached register state */
1292 deactivate_mm(tsk, mm);
1295 * Signal userspace if we're not exiting with a core dump
1296 * because we want to leave the value intact for debugging
1299 if (tsk->clear_child_tid) {
1300 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1301 atomic_read(&mm->mm_users) > 1) {
1303 * We don't check the error code - if userspace has
1304 * not set up a proper pointer then tough luck.
1306 put_user(0, tsk->clear_child_tid);
1307 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1308 1, NULL, NULL, 0, 0);
1310 tsk->clear_child_tid = NULL;
1314 * All done, finally we can wake up parent and return this mm to him.
1315 * Also kthread_stop() uses this completion for synchronization.
1317 if (tsk->vfork_done)
1318 complete_vfork_done(tsk);
1322 * dup_mm() - duplicates an existing mm structure
1323 * @tsk: the task_struct with which the new mm will be associated.
1324 * @oldmm: the mm to duplicate.
1326 * Allocates a new mm structure and duplicates the provided @oldmm structure
1329 * Return: the duplicated mm or NULL on failure.
1331 static struct mm_struct *dup_mm(struct task_struct *tsk,
1332 struct mm_struct *oldmm)
1334 struct mm_struct *mm;
1341 memcpy(mm, oldmm, sizeof(*mm));
1343 if (!mm_init(mm, tsk, mm->user_ns))
1346 err = dup_mmap(mm, oldmm);
1350 mm->hiwater_rss = get_mm_rss(mm);
1351 mm->hiwater_vm = mm->total_vm;
1353 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1359 /* don't put binfmt in mmput, we haven't got module yet */
1361 mm_init_owner(mm, NULL);
1368 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1370 struct mm_struct *mm, *oldmm;
1373 tsk->min_flt = tsk->maj_flt = 0;
1374 tsk->nvcsw = tsk->nivcsw = 0;
1375 #ifdef CONFIG_DETECT_HUNG_TASK
1376 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1377 tsk->last_switch_time = 0;
1381 tsk->active_mm = NULL;
1384 * Are we cloning a kernel thread?
1386 * We need to steal a active VM for that..
1388 oldmm = current->mm;
1392 /* initialize the new vmacache entries */
1393 vmacache_flush(tsk);
1395 if (clone_flags & CLONE_VM) {
1402 mm = dup_mm(tsk, current->mm);
1408 tsk->active_mm = mm;
1415 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1417 struct fs_struct *fs = current->fs;
1418 if (clone_flags & CLONE_FS) {
1419 /* tsk->fs is already what we want */
1420 spin_lock(&fs->lock);
1422 spin_unlock(&fs->lock);
1426 spin_unlock(&fs->lock);
1429 tsk->fs = copy_fs_struct(fs);
1435 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1437 struct files_struct *oldf, *newf;
1441 * A background process may not have any files ...
1443 oldf = current->files;
1447 if (clone_flags & CLONE_FILES) {
1448 atomic_inc(&oldf->count);
1452 newf = dup_fd(oldf, &error);
1462 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1465 struct io_context *ioc = current->io_context;
1466 struct io_context *new_ioc;
1471 * Share io context with parent, if CLONE_IO is set
1473 if (clone_flags & CLONE_IO) {
1475 tsk->io_context = ioc;
1476 } else if (ioprio_valid(ioc->ioprio)) {
1477 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1478 if (unlikely(!new_ioc))
1481 new_ioc->ioprio = ioc->ioprio;
1482 put_io_context(new_ioc);
1488 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1490 struct sighand_struct *sig;
1492 if (clone_flags & CLONE_SIGHAND) {
1493 refcount_inc(¤t->sighand->count);
1496 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1497 rcu_assign_pointer(tsk->sighand, sig);
1501 refcount_set(&sig->count, 1);
1502 spin_lock_irq(¤t->sighand->siglock);
1503 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1504 spin_unlock_irq(¤t->sighand->siglock);
1508 void __cleanup_sighand(struct sighand_struct *sighand)
1510 if (refcount_dec_and_test(&sighand->count)) {
1511 signalfd_cleanup(sighand);
1513 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1514 * without an RCU grace period, see __lock_task_sighand().
1516 kmem_cache_free(sighand_cachep, sighand);
1520 #ifdef CONFIG_POSIX_TIMERS
1522 * Initialize POSIX timer handling for a thread group.
1524 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1526 unsigned long cpu_limit;
1528 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1529 if (cpu_limit != RLIM_INFINITY) {
1530 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1531 sig->cputimer.running = true;
1534 /* The timer lists. */
1535 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1536 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1537 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1540 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1543 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1545 struct signal_struct *sig;
1547 if (clone_flags & CLONE_THREAD)
1550 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1555 sig->nr_threads = 1;
1556 atomic_set(&sig->live, 1);
1557 refcount_set(&sig->sigcnt, 1);
1559 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1560 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1561 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1563 init_waitqueue_head(&sig->wait_chldexit);
1564 sig->curr_target = tsk;
1565 init_sigpending(&sig->shared_pending);
1566 INIT_HLIST_HEAD(&sig->multiprocess);
1567 seqlock_init(&sig->stats_lock);
1568 prev_cputime_init(&sig->prev_cputime);
1570 #ifdef CONFIG_POSIX_TIMERS
1571 INIT_LIST_HEAD(&sig->posix_timers);
1572 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1573 sig->real_timer.function = it_real_fn;
1576 task_lock(current->group_leader);
1577 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1578 task_unlock(current->group_leader);
1580 posix_cpu_timers_init_group(sig);
1582 tty_audit_fork(sig);
1583 sched_autogroup_fork(sig);
1585 sig->oom_score_adj = current->signal->oom_score_adj;
1586 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1588 mutex_init(&sig->cred_guard_mutex);
1593 static void copy_seccomp(struct task_struct *p)
1595 #ifdef CONFIG_SECCOMP
1597 * Must be called with sighand->lock held, which is common to
1598 * all threads in the group. Holding cred_guard_mutex is not
1599 * needed because this new task is not yet running and cannot
1602 assert_spin_locked(¤t->sighand->siglock);
1604 /* Ref-count the new filter user, and assign it. */
1605 get_seccomp_filter(current);
1606 p->seccomp = current->seccomp;
1609 * Explicitly enable no_new_privs here in case it got set
1610 * between the task_struct being duplicated and holding the
1611 * sighand lock. The seccomp state and nnp must be in sync.
1613 if (task_no_new_privs(current))
1614 task_set_no_new_privs(p);
1617 * If the parent gained a seccomp mode after copying thread
1618 * flags and between before we held the sighand lock, we have
1619 * to manually enable the seccomp thread flag here.
1621 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1622 set_tsk_thread_flag(p, TIF_SECCOMP);
1626 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1628 current->clear_child_tid = tidptr;
1630 return task_pid_vnr(current);
1633 static void rt_mutex_init_task(struct task_struct *p)
1635 raw_spin_lock_init(&p->pi_lock);
1636 #ifdef CONFIG_RT_MUTEXES
1637 p->pi_waiters = RB_ROOT_CACHED;
1638 p->pi_top_task = NULL;
1639 p->pi_blocked_on = NULL;
1643 #ifdef CONFIG_POSIX_TIMERS
1645 * Initialize POSIX timer handling for a single task.
1647 static void posix_cpu_timers_init(struct task_struct *tsk)
1649 tsk->cputime_expires.prof_exp = 0;
1650 tsk->cputime_expires.virt_exp = 0;
1651 tsk->cputime_expires.sched_exp = 0;
1652 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1653 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1654 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1657 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1660 static inline void init_task_pid_links(struct task_struct *task)
1664 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1665 INIT_HLIST_NODE(&task->pid_links[type]);
1670 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1672 if (type == PIDTYPE_PID)
1673 task->thread_pid = pid;
1675 task->signal->pids[type] = pid;
1678 static inline void rcu_copy_process(struct task_struct *p)
1680 #ifdef CONFIG_PREEMPT_RCU
1681 p->rcu_read_lock_nesting = 0;
1682 p->rcu_read_unlock_special.s = 0;
1683 p->rcu_blocked_node = NULL;
1684 INIT_LIST_HEAD(&p->rcu_node_entry);
1685 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1686 #ifdef CONFIG_TASKS_RCU
1687 p->rcu_tasks_holdout = false;
1688 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1689 p->rcu_tasks_idle_cpu = -1;
1690 #endif /* #ifdef CONFIG_TASKS_RCU */
1693 struct pid *pidfd_pid(const struct file *file)
1695 if (file->f_op == &pidfd_fops)
1696 return file->private_data;
1698 return ERR_PTR(-EBADF);
1701 static int pidfd_release(struct inode *inode, struct file *file)
1703 struct pid *pid = file->private_data;
1705 file->private_data = NULL;
1710 #ifdef CONFIG_PROC_FS
1711 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1713 struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1714 struct pid *pid = f->private_data;
1716 seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1722 * Poll support for process exit notification.
1724 static unsigned int pidfd_poll(struct file *file, struct poll_table_struct *pts)
1726 struct task_struct *task;
1727 struct pid *pid = file->private_data;
1730 poll_wait(file, &pid->wait_pidfd, pts);
1733 task = pid_task(pid, PIDTYPE_PID);
1735 * Inform pollers only when the whole thread group exits.
1736 * If the thread group leader exits before all other threads in the
1737 * group, then poll(2) should block, similar to the wait(2) family.
1739 if (!task || (task->exit_state && thread_group_empty(task)))
1740 poll_flags = POLLIN | POLLRDNORM;
1746 const struct file_operations pidfd_fops = {
1747 .release = pidfd_release,
1749 #ifdef CONFIG_PROC_FS
1750 .show_fdinfo = pidfd_show_fdinfo,
1754 static void __delayed_free_task(struct rcu_head *rhp)
1756 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1761 static __always_inline void delayed_free_task(struct task_struct *tsk)
1763 if (IS_ENABLED(CONFIG_MEMCG))
1764 call_rcu(&tsk->rcu, __delayed_free_task);
1770 * This creates a new process as a copy of the old one,
1771 * but does not actually start it yet.
1773 * It copies the registers, and all the appropriate
1774 * parts of the process environment (as per the clone
1775 * flags). The actual kick-off is left to the caller.
1777 static __latent_entropy struct task_struct *copy_process(
1781 struct kernel_clone_args *args)
1783 int pidfd = -1, retval;
1784 struct task_struct *p;
1785 struct multiprocess_signals delayed;
1786 struct file *pidfile = NULL;
1787 u64 clone_flags = args->flags;
1790 * Don't allow sharing the root directory with processes in a different
1793 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1794 return ERR_PTR(-EINVAL);
1796 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1797 return ERR_PTR(-EINVAL);
1800 * Thread groups must share signals as well, and detached threads
1801 * can only be started up within the thread group.
1803 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1804 return ERR_PTR(-EINVAL);
1807 * Shared signal handlers imply shared VM. By way of the above,
1808 * thread groups also imply shared VM. Blocking this case allows
1809 * for various simplifications in other code.
1811 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1812 return ERR_PTR(-EINVAL);
1815 * Siblings of global init remain as zombies on exit since they are
1816 * not reaped by their parent (swapper). To solve this and to avoid
1817 * multi-rooted process trees, prevent global and container-inits
1818 * from creating siblings.
1820 if ((clone_flags & CLONE_PARENT) &&
1821 current->signal->flags & SIGNAL_UNKILLABLE)
1822 return ERR_PTR(-EINVAL);
1825 * If the new process will be in a different pid or user namespace
1826 * do not allow it to share a thread group with the forking task.
1828 if (clone_flags & CLONE_THREAD) {
1829 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1830 (task_active_pid_ns(current) !=
1831 current->nsproxy->pid_ns_for_children))
1832 return ERR_PTR(-EINVAL);
1835 if (clone_flags & CLONE_PIDFD) {
1837 * - CLONE_DETACHED is blocked so that we can potentially
1838 * reuse it later for CLONE_PIDFD.
1839 * - CLONE_THREAD is blocked until someone really needs it.
1841 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1842 return ERR_PTR(-EINVAL);
1846 * Force any signals received before this point to be delivered
1847 * before the fork happens. Collect up signals sent to multiple
1848 * processes that happen during the fork and delay them so that
1849 * they appear to happen after the fork.
1851 sigemptyset(&delayed.signal);
1852 INIT_HLIST_NODE(&delayed.node);
1854 spin_lock_irq(¤t->sighand->siglock);
1855 if (!(clone_flags & CLONE_THREAD))
1856 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1857 recalc_sigpending();
1858 spin_unlock_irq(¤t->sighand->siglock);
1859 retval = -ERESTARTNOINTR;
1860 if (signal_pending(current))
1864 p = dup_task_struct(current, node);
1869 * This _must_ happen before we call free_task(), i.e. before we jump
1870 * to any of the bad_fork_* labels. This is to avoid freeing
1871 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1872 * kernel threads (PF_KTHREAD).
1874 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1876 * Clear TID on mm_release()?
1878 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1880 ftrace_graph_init_task(p);
1882 rt_mutex_init_task(p);
1884 #ifdef CONFIG_PROVE_LOCKING
1885 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1886 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1889 if (atomic_read(&p->real_cred->user->processes) >=
1890 task_rlimit(p, RLIMIT_NPROC)) {
1891 if (p->real_cred->user != INIT_USER &&
1892 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1895 current->flags &= ~PF_NPROC_EXCEEDED;
1897 retval = copy_creds(p, clone_flags);
1902 * If multiple threads are within copy_process(), then this check
1903 * triggers too late. This doesn't hurt, the check is only there
1904 * to stop root fork bombs.
1907 if (nr_threads >= max_threads)
1908 goto bad_fork_cleanup_count;
1910 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1911 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1912 p->flags |= PF_FORKNOEXEC;
1913 INIT_LIST_HEAD(&p->children);
1914 INIT_LIST_HEAD(&p->sibling);
1915 rcu_copy_process(p);
1916 p->vfork_done = NULL;
1917 spin_lock_init(&p->alloc_lock);
1919 init_sigpending(&p->pending);
1921 p->utime = p->stime = p->gtime = 0;
1922 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1923 p->utimescaled = p->stimescaled = 0;
1925 prev_cputime_init(&p->prev_cputime);
1927 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1928 seqcount_init(&p->vtime.seqcount);
1929 p->vtime.starttime = 0;
1930 p->vtime.state = VTIME_INACTIVE;
1933 #if defined(SPLIT_RSS_COUNTING)
1934 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1937 p->default_timer_slack_ns = current->timer_slack_ns;
1943 task_io_accounting_init(&p->ioac);
1944 acct_clear_integrals(p);
1946 posix_cpu_timers_init(p);
1948 p->io_context = NULL;
1949 audit_set_context(p, NULL);
1952 p->mempolicy = mpol_dup(p->mempolicy);
1953 if (IS_ERR(p->mempolicy)) {
1954 retval = PTR_ERR(p->mempolicy);
1955 p->mempolicy = NULL;
1956 goto bad_fork_cleanup_threadgroup_lock;
1959 #ifdef CONFIG_CPUSETS
1960 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1961 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1962 seqcount_init(&p->mems_allowed_seq);
1964 #ifdef CONFIG_TRACE_IRQFLAGS
1966 p->hardirqs_enabled = 0;
1967 p->hardirq_enable_ip = 0;
1968 p->hardirq_enable_event = 0;
1969 p->hardirq_disable_ip = _THIS_IP_;
1970 p->hardirq_disable_event = 0;
1971 p->softirqs_enabled = 1;
1972 p->softirq_enable_ip = _THIS_IP_;
1973 p->softirq_enable_event = 0;
1974 p->softirq_disable_ip = 0;
1975 p->softirq_disable_event = 0;
1976 p->hardirq_context = 0;
1977 p->softirq_context = 0;
1980 p->pagefault_disabled = 0;
1982 #ifdef CONFIG_LOCKDEP
1983 lockdep_init_task(p);
1986 #ifdef CONFIG_DEBUG_MUTEXES
1987 p->blocked_on = NULL; /* not blocked yet */
1989 #ifdef CONFIG_BCACHE
1990 p->sequential_io = 0;
1991 p->sequential_io_avg = 0;
1994 /* Perform scheduler related setup. Assign this task to a CPU. */
1995 retval = sched_fork(clone_flags, p);
1997 goto bad_fork_cleanup_policy;
1999 retval = perf_event_init_task(p);
2001 goto bad_fork_cleanup_policy;
2002 retval = audit_alloc(p);
2004 goto bad_fork_cleanup_perf;
2005 /* copy all the process information */
2007 retval = security_task_alloc(p, clone_flags);
2009 goto bad_fork_cleanup_audit;
2010 retval = copy_semundo(clone_flags, p);
2012 goto bad_fork_cleanup_security;
2013 retval = copy_files(clone_flags, p);
2015 goto bad_fork_cleanup_semundo;
2016 retval = copy_fs(clone_flags, p);
2018 goto bad_fork_cleanup_files;
2019 retval = copy_sighand(clone_flags, p);
2021 goto bad_fork_cleanup_fs;
2022 retval = copy_signal(clone_flags, p);
2024 goto bad_fork_cleanup_sighand;
2025 retval = copy_mm(clone_flags, p);
2027 goto bad_fork_cleanup_signal;
2028 retval = copy_namespaces(clone_flags, p);
2030 goto bad_fork_cleanup_mm;
2031 retval = copy_io(clone_flags, p);
2033 goto bad_fork_cleanup_namespaces;
2034 retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2037 goto bad_fork_cleanup_io;
2039 stackleak_task_init(p);
2041 if (pid != &init_struct_pid) {
2042 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
2044 retval = PTR_ERR(pid);
2045 goto bad_fork_cleanup_thread;
2050 * This has to happen after we've potentially unshared the file
2051 * descriptor table (so that the pidfd doesn't leak into the child
2052 * if the fd table isn't shared).
2054 if (clone_flags & CLONE_PIDFD) {
2055 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2057 goto bad_fork_free_pid;
2061 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2062 O_RDWR | O_CLOEXEC);
2063 if (IS_ERR(pidfile)) {
2064 put_unused_fd(pidfd);
2065 retval = PTR_ERR(pidfile);
2066 goto bad_fork_free_pid;
2068 get_pid(pid); /* held by pidfile now */
2070 retval = put_user(pidfd, args->pidfd);
2072 goto bad_fork_put_pidfd;
2079 p->robust_list = NULL;
2080 #ifdef CONFIG_COMPAT
2081 p->compat_robust_list = NULL;
2083 INIT_LIST_HEAD(&p->pi_state_list);
2084 p->pi_state_cache = NULL;
2087 * sigaltstack should be cleared when sharing the same VM
2089 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2093 * Syscall tracing and stepping should be turned off in the
2094 * child regardless of CLONE_PTRACE.
2096 user_disable_single_step(p);
2097 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2098 #ifdef TIF_SYSCALL_EMU
2099 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2101 clear_tsk_latency_tracing(p);
2103 /* ok, now we should be set up.. */
2104 p->pid = pid_nr(pid);
2105 if (clone_flags & CLONE_THREAD) {
2106 p->exit_signal = -1;
2107 p->group_leader = current->group_leader;
2108 p->tgid = current->tgid;
2110 if (clone_flags & CLONE_PARENT)
2111 p->exit_signal = current->group_leader->exit_signal;
2113 p->exit_signal = args->exit_signal;
2114 p->group_leader = p;
2119 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2120 p->dirty_paused_when = 0;
2122 p->pdeath_signal = 0;
2123 INIT_LIST_HEAD(&p->thread_group);
2124 p->task_works = NULL;
2126 cgroup_threadgroup_change_begin(current);
2128 * Ensure that the cgroup subsystem policies allow the new process to be
2129 * forked. It should be noted the the new process's css_set can be changed
2130 * between here and cgroup_post_fork() if an organisation operation is in
2133 retval = cgroup_can_fork(p);
2135 goto bad_fork_cgroup_threadgroup_change_end;
2138 * From this point on we must avoid any synchronous user-space
2139 * communication until we take the tasklist-lock. In particular, we do
2140 * not want user-space to be able to predict the process start-time by
2141 * stalling fork(2) after we recorded the start_time but before it is
2142 * visible to the system.
2145 p->start_time = ktime_get_ns();
2146 p->real_start_time = ktime_get_boottime_ns();
2149 * Make it visible to the rest of the system, but dont wake it up yet.
2150 * Need tasklist lock for parent etc handling!
2152 write_lock_irq(&tasklist_lock);
2154 /* CLONE_PARENT re-uses the old parent */
2155 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2156 p->real_parent = current->real_parent;
2157 p->parent_exec_id = current->parent_exec_id;
2159 p->real_parent = current;
2160 p->parent_exec_id = current->self_exec_id;
2163 klp_copy_process(p);
2165 spin_lock(¤t->sighand->siglock);
2168 * Copy seccomp details explicitly here, in case they were changed
2169 * before holding sighand lock.
2173 rseq_fork(p, clone_flags);
2175 /* Don't start children in a dying pid namespace */
2176 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2178 goto bad_fork_cancel_cgroup;
2181 /* Let kill terminate clone/fork in the middle */
2182 if (fatal_signal_pending(current)) {
2184 goto bad_fork_cancel_cgroup;
2187 /* past the last point of failure */
2189 fd_install(pidfd, pidfile);
2191 init_task_pid_links(p);
2192 if (likely(p->pid)) {
2193 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2195 init_task_pid(p, PIDTYPE_PID, pid);
2196 if (thread_group_leader(p)) {
2197 init_task_pid(p, PIDTYPE_TGID, pid);
2198 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2199 init_task_pid(p, PIDTYPE_SID, task_session(current));
2201 if (is_child_reaper(pid)) {
2202 ns_of_pid(pid)->child_reaper = p;
2203 p->signal->flags |= SIGNAL_UNKILLABLE;
2205 p->signal->shared_pending.signal = delayed.signal;
2206 p->signal->tty = tty_kref_get(current->signal->tty);
2208 * Inherit has_child_subreaper flag under the same
2209 * tasklist_lock with adding child to the process tree
2210 * for propagate_has_child_subreaper optimization.
2212 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2213 p->real_parent->signal->is_child_subreaper;
2214 list_add_tail(&p->sibling, &p->real_parent->children);
2215 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2216 attach_pid(p, PIDTYPE_TGID);
2217 attach_pid(p, PIDTYPE_PGID);
2218 attach_pid(p, PIDTYPE_SID);
2219 __this_cpu_inc(process_counts);
2221 current->signal->nr_threads++;
2222 atomic_inc(¤t->signal->live);
2223 refcount_inc(¤t->signal->sigcnt);
2224 task_join_group_stop(p);
2225 list_add_tail_rcu(&p->thread_group,
2226 &p->group_leader->thread_group);
2227 list_add_tail_rcu(&p->thread_node,
2228 &p->signal->thread_head);
2230 attach_pid(p, PIDTYPE_PID);
2234 hlist_del_init(&delayed.node);
2235 spin_unlock(¤t->sighand->siglock);
2236 syscall_tracepoint_update(p);
2237 write_unlock_irq(&tasklist_lock);
2239 proc_fork_connector(p);
2240 cgroup_post_fork(p);
2241 cgroup_threadgroup_change_end(current);
2244 trace_task_newtask(p, clone_flags);
2245 uprobe_copy_process(p, clone_flags);
2249 bad_fork_cancel_cgroup:
2250 spin_unlock(¤t->sighand->siglock);
2251 write_unlock_irq(&tasklist_lock);
2252 cgroup_cancel_fork(p);
2253 bad_fork_cgroup_threadgroup_change_end:
2254 cgroup_threadgroup_change_end(current);
2256 if (clone_flags & CLONE_PIDFD) {
2258 put_unused_fd(pidfd);
2261 if (pid != &init_struct_pid)
2263 bad_fork_cleanup_thread:
2265 bad_fork_cleanup_io:
2268 bad_fork_cleanup_namespaces:
2269 exit_task_namespaces(p);
2270 bad_fork_cleanup_mm:
2272 mm_clear_owner(p->mm, p);
2275 bad_fork_cleanup_signal:
2276 if (!(clone_flags & CLONE_THREAD))
2277 free_signal_struct(p->signal);
2278 bad_fork_cleanup_sighand:
2279 __cleanup_sighand(p->sighand);
2280 bad_fork_cleanup_fs:
2281 exit_fs(p); /* blocking */
2282 bad_fork_cleanup_files:
2283 exit_files(p); /* blocking */
2284 bad_fork_cleanup_semundo:
2286 bad_fork_cleanup_security:
2287 security_task_free(p);
2288 bad_fork_cleanup_audit:
2290 bad_fork_cleanup_perf:
2291 perf_event_free_task(p);
2292 bad_fork_cleanup_policy:
2293 lockdep_free_task(p);
2295 mpol_put(p->mempolicy);
2296 bad_fork_cleanup_threadgroup_lock:
2298 delayacct_tsk_free(p);
2299 bad_fork_cleanup_count:
2300 atomic_dec(&p->cred->user->processes);
2303 p->state = TASK_DEAD;
2305 delayed_free_task(p);
2307 spin_lock_irq(¤t->sighand->siglock);
2308 hlist_del_init(&delayed.node);
2309 spin_unlock_irq(¤t->sighand->siglock);
2310 return ERR_PTR(retval);
2313 static inline void init_idle_pids(struct task_struct *idle)
2317 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2318 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2319 init_task_pid(idle, type, &init_struct_pid);
2323 struct task_struct *fork_idle(int cpu)
2325 struct task_struct *task;
2326 struct kernel_clone_args args = {
2330 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2331 if (!IS_ERR(task)) {
2332 init_idle_pids(task);
2333 init_idle(task, cpu);
2339 struct mm_struct *copy_init_mm(void)
2341 return dup_mm(NULL, &init_mm);
2345 * Ok, this is the main fork-routine.
2347 * It copies the process, and if successful kick-starts
2348 * it and waits for it to finish using the VM if required.
2350 * args->exit_signal is expected to be checked for sanity by the caller.
2352 long _do_fork(struct kernel_clone_args *args)
2354 u64 clone_flags = args->flags;
2355 struct completion vfork;
2357 struct task_struct *p;
2362 * Determine whether and which event to report to ptracer. When
2363 * called from kernel_thread or CLONE_UNTRACED is explicitly
2364 * requested, no event is reported; otherwise, report if the event
2365 * for the type of forking is enabled.
2367 if (!(clone_flags & CLONE_UNTRACED)) {
2368 if (clone_flags & CLONE_VFORK)
2369 trace = PTRACE_EVENT_VFORK;
2370 else if (args->exit_signal != SIGCHLD)
2371 trace = PTRACE_EVENT_CLONE;
2373 trace = PTRACE_EVENT_FORK;
2375 if (likely(!ptrace_event_enabled(current, trace)))
2379 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2380 add_latent_entropy();
2386 * Do this prior waking up the new thread - the thread pointer
2387 * might get invalid after that point, if the thread exits quickly.
2389 trace_sched_process_fork(current, p);
2391 pid = get_task_pid(p, PIDTYPE_PID);
2394 if (clone_flags & CLONE_PARENT_SETTID)
2395 put_user(nr, args->parent_tid);
2397 if (clone_flags & CLONE_VFORK) {
2398 p->vfork_done = &vfork;
2399 init_completion(&vfork);
2403 wake_up_new_task(p);
2405 /* forking complete and child started to run, tell ptracer */
2406 if (unlikely(trace))
2407 ptrace_event_pid(trace, pid);
2409 if (clone_flags & CLONE_VFORK) {
2410 if (!wait_for_vfork_done(p, &vfork))
2411 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2418 bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
2420 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2421 if ((kargs->flags & CLONE_PIDFD) &&
2422 (kargs->flags & CLONE_PARENT_SETTID))
2428 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2429 /* For compatibility with architectures that call do_fork directly rather than
2430 * using the syscall entry points below. */
2431 long do_fork(unsigned long clone_flags,
2432 unsigned long stack_start,
2433 unsigned long stack_size,
2434 int __user *parent_tidptr,
2435 int __user *child_tidptr)
2437 struct kernel_clone_args args = {
2438 .flags = (clone_flags & ~CSIGNAL),
2439 .pidfd = parent_tidptr,
2440 .child_tid = child_tidptr,
2441 .parent_tid = parent_tidptr,
2442 .exit_signal = (clone_flags & CSIGNAL),
2443 .stack = stack_start,
2444 .stack_size = stack_size,
2447 if (!legacy_clone_args_valid(&args))
2450 return _do_fork(&args);
2455 * Create a kernel thread.
2457 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2459 struct kernel_clone_args args = {
2460 .flags = ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2461 .exit_signal = (flags & CSIGNAL),
2462 .stack = (unsigned long)fn,
2463 .stack_size = (unsigned long)arg,
2466 return _do_fork(&args);
2469 #ifdef __ARCH_WANT_SYS_FORK
2470 SYSCALL_DEFINE0(fork)
2473 struct kernel_clone_args args = {
2474 .exit_signal = SIGCHLD,
2477 return _do_fork(&args);
2479 /* can not support in nommu mode */
2485 #ifdef __ARCH_WANT_SYS_VFORK
2486 SYSCALL_DEFINE0(vfork)
2488 struct kernel_clone_args args = {
2489 .flags = CLONE_VFORK | CLONE_VM,
2490 .exit_signal = SIGCHLD,
2493 return _do_fork(&args);
2497 #ifdef __ARCH_WANT_SYS_CLONE
2498 #ifdef CONFIG_CLONE_BACKWARDS
2499 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2500 int __user *, parent_tidptr,
2502 int __user *, child_tidptr)
2503 #elif defined(CONFIG_CLONE_BACKWARDS2)
2504 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2505 int __user *, parent_tidptr,
2506 int __user *, child_tidptr,
2508 #elif defined(CONFIG_CLONE_BACKWARDS3)
2509 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2511 int __user *, parent_tidptr,
2512 int __user *, child_tidptr,
2515 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2516 int __user *, parent_tidptr,
2517 int __user *, child_tidptr,
2521 struct kernel_clone_args args = {
2522 .flags = (clone_flags & ~CSIGNAL),
2523 .pidfd = parent_tidptr,
2524 .child_tid = child_tidptr,
2525 .parent_tid = parent_tidptr,
2526 .exit_signal = (clone_flags & CSIGNAL),
2531 if (!legacy_clone_args_valid(&args))
2534 return _do_fork(&args);
2538 #ifdef __ARCH_WANT_SYS_CLONE3
2539 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2540 struct clone_args __user *uargs,
2543 struct clone_args args;
2545 if (unlikely(size > PAGE_SIZE))
2548 if (unlikely(size < sizeof(struct clone_args)))
2551 if (unlikely(!access_ok(uargs, size)))
2554 if (size > sizeof(struct clone_args)) {
2555 unsigned char __user *addr;
2556 unsigned char __user *end;
2559 addr = (void __user *)uargs + sizeof(struct clone_args);
2560 end = (void __user *)uargs + size;
2562 for (; addr < end; addr++) {
2563 if (get_user(val, addr))
2569 size = sizeof(struct clone_args);
2572 if (copy_from_user(&args, uargs, size))
2576 * Verify that higher 32bits of exit_signal are unset and that
2577 * it is a valid signal
2579 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2580 !valid_signal(args.exit_signal)))
2583 *kargs = (struct kernel_clone_args){
2584 .flags = args.flags,
2585 .pidfd = u64_to_user_ptr(args.pidfd),
2586 .child_tid = u64_to_user_ptr(args.child_tid),
2587 .parent_tid = u64_to_user_ptr(args.parent_tid),
2588 .exit_signal = args.exit_signal,
2589 .stack = args.stack,
2590 .stack_size = args.stack_size,
2597 static bool clone3_args_valid(const struct kernel_clone_args *kargs)
2600 * All lower bits of the flag word are taken.
2601 * Verify that no other unknown flags are passed along.
2603 if (kargs->flags & ~CLONE_LEGACY_FLAGS)
2607 * - make the CLONE_DETACHED bit reuseable for clone3
2608 * - make the CSIGNAL bits reuseable for clone3
2610 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2613 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2620 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2624 struct kernel_clone_args kargs;
2626 err = copy_clone_args_from_user(&kargs, uargs, size);
2630 if (!clone3_args_valid(&kargs))
2633 return _do_fork(&kargs);
2637 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2639 struct task_struct *leader, *parent, *child;
2642 read_lock(&tasklist_lock);
2643 leader = top = top->group_leader;
2645 for_each_thread(leader, parent) {
2646 list_for_each_entry(child, &parent->children, sibling) {
2647 res = visitor(child, data);
2659 if (leader != top) {
2661 parent = child->real_parent;
2662 leader = parent->group_leader;
2666 read_unlock(&tasklist_lock);
2669 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2670 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2673 static void sighand_ctor(void *data)
2675 struct sighand_struct *sighand = data;
2677 spin_lock_init(&sighand->siglock);
2678 init_waitqueue_head(&sighand->signalfd_wqh);
2681 void __init proc_caches_init(void)
2683 unsigned int mm_size;
2685 sighand_cachep = kmem_cache_create("sighand_cache",
2686 sizeof(struct sighand_struct), 0,
2687 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2688 SLAB_ACCOUNT, sighand_ctor);
2689 signal_cachep = kmem_cache_create("signal_cache",
2690 sizeof(struct signal_struct), 0,
2691 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2693 files_cachep = kmem_cache_create("files_cache",
2694 sizeof(struct files_struct), 0,
2695 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2697 fs_cachep = kmem_cache_create("fs_cache",
2698 sizeof(struct fs_struct), 0,
2699 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2703 * The mm_cpumask is located at the end of mm_struct, and is
2704 * dynamically sized based on the maximum CPU number this system
2705 * can have, taking hotplug into account (nr_cpu_ids).
2707 mm_size = sizeof(struct mm_struct) + cpumask_size();
2709 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2710 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2711 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2712 offsetof(struct mm_struct, saved_auxv),
2713 sizeof_field(struct mm_struct, saved_auxv),
2715 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2717 nsproxy_cache_init();
2721 * Check constraints on flags passed to the unshare system call.
2723 static int check_unshare_flags(unsigned long unshare_flags)
2725 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2726 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2727 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2728 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2731 * Not implemented, but pretend it works if there is nothing
2732 * to unshare. Note that unsharing the address space or the
2733 * signal handlers also need to unshare the signal queues (aka
2736 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2737 if (!thread_group_empty(current))
2740 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2741 if (refcount_read(¤t->sighand->count) > 1)
2744 if (unshare_flags & CLONE_VM) {
2745 if (!current_is_single_threaded())
2753 * Unshare the filesystem structure if it is being shared
2755 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2757 struct fs_struct *fs = current->fs;
2759 if (!(unshare_flags & CLONE_FS) || !fs)
2762 /* don't need lock here; in the worst case we'll do useless copy */
2766 *new_fsp = copy_fs_struct(fs);
2774 * Unshare file descriptor table if it is being shared
2776 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2778 struct files_struct *fd = current->files;
2781 if ((unshare_flags & CLONE_FILES) &&
2782 (fd && atomic_read(&fd->count) > 1)) {
2783 *new_fdp = dup_fd(fd, &error);
2792 * unshare allows a process to 'unshare' part of the process
2793 * context which was originally shared using clone. copy_*
2794 * functions used by do_fork() cannot be used here directly
2795 * because they modify an inactive task_struct that is being
2796 * constructed. Here we are modifying the current, active,
2799 int ksys_unshare(unsigned long unshare_flags)
2801 struct fs_struct *fs, *new_fs = NULL;
2802 struct files_struct *fd, *new_fd = NULL;
2803 struct cred *new_cred = NULL;
2804 struct nsproxy *new_nsproxy = NULL;
2809 * If unsharing a user namespace must also unshare the thread group
2810 * and unshare the filesystem root and working directories.
2812 if (unshare_flags & CLONE_NEWUSER)
2813 unshare_flags |= CLONE_THREAD | CLONE_FS;
2815 * If unsharing vm, must also unshare signal handlers.
2817 if (unshare_flags & CLONE_VM)
2818 unshare_flags |= CLONE_SIGHAND;
2820 * If unsharing a signal handlers, must also unshare the signal queues.
2822 if (unshare_flags & CLONE_SIGHAND)
2823 unshare_flags |= CLONE_THREAD;
2825 * If unsharing namespace, must also unshare filesystem information.
2827 if (unshare_flags & CLONE_NEWNS)
2828 unshare_flags |= CLONE_FS;
2830 err = check_unshare_flags(unshare_flags);
2832 goto bad_unshare_out;
2834 * CLONE_NEWIPC must also detach from the undolist: after switching
2835 * to a new ipc namespace, the semaphore arrays from the old
2836 * namespace are unreachable.
2838 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2840 err = unshare_fs(unshare_flags, &new_fs);
2842 goto bad_unshare_out;
2843 err = unshare_fd(unshare_flags, &new_fd);
2845 goto bad_unshare_cleanup_fs;
2846 err = unshare_userns(unshare_flags, &new_cred);
2848 goto bad_unshare_cleanup_fd;
2849 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2852 goto bad_unshare_cleanup_cred;
2854 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2857 * CLONE_SYSVSEM is equivalent to sys_exit().
2861 if (unshare_flags & CLONE_NEWIPC) {
2862 /* Orphan segments in old ns (see sem above). */
2864 shm_init_task(current);
2868 switch_task_namespaces(current, new_nsproxy);
2874 spin_lock(&fs->lock);
2875 current->fs = new_fs;
2880 spin_unlock(&fs->lock);
2884 fd = current->files;
2885 current->files = new_fd;
2889 task_unlock(current);
2892 /* Install the new user namespace */
2893 commit_creds(new_cred);
2898 perf_event_namespaces(current);
2900 bad_unshare_cleanup_cred:
2903 bad_unshare_cleanup_fd:
2905 put_files_struct(new_fd);
2907 bad_unshare_cleanup_fs:
2909 free_fs_struct(new_fs);
2915 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2917 return ksys_unshare(unshare_flags);
2921 * Helper to unshare the files of the current task.
2922 * We don't want to expose copy_files internals to
2923 * the exec layer of the kernel.
2926 int unshare_files(struct files_struct **displaced)
2928 struct task_struct *task = current;
2929 struct files_struct *copy = NULL;
2932 error = unshare_fd(CLONE_FILES, ©);
2933 if (error || !copy) {
2937 *displaced = task->files;
2944 int sysctl_max_threads(struct ctl_table *table, int write,
2945 void __user *buffer, size_t *lenp, loff_t *ppos)
2949 int threads = max_threads;
2950 int min = MIN_THREADS;
2951 int max = MAX_THREADS;
2958 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2962 set_max_threads(threads);