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/kmsan.h>
41 #include <linux/binfmts.h>
42 #include <linux/mman.h>
43 #include <linux/mmu_notifier.h>
46 #include <linux/mm_inline.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/syscall_user_dispatch.h>
57 #include <linux/jiffies.h>
58 #include <linux/futex.h>
59 #include <linux/compat.h>
60 #include <linux/kthread.h>
61 #include <linux/task_io_accounting_ops.h>
62 #include <linux/rcupdate.h>
63 #include <linux/ptrace.h>
64 #include <linux/mount.h>
65 #include <linux/audit.h>
66 #include <linux/memcontrol.h>
67 #include <linux/ftrace.h>
68 #include <linux/proc_fs.h>
69 #include <linux/profile.h>
70 #include <linux/rmap.h>
71 #include <linux/ksm.h>
72 #include <linux/acct.h>
73 #include <linux/userfaultfd_k.h>
74 #include <linux/tsacct_kern.h>
75 #include <linux/cn_proc.h>
76 #include <linux/freezer.h>
77 #include <linux/delayacct.h>
78 #include <linux/taskstats_kern.h>
79 #include <linux/tty.h>
80 #include <linux/fs_struct.h>
81 #include <linux/magic.h>
82 #include <linux/perf_event.h>
83 #include <linux/posix-timers.h>
84 #include <linux/user-return-notifier.h>
85 #include <linux/oom.h>
86 #include <linux/khugepaged.h>
87 #include <linux/signalfd.h>
88 #include <linux/uprobes.h>
89 #include <linux/aio.h>
90 #include <linux/compiler.h>
91 #include <linux/sysctl.h>
92 #include <linux/kcov.h>
93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
97 #include <linux/scs.h>
98 #include <linux/io_uring.h>
99 #include <linux/bpf.h>
100 #include <linux/stackprotector.h>
101 #include <linux/user_events.h>
102 #include <linux/iommu.h>
103 #include <linux/rseq.h>
105 #include <asm/pgalloc.h>
106 #include <linux/uaccess.h>
107 #include <asm/mmu_context.h>
108 #include <asm/cacheflush.h>
109 #include <asm/tlbflush.h>
111 #include <trace/events/sched.h>
113 #define CREATE_TRACE_POINTS
114 #include <trace/events/task.h>
117 * Minimum number of threads to boot the kernel
119 #define MIN_THREADS 20
122 * Maximum number of threads
124 #define MAX_THREADS FUTEX_TID_MASK
127 * Protected counters by write_lock_irq(&tasklist_lock)
129 unsigned long total_forks; /* Handle normal Linux uptimes. */
130 int nr_threads; /* The idle threads do not count.. */
132 static int max_threads; /* tunable limit on nr_threads */
134 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
136 static const char * const resident_page_types[] = {
137 NAMED_ARRAY_INDEX(MM_FILEPAGES),
138 NAMED_ARRAY_INDEX(MM_ANONPAGES),
139 NAMED_ARRAY_INDEX(MM_SWAPENTS),
140 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
143 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
145 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
147 #ifdef CONFIG_PROVE_RCU
148 int lockdep_tasklist_lock_is_held(void)
150 return lockdep_is_held(&tasklist_lock);
152 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
153 #endif /* #ifdef CONFIG_PROVE_RCU */
155 int nr_processes(void)
160 for_each_possible_cpu(cpu)
161 total += per_cpu(process_counts, cpu);
166 void __weak arch_release_task_struct(struct task_struct *tsk)
170 static struct kmem_cache *task_struct_cachep;
172 static inline struct task_struct *alloc_task_struct_node(int node)
174 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
177 static inline void free_task_struct(struct task_struct *tsk)
179 kmem_cache_free(task_struct_cachep, tsk);
183 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
184 * kmemcache based allocator.
186 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
188 # ifdef CONFIG_VMAP_STACK
190 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
191 * flush. Try to minimize the number of calls by caching stacks.
193 #define NR_CACHED_STACKS 2
194 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
198 struct vm_struct *stack_vm_area;
201 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
205 for (i = 0; i < NR_CACHED_STACKS; i++) {
206 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
213 static void thread_stack_free_rcu(struct rcu_head *rh)
215 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
217 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
223 static void thread_stack_delayed_free(struct task_struct *tsk)
225 struct vm_stack *vm_stack = tsk->stack;
227 vm_stack->stack_vm_area = tsk->stack_vm_area;
228 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
231 static int free_vm_stack_cache(unsigned int cpu)
233 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
236 for (i = 0; i < NR_CACHED_STACKS; i++) {
237 struct vm_struct *vm_stack = cached_vm_stacks[i];
242 vfree(vm_stack->addr);
243 cached_vm_stacks[i] = NULL;
249 static int memcg_charge_kernel_stack(struct vm_struct *vm)
255 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
257 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
258 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
265 for (i = 0; i < nr_charged; i++)
266 memcg_kmem_uncharge_page(vm->pages[i], 0);
270 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
272 struct vm_struct *vm;
276 for (i = 0; i < NR_CACHED_STACKS; i++) {
279 s = this_cpu_xchg(cached_stacks[i], NULL);
284 /* Reset stack metadata. */
285 kasan_unpoison_range(s->addr, THREAD_SIZE);
287 stack = kasan_reset_tag(s->addr);
289 /* Clear stale pointers from reused stack. */
290 memset(stack, 0, THREAD_SIZE);
292 if (memcg_charge_kernel_stack(s)) {
297 tsk->stack_vm_area = s;
303 * Allocated stacks are cached and later reused by new threads,
304 * so memcg accounting is performed manually on assigning/releasing
305 * stacks to tasks. Drop __GFP_ACCOUNT.
307 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
308 VMALLOC_START, VMALLOC_END,
309 THREADINFO_GFP & ~__GFP_ACCOUNT,
311 0, node, __builtin_return_address(0));
315 vm = find_vm_area(stack);
316 if (memcg_charge_kernel_stack(vm)) {
321 * We can't call find_vm_area() in interrupt context, and
322 * free_thread_stack() can be called in interrupt context,
323 * so cache the vm_struct.
325 tsk->stack_vm_area = vm;
326 stack = kasan_reset_tag(stack);
331 static void free_thread_stack(struct task_struct *tsk)
333 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
334 thread_stack_delayed_free(tsk);
337 tsk->stack_vm_area = NULL;
340 # else /* !CONFIG_VMAP_STACK */
342 static void thread_stack_free_rcu(struct rcu_head *rh)
344 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
347 static void thread_stack_delayed_free(struct task_struct *tsk)
349 struct rcu_head *rh = tsk->stack;
351 call_rcu(rh, thread_stack_free_rcu);
354 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
356 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
360 tsk->stack = kasan_reset_tag(page_address(page));
366 static void free_thread_stack(struct task_struct *tsk)
368 thread_stack_delayed_free(tsk);
372 # endif /* CONFIG_VMAP_STACK */
373 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
375 static struct kmem_cache *thread_stack_cache;
377 static void thread_stack_free_rcu(struct rcu_head *rh)
379 kmem_cache_free(thread_stack_cache, rh);
382 static void thread_stack_delayed_free(struct task_struct *tsk)
384 struct rcu_head *rh = tsk->stack;
386 call_rcu(rh, thread_stack_free_rcu);
389 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
391 unsigned long *stack;
392 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
393 stack = kasan_reset_tag(stack);
395 return stack ? 0 : -ENOMEM;
398 static void free_thread_stack(struct task_struct *tsk)
400 thread_stack_delayed_free(tsk);
404 void thread_stack_cache_init(void)
406 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
407 THREAD_SIZE, THREAD_SIZE, 0, 0,
409 BUG_ON(thread_stack_cache == NULL);
412 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
414 /* SLAB cache for signal_struct structures (tsk->signal) */
415 static struct kmem_cache *signal_cachep;
417 /* SLAB cache for sighand_struct structures (tsk->sighand) */
418 struct kmem_cache *sighand_cachep;
420 /* SLAB cache for files_struct structures (tsk->files) */
421 struct kmem_cache *files_cachep;
423 /* SLAB cache for fs_struct structures (tsk->fs) */
424 struct kmem_cache *fs_cachep;
426 /* SLAB cache for vm_area_struct structures */
427 static struct kmem_cache *vm_area_cachep;
429 /* SLAB cache for mm_struct structures (tsk->mm) */
430 static struct kmem_cache *mm_cachep;
432 #ifdef CONFIG_PER_VMA_LOCK
434 /* SLAB cache for vm_area_struct.lock */
435 static struct kmem_cache *vma_lock_cachep;
437 static bool vma_lock_alloc(struct vm_area_struct *vma)
439 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
443 init_rwsem(&vma->vm_lock->lock);
444 vma->vm_lock_seq = -1;
449 static inline void vma_lock_free(struct vm_area_struct *vma)
451 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
454 #else /* CONFIG_PER_VMA_LOCK */
456 static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
457 static inline void vma_lock_free(struct vm_area_struct *vma) {}
459 #endif /* CONFIG_PER_VMA_LOCK */
461 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
463 struct vm_area_struct *vma;
465 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
470 if (!vma_lock_alloc(vma)) {
471 kmem_cache_free(vm_area_cachep, vma);
478 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
480 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
485 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
486 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
488 * orig->shared.rb may be modified concurrently, but the clone
489 * will be reinitialized.
491 data_race(memcpy(new, orig, sizeof(*new)));
492 if (!vma_lock_alloc(new)) {
493 kmem_cache_free(vm_area_cachep, new);
496 INIT_LIST_HEAD(&new->anon_vma_chain);
497 vma_numab_state_init(new);
498 dup_anon_vma_name(orig, new);
503 void __vm_area_free(struct vm_area_struct *vma)
505 vma_numab_state_free(vma);
506 free_anon_vma_name(vma);
508 kmem_cache_free(vm_area_cachep, vma);
511 #ifdef CONFIG_PER_VMA_LOCK
512 static void vm_area_free_rcu_cb(struct rcu_head *head)
514 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
517 /* The vma should not be locked while being destroyed. */
518 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
523 void vm_area_free(struct vm_area_struct *vma)
525 #ifdef CONFIG_PER_VMA_LOCK
526 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
532 static void account_kernel_stack(struct task_struct *tsk, int account)
534 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
535 struct vm_struct *vm = task_stack_vm_area(tsk);
538 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
539 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
540 account * (PAGE_SIZE / 1024));
542 void *stack = task_stack_page(tsk);
544 /* All stack pages are in the same node. */
545 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
546 account * (THREAD_SIZE / 1024));
550 void exit_task_stack_account(struct task_struct *tsk)
552 account_kernel_stack(tsk, -1);
554 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
555 struct vm_struct *vm;
558 vm = task_stack_vm_area(tsk);
559 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
560 memcg_kmem_uncharge_page(vm->pages[i], 0);
564 static void release_task_stack(struct task_struct *tsk)
566 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
567 return; /* Better to leak the stack than to free prematurely */
569 free_thread_stack(tsk);
572 #ifdef CONFIG_THREAD_INFO_IN_TASK
573 void put_task_stack(struct task_struct *tsk)
575 if (refcount_dec_and_test(&tsk->stack_refcount))
576 release_task_stack(tsk);
580 void free_task(struct task_struct *tsk)
582 #ifdef CONFIG_SECCOMP
583 WARN_ON_ONCE(tsk->seccomp.filter);
585 release_user_cpus_ptr(tsk);
588 #ifndef CONFIG_THREAD_INFO_IN_TASK
590 * The task is finally done with both the stack and thread_info,
593 release_task_stack(tsk);
596 * If the task had a separate stack allocation, it should be gone
599 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
601 rt_mutex_debug_task_free(tsk);
602 ftrace_graph_exit_task(tsk);
603 arch_release_task_struct(tsk);
604 if (tsk->flags & PF_KTHREAD)
605 free_kthread_struct(tsk);
606 bpf_task_storage_free(tsk);
607 free_task_struct(tsk);
609 EXPORT_SYMBOL(free_task);
611 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
613 struct file *exe_file;
615 exe_file = get_mm_exe_file(oldmm);
616 RCU_INIT_POINTER(mm->exe_file, exe_file);
618 * We depend on the oldmm having properly denied write access to the
621 if (exe_file && deny_write_access(exe_file))
622 pr_warn_once("deny_write_access() failed in %s\n", __func__);
626 static __latent_entropy int dup_mmap(struct mm_struct *mm,
627 struct mm_struct *oldmm)
629 struct vm_area_struct *mpnt, *tmp;
631 unsigned long charge = 0;
633 VMA_ITERATOR(vmi, mm, 0);
635 uprobe_start_dup_mmap();
636 if (mmap_write_lock_killable(oldmm)) {
638 goto fail_uprobe_end;
640 flush_cache_dup_mm(oldmm);
641 uprobe_dup_mmap(oldmm, mm);
643 * Not linked in yet - no deadlock potential:
645 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
647 /* No ordering required: file already has been exposed. */
648 dup_mm_exe_file(mm, oldmm);
650 mm->total_vm = oldmm->total_vm;
651 mm->data_vm = oldmm->data_vm;
652 mm->exec_vm = oldmm->exec_vm;
653 mm->stack_vm = oldmm->stack_vm;
655 retval = ksm_fork(mm, oldmm);
658 khugepaged_fork(mm, oldmm);
660 /* Use __mt_dup() to efficiently build an identical maple tree. */
661 retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
662 if (unlikely(retval))
665 mt_clear_in_rcu(vmi.mas.tree);
666 for_each_vma(vmi, mpnt) {
669 vma_start_write(mpnt);
670 if (mpnt->vm_flags & VM_DONTCOPY) {
671 retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
672 mpnt->vm_end, GFP_KERNEL);
676 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
681 * Don't duplicate many vmas if we've been oom-killed (for
684 if (fatal_signal_pending(current)) {
688 if (mpnt->vm_flags & VM_ACCOUNT) {
689 unsigned long len = vma_pages(mpnt);
691 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
695 tmp = vm_area_dup(mpnt);
698 retval = vma_dup_policy(mpnt, tmp);
700 goto fail_nomem_policy;
702 retval = dup_userfaultfd(tmp, &uf);
704 goto fail_nomem_anon_vma_fork;
705 if (tmp->vm_flags & VM_WIPEONFORK) {
707 * VM_WIPEONFORK gets a clean slate in the child.
708 * Don't prepare anon_vma until fault since we don't
709 * copy page for current vma.
711 tmp->anon_vma = NULL;
712 } else if (anon_vma_fork(tmp, mpnt))
713 goto fail_nomem_anon_vma_fork;
714 vm_flags_clear(tmp, VM_LOCKED_MASK);
717 struct address_space *mapping = file->f_mapping;
720 i_mmap_lock_write(mapping);
721 if (vma_is_shared_maywrite(tmp))
722 mapping_allow_writable(mapping);
723 flush_dcache_mmap_lock(mapping);
724 /* insert tmp into the share list, just after mpnt */
725 vma_interval_tree_insert_after(tmp, mpnt,
727 flush_dcache_mmap_unlock(mapping);
728 i_mmap_unlock_write(mapping);
732 * Copy/update hugetlb private vma information.
734 if (is_vm_hugetlb_page(tmp))
735 hugetlb_dup_vma_private(tmp);
738 * Link the vma into the MT. After using __mt_dup(), memory
739 * allocation is not necessary here, so it cannot fail.
741 vma_iter_bulk_store(&vmi, tmp);
744 if (!(tmp->vm_flags & VM_WIPEONFORK))
745 retval = copy_page_range(tmp, mpnt);
747 if (tmp->vm_ops && tmp->vm_ops->open)
748 tmp->vm_ops->open(tmp);
751 mpnt = vma_next(&vmi);
755 /* a new mm has just been created */
756 retval = arch_dup_mmap(oldmm, mm);
760 mt_set_in_rcu(vmi.mas.tree);
763 * The entire maple tree has already been duplicated. If the
764 * mmap duplication fails, mark the failure point with
765 * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered,
766 * stop releasing VMAs that have not been duplicated after this
769 mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1);
770 mas_store(&vmi.mas, XA_ZERO_ENTRY);
773 mmap_write_unlock(mm);
775 mmap_write_unlock(oldmm);
776 dup_userfaultfd_complete(&uf);
778 uprobe_end_dup_mmap();
781 fail_nomem_anon_vma_fork:
782 mpol_put(vma_policy(tmp));
787 vm_unacct_memory(charge);
791 static inline int mm_alloc_pgd(struct mm_struct *mm)
793 mm->pgd = pgd_alloc(mm);
794 if (unlikely(!mm->pgd))
799 static inline void mm_free_pgd(struct mm_struct *mm)
801 pgd_free(mm, mm->pgd);
804 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
806 mmap_write_lock(oldmm);
807 dup_mm_exe_file(mm, oldmm);
808 mmap_write_unlock(oldmm);
811 #define mm_alloc_pgd(mm) (0)
812 #define mm_free_pgd(mm)
813 #endif /* CONFIG_MMU */
815 static void check_mm(struct mm_struct *mm)
819 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
820 "Please make sure 'struct resident_page_types[]' is updated as well");
822 for (i = 0; i < NR_MM_COUNTERS; i++) {
823 long x = percpu_counter_sum(&mm->rss_stat[i]);
826 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
827 mm, resident_page_types[i], x);
830 if (mm_pgtables_bytes(mm))
831 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
832 mm_pgtables_bytes(mm));
834 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
835 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
839 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
840 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
842 static void do_check_lazy_tlb(void *arg)
844 struct mm_struct *mm = arg;
846 WARN_ON_ONCE(current->active_mm == mm);
849 static void do_shoot_lazy_tlb(void *arg)
851 struct mm_struct *mm = arg;
853 if (current->active_mm == mm) {
854 WARN_ON_ONCE(current->mm);
855 current->active_mm = &init_mm;
856 switch_mm(mm, &init_mm, current);
860 static void cleanup_lazy_tlbs(struct mm_struct *mm)
862 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
864 * In this case, lazy tlb mms are refounted and would not reach
865 * __mmdrop until all CPUs have switched away and mmdrop()ed.
871 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
872 * requires lazy mm users to switch to another mm when the refcount
873 * drops to zero, before the mm is freed. This requires IPIs here to
874 * switch kernel threads to init_mm.
876 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
877 * switch with the final userspace teardown TLB flush which leaves the
878 * mm lazy on this CPU but no others, reducing the need for additional
879 * IPIs here. There are cases where a final IPI is still required here,
880 * such as the final mmdrop being performed on a different CPU than the
881 * one exiting, or kernel threads using the mm when userspace exits.
883 * IPI overheads have not found to be expensive, but they could be
884 * reduced in a number of possible ways, for example (roughly
885 * increasing order of complexity):
886 * - The last lazy reference created by exit_mm() could instead switch
887 * to init_mm, however it's probable this will run on the same CPU
888 * immediately afterwards, so this may not reduce IPIs much.
889 * - A batch of mms requiring IPIs could be gathered and freed at once.
890 * - CPUs store active_mm where it can be remotely checked without a
891 * lock, to filter out false-positives in the cpumask.
892 * - After mm_users or mm_count reaches zero, switching away from the
893 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
894 * with some batching or delaying of the final IPIs.
895 * - A delayed freeing and RCU-like quiescing sequence based on mm
896 * switching to avoid IPIs completely.
898 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
899 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
900 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
904 * Called when the last reference to the mm
905 * is dropped: either by a lazy thread or by
906 * mmput. Free the page directory and the mm.
908 void __mmdrop(struct mm_struct *mm)
910 BUG_ON(mm == &init_mm);
911 WARN_ON_ONCE(mm == current->mm);
913 /* Ensure no CPUs are using this as their lazy tlb mm */
914 cleanup_lazy_tlbs(mm);
916 WARN_ON_ONCE(mm == current->active_mm);
919 mmu_notifier_subscriptions_destroy(mm);
921 put_user_ns(mm->user_ns);
924 percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
928 EXPORT_SYMBOL_GPL(__mmdrop);
930 static void mmdrop_async_fn(struct work_struct *work)
932 struct mm_struct *mm;
934 mm = container_of(work, struct mm_struct, async_put_work);
938 static void mmdrop_async(struct mm_struct *mm)
940 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
941 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
942 schedule_work(&mm->async_put_work);
946 static inline void free_signal_struct(struct signal_struct *sig)
948 taskstats_tgid_free(sig);
949 sched_autogroup_exit(sig);
951 * __mmdrop is not safe to call from softirq context on x86 due to
952 * pgd_dtor so postpone it to the async context
955 mmdrop_async(sig->oom_mm);
956 kmem_cache_free(signal_cachep, sig);
959 static inline void put_signal_struct(struct signal_struct *sig)
961 if (refcount_dec_and_test(&sig->sigcnt))
962 free_signal_struct(sig);
965 void __put_task_struct(struct task_struct *tsk)
967 WARN_ON(!tsk->exit_state);
968 WARN_ON(refcount_read(&tsk->usage));
969 WARN_ON(tsk == current);
973 task_numa_free(tsk, true);
974 security_task_free(tsk);
976 delayacct_tsk_free(tsk);
977 put_signal_struct(tsk->signal);
978 sched_core_free(tsk);
981 EXPORT_SYMBOL_GPL(__put_task_struct);
983 void __put_task_struct_rcu_cb(struct rcu_head *rhp)
985 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
987 __put_task_struct(task);
989 EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
991 void __init __weak arch_task_cache_init(void) { }
996 static void set_max_threads(unsigned int max_threads_suggested)
999 unsigned long nr_pages = totalram_pages();
1002 * The number of threads shall be limited such that the thread
1003 * structures may only consume a small part of the available memory.
1005 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1006 threads = MAX_THREADS;
1008 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1009 (u64) THREAD_SIZE * 8UL);
1011 if (threads > max_threads_suggested)
1012 threads = max_threads_suggested;
1014 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1017 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1018 /* Initialized by the architecture: */
1019 int arch_task_struct_size __read_mostly;
1022 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1024 /* Fetch thread_struct whitelist for the architecture. */
1025 arch_thread_struct_whitelist(offset, size);
1028 * Handle zero-sized whitelist or empty thread_struct, otherwise
1029 * adjust offset to position of thread_struct in task_struct.
1031 if (unlikely(*size == 0))
1034 *offset += offsetof(struct task_struct, thread);
1037 void __init fork_init(void)
1040 #ifndef ARCH_MIN_TASKALIGN
1041 #define ARCH_MIN_TASKALIGN 0
1043 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1044 unsigned long useroffset, usersize;
1046 /* create a slab on which task_structs can be allocated */
1047 task_struct_whitelist(&useroffset, &usersize);
1048 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1049 arch_task_struct_size, align,
1050 SLAB_PANIC|SLAB_ACCOUNT,
1051 useroffset, usersize, NULL);
1053 /* do the arch specific task caches init */
1054 arch_task_cache_init();
1056 set_max_threads(MAX_THREADS);
1058 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1059 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1060 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1061 init_task.signal->rlim[RLIMIT_NPROC];
1063 for (i = 0; i < UCOUNT_COUNTS; i++)
1064 init_user_ns.ucount_max[i] = max_threads/2;
1066 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1067 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1068 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1069 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1071 #ifdef CONFIG_VMAP_STACK
1072 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1073 NULL, free_vm_stack_cache);
1078 lockdep_init_task(&init_task);
1082 int __weak arch_dup_task_struct(struct task_struct *dst,
1083 struct task_struct *src)
1089 void set_task_stack_end_magic(struct task_struct *tsk)
1091 unsigned long *stackend;
1093 stackend = end_of_stack(tsk);
1094 *stackend = STACK_END_MAGIC; /* for overflow detection */
1097 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1099 struct task_struct *tsk;
1102 if (node == NUMA_NO_NODE)
1103 node = tsk_fork_get_node(orig);
1104 tsk = alloc_task_struct_node(node);
1108 err = arch_dup_task_struct(tsk, orig);
1112 err = alloc_thread_stack_node(tsk, node);
1116 #ifdef CONFIG_THREAD_INFO_IN_TASK
1117 refcount_set(&tsk->stack_refcount, 1);
1119 account_kernel_stack(tsk, 1);
1121 err = scs_prepare(tsk, node);
1125 #ifdef CONFIG_SECCOMP
1127 * We must handle setting up seccomp filters once we're under
1128 * the sighand lock in case orig has changed between now and
1129 * then. Until then, filter must be NULL to avoid messing up
1130 * the usage counts on the error path calling free_task.
1132 tsk->seccomp.filter = NULL;
1135 setup_thread_stack(tsk, orig);
1136 clear_user_return_notifier(tsk);
1137 clear_tsk_need_resched(tsk);
1138 set_task_stack_end_magic(tsk);
1139 clear_syscall_work_syscall_user_dispatch(tsk);
1141 #ifdef CONFIG_STACKPROTECTOR
1142 tsk->stack_canary = get_random_canary();
1144 if (orig->cpus_ptr == &orig->cpus_mask)
1145 tsk->cpus_ptr = &tsk->cpus_mask;
1146 dup_user_cpus_ptr(tsk, orig, node);
1149 * One for the user space visible state that goes away when reaped.
1150 * One for the scheduler.
1152 refcount_set(&tsk->rcu_users, 2);
1153 /* One for the rcu users */
1154 refcount_set(&tsk->usage, 1);
1155 #ifdef CONFIG_BLK_DEV_IO_TRACE
1156 tsk->btrace_seq = 0;
1158 tsk->splice_pipe = NULL;
1159 tsk->task_frag.page = NULL;
1160 tsk->wake_q.next = NULL;
1161 tsk->worker_private = NULL;
1163 kcov_task_init(tsk);
1164 kmsan_task_create(tsk);
1165 kmap_local_fork(tsk);
1167 #ifdef CONFIG_FAULT_INJECTION
1171 #ifdef CONFIG_BLK_CGROUP
1172 tsk->throttle_disk = NULL;
1173 tsk->use_memdelay = 0;
1176 #ifdef CONFIG_ARCH_HAS_CPU_PASID
1177 tsk->pasid_activated = 0;
1181 tsk->active_memcg = NULL;
1184 #ifdef CONFIG_CPU_SUP_INTEL
1185 tsk->reported_split_lock = 0;
1188 #ifdef CONFIG_SCHED_MM_CID
1190 tsk->last_mm_cid = -1;
1191 tsk->mm_cid_active = 0;
1192 tsk->migrate_from_cpu = -1;
1197 exit_task_stack_account(tsk);
1198 free_thread_stack(tsk);
1200 free_task_struct(tsk);
1204 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1206 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1208 static int __init coredump_filter_setup(char *s)
1210 default_dump_filter =
1211 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1212 MMF_DUMP_FILTER_MASK;
1216 __setup("coredump_filter=", coredump_filter_setup);
1218 #include <linux/init_task.h>
1220 static void mm_init_aio(struct mm_struct *mm)
1223 spin_lock_init(&mm->ioctx_lock);
1224 mm->ioctx_table = NULL;
1228 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1229 struct task_struct *p)
1233 WRITE_ONCE(mm->owner, NULL);
1237 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1244 static void mm_init_uprobes_state(struct mm_struct *mm)
1246 #ifdef CONFIG_UPROBES
1247 mm->uprobes_state.xol_area = NULL;
1251 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1252 struct user_namespace *user_ns)
1254 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1255 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1256 atomic_set(&mm->mm_users, 1);
1257 atomic_set(&mm->mm_count, 1);
1258 seqcount_init(&mm->write_protect_seq);
1260 INIT_LIST_HEAD(&mm->mmlist);
1261 #ifdef CONFIG_PER_VMA_LOCK
1262 mm->mm_lock_seq = 0;
1264 mm_pgtables_bytes_init(mm);
1267 atomic64_set(&mm->pinned_vm, 0);
1268 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1269 spin_lock_init(&mm->page_table_lock);
1270 spin_lock_init(&mm->arg_lock);
1271 mm_init_cpumask(mm);
1273 mm_init_owner(mm, p);
1275 RCU_INIT_POINTER(mm->exe_file, NULL);
1276 mmu_notifier_subscriptions_init(mm);
1277 init_tlb_flush_pending(mm);
1278 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1279 mm->pmd_huge_pte = NULL;
1281 mm_init_uprobes_state(mm);
1282 hugetlb_count_init(mm);
1285 mm->flags = mmf_init_flags(current->mm->flags);
1286 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1288 mm->flags = default_dump_filter;
1292 if (mm_alloc_pgd(mm))
1295 if (init_new_context(p, mm))
1296 goto fail_nocontext;
1298 if (mm_alloc_cid(mm))
1301 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1305 mm->user_ns = get_user_ns(user_ns);
1306 lru_gen_init_mm(mm);
1312 destroy_context(mm);
1321 * Allocate and initialize an mm_struct.
1323 struct mm_struct *mm_alloc(void)
1325 struct mm_struct *mm;
1331 memset(mm, 0, sizeof(*mm));
1332 return mm_init(mm, current, current_user_ns());
1335 static inline void __mmput(struct mm_struct *mm)
1337 VM_BUG_ON(atomic_read(&mm->mm_users));
1339 uprobe_clear_state(mm);
1342 khugepaged_exit(mm); /* must run before exit_mmap */
1344 mm_put_huge_zero_page(mm);
1345 set_mm_exe_file(mm, NULL);
1346 if (!list_empty(&mm->mmlist)) {
1347 spin_lock(&mmlist_lock);
1348 list_del(&mm->mmlist);
1349 spin_unlock(&mmlist_lock);
1352 module_put(mm->binfmt->module);
1358 * Decrement the use count and release all resources for an mm.
1360 void mmput(struct mm_struct *mm)
1364 if (atomic_dec_and_test(&mm->mm_users))
1367 EXPORT_SYMBOL_GPL(mmput);
1370 static void mmput_async_fn(struct work_struct *work)
1372 struct mm_struct *mm = container_of(work, struct mm_struct,
1378 void mmput_async(struct mm_struct *mm)
1380 if (atomic_dec_and_test(&mm->mm_users)) {
1381 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1382 schedule_work(&mm->async_put_work);
1385 EXPORT_SYMBOL_GPL(mmput_async);
1389 * set_mm_exe_file - change a reference to the mm's executable file
1390 * @mm: The mm to change.
1391 * @new_exe_file: The new file to use.
1393 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1395 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1396 * invocations: in mmput() nobody alive left, in execve it happens before
1397 * the new mm is made visible to anyone.
1399 * Can only fail if new_exe_file != NULL.
1401 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1403 struct file *old_exe_file;
1406 * It is safe to dereference the exe_file without RCU as
1407 * this function is only called if nobody else can access
1408 * this mm -- see comment above for justification.
1410 old_exe_file = rcu_dereference_raw(mm->exe_file);
1414 * We expect the caller (i.e., sys_execve) to already denied
1415 * write access, so this is unlikely to fail.
1417 if (unlikely(deny_write_access(new_exe_file)))
1419 get_file(new_exe_file);
1421 rcu_assign_pointer(mm->exe_file, new_exe_file);
1423 allow_write_access(old_exe_file);
1430 * replace_mm_exe_file - replace a reference to the mm's executable file
1431 * @mm: The mm to change.
1432 * @new_exe_file: The new file to use.
1434 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1436 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1438 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1440 struct vm_area_struct *vma;
1441 struct file *old_exe_file;
1444 /* Forbid mm->exe_file change if old file still mapped. */
1445 old_exe_file = get_mm_exe_file(mm);
1447 VMA_ITERATOR(vmi, mm, 0);
1449 for_each_vma(vmi, vma) {
1452 if (path_equal(&vma->vm_file->f_path,
1453 &old_exe_file->f_path)) {
1458 mmap_read_unlock(mm);
1464 ret = deny_write_access(new_exe_file);
1467 get_file(new_exe_file);
1469 /* set the new file */
1470 mmap_write_lock(mm);
1471 old_exe_file = rcu_dereference_raw(mm->exe_file);
1472 rcu_assign_pointer(mm->exe_file, new_exe_file);
1473 mmap_write_unlock(mm);
1476 allow_write_access(old_exe_file);
1483 * get_mm_exe_file - acquire a reference to the mm's executable file
1484 * @mm: The mm of interest.
1486 * Returns %NULL if mm has no associated executable file.
1487 * User must release file via fput().
1489 struct file *get_mm_exe_file(struct mm_struct *mm)
1491 struct file *exe_file;
1494 exe_file = get_file_rcu(&mm->exe_file);
1500 * get_task_exe_file - acquire a reference to the task's executable file
1503 * Returns %NULL if task's mm (if any) has no associated executable file or
1504 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1505 * User must release file via fput().
1507 struct file *get_task_exe_file(struct task_struct *task)
1509 struct file *exe_file = NULL;
1510 struct mm_struct *mm;
1515 if (!(task->flags & PF_KTHREAD))
1516 exe_file = get_mm_exe_file(mm);
1523 * get_task_mm - acquire a reference to the task's mm
1526 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1527 * this kernel workthread has transiently adopted a user mm with use_mm,
1528 * to do its AIO) is not set and if so returns a reference to it, after
1529 * bumping up the use count. User must release the mm via mmput()
1530 * after use. Typically used by /proc and ptrace.
1532 struct mm_struct *get_task_mm(struct task_struct *task)
1534 struct mm_struct *mm;
1539 if (task->flags & PF_KTHREAD)
1547 EXPORT_SYMBOL_GPL(get_task_mm);
1549 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1551 struct mm_struct *mm;
1554 err = down_read_killable(&task->signal->exec_update_lock);
1556 return ERR_PTR(err);
1558 mm = get_task_mm(task);
1559 if (mm && mm != current->mm &&
1560 !ptrace_may_access(task, mode)) {
1562 mm = ERR_PTR(-EACCES);
1564 up_read(&task->signal->exec_update_lock);
1569 static void complete_vfork_done(struct task_struct *tsk)
1571 struct completion *vfork;
1574 vfork = tsk->vfork_done;
1575 if (likely(vfork)) {
1576 tsk->vfork_done = NULL;
1582 static int wait_for_vfork_done(struct task_struct *child,
1583 struct completion *vfork)
1585 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1588 cgroup_enter_frozen();
1589 killed = wait_for_completion_state(vfork, state);
1590 cgroup_leave_frozen(false);
1594 child->vfork_done = NULL;
1598 put_task_struct(child);
1602 /* Please note the differences between mmput and mm_release.
1603 * mmput is called whenever we stop holding onto a mm_struct,
1604 * error success whatever.
1606 * mm_release is called after a mm_struct has been removed
1607 * from the current process.
1609 * This difference is important for error handling, when we
1610 * only half set up a mm_struct for a new process and need to restore
1611 * the old one. Because we mmput the new mm_struct before
1612 * restoring the old one. . .
1613 * Eric Biederman 10 January 1998
1615 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1617 uprobe_free_utask(tsk);
1619 /* Get rid of any cached register state */
1620 deactivate_mm(tsk, mm);
1623 * Signal userspace if we're not exiting with a core dump
1624 * because we want to leave the value intact for debugging
1627 if (tsk->clear_child_tid) {
1628 if (atomic_read(&mm->mm_users) > 1) {
1630 * We don't check the error code - if userspace has
1631 * not set up a proper pointer then tough luck.
1633 put_user(0, tsk->clear_child_tid);
1634 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1635 1, NULL, NULL, 0, 0);
1637 tsk->clear_child_tid = NULL;
1641 * All done, finally we can wake up parent and return this mm to him.
1642 * Also kthread_stop() uses this completion for synchronization.
1644 if (tsk->vfork_done)
1645 complete_vfork_done(tsk);
1648 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1650 futex_exit_release(tsk);
1651 mm_release(tsk, mm);
1654 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1656 futex_exec_release(tsk);
1657 mm_release(tsk, mm);
1661 * dup_mm() - duplicates an existing mm structure
1662 * @tsk: the task_struct with which the new mm will be associated.
1663 * @oldmm: the mm to duplicate.
1665 * Allocates a new mm structure and duplicates the provided @oldmm structure
1668 * Return: the duplicated mm or NULL on failure.
1670 static struct mm_struct *dup_mm(struct task_struct *tsk,
1671 struct mm_struct *oldmm)
1673 struct mm_struct *mm;
1680 memcpy(mm, oldmm, sizeof(*mm));
1682 if (!mm_init(mm, tsk, mm->user_ns))
1685 err = dup_mmap(mm, oldmm);
1689 mm->hiwater_rss = get_mm_rss(mm);
1690 mm->hiwater_vm = mm->total_vm;
1692 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1698 /* don't put binfmt in mmput, we haven't got module yet */
1700 mm_init_owner(mm, NULL);
1707 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1709 struct mm_struct *mm, *oldmm;
1711 tsk->min_flt = tsk->maj_flt = 0;
1712 tsk->nvcsw = tsk->nivcsw = 0;
1713 #ifdef CONFIG_DETECT_HUNG_TASK
1714 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1715 tsk->last_switch_time = 0;
1719 tsk->active_mm = NULL;
1722 * Are we cloning a kernel thread?
1724 * We need to steal a active VM for that..
1726 oldmm = current->mm;
1730 if (clone_flags & CLONE_VM) {
1734 mm = dup_mm(tsk, current->mm);
1740 tsk->active_mm = mm;
1741 sched_mm_cid_fork(tsk);
1745 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1747 struct fs_struct *fs = current->fs;
1748 if (clone_flags & CLONE_FS) {
1749 /* tsk->fs is already what we want */
1750 spin_lock(&fs->lock);
1752 spin_unlock(&fs->lock);
1756 spin_unlock(&fs->lock);
1759 tsk->fs = copy_fs_struct(fs);
1765 static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1768 struct files_struct *oldf, *newf;
1772 * A background process may not have any files ...
1774 oldf = current->files;
1783 if (clone_flags & CLONE_FILES) {
1784 atomic_inc(&oldf->count);
1788 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1798 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1800 struct sighand_struct *sig;
1802 if (clone_flags & CLONE_SIGHAND) {
1803 refcount_inc(¤t->sighand->count);
1806 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1807 RCU_INIT_POINTER(tsk->sighand, sig);
1811 refcount_set(&sig->count, 1);
1812 spin_lock_irq(¤t->sighand->siglock);
1813 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1814 spin_unlock_irq(¤t->sighand->siglock);
1816 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1817 if (clone_flags & CLONE_CLEAR_SIGHAND)
1818 flush_signal_handlers(tsk, 0);
1823 void __cleanup_sighand(struct sighand_struct *sighand)
1825 if (refcount_dec_and_test(&sighand->count)) {
1826 signalfd_cleanup(sighand);
1828 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1829 * without an RCU grace period, see __lock_task_sighand().
1831 kmem_cache_free(sighand_cachep, sighand);
1836 * Initialize POSIX timer handling for a thread group.
1838 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1840 struct posix_cputimers *pct = &sig->posix_cputimers;
1841 unsigned long cpu_limit;
1843 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1844 posix_cputimers_group_init(pct, cpu_limit);
1847 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1849 struct signal_struct *sig;
1851 if (clone_flags & CLONE_THREAD)
1854 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1859 sig->nr_threads = 1;
1860 sig->quick_threads = 1;
1861 atomic_set(&sig->live, 1);
1862 refcount_set(&sig->sigcnt, 1);
1864 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1865 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1866 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1868 init_waitqueue_head(&sig->wait_chldexit);
1869 sig->curr_target = tsk;
1870 init_sigpending(&sig->shared_pending);
1871 INIT_HLIST_HEAD(&sig->multiprocess);
1872 seqlock_init(&sig->stats_lock);
1873 prev_cputime_init(&sig->prev_cputime);
1875 #ifdef CONFIG_POSIX_TIMERS
1876 INIT_LIST_HEAD(&sig->posix_timers);
1877 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1878 sig->real_timer.function = it_real_fn;
1881 task_lock(current->group_leader);
1882 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1883 task_unlock(current->group_leader);
1885 posix_cpu_timers_init_group(sig);
1887 tty_audit_fork(sig);
1888 sched_autogroup_fork(sig);
1890 sig->oom_score_adj = current->signal->oom_score_adj;
1891 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1893 mutex_init(&sig->cred_guard_mutex);
1894 init_rwsem(&sig->exec_update_lock);
1899 static void copy_seccomp(struct task_struct *p)
1901 #ifdef CONFIG_SECCOMP
1903 * Must be called with sighand->lock held, which is common to
1904 * all threads in the group. Holding cred_guard_mutex is not
1905 * needed because this new task is not yet running and cannot
1908 assert_spin_locked(¤t->sighand->siglock);
1910 /* Ref-count the new filter user, and assign it. */
1911 get_seccomp_filter(current);
1912 p->seccomp = current->seccomp;
1915 * Explicitly enable no_new_privs here in case it got set
1916 * between the task_struct being duplicated and holding the
1917 * sighand lock. The seccomp state and nnp must be in sync.
1919 if (task_no_new_privs(current))
1920 task_set_no_new_privs(p);
1923 * If the parent gained a seccomp mode after copying thread
1924 * flags and between before we held the sighand lock, we have
1925 * to manually enable the seccomp thread flag here.
1927 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1928 set_task_syscall_work(p, SECCOMP);
1932 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1934 current->clear_child_tid = tidptr;
1936 return task_pid_vnr(current);
1939 static void rt_mutex_init_task(struct task_struct *p)
1941 raw_spin_lock_init(&p->pi_lock);
1942 #ifdef CONFIG_RT_MUTEXES
1943 p->pi_waiters = RB_ROOT_CACHED;
1944 p->pi_top_task = NULL;
1945 p->pi_blocked_on = NULL;
1949 static inline void init_task_pid_links(struct task_struct *task)
1953 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1954 INIT_HLIST_NODE(&task->pid_links[type]);
1958 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1960 if (type == PIDTYPE_PID)
1961 task->thread_pid = pid;
1963 task->signal->pids[type] = pid;
1966 static inline void rcu_copy_process(struct task_struct *p)
1968 #ifdef CONFIG_PREEMPT_RCU
1969 p->rcu_read_lock_nesting = 0;
1970 p->rcu_read_unlock_special.s = 0;
1971 p->rcu_blocked_node = NULL;
1972 INIT_LIST_HEAD(&p->rcu_node_entry);
1973 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1974 #ifdef CONFIG_TASKS_RCU
1975 p->rcu_tasks_holdout = false;
1976 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1977 p->rcu_tasks_idle_cpu = -1;
1978 #endif /* #ifdef CONFIG_TASKS_RCU */
1979 #ifdef CONFIG_TASKS_TRACE_RCU
1980 p->trc_reader_nesting = 0;
1981 p->trc_reader_special.s = 0;
1982 INIT_LIST_HEAD(&p->trc_holdout_list);
1983 INIT_LIST_HEAD(&p->trc_blkd_node);
1984 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1987 struct pid *pidfd_pid(const struct file *file)
1989 if (file->f_op == &pidfd_fops)
1990 return file->private_data;
1992 return ERR_PTR(-EBADF);
1995 static int pidfd_release(struct inode *inode, struct file *file)
1997 struct pid *pid = file->private_data;
1999 file->private_data = NULL;
2004 #ifdef CONFIG_PROC_FS
2006 * pidfd_show_fdinfo - print information about a pidfd
2007 * @m: proc fdinfo file
2008 * @f: file referencing a pidfd
2011 * This function will print the pid that a given pidfd refers to in the
2012 * pid namespace of the procfs instance.
2013 * If the pid namespace of the process is not a descendant of the pid
2014 * namespace of the procfs instance 0 will be shown as its pid. This is
2015 * similar to calling getppid() on a process whose parent is outside of
2016 * its pid namespace.
2019 * If pid namespaces are supported then this function will also print
2020 * the pid of a given pidfd refers to for all descendant pid namespaces
2021 * starting from the current pid namespace of the instance, i.e. the
2022 * Pid field and the first entry in the NSpid field will be identical.
2023 * If the pid namespace of the process is not a descendant of the pid
2024 * namespace of the procfs instance 0 will be shown as its first NSpid
2025 * entry and no others will be shown.
2026 * Note that this differs from the Pid and NSpid fields in
2027 * /proc/<pid>/status where Pid and NSpid are always shown relative to
2028 * the pid namespace of the procfs instance. The difference becomes
2029 * obvious when sending around a pidfd between pid namespaces from a
2030 * different branch of the tree, i.e. where no ancestral relation is
2031 * present between the pid namespaces:
2032 * - create two new pid namespaces ns1 and ns2 in the initial pid
2033 * namespace (also take care to create new mount namespaces in the
2034 * new pid namespace and mount procfs)
2035 * - create a process with a pidfd in ns1
2036 * - send pidfd from ns1 to ns2
2037 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
2038 * have exactly one entry, which is 0
2040 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
2042 struct pid *pid = f->private_data;
2043 struct pid_namespace *ns;
2046 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
2047 ns = proc_pid_ns(file_inode(m->file)->i_sb);
2048 nr = pid_nr_ns(pid, ns);
2051 seq_put_decimal_ll(m, "Pid:\t", nr);
2053 #ifdef CONFIG_PID_NS
2054 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
2058 /* If nr is non-zero it means that 'pid' is valid and that
2059 * ns, i.e. the pid namespace associated with the procfs
2060 * instance, is in the pid namespace hierarchy of pid.
2061 * Start at one below the already printed level.
2063 for (i = ns->level + 1; i <= pid->level; i++)
2064 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2072 * Poll support for process exit notification.
2074 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2076 struct pid *pid = file->private_data;
2077 __poll_t poll_flags = 0;
2079 poll_wait(file, &pid->wait_pidfd, pts);
2082 * Inform pollers only when the whole thread group exits.
2083 * If the thread group leader exits before all other threads in the
2084 * group, then poll(2) should block, similar to the wait(2) family.
2086 if (thread_group_exited(pid))
2087 poll_flags = EPOLLIN | EPOLLRDNORM;
2092 const struct file_operations pidfd_fops = {
2093 .release = pidfd_release,
2095 #ifdef CONFIG_PROC_FS
2096 .show_fdinfo = pidfd_show_fdinfo,
2101 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2102 * @pid: the struct pid for which to create a pidfd
2103 * @flags: flags of the new @pidfd
2104 * @ret: Where to return the file for the pidfd.
2106 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2107 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2109 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2110 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2111 * pidfd file are prepared.
2113 * If this function returns successfully the caller is responsible to either
2114 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2115 * order to install the pidfd into its file descriptor table or they must use
2116 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2119 * This function is useful when a pidfd must already be reserved but there
2120 * might still be points of failure afterwards and the caller wants to ensure
2121 * that no pidfd is leaked into its file descriptor table.
2123 * Return: On success, a reserved pidfd is returned from the function and a new
2124 * pidfd file is returned in the last argument to the function. On
2125 * error, a negative error code is returned from the function and the
2126 * last argument remains unchanged.
2128 static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2131 struct file *pidfd_file;
2133 if (flags & ~(O_NONBLOCK | O_RDWR | O_CLOEXEC))
2136 pidfd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2140 pidfd_file = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2141 flags | O_RDWR | O_CLOEXEC);
2142 if (IS_ERR(pidfd_file)) {
2143 put_unused_fd(pidfd);
2144 return PTR_ERR(pidfd_file);
2146 get_pid(pid); /* held by pidfd_file now */
2152 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2153 * @pid: the struct pid for which to create a pidfd
2154 * @flags: flags of the new @pidfd
2155 * @ret: Where to return the pidfd.
2157 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2158 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2160 * The helper verifies that @pid is used as a thread group leader.
2162 * If this function returns successfully the caller is responsible to either
2163 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2164 * order to install the pidfd into its file descriptor table or they must use
2165 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2168 * This function is useful when a pidfd must already be reserved but there
2169 * might still be points of failure afterwards and the caller wants to ensure
2170 * that no pidfd is leaked into its file descriptor table.
2172 * Return: On success, a reserved pidfd is returned from the function and a new
2173 * pidfd file is returned in the last argument to the function. On
2174 * error, a negative error code is returned from the function and the
2175 * last argument remains unchanged.
2177 int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2179 if (!pid || !pid_has_task(pid, PIDTYPE_TGID))
2182 return __pidfd_prepare(pid, flags, ret);
2185 static void __delayed_free_task(struct rcu_head *rhp)
2187 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2192 static __always_inline void delayed_free_task(struct task_struct *tsk)
2194 if (IS_ENABLED(CONFIG_MEMCG))
2195 call_rcu(&tsk->rcu, __delayed_free_task);
2200 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2202 /* Skip if kernel thread */
2206 /* Skip if spawning a thread or using vfork */
2207 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2210 /* We need to synchronize with __set_oom_adj */
2211 mutex_lock(&oom_adj_mutex);
2212 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2213 /* Update the values in case they were changed after copy_signal */
2214 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2215 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2216 mutex_unlock(&oom_adj_mutex);
2220 static void rv_task_fork(struct task_struct *p)
2224 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2225 p->rv[i].da_mon.monitoring = false;
2228 #define rv_task_fork(p) do {} while (0)
2232 * This creates a new process as a copy of the old one,
2233 * but does not actually start it yet.
2235 * It copies the registers, and all the appropriate
2236 * parts of the process environment (as per the clone
2237 * flags). The actual kick-off is left to the caller.
2239 __latent_entropy struct task_struct *copy_process(
2243 struct kernel_clone_args *args)
2245 int pidfd = -1, retval;
2246 struct task_struct *p;
2247 struct multiprocess_signals delayed;
2248 struct file *pidfile = NULL;
2249 const u64 clone_flags = args->flags;
2250 struct nsproxy *nsp = current->nsproxy;
2253 * Don't allow sharing the root directory with processes in a different
2256 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2257 return ERR_PTR(-EINVAL);
2259 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2260 return ERR_PTR(-EINVAL);
2263 * Thread groups must share signals as well, and detached threads
2264 * can only be started up within the thread group.
2266 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2267 return ERR_PTR(-EINVAL);
2270 * Shared signal handlers imply shared VM. By way of the above,
2271 * thread groups also imply shared VM. Blocking this case allows
2272 * for various simplifications in other code.
2274 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2275 return ERR_PTR(-EINVAL);
2278 * Siblings of global init remain as zombies on exit since they are
2279 * not reaped by their parent (swapper). To solve this and to avoid
2280 * multi-rooted process trees, prevent global and container-inits
2281 * from creating siblings.
2283 if ((clone_flags & CLONE_PARENT) &&
2284 current->signal->flags & SIGNAL_UNKILLABLE)
2285 return ERR_PTR(-EINVAL);
2288 * If the new process will be in a different pid or user namespace
2289 * do not allow it to share a thread group with the forking task.
2291 if (clone_flags & CLONE_THREAD) {
2292 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2293 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2294 return ERR_PTR(-EINVAL);
2297 if (clone_flags & CLONE_PIDFD) {
2299 * - CLONE_DETACHED is blocked so that we can potentially
2300 * reuse it later for CLONE_PIDFD.
2301 * - CLONE_THREAD is blocked until someone really needs it.
2303 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2304 return ERR_PTR(-EINVAL);
2308 * Force any signals received before this point to be delivered
2309 * before the fork happens. Collect up signals sent to multiple
2310 * processes that happen during the fork and delay them so that
2311 * they appear to happen after the fork.
2313 sigemptyset(&delayed.signal);
2314 INIT_HLIST_NODE(&delayed.node);
2316 spin_lock_irq(¤t->sighand->siglock);
2317 if (!(clone_flags & CLONE_THREAD))
2318 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2319 recalc_sigpending();
2320 spin_unlock_irq(¤t->sighand->siglock);
2321 retval = -ERESTARTNOINTR;
2322 if (task_sigpending(current))
2326 p = dup_task_struct(current, node);
2329 p->flags &= ~PF_KTHREAD;
2331 p->flags |= PF_KTHREAD;
2332 if (args->user_worker) {
2334 * Mark us a user worker, and block any signal that isn't
2337 p->flags |= PF_USER_WORKER;
2338 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2340 if (args->io_thread)
2341 p->flags |= PF_IO_WORKER;
2344 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2346 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2348 * Clear TID on mm_release()?
2350 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2352 ftrace_graph_init_task(p);
2354 rt_mutex_init_task(p);
2356 lockdep_assert_irqs_enabled();
2357 #ifdef CONFIG_PROVE_LOCKING
2358 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2360 retval = copy_creds(p, clone_flags);
2365 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2366 if (p->real_cred->user != INIT_USER &&
2367 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2368 goto bad_fork_cleanup_count;
2370 current->flags &= ~PF_NPROC_EXCEEDED;
2373 * If multiple threads are within copy_process(), then this check
2374 * triggers too late. This doesn't hurt, the check is only there
2375 * to stop root fork bombs.
2378 if (data_race(nr_threads >= max_threads))
2379 goto bad_fork_cleanup_count;
2381 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2382 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2383 p->flags |= PF_FORKNOEXEC;
2384 INIT_LIST_HEAD(&p->children);
2385 INIT_LIST_HEAD(&p->sibling);
2386 rcu_copy_process(p);
2387 p->vfork_done = NULL;
2388 spin_lock_init(&p->alloc_lock);
2390 init_sigpending(&p->pending);
2392 p->utime = p->stime = p->gtime = 0;
2393 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2394 p->utimescaled = p->stimescaled = 0;
2396 prev_cputime_init(&p->prev_cputime);
2398 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2399 seqcount_init(&p->vtime.seqcount);
2400 p->vtime.starttime = 0;
2401 p->vtime.state = VTIME_INACTIVE;
2404 #ifdef CONFIG_IO_URING
2408 p->default_timer_slack_ns = current->timer_slack_ns;
2414 task_io_accounting_init(&p->ioac);
2415 acct_clear_integrals(p);
2417 posix_cputimers_init(&p->posix_cputimers);
2419 p->io_context = NULL;
2420 audit_set_context(p, NULL);
2422 if (args->kthread) {
2423 if (!set_kthread_struct(p))
2424 goto bad_fork_cleanup_delayacct;
2427 p->mempolicy = mpol_dup(p->mempolicy);
2428 if (IS_ERR(p->mempolicy)) {
2429 retval = PTR_ERR(p->mempolicy);
2430 p->mempolicy = NULL;
2431 goto bad_fork_cleanup_delayacct;
2434 #ifdef CONFIG_CPUSETS
2435 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2436 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2437 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2439 #ifdef CONFIG_TRACE_IRQFLAGS
2440 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2441 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2442 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2443 p->softirqs_enabled = 1;
2444 p->softirq_context = 0;
2447 p->pagefault_disabled = 0;
2449 #ifdef CONFIG_LOCKDEP
2450 lockdep_init_task(p);
2453 #ifdef CONFIG_DEBUG_MUTEXES
2454 p->blocked_on = NULL; /* not blocked yet */
2456 #ifdef CONFIG_BCACHE
2457 p->sequential_io = 0;
2458 p->sequential_io_avg = 0;
2460 #ifdef CONFIG_BPF_SYSCALL
2461 RCU_INIT_POINTER(p->bpf_storage, NULL);
2465 /* Perform scheduler related setup. Assign this task to a CPU. */
2466 retval = sched_fork(clone_flags, p);
2468 goto bad_fork_cleanup_policy;
2470 retval = perf_event_init_task(p, clone_flags);
2472 goto bad_fork_cleanup_policy;
2473 retval = audit_alloc(p);
2475 goto bad_fork_cleanup_perf;
2476 /* copy all the process information */
2478 retval = security_task_alloc(p, clone_flags);
2480 goto bad_fork_cleanup_audit;
2481 retval = copy_semundo(clone_flags, p);
2483 goto bad_fork_cleanup_security;
2484 retval = copy_files(clone_flags, p, args->no_files);
2486 goto bad_fork_cleanup_semundo;
2487 retval = copy_fs(clone_flags, p);
2489 goto bad_fork_cleanup_files;
2490 retval = copy_sighand(clone_flags, p);
2492 goto bad_fork_cleanup_fs;
2493 retval = copy_signal(clone_flags, p);
2495 goto bad_fork_cleanup_sighand;
2496 retval = copy_mm(clone_flags, p);
2498 goto bad_fork_cleanup_signal;
2499 retval = copy_namespaces(clone_flags, p);
2501 goto bad_fork_cleanup_mm;
2502 retval = copy_io(clone_flags, p);
2504 goto bad_fork_cleanup_namespaces;
2505 retval = copy_thread(p, args);
2507 goto bad_fork_cleanup_io;
2509 stackleak_task_init(p);
2511 if (pid != &init_struct_pid) {
2512 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2513 args->set_tid_size);
2515 retval = PTR_ERR(pid);
2516 goto bad_fork_cleanup_thread;
2521 * This has to happen after we've potentially unshared the file
2522 * descriptor table (so that the pidfd doesn't leak into the child
2523 * if the fd table isn't shared).
2525 if (clone_flags & CLONE_PIDFD) {
2526 /* Note that no task has been attached to @pid yet. */
2527 retval = __pidfd_prepare(pid, O_RDWR | O_CLOEXEC, &pidfile);
2529 goto bad_fork_free_pid;
2532 retval = put_user(pidfd, args->pidfd);
2534 goto bad_fork_put_pidfd;
2543 * sigaltstack should be cleared when sharing the same VM
2545 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2549 * Syscall tracing and stepping should be turned off in the
2550 * child regardless of CLONE_PTRACE.
2552 user_disable_single_step(p);
2553 clear_task_syscall_work(p, SYSCALL_TRACE);
2554 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2555 clear_task_syscall_work(p, SYSCALL_EMU);
2557 clear_tsk_latency_tracing(p);
2559 /* ok, now we should be set up.. */
2560 p->pid = pid_nr(pid);
2561 if (clone_flags & CLONE_THREAD) {
2562 p->group_leader = current->group_leader;
2563 p->tgid = current->tgid;
2565 p->group_leader = p;
2570 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2571 p->dirty_paused_when = 0;
2573 p->pdeath_signal = 0;
2574 p->task_works = NULL;
2575 clear_posix_cputimers_work(p);
2577 #ifdef CONFIG_KRETPROBES
2578 p->kretprobe_instances.first = NULL;
2580 #ifdef CONFIG_RETHOOK
2581 p->rethooks.first = NULL;
2585 * Ensure that the cgroup subsystem policies allow the new process to be
2586 * forked. It should be noted that the new process's css_set can be changed
2587 * between here and cgroup_post_fork() if an organisation operation is in
2590 retval = cgroup_can_fork(p, args);
2592 goto bad_fork_put_pidfd;
2595 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2596 * the new task on the correct runqueue. All this *before* the task
2599 * This isn't part of ->can_fork() because while the re-cloning is
2600 * cgroup specific, it unconditionally needs to place the task on a
2603 sched_cgroup_fork(p, args);
2606 * From this point on we must avoid any synchronous user-space
2607 * communication until we take the tasklist-lock. In particular, we do
2608 * not want user-space to be able to predict the process start-time by
2609 * stalling fork(2) after we recorded the start_time but before it is
2610 * visible to the system.
2613 p->start_time = ktime_get_ns();
2614 p->start_boottime = ktime_get_boottime_ns();
2617 * Make it visible to the rest of the system, but dont wake it up yet.
2618 * Need tasklist lock for parent etc handling!
2620 write_lock_irq(&tasklist_lock);
2622 /* CLONE_PARENT re-uses the old parent */
2623 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2624 p->real_parent = current->real_parent;
2625 p->parent_exec_id = current->parent_exec_id;
2626 if (clone_flags & CLONE_THREAD)
2627 p->exit_signal = -1;
2629 p->exit_signal = current->group_leader->exit_signal;
2631 p->real_parent = current;
2632 p->parent_exec_id = current->self_exec_id;
2633 p->exit_signal = args->exit_signal;
2636 klp_copy_process(p);
2640 spin_lock(¤t->sighand->siglock);
2644 rseq_fork(p, clone_flags);
2646 /* Don't start children in a dying pid namespace */
2647 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2649 goto bad_fork_cancel_cgroup;
2652 /* Let kill terminate clone/fork in the middle */
2653 if (fatal_signal_pending(current)) {
2655 goto bad_fork_cancel_cgroup;
2658 /* No more failure paths after this point. */
2661 * Copy seccomp details explicitly here, in case they were changed
2662 * before holding sighand lock.
2666 init_task_pid_links(p);
2667 if (likely(p->pid)) {
2668 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2670 init_task_pid(p, PIDTYPE_PID, pid);
2671 if (thread_group_leader(p)) {
2672 init_task_pid(p, PIDTYPE_TGID, pid);
2673 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2674 init_task_pid(p, PIDTYPE_SID, task_session(current));
2676 if (is_child_reaper(pid)) {
2677 ns_of_pid(pid)->child_reaper = p;
2678 p->signal->flags |= SIGNAL_UNKILLABLE;
2680 p->signal->shared_pending.signal = delayed.signal;
2681 p->signal->tty = tty_kref_get(current->signal->tty);
2683 * Inherit has_child_subreaper flag under the same
2684 * tasklist_lock with adding child to the process tree
2685 * for propagate_has_child_subreaper optimization.
2687 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2688 p->real_parent->signal->is_child_subreaper;
2689 list_add_tail(&p->sibling, &p->real_parent->children);
2690 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2691 attach_pid(p, PIDTYPE_TGID);
2692 attach_pid(p, PIDTYPE_PGID);
2693 attach_pid(p, PIDTYPE_SID);
2694 __this_cpu_inc(process_counts);
2696 current->signal->nr_threads++;
2697 current->signal->quick_threads++;
2698 atomic_inc(¤t->signal->live);
2699 refcount_inc(¤t->signal->sigcnt);
2700 task_join_group_stop(p);
2701 list_add_tail_rcu(&p->thread_node,
2702 &p->signal->thread_head);
2704 attach_pid(p, PIDTYPE_PID);
2708 hlist_del_init(&delayed.node);
2709 spin_unlock(¤t->sighand->siglock);
2710 syscall_tracepoint_update(p);
2711 write_unlock_irq(&tasklist_lock);
2714 fd_install(pidfd, pidfile);
2716 proc_fork_connector(p);
2718 cgroup_post_fork(p, args);
2721 trace_task_newtask(p, clone_flags);
2722 uprobe_copy_process(p, clone_flags);
2723 user_events_fork(p, clone_flags);
2725 copy_oom_score_adj(clone_flags, p);
2729 bad_fork_cancel_cgroup:
2731 spin_unlock(¤t->sighand->siglock);
2732 write_unlock_irq(&tasklist_lock);
2733 cgroup_cancel_fork(p, args);
2735 if (clone_flags & CLONE_PIDFD) {
2737 put_unused_fd(pidfd);
2740 if (pid != &init_struct_pid)
2742 bad_fork_cleanup_thread:
2744 bad_fork_cleanup_io:
2747 bad_fork_cleanup_namespaces:
2748 exit_task_namespaces(p);
2749 bad_fork_cleanup_mm:
2751 mm_clear_owner(p->mm, p);
2754 bad_fork_cleanup_signal:
2755 if (!(clone_flags & CLONE_THREAD))
2756 free_signal_struct(p->signal);
2757 bad_fork_cleanup_sighand:
2758 __cleanup_sighand(p->sighand);
2759 bad_fork_cleanup_fs:
2760 exit_fs(p); /* blocking */
2761 bad_fork_cleanup_files:
2762 exit_files(p); /* blocking */
2763 bad_fork_cleanup_semundo:
2765 bad_fork_cleanup_security:
2766 security_task_free(p);
2767 bad_fork_cleanup_audit:
2769 bad_fork_cleanup_perf:
2770 perf_event_free_task(p);
2771 bad_fork_cleanup_policy:
2772 lockdep_free_task(p);
2774 mpol_put(p->mempolicy);
2776 bad_fork_cleanup_delayacct:
2777 delayacct_tsk_free(p);
2778 bad_fork_cleanup_count:
2779 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2782 WRITE_ONCE(p->__state, TASK_DEAD);
2783 exit_task_stack_account(p);
2785 delayed_free_task(p);
2787 spin_lock_irq(¤t->sighand->siglock);
2788 hlist_del_init(&delayed.node);
2789 spin_unlock_irq(¤t->sighand->siglock);
2790 return ERR_PTR(retval);
2793 static inline void init_idle_pids(struct task_struct *idle)
2797 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2798 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2799 init_task_pid(idle, type, &init_struct_pid);
2803 static int idle_dummy(void *dummy)
2805 /* This function is never called */
2809 struct task_struct * __init fork_idle(int cpu)
2811 struct task_struct *task;
2812 struct kernel_clone_args args = {
2820 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2821 if (!IS_ERR(task)) {
2822 init_idle_pids(task);
2823 init_idle(task, cpu);
2830 * This is like kernel_clone(), but shaved down and tailored to just
2831 * creating io_uring workers. It returns a created task, or an error pointer.
2832 * The returned task is inactive, and the caller must fire it up through
2833 * wake_up_new_task(p). All signals are blocked in the created task.
2835 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2837 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2839 struct kernel_clone_args args = {
2840 .flags = ((lower_32_bits(flags) | CLONE_VM |
2841 CLONE_UNTRACED) & ~CSIGNAL),
2842 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2849 return copy_process(NULL, 0, node, &args);
2853 * Ok, this is the main fork-routine.
2855 * It copies the process, and if successful kick-starts
2856 * it and waits for it to finish using the VM if required.
2858 * args->exit_signal is expected to be checked for sanity by the caller.
2860 pid_t kernel_clone(struct kernel_clone_args *args)
2862 u64 clone_flags = args->flags;
2863 struct completion vfork;
2865 struct task_struct *p;
2870 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2871 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2872 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2873 * field in struct clone_args and it still doesn't make sense to have
2874 * them both point at the same memory location. Performing this check
2875 * here has the advantage that we don't need to have a separate helper
2876 * to check for legacy clone().
2878 if ((args->flags & CLONE_PIDFD) &&
2879 (args->flags & CLONE_PARENT_SETTID) &&
2880 (args->pidfd == args->parent_tid))
2884 * Determine whether and which event to report to ptracer. When
2885 * called from kernel_thread or CLONE_UNTRACED is explicitly
2886 * requested, no event is reported; otherwise, report if the event
2887 * for the type of forking is enabled.
2889 if (!(clone_flags & CLONE_UNTRACED)) {
2890 if (clone_flags & CLONE_VFORK)
2891 trace = PTRACE_EVENT_VFORK;
2892 else if (args->exit_signal != SIGCHLD)
2893 trace = PTRACE_EVENT_CLONE;
2895 trace = PTRACE_EVENT_FORK;
2897 if (likely(!ptrace_event_enabled(current, trace)))
2901 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2902 add_latent_entropy();
2908 * Do this prior waking up the new thread - the thread pointer
2909 * might get invalid after that point, if the thread exits quickly.
2911 trace_sched_process_fork(current, p);
2913 pid = get_task_pid(p, PIDTYPE_PID);
2916 if (clone_flags & CLONE_PARENT_SETTID)
2917 put_user(nr, args->parent_tid);
2919 if (clone_flags & CLONE_VFORK) {
2920 p->vfork_done = &vfork;
2921 init_completion(&vfork);
2925 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2926 /* lock the task to synchronize with memcg migration */
2928 lru_gen_add_mm(p->mm);
2932 wake_up_new_task(p);
2934 /* forking complete and child started to run, tell ptracer */
2935 if (unlikely(trace))
2936 ptrace_event_pid(trace, pid);
2938 if (clone_flags & CLONE_VFORK) {
2939 if (!wait_for_vfork_done(p, &vfork))
2940 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2948 * Create a kernel thread.
2950 pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2951 unsigned long flags)
2953 struct kernel_clone_args args = {
2954 .flags = ((lower_32_bits(flags) | CLONE_VM |
2955 CLONE_UNTRACED) & ~CSIGNAL),
2956 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2963 return kernel_clone(&args);
2967 * Create a user mode thread.
2969 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2971 struct kernel_clone_args args = {
2972 .flags = ((lower_32_bits(flags) | CLONE_VM |
2973 CLONE_UNTRACED) & ~CSIGNAL),
2974 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2979 return kernel_clone(&args);
2982 #ifdef __ARCH_WANT_SYS_FORK
2983 SYSCALL_DEFINE0(fork)
2986 struct kernel_clone_args args = {
2987 .exit_signal = SIGCHLD,
2990 return kernel_clone(&args);
2992 /* can not support in nommu mode */
2998 #ifdef __ARCH_WANT_SYS_VFORK
2999 SYSCALL_DEFINE0(vfork)
3001 struct kernel_clone_args args = {
3002 .flags = CLONE_VFORK | CLONE_VM,
3003 .exit_signal = SIGCHLD,
3006 return kernel_clone(&args);
3010 #ifdef __ARCH_WANT_SYS_CLONE
3011 #ifdef CONFIG_CLONE_BACKWARDS
3012 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3013 int __user *, parent_tidptr,
3015 int __user *, child_tidptr)
3016 #elif defined(CONFIG_CLONE_BACKWARDS2)
3017 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
3018 int __user *, parent_tidptr,
3019 int __user *, child_tidptr,
3021 #elif defined(CONFIG_CLONE_BACKWARDS3)
3022 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
3024 int __user *, parent_tidptr,
3025 int __user *, child_tidptr,
3028 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3029 int __user *, parent_tidptr,
3030 int __user *, child_tidptr,
3034 struct kernel_clone_args args = {
3035 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
3036 .pidfd = parent_tidptr,
3037 .child_tid = child_tidptr,
3038 .parent_tid = parent_tidptr,
3039 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
3044 return kernel_clone(&args);
3048 #ifdef __ARCH_WANT_SYS_CLONE3
3050 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
3051 struct clone_args __user *uargs,
3055 struct clone_args args;
3056 pid_t *kset_tid = kargs->set_tid;
3058 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
3059 CLONE_ARGS_SIZE_VER0);
3060 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
3061 CLONE_ARGS_SIZE_VER1);
3062 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
3063 CLONE_ARGS_SIZE_VER2);
3064 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
3066 if (unlikely(usize > PAGE_SIZE))
3068 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
3071 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
3075 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
3078 if (unlikely(!args.set_tid && args.set_tid_size > 0))
3081 if (unlikely(args.set_tid && args.set_tid_size == 0))
3085 * Verify that higher 32bits of exit_signal are unset and that
3086 * it is a valid signal
3088 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3089 !valid_signal(args.exit_signal)))
3092 if ((args.flags & CLONE_INTO_CGROUP) &&
3093 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3096 *kargs = (struct kernel_clone_args){
3097 .flags = args.flags,
3098 .pidfd = u64_to_user_ptr(args.pidfd),
3099 .child_tid = u64_to_user_ptr(args.child_tid),
3100 .parent_tid = u64_to_user_ptr(args.parent_tid),
3101 .exit_signal = args.exit_signal,
3102 .stack = args.stack,
3103 .stack_size = args.stack_size,
3105 .set_tid_size = args.set_tid_size,
3106 .cgroup = args.cgroup,
3110 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3111 (kargs->set_tid_size * sizeof(pid_t))))
3114 kargs->set_tid = kset_tid;
3120 * clone3_stack_valid - check and prepare stack
3121 * @kargs: kernel clone args
3123 * Verify that the stack arguments userspace gave us are sane.
3124 * In addition, set the stack direction for userspace since it's easy for us to
3127 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3129 if (kargs->stack == 0) {
3130 if (kargs->stack_size > 0)
3133 if (kargs->stack_size == 0)
3136 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3139 #if !defined(CONFIG_STACK_GROWSUP)
3140 kargs->stack += kargs->stack_size;
3147 static bool clone3_args_valid(struct kernel_clone_args *kargs)
3149 /* Verify that no unknown flags are passed along. */
3151 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3155 * - make the CLONE_DETACHED bit reusable for clone3
3156 * - make the CSIGNAL bits reusable for clone3
3158 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3161 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3162 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3165 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3169 if (!clone3_stack_valid(kargs))
3176 * sys_clone3 - create a new process with specific properties
3177 * @uargs: argument structure
3178 * @size: size of @uargs
3180 * clone3() is the extensible successor to clone()/clone2().
3181 * It takes a struct as argument that is versioned by its size.
3183 * Return: On success, a positive PID for the child process.
3184 * On error, a negative errno number.
3186 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3190 struct kernel_clone_args kargs;
3191 pid_t set_tid[MAX_PID_NS_LEVEL];
3193 kargs.set_tid = set_tid;
3195 err = copy_clone_args_from_user(&kargs, uargs, size);
3199 if (!clone3_args_valid(&kargs))
3202 return kernel_clone(&kargs);
3206 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3208 struct task_struct *leader, *parent, *child;
3211 read_lock(&tasklist_lock);
3212 leader = top = top->group_leader;
3214 for_each_thread(leader, parent) {
3215 list_for_each_entry(child, &parent->children, sibling) {
3216 res = visitor(child, data);
3228 if (leader != top) {
3230 parent = child->real_parent;
3231 leader = parent->group_leader;
3235 read_unlock(&tasklist_lock);
3238 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3239 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3242 static void sighand_ctor(void *data)
3244 struct sighand_struct *sighand = data;
3246 spin_lock_init(&sighand->siglock);
3247 init_waitqueue_head(&sighand->signalfd_wqh);
3250 void __init mm_cache_init(void)
3252 unsigned int mm_size;
3255 * The mm_cpumask is located at the end of mm_struct, and is
3256 * dynamically sized based on the maximum CPU number this system
3257 * can have, taking hotplug into account (nr_cpu_ids).
3259 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3261 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3262 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3263 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3264 offsetof(struct mm_struct, saved_auxv),
3265 sizeof_field(struct mm_struct, saved_auxv),
3269 void __init proc_caches_init(void)
3271 sighand_cachep = kmem_cache_create("sighand_cache",
3272 sizeof(struct sighand_struct), 0,
3273 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3274 SLAB_ACCOUNT, sighand_ctor);
3275 signal_cachep = kmem_cache_create("signal_cache",
3276 sizeof(struct signal_struct), 0,
3277 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3279 files_cachep = kmem_cache_create("files_cache",
3280 sizeof(struct files_struct), 0,
3281 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3283 fs_cachep = kmem_cache_create("fs_cache",
3284 sizeof(struct fs_struct), 0,
3285 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3288 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3289 #ifdef CONFIG_PER_VMA_LOCK
3290 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3293 nsproxy_cache_init();
3297 * Check constraints on flags passed to the unshare system call.
3299 static int check_unshare_flags(unsigned long unshare_flags)
3301 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3302 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3303 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3304 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3308 * Not implemented, but pretend it works if there is nothing
3309 * to unshare. Note that unsharing the address space or the
3310 * signal handlers also need to unshare the signal queues (aka
3313 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3314 if (!thread_group_empty(current))
3317 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3318 if (refcount_read(¤t->sighand->count) > 1)
3321 if (unshare_flags & CLONE_VM) {
3322 if (!current_is_single_threaded())
3330 * Unshare the filesystem structure if it is being shared
3332 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3334 struct fs_struct *fs = current->fs;
3336 if (!(unshare_flags & CLONE_FS) || !fs)
3339 /* don't need lock here; in the worst case we'll do useless copy */
3343 *new_fsp = copy_fs_struct(fs);
3351 * Unshare file descriptor table if it is being shared
3353 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3354 struct files_struct **new_fdp)
3356 struct files_struct *fd = current->files;
3359 if ((unshare_flags & CLONE_FILES) &&
3360 (fd && atomic_read(&fd->count) > 1)) {
3361 *new_fdp = dup_fd(fd, max_fds, &error);
3370 * unshare allows a process to 'unshare' part of the process
3371 * context which was originally shared using clone. copy_*
3372 * functions used by kernel_clone() cannot be used here directly
3373 * because they modify an inactive task_struct that is being
3374 * constructed. Here we are modifying the current, active,
3377 int ksys_unshare(unsigned long unshare_flags)
3379 struct fs_struct *fs, *new_fs = NULL;
3380 struct files_struct *new_fd = NULL;
3381 struct cred *new_cred = NULL;
3382 struct nsproxy *new_nsproxy = NULL;
3387 * If unsharing a user namespace must also unshare the thread group
3388 * and unshare the filesystem root and working directories.
3390 if (unshare_flags & CLONE_NEWUSER)
3391 unshare_flags |= CLONE_THREAD | CLONE_FS;
3393 * If unsharing vm, must also unshare signal handlers.
3395 if (unshare_flags & CLONE_VM)
3396 unshare_flags |= CLONE_SIGHAND;
3398 * If unsharing a signal handlers, must also unshare the signal queues.
3400 if (unshare_flags & CLONE_SIGHAND)
3401 unshare_flags |= CLONE_THREAD;
3403 * If unsharing namespace, must also unshare filesystem information.
3405 if (unshare_flags & CLONE_NEWNS)
3406 unshare_flags |= CLONE_FS;
3408 err = check_unshare_flags(unshare_flags);
3410 goto bad_unshare_out;
3412 * CLONE_NEWIPC must also detach from the undolist: after switching
3413 * to a new ipc namespace, the semaphore arrays from the old
3414 * namespace are unreachable.
3416 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3418 err = unshare_fs(unshare_flags, &new_fs);
3420 goto bad_unshare_out;
3421 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3423 goto bad_unshare_cleanup_fs;
3424 err = unshare_userns(unshare_flags, &new_cred);
3426 goto bad_unshare_cleanup_fd;
3427 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3430 goto bad_unshare_cleanup_cred;
3433 err = set_cred_ucounts(new_cred);
3435 goto bad_unshare_cleanup_cred;
3438 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3441 * CLONE_SYSVSEM is equivalent to sys_exit().
3445 if (unshare_flags & CLONE_NEWIPC) {
3446 /* Orphan segments in old ns (see sem above). */
3448 shm_init_task(current);
3452 switch_task_namespaces(current, new_nsproxy);
3458 spin_lock(&fs->lock);
3459 current->fs = new_fs;
3464 spin_unlock(&fs->lock);
3468 swap(current->files, new_fd);
3470 task_unlock(current);
3473 /* Install the new user namespace */
3474 commit_creds(new_cred);
3479 perf_event_namespaces(current);
3481 bad_unshare_cleanup_cred:
3484 bad_unshare_cleanup_fd:
3486 put_files_struct(new_fd);
3488 bad_unshare_cleanup_fs:
3490 free_fs_struct(new_fs);
3496 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3498 return ksys_unshare(unshare_flags);
3502 * Helper to unshare the files of the current task.
3503 * We don't want to expose copy_files internals to
3504 * the exec layer of the kernel.
3507 int unshare_files(void)
3509 struct task_struct *task = current;
3510 struct files_struct *old, *copy = NULL;
3513 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3521 put_files_struct(old);
3525 int sysctl_max_threads(struct ctl_table *table, int write,
3526 void *buffer, size_t *lenp, loff_t *ppos)
3530 int threads = max_threads;
3532 int max = MAX_THREADS;
3539 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3543 max_threads = threads;