1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
15 #include <linux/anon_inodes.h>
16 #include <linux/slab.h>
17 #include <linux/sched/autogroup.h>
18 #include <linux/sched/mm.h>
19 #include <linux/sched/coredump.h>
20 #include <linux/sched/user.h>
21 #include <linux/sched/numa_balancing.h>
22 #include <linux/sched/stat.h>
23 #include <linux/sched/task.h>
24 #include <linux/sched/task_stack.h>
25 #include <linux/sched/cputime.h>
26 #include <linux/seq_file.h>
27 #include <linux/rtmutex.h>
28 #include <linux/init.h>
29 #include <linux/unistd.h>
30 #include <linux/module.h>
31 #include <linux/vmalloc.h>
32 #include <linux/completion.h>
33 #include <linux/personality.h>
34 #include <linux/mempolicy.h>
35 #include <linux/sem.h>
36 #include <linux/file.h>
37 #include <linux/fdtable.h>
38 #include <linux/iocontext.h>
39 #include <linux/key.h>
40 #include <linux/binfmts.h>
41 #include <linux/mman.h>
42 #include <linux/mmu_notifier.h>
45 #include <linux/vmacache.h>
46 #include <linux/nsproxy.h>
47 #include <linux/capability.h>
48 #include <linux/cpu.h>
49 #include <linux/cgroup.h>
50 #include <linux/security.h>
51 #include <linux/hugetlb.h>
52 #include <linux/seccomp.h>
53 #include <linux/swap.h>
54 #include <linux/syscalls.h>
55 #include <linux/jiffies.h>
56 #include <linux/futex.h>
57 #include <linux/compat.h>
58 #include <linux/kthread.h>
59 #include <linux/task_io_accounting_ops.h>
60 #include <linux/rcupdate.h>
61 #include <linux/ptrace.h>
62 #include <linux/mount.h>
63 #include <linux/audit.h>
64 #include <linux/memcontrol.h>
65 #include <linux/ftrace.h>
66 #include <linux/proc_fs.h>
67 #include <linux/profile.h>
68 #include <linux/rmap.h>
69 #include <linux/ksm.h>
70 #include <linux/acct.h>
71 #include <linux/userfaultfd_k.h>
72 #include <linux/tsacct_kern.h>
73 #include <linux/cn_proc.h>
74 #include <linux/freezer.h>
75 #include <linux/delayacct.h>
76 #include <linux/taskstats_kern.h>
77 #include <linux/random.h>
78 #include <linux/tty.h>
79 #include <linux/blkdev.h>
80 #include <linux/fs_struct.h>
81 #include <linux/magic.h>
82 #include <linux/perf_event.h>
83 #include <linux/posix-timers.h>
84 #include <linux/user-return-notifier.h>
85 #include <linux/oom.h>
86 #include <linux/khugepaged.h>
87 #include <linux/signalfd.h>
88 #include <linux/uprobes.h>
89 #include <linux/aio.h>
90 #include <linux/compiler.h>
91 #include <linux/sysctl.h>
92 #include <linux/kcov.h>
93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
97 #include <linux/scs.h>
98 #include <linux/io_uring.h>
100 #include <asm/pgalloc.h>
101 #include <linux/uaccess.h>
102 #include <asm/mmu_context.h>
103 #include <asm/cacheflush.h>
104 #include <asm/tlbflush.h>
106 #include <trace/events/sched.h>
108 #define CREATE_TRACE_POINTS
109 #include <trace/events/task.h>
112 * Minimum number of threads to boot the kernel
114 #define MIN_THREADS 20
117 * Maximum number of threads
119 #define MAX_THREADS FUTEX_TID_MASK
122 * Protected counters by write_lock_irq(&tasklist_lock)
124 unsigned long total_forks; /* Handle normal Linux uptimes. */
125 int nr_threads; /* The idle threads do not count.. */
127 static int max_threads; /* tunable limit on nr_threads */
129 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
131 static const char * const resident_page_types[] = {
132 NAMED_ARRAY_INDEX(MM_FILEPAGES),
133 NAMED_ARRAY_INDEX(MM_ANONPAGES),
134 NAMED_ARRAY_INDEX(MM_SWAPENTS),
135 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
138 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
140 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
142 #ifdef CONFIG_PROVE_RCU
143 int lockdep_tasklist_lock_is_held(void)
145 return lockdep_is_held(&tasklist_lock);
147 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
148 #endif /* #ifdef CONFIG_PROVE_RCU */
150 int nr_processes(void)
155 for_each_possible_cpu(cpu)
156 total += per_cpu(process_counts, cpu);
161 void __weak arch_release_task_struct(struct task_struct *tsk)
165 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
166 static struct kmem_cache *task_struct_cachep;
168 static inline struct task_struct *alloc_task_struct_node(int node)
170 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
173 static inline void free_task_struct(struct task_struct *tsk)
175 kmem_cache_free(task_struct_cachep, tsk);
179 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
182 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
183 * kmemcache based allocator.
185 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
187 #ifdef CONFIG_VMAP_STACK
189 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
190 * flush. Try to minimize the number of calls by caching stacks.
192 #define NR_CACHED_STACKS 2
193 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
195 static int free_vm_stack_cache(unsigned int cpu)
197 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
200 for (i = 0; i < NR_CACHED_STACKS; i++) {
201 struct vm_struct *vm_stack = cached_vm_stacks[i];
206 vfree(vm_stack->addr);
207 cached_vm_stacks[i] = NULL;
214 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
216 #ifdef CONFIG_VMAP_STACK
220 for (i = 0; i < NR_CACHED_STACKS; i++) {
223 s = this_cpu_xchg(cached_stacks[i], NULL);
228 /* Clear the KASAN shadow of the stack. */
229 kasan_unpoison_shadow(s->addr, THREAD_SIZE);
231 /* Clear stale pointers from reused stack. */
232 memset(s->addr, 0, THREAD_SIZE);
234 tsk->stack_vm_area = s;
235 tsk->stack = s->addr;
240 * Allocated stacks are cached and later reused by new threads,
241 * so memcg accounting is performed manually on assigning/releasing
242 * stacks to tasks. Drop __GFP_ACCOUNT.
244 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
245 VMALLOC_START, VMALLOC_END,
246 THREADINFO_GFP & ~__GFP_ACCOUNT,
248 0, node, __builtin_return_address(0));
251 * We can't call find_vm_area() in interrupt context, and
252 * free_thread_stack() can be called in interrupt context,
253 * so cache the vm_struct.
256 tsk->stack_vm_area = find_vm_area(stack);
261 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
265 tsk->stack = kasan_reset_tag(page_address(page));
272 static inline void free_thread_stack(struct task_struct *tsk)
274 #ifdef CONFIG_VMAP_STACK
275 struct vm_struct *vm = task_stack_vm_area(tsk);
280 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
281 memcg_kmem_uncharge_page(vm->pages[i], 0);
283 for (i = 0; i < NR_CACHED_STACKS; i++) {
284 if (this_cpu_cmpxchg(cached_stacks[i],
285 NULL, tsk->stack_vm_area) != NULL)
291 vfree_atomic(tsk->stack);
296 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
299 static struct kmem_cache *thread_stack_cache;
301 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
304 unsigned long *stack;
305 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
306 stack = kasan_reset_tag(stack);
311 static void free_thread_stack(struct task_struct *tsk)
313 kmem_cache_free(thread_stack_cache, tsk->stack);
316 void thread_stack_cache_init(void)
318 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
319 THREAD_SIZE, THREAD_SIZE, 0, 0,
321 BUG_ON(thread_stack_cache == NULL);
326 /* SLAB cache for signal_struct structures (tsk->signal) */
327 static struct kmem_cache *signal_cachep;
329 /* SLAB cache for sighand_struct structures (tsk->sighand) */
330 struct kmem_cache *sighand_cachep;
332 /* SLAB cache for files_struct structures (tsk->files) */
333 struct kmem_cache *files_cachep;
335 /* SLAB cache for fs_struct structures (tsk->fs) */
336 struct kmem_cache *fs_cachep;
338 /* SLAB cache for vm_area_struct structures */
339 static struct kmem_cache *vm_area_cachep;
341 /* SLAB cache for mm_struct structures (tsk->mm) */
342 static struct kmem_cache *mm_cachep;
344 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
346 struct vm_area_struct *vma;
348 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
354 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
356 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
359 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
360 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
362 * orig->shared.rb may be modified concurrently, but the clone
363 * will be reinitialized.
365 *new = data_race(*orig);
366 INIT_LIST_HEAD(&new->anon_vma_chain);
367 new->vm_next = new->vm_prev = NULL;
372 void vm_area_free(struct vm_area_struct *vma)
374 kmem_cache_free(vm_area_cachep, vma);
377 static void account_kernel_stack(struct task_struct *tsk, int account)
379 void *stack = task_stack_page(tsk);
380 struct vm_struct *vm = task_stack_vm_area(tsk);
383 /* All stack pages are in the same node. */
385 mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
386 account * (THREAD_SIZE / 1024));
388 mod_lruvec_slab_state(stack, NR_KERNEL_STACK_KB,
389 account * (THREAD_SIZE / 1024));
392 static int memcg_charge_kernel_stack(struct task_struct *tsk)
394 #ifdef CONFIG_VMAP_STACK
395 struct vm_struct *vm = task_stack_vm_area(tsk);
398 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
403 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
405 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
407 * If memcg_kmem_charge_page() fails, page->mem_cgroup
408 * pointer is NULL, and memcg_kmem_uncharge_page() in
409 * free_thread_stack() will ignore this page.
411 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
421 static void release_task_stack(struct task_struct *tsk)
423 if (WARN_ON(tsk->state != TASK_DEAD))
424 return; /* Better to leak the stack than to free prematurely */
426 account_kernel_stack(tsk, -1);
427 free_thread_stack(tsk);
429 #ifdef CONFIG_VMAP_STACK
430 tsk->stack_vm_area = NULL;
434 #ifdef CONFIG_THREAD_INFO_IN_TASK
435 void put_task_stack(struct task_struct *tsk)
437 if (refcount_dec_and_test(&tsk->stack_refcount))
438 release_task_stack(tsk);
442 void free_task(struct task_struct *tsk)
444 #ifdef CONFIG_SECCOMP
445 WARN_ON_ONCE(tsk->seccomp.filter);
449 #ifndef CONFIG_THREAD_INFO_IN_TASK
451 * The task is finally done with both the stack and thread_info,
454 release_task_stack(tsk);
457 * If the task had a separate stack allocation, it should be gone
460 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
462 rt_mutex_debug_task_free(tsk);
463 ftrace_graph_exit_task(tsk);
464 arch_release_task_struct(tsk);
465 if (tsk->flags & PF_KTHREAD)
466 free_kthread_struct(tsk);
467 free_task_struct(tsk);
469 EXPORT_SYMBOL(free_task);
472 static __latent_entropy int dup_mmap(struct mm_struct *mm,
473 struct mm_struct *oldmm)
475 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
476 struct rb_node **rb_link, *rb_parent;
478 unsigned long charge;
481 uprobe_start_dup_mmap();
482 if (mmap_write_lock_killable(oldmm)) {
484 goto fail_uprobe_end;
486 flush_cache_dup_mm(oldmm);
487 uprobe_dup_mmap(oldmm, mm);
489 * Not linked in yet - no deadlock potential:
491 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
493 /* No ordering required: file already has been exposed. */
494 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
496 mm->total_vm = oldmm->total_vm;
497 mm->data_vm = oldmm->data_vm;
498 mm->exec_vm = oldmm->exec_vm;
499 mm->stack_vm = oldmm->stack_vm;
501 rb_link = &mm->mm_rb.rb_node;
504 retval = ksm_fork(mm, oldmm);
507 retval = khugepaged_fork(mm, oldmm);
512 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
515 if (mpnt->vm_flags & VM_DONTCOPY) {
516 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
521 * Don't duplicate many vmas if we've been oom-killed (for
524 if (fatal_signal_pending(current)) {
528 if (mpnt->vm_flags & VM_ACCOUNT) {
529 unsigned long len = vma_pages(mpnt);
531 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
535 tmp = vm_area_dup(mpnt);
538 retval = vma_dup_policy(mpnt, tmp);
540 goto fail_nomem_policy;
542 retval = dup_userfaultfd(tmp, &uf);
544 goto fail_nomem_anon_vma_fork;
545 if (tmp->vm_flags & VM_WIPEONFORK) {
547 * VM_WIPEONFORK gets a clean slate in the child.
548 * Don't prepare anon_vma until fault since we don't
549 * copy page for current vma.
551 tmp->anon_vma = NULL;
552 } else if (anon_vma_fork(tmp, mpnt))
553 goto fail_nomem_anon_vma_fork;
554 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
557 struct inode *inode = file_inode(file);
558 struct address_space *mapping = file->f_mapping;
561 if (tmp->vm_flags & VM_DENYWRITE)
562 put_write_access(inode);
563 i_mmap_lock_write(mapping);
564 if (tmp->vm_flags & VM_SHARED)
565 mapping_allow_writable(mapping);
566 flush_dcache_mmap_lock(mapping);
567 /* insert tmp into the share list, just after mpnt */
568 vma_interval_tree_insert_after(tmp, mpnt,
570 flush_dcache_mmap_unlock(mapping);
571 i_mmap_unlock_write(mapping);
575 * Clear hugetlb-related page reserves for children. This only
576 * affects MAP_PRIVATE mappings. Faults generated by the child
577 * are not guaranteed to succeed, even if read-only
579 if (is_vm_hugetlb_page(tmp))
580 reset_vma_resv_huge_pages(tmp);
583 * Link in the new vma and copy the page table entries.
586 pprev = &tmp->vm_next;
590 __vma_link_rb(mm, tmp, rb_link, rb_parent);
591 rb_link = &tmp->vm_rb.rb_right;
592 rb_parent = &tmp->vm_rb;
595 if (!(tmp->vm_flags & VM_WIPEONFORK))
596 retval = copy_page_range(tmp, mpnt);
598 if (tmp->vm_ops && tmp->vm_ops->open)
599 tmp->vm_ops->open(tmp);
604 /* a new mm has just been created */
605 retval = arch_dup_mmap(oldmm, mm);
607 mmap_write_unlock(mm);
609 mmap_write_unlock(oldmm);
610 dup_userfaultfd_complete(&uf);
612 uprobe_end_dup_mmap();
614 fail_nomem_anon_vma_fork:
615 mpol_put(vma_policy(tmp));
620 vm_unacct_memory(charge);
624 static inline int mm_alloc_pgd(struct mm_struct *mm)
626 mm->pgd = pgd_alloc(mm);
627 if (unlikely(!mm->pgd))
632 static inline void mm_free_pgd(struct mm_struct *mm)
634 pgd_free(mm, mm->pgd);
637 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
639 mmap_write_lock(oldmm);
640 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
641 mmap_write_unlock(oldmm);
644 #define mm_alloc_pgd(mm) (0)
645 #define mm_free_pgd(mm)
646 #endif /* CONFIG_MMU */
648 static void check_mm(struct mm_struct *mm)
652 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
653 "Please make sure 'struct resident_page_types[]' is updated as well");
655 for (i = 0; i < NR_MM_COUNTERS; i++) {
656 long x = atomic_long_read(&mm->rss_stat.count[i]);
659 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
660 mm, resident_page_types[i], x);
663 if (mm_pgtables_bytes(mm))
664 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
665 mm_pgtables_bytes(mm));
667 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
668 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
672 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
673 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
676 * Called when the last reference to the mm
677 * is dropped: either by a lazy thread or by
678 * mmput. Free the page directory and the mm.
680 void __mmdrop(struct mm_struct *mm)
682 BUG_ON(mm == &init_mm);
683 WARN_ON_ONCE(mm == current->mm);
684 WARN_ON_ONCE(mm == current->active_mm);
687 mmu_notifier_subscriptions_destroy(mm);
689 put_user_ns(mm->user_ns);
692 EXPORT_SYMBOL_GPL(__mmdrop);
694 static void mmdrop_async_fn(struct work_struct *work)
696 struct mm_struct *mm;
698 mm = container_of(work, struct mm_struct, async_put_work);
702 static void mmdrop_async(struct mm_struct *mm)
704 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
705 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
706 schedule_work(&mm->async_put_work);
710 static inline void free_signal_struct(struct signal_struct *sig)
712 taskstats_tgid_free(sig);
713 sched_autogroup_exit(sig);
715 * __mmdrop is not safe to call from softirq context on x86 due to
716 * pgd_dtor so postpone it to the async context
719 mmdrop_async(sig->oom_mm);
720 kmem_cache_free(signal_cachep, sig);
723 static inline void put_signal_struct(struct signal_struct *sig)
725 if (refcount_dec_and_test(&sig->sigcnt))
726 free_signal_struct(sig);
729 void __put_task_struct(struct task_struct *tsk)
731 WARN_ON(!tsk->exit_state);
732 WARN_ON(refcount_read(&tsk->usage));
733 WARN_ON(tsk == current);
737 task_numa_free(tsk, true);
738 security_task_free(tsk);
740 delayacct_tsk_free(tsk);
741 put_signal_struct(tsk->signal);
743 if (!profile_handoff_task(tsk))
746 EXPORT_SYMBOL_GPL(__put_task_struct);
748 void __put_task_struct_rcu_cb(struct rcu_head *rhp)
750 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
752 __put_task_struct(task);
754 EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
756 void __init __weak arch_task_cache_init(void) { }
761 static void set_max_threads(unsigned int max_threads_suggested)
764 unsigned long nr_pages = totalram_pages();
767 * The number of threads shall be limited such that the thread
768 * structures may only consume a small part of the available memory.
770 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
771 threads = MAX_THREADS;
773 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
774 (u64) THREAD_SIZE * 8UL);
776 if (threads > max_threads_suggested)
777 threads = max_threads_suggested;
779 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
782 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
783 /* Initialized by the architecture: */
784 int arch_task_struct_size __read_mostly;
787 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
788 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
790 /* Fetch thread_struct whitelist for the architecture. */
791 arch_thread_struct_whitelist(offset, size);
794 * Handle zero-sized whitelist or empty thread_struct, otherwise
795 * adjust offset to position of thread_struct in task_struct.
797 if (unlikely(*size == 0))
800 *offset += offsetof(struct task_struct, thread);
802 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
804 void __init fork_init(void)
807 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
808 #ifndef ARCH_MIN_TASKALIGN
809 #define ARCH_MIN_TASKALIGN 0
811 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
812 unsigned long useroffset, usersize;
814 /* create a slab on which task_structs can be allocated */
815 task_struct_whitelist(&useroffset, &usersize);
816 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
817 arch_task_struct_size, align,
818 SLAB_PANIC|SLAB_ACCOUNT,
819 useroffset, usersize, NULL);
822 /* do the arch specific task caches init */
823 arch_task_cache_init();
825 set_max_threads(MAX_THREADS);
827 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
828 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
829 init_task.signal->rlim[RLIMIT_SIGPENDING] =
830 init_task.signal->rlim[RLIMIT_NPROC];
832 for (i = 0; i < UCOUNT_COUNTS; i++) {
833 init_user_ns.ucount_max[i] = max_threads/2;
836 #ifdef CONFIG_VMAP_STACK
837 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
838 NULL, free_vm_stack_cache);
843 lockdep_init_task(&init_task);
847 int __weak arch_dup_task_struct(struct task_struct *dst,
848 struct task_struct *src)
854 void set_task_stack_end_magic(struct task_struct *tsk)
856 unsigned long *stackend;
858 stackend = end_of_stack(tsk);
859 *stackend = STACK_END_MAGIC; /* for overflow detection */
862 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
864 struct task_struct *tsk;
865 unsigned long *stack;
866 struct vm_struct *stack_vm_area __maybe_unused;
869 if (node == NUMA_NO_NODE)
870 node = tsk_fork_get_node(orig);
871 tsk = alloc_task_struct_node(node);
875 stack = alloc_thread_stack_node(tsk, node);
879 if (memcg_charge_kernel_stack(tsk))
882 stack_vm_area = task_stack_vm_area(tsk);
884 err = arch_dup_task_struct(tsk, orig);
887 * arch_dup_task_struct() clobbers the stack-related fields. Make
888 * sure they're properly initialized before using any stack-related
892 #ifdef CONFIG_VMAP_STACK
893 tsk->stack_vm_area = stack_vm_area;
895 #ifdef CONFIG_THREAD_INFO_IN_TASK
896 refcount_set(&tsk->stack_refcount, 1);
902 err = scs_prepare(tsk, node);
906 #ifdef CONFIG_SECCOMP
908 * We must handle setting up seccomp filters once we're under
909 * the sighand lock in case orig has changed between now and
910 * then. Until then, filter must be NULL to avoid messing up
911 * the usage counts on the error path calling free_task.
913 tsk->seccomp.filter = NULL;
916 setup_thread_stack(tsk, orig);
917 clear_user_return_notifier(tsk);
918 clear_tsk_need_resched(tsk);
919 set_task_stack_end_magic(tsk);
921 #ifdef CONFIG_STACKPROTECTOR
922 tsk->stack_canary = get_random_canary();
924 if (orig->cpus_ptr == &orig->cpus_mask)
925 tsk->cpus_ptr = &tsk->cpus_mask;
928 * One for the user space visible state that goes away when reaped.
929 * One for the scheduler.
931 refcount_set(&tsk->rcu_users, 2);
932 /* One for the rcu users */
933 refcount_set(&tsk->usage, 1);
934 #ifdef CONFIG_BLK_DEV_IO_TRACE
937 tsk->splice_pipe = NULL;
938 tsk->task_frag.page = NULL;
939 tsk->wake_q.next = NULL;
940 tsk->pf_io_worker = NULL;
942 account_kernel_stack(tsk, 1);
946 #ifdef CONFIG_FAULT_INJECTION
950 #ifdef CONFIG_BLK_CGROUP
951 tsk->throttle_queue = NULL;
952 tsk->use_memdelay = 0;
956 tsk->active_memcg = NULL;
961 free_thread_stack(tsk);
963 free_task_struct(tsk);
967 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
969 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
971 static int __init coredump_filter_setup(char *s)
973 default_dump_filter =
974 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
975 MMF_DUMP_FILTER_MASK;
979 __setup("coredump_filter=", coredump_filter_setup);
981 #include <linux/init_task.h>
983 static void mm_init_aio(struct mm_struct *mm)
986 spin_lock_init(&mm->ioctx_lock);
987 mm->ioctx_table = NULL;
991 static __always_inline void mm_clear_owner(struct mm_struct *mm,
992 struct task_struct *p)
996 WRITE_ONCE(mm->owner, NULL);
1000 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1007 static void mm_init_pasid(struct mm_struct *mm)
1009 #ifdef CONFIG_IOMMU_SUPPORT
1010 mm->pasid = INIT_PASID;
1014 static void mm_init_uprobes_state(struct mm_struct *mm)
1016 #ifdef CONFIG_UPROBES
1017 mm->uprobes_state.xol_area = NULL;
1021 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1022 struct user_namespace *user_ns)
1025 mm->mm_rb = RB_ROOT;
1026 mm->vmacache_seqnum = 0;
1027 atomic_set(&mm->mm_users, 1);
1028 atomic_set(&mm->mm_count, 1);
1029 seqcount_init(&mm->write_protect_seq);
1031 INIT_LIST_HEAD(&mm->mmlist);
1032 mm->core_state = NULL;
1033 mm_pgtables_bytes_init(mm);
1036 atomic_set(&mm->has_pinned, 0);
1037 atomic64_set(&mm->pinned_vm, 0);
1038 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1039 spin_lock_init(&mm->page_table_lock);
1040 spin_lock_init(&mm->arg_lock);
1041 mm_init_cpumask(mm);
1043 mm_init_owner(mm, p);
1045 RCU_INIT_POINTER(mm->exe_file, NULL);
1046 mmu_notifier_subscriptions_init(mm);
1047 init_tlb_flush_pending(mm);
1048 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1049 mm->pmd_huge_pte = NULL;
1051 mm_init_uprobes_state(mm);
1052 hugetlb_count_init(mm);
1055 mm->flags = current->mm->flags & MMF_INIT_MASK;
1056 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1058 mm->flags = default_dump_filter;
1062 if (mm_alloc_pgd(mm))
1065 if (init_new_context(p, mm))
1066 goto fail_nocontext;
1068 mm->user_ns = get_user_ns(user_ns);
1079 * Allocate and initialize an mm_struct.
1081 struct mm_struct *mm_alloc(void)
1083 struct mm_struct *mm;
1089 memset(mm, 0, sizeof(*mm));
1090 return mm_init(mm, current, current_user_ns());
1093 static inline void __mmput(struct mm_struct *mm)
1095 VM_BUG_ON(atomic_read(&mm->mm_users));
1097 uprobe_clear_state(mm);
1100 khugepaged_exit(mm); /* must run before exit_mmap */
1102 mm_put_huge_zero_page(mm);
1103 set_mm_exe_file(mm, NULL);
1104 if (!list_empty(&mm->mmlist)) {
1105 spin_lock(&mmlist_lock);
1106 list_del(&mm->mmlist);
1107 spin_unlock(&mmlist_lock);
1110 module_put(mm->binfmt->module);
1115 * Decrement the use count and release all resources for an mm.
1117 void mmput(struct mm_struct *mm)
1121 if (atomic_dec_and_test(&mm->mm_users))
1124 EXPORT_SYMBOL_GPL(mmput);
1127 static void mmput_async_fn(struct work_struct *work)
1129 struct mm_struct *mm = container_of(work, struct mm_struct,
1135 void mmput_async(struct mm_struct *mm)
1137 if (atomic_dec_and_test(&mm->mm_users)) {
1138 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1139 schedule_work(&mm->async_put_work);
1142 EXPORT_SYMBOL_GPL(mmput_async);
1146 * set_mm_exe_file - change a reference to the mm's executable file
1148 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1150 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1151 * invocations: in mmput() nobody alive left, in execve task is single
1152 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1153 * mm->exe_file, but does so without using set_mm_exe_file() in order
1154 * to do avoid the need for any locks.
1156 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1158 struct file *old_exe_file;
1161 * It is safe to dereference the exe_file without RCU as
1162 * this function is only called if nobody else can access
1163 * this mm -- see comment above for justification.
1165 old_exe_file = rcu_dereference_raw(mm->exe_file);
1168 get_file(new_exe_file);
1169 rcu_assign_pointer(mm->exe_file, new_exe_file);
1175 * get_mm_exe_file - acquire a reference to the mm's executable file
1177 * Returns %NULL if mm has no associated executable file.
1178 * User must release file via fput().
1180 struct file *get_mm_exe_file(struct mm_struct *mm)
1182 struct file *exe_file;
1185 exe_file = rcu_dereference(mm->exe_file);
1186 if (exe_file && !get_file_rcu(exe_file))
1191 EXPORT_SYMBOL(get_mm_exe_file);
1194 * get_task_exe_file - acquire a reference to the task's executable file
1196 * Returns %NULL if task's mm (if any) has no associated executable file or
1197 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1198 * User must release file via fput().
1200 struct file *get_task_exe_file(struct task_struct *task)
1202 struct file *exe_file = NULL;
1203 struct mm_struct *mm;
1208 if (!(task->flags & PF_KTHREAD))
1209 exe_file = get_mm_exe_file(mm);
1214 EXPORT_SYMBOL(get_task_exe_file);
1217 * get_task_mm - acquire a reference to the task's mm
1219 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1220 * this kernel workthread has transiently adopted a user mm with use_mm,
1221 * to do its AIO) is not set and if so returns a reference to it, after
1222 * bumping up the use count. User must release the mm via mmput()
1223 * after use. Typically used by /proc and ptrace.
1225 struct mm_struct *get_task_mm(struct task_struct *task)
1227 struct mm_struct *mm;
1232 if (task->flags & PF_KTHREAD)
1240 EXPORT_SYMBOL_GPL(get_task_mm);
1242 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1244 struct mm_struct *mm;
1247 err = down_read_killable(&task->signal->exec_update_lock);
1249 return ERR_PTR(err);
1251 mm = get_task_mm(task);
1252 if (mm && mm != current->mm &&
1253 !ptrace_may_access(task, mode)) {
1255 mm = ERR_PTR(-EACCES);
1257 up_read(&task->signal->exec_update_lock);
1262 static void complete_vfork_done(struct task_struct *tsk)
1264 struct completion *vfork;
1267 vfork = tsk->vfork_done;
1268 if (likely(vfork)) {
1269 tsk->vfork_done = NULL;
1275 static int wait_for_vfork_done(struct task_struct *child,
1276 struct completion *vfork)
1280 freezer_do_not_count();
1281 cgroup_enter_frozen();
1282 killed = wait_for_completion_killable(vfork);
1283 cgroup_leave_frozen(false);
1288 child->vfork_done = NULL;
1292 put_task_struct(child);
1296 /* Please note the differences between mmput and mm_release.
1297 * mmput is called whenever we stop holding onto a mm_struct,
1298 * error success whatever.
1300 * mm_release is called after a mm_struct has been removed
1301 * from the current process.
1303 * This difference is important for error handling, when we
1304 * only half set up a mm_struct for a new process and need to restore
1305 * the old one. Because we mmput the new mm_struct before
1306 * restoring the old one. . .
1307 * Eric Biederman 10 January 1998
1309 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1311 uprobe_free_utask(tsk);
1313 /* Get rid of any cached register state */
1314 deactivate_mm(tsk, mm);
1317 * Signal userspace if we're not exiting with a core dump
1318 * because we want to leave the value intact for debugging
1321 if (tsk->clear_child_tid) {
1322 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1323 atomic_read(&mm->mm_users) > 1) {
1325 * We don't check the error code - if userspace has
1326 * not set up a proper pointer then tough luck.
1328 put_user(0, tsk->clear_child_tid);
1329 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1330 1, NULL, NULL, 0, 0);
1332 tsk->clear_child_tid = NULL;
1336 * All done, finally we can wake up parent and return this mm to him.
1337 * Also kthread_stop() uses this completion for synchronization.
1339 if (tsk->vfork_done)
1340 complete_vfork_done(tsk);
1343 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1345 futex_exit_release(tsk);
1346 mm_release(tsk, mm);
1349 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1351 futex_exec_release(tsk);
1352 mm_release(tsk, mm);
1356 * dup_mm() - duplicates an existing mm structure
1357 * @tsk: the task_struct with which the new mm will be associated.
1358 * @oldmm: the mm to duplicate.
1360 * Allocates a new mm structure and duplicates the provided @oldmm structure
1363 * Return: the duplicated mm or NULL on failure.
1365 static struct mm_struct *dup_mm(struct task_struct *tsk,
1366 struct mm_struct *oldmm)
1368 struct mm_struct *mm;
1375 memcpy(mm, oldmm, sizeof(*mm));
1377 if (!mm_init(mm, tsk, mm->user_ns))
1380 err = dup_mmap(mm, oldmm);
1384 mm->hiwater_rss = get_mm_rss(mm);
1385 mm->hiwater_vm = mm->total_vm;
1387 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1393 /* don't put binfmt in mmput, we haven't got module yet */
1395 mm_init_owner(mm, NULL);
1402 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1404 struct mm_struct *mm, *oldmm;
1407 tsk->min_flt = tsk->maj_flt = 0;
1408 tsk->nvcsw = tsk->nivcsw = 0;
1409 #ifdef CONFIG_DETECT_HUNG_TASK
1410 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1411 tsk->last_switch_time = 0;
1415 tsk->active_mm = NULL;
1418 * Are we cloning a kernel thread?
1420 * We need to steal a active VM for that..
1422 oldmm = current->mm;
1426 /* initialize the new vmacache entries */
1427 vmacache_flush(tsk);
1429 if (clone_flags & CLONE_VM) {
1436 mm = dup_mm(tsk, current->mm);
1442 tsk->active_mm = mm;
1449 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1451 struct fs_struct *fs = current->fs;
1452 if (clone_flags & CLONE_FS) {
1453 /* tsk->fs is already what we want */
1454 spin_lock(&fs->lock);
1456 spin_unlock(&fs->lock);
1460 spin_unlock(&fs->lock);
1463 tsk->fs = copy_fs_struct(fs);
1469 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1471 struct files_struct *oldf, *newf;
1475 * A background process may not have any files ...
1477 oldf = current->files;
1481 if (clone_flags & CLONE_FILES) {
1482 atomic_inc(&oldf->count);
1486 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1496 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1499 struct io_context *ioc = current->io_context;
1500 struct io_context *new_ioc;
1505 * Share io context with parent, if CLONE_IO is set
1507 if (clone_flags & CLONE_IO) {
1509 tsk->io_context = ioc;
1510 } else if (ioprio_valid(ioc->ioprio)) {
1511 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1512 if (unlikely(!new_ioc))
1515 new_ioc->ioprio = ioc->ioprio;
1516 put_io_context(new_ioc);
1522 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1524 struct sighand_struct *sig;
1526 if (clone_flags & CLONE_SIGHAND) {
1527 refcount_inc(¤t->sighand->count);
1530 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1531 RCU_INIT_POINTER(tsk->sighand, sig);
1535 refcount_set(&sig->count, 1);
1536 spin_lock_irq(¤t->sighand->siglock);
1537 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1538 spin_unlock_irq(¤t->sighand->siglock);
1540 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1541 if (clone_flags & CLONE_CLEAR_SIGHAND)
1542 flush_signal_handlers(tsk, 0);
1547 void __cleanup_sighand(struct sighand_struct *sighand)
1549 if (refcount_dec_and_test(&sighand->count)) {
1550 signalfd_cleanup(sighand);
1552 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1553 * without an RCU grace period, see __lock_task_sighand().
1555 kmem_cache_free(sighand_cachep, sighand);
1560 * Initialize POSIX timer handling for a thread group.
1562 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1564 struct posix_cputimers *pct = &sig->posix_cputimers;
1565 unsigned long cpu_limit;
1567 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1568 posix_cputimers_group_init(pct, cpu_limit);
1571 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1573 struct signal_struct *sig;
1575 if (clone_flags & CLONE_THREAD)
1578 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1583 sig->nr_threads = 1;
1584 atomic_set(&sig->live, 1);
1585 refcount_set(&sig->sigcnt, 1);
1587 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1588 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1589 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1591 init_waitqueue_head(&sig->wait_chldexit);
1592 sig->curr_target = tsk;
1593 init_sigpending(&sig->shared_pending);
1594 INIT_HLIST_HEAD(&sig->multiprocess);
1595 seqlock_init(&sig->stats_lock);
1596 prev_cputime_init(&sig->prev_cputime);
1598 #ifdef CONFIG_POSIX_TIMERS
1599 INIT_LIST_HEAD(&sig->posix_timers);
1600 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1601 sig->real_timer.function = it_real_fn;
1604 task_lock(current->group_leader);
1605 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1606 task_unlock(current->group_leader);
1608 posix_cpu_timers_init_group(sig);
1610 tty_audit_fork(sig);
1611 sched_autogroup_fork(sig);
1613 sig->oom_score_adj = current->signal->oom_score_adj;
1614 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1616 mutex_init(&sig->cred_guard_mutex);
1617 init_rwsem(&sig->exec_update_lock);
1622 static void copy_seccomp(struct task_struct *p)
1624 #ifdef CONFIG_SECCOMP
1626 * Must be called with sighand->lock held, which is common to
1627 * all threads in the group. Holding cred_guard_mutex is not
1628 * needed because this new task is not yet running and cannot
1631 assert_spin_locked(¤t->sighand->siglock);
1633 /* Ref-count the new filter user, and assign it. */
1634 get_seccomp_filter(current);
1635 p->seccomp = current->seccomp;
1638 * Explicitly enable no_new_privs here in case it got set
1639 * between the task_struct being duplicated and holding the
1640 * sighand lock. The seccomp state and nnp must be in sync.
1642 if (task_no_new_privs(current))
1643 task_set_no_new_privs(p);
1646 * If the parent gained a seccomp mode after copying thread
1647 * flags and between before we held the sighand lock, we have
1648 * to manually enable the seccomp thread flag here.
1650 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1651 set_tsk_thread_flag(p, TIF_SECCOMP);
1655 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1657 current->clear_child_tid = tidptr;
1659 return task_pid_vnr(current);
1662 static void rt_mutex_init_task(struct task_struct *p)
1664 raw_spin_lock_init(&p->pi_lock);
1665 #ifdef CONFIG_RT_MUTEXES
1666 p->pi_waiters = RB_ROOT_CACHED;
1667 p->pi_top_task = NULL;
1668 p->pi_blocked_on = NULL;
1672 static inline void init_task_pid_links(struct task_struct *task)
1676 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1677 INIT_HLIST_NODE(&task->pid_links[type]);
1682 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1684 if (type == PIDTYPE_PID)
1685 task->thread_pid = pid;
1687 task->signal->pids[type] = pid;
1690 static inline void rcu_copy_process(struct task_struct *p)
1692 #ifdef CONFIG_PREEMPT_RCU
1693 p->rcu_read_lock_nesting = 0;
1694 p->rcu_read_unlock_special.s = 0;
1695 p->rcu_blocked_node = NULL;
1696 INIT_LIST_HEAD(&p->rcu_node_entry);
1697 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1698 #ifdef CONFIG_TASKS_RCU
1699 p->rcu_tasks_holdout = false;
1700 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1701 p->rcu_tasks_idle_cpu = -1;
1702 #endif /* #ifdef CONFIG_TASKS_RCU */
1703 #ifdef CONFIG_TASKS_TRACE_RCU
1704 p->trc_reader_nesting = 0;
1705 p->trc_reader_special.s = 0;
1706 INIT_LIST_HEAD(&p->trc_holdout_list);
1707 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1710 struct pid *pidfd_pid(const struct file *file)
1712 if (file->f_op == &pidfd_fops)
1713 return file->private_data;
1715 return ERR_PTR(-EBADF);
1718 static int pidfd_release(struct inode *inode, struct file *file)
1720 struct pid *pid = file->private_data;
1722 file->private_data = NULL;
1727 #ifdef CONFIG_PROC_FS
1729 * pidfd_show_fdinfo - print information about a pidfd
1730 * @m: proc fdinfo file
1731 * @f: file referencing a pidfd
1734 * This function will print the pid that a given pidfd refers to in the
1735 * pid namespace of the procfs instance.
1736 * If the pid namespace of the process is not a descendant of the pid
1737 * namespace of the procfs instance 0 will be shown as its pid. This is
1738 * similar to calling getppid() on a process whose parent is outside of
1739 * its pid namespace.
1742 * If pid namespaces are supported then this function will also print
1743 * the pid of a given pidfd refers to for all descendant pid namespaces
1744 * starting from the current pid namespace of the instance, i.e. the
1745 * Pid field and the first entry in the NSpid field will be identical.
1746 * If the pid namespace of the process is not a descendant of the pid
1747 * namespace of the procfs instance 0 will be shown as its first NSpid
1748 * entry and no others will be shown.
1749 * Note that this differs from the Pid and NSpid fields in
1750 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1751 * the pid namespace of the procfs instance. The difference becomes
1752 * obvious when sending around a pidfd between pid namespaces from a
1753 * different branch of the tree, i.e. where no ancestoral relation is
1754 * present between the pid namespaces:
1755 * - create two new pid namespaces ns1 and ns2 in the initial pid
1756 * namespace (also take care to create new mount namespaces in the
1757 * new pid namespace and mount procfs)
1758 * - create a process with a pidfd in ns1
1759 * - send pidfd from ns1 to ns2
1760 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1761 * have exactly one entry, which is 0
1763 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1765 struct pid *pid = f->private_data;
1766 struct pid_namespace *ns;
1769 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1770 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1771 nr = pid_nr_ns(pid, ns);
1774 seq_put_decimal_ll(m, "Pid:\t", nr);
1776 #ifdef CONFIG_PID_NS
1777 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1781 /* If nr is non-zero it means that 'pid' is valid and that
1782 * ns, i.e. the pid namespace associated with the procfs
1783 * instance, is in the pid namespace hierarchy of pid.
1784 * Start at one below the already printed level.
1786 for (i = ns->level + 1; i <= pid->level; i++)
1787 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1795 * Poll support for process exit notification.
1797 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1799 struct pid *pid = file->private_data;
1800 __poll_t poll_flags = 0;
1802 poll_wait(file, &pid->wait_pidfd, pts);
1805 * Inform pollers only when the whole thread group exits.
1806 * If the thread group leader exits before all other threads in the
1807 * group, then poll(2) should block, similar to the wait(2) family.
1809 if (thread_group_exited(pid))
1810 poll_flags = EPOLLIN | EPOLLRDNORM;
1815 const struct file_operations pidfd_fops = {
1816 .release = pidfd_release,
1818 #ifdef CONFIG_PROC_FS
1819 .show_fdinfo = pidfd_show_fdinfo,
1823 static void __delayed_free_task(struct rcu_head *rhp)
1825 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1830 static __always_inline void delayed_free_task(struct task_struct *tsk)
1832 if (IS_ENABLED(CONFIG_MEMCG))
1833 call_rcu(&tsk->rcu, __delayed_free_task);
1838 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1840 /* Skip if kernel thread */
1844 /* Skip if spawning a thread or using vfork */
1845 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1848 /* We need to synchronize with __set_oom_adj */
1849 mutex_lock(&oom_adj_mutex);
1850 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1851 /* Update the values in case they were changed after copy_signal */
1852 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1853 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1854 mutex_unlock(&oom_adj_mutex);
1858 * This creates a new process as a copy of the old one,
1859 * but does not actually start it yet.
1861 * It copies the registers, and all the appropriate
1862 * parts of the process environment (as per the clone
1863 * flags). The actual kick-off is left to the caller.
1865 static __latent_entropy struct task_struct *copy_process(
1869 struct kernel_clone_args *args)
1871 int pidfd = -1, retval;
1872 struct task_struct *p;
1873 struct multiprocess_signals delayed;
1874 struct file *pidfile = NULL;
1875 u64 clone_flags = args->flags;
1876 struct nsproxy *nsp = current->nsproxy;
1879 * Don't allow sharing the root directory with processes in a different
1882 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1883 return ERR_PTR(-EINVAL);
1885 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1886 return ERR_PTR(-EINVAL);
1889 * Thread groups must share signals as well, and detached threads
1890 * can only be started up within the thread group.
1892 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1893 return ERR_PTR(-EINVAL);
1896 * Shared signal handlers imply shared VM. By way of the above,
1897 * thread groups also imply shared VM. Blocking this case allows
1898 * for various simplifications in other code.
1900 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1901 return ERR_PTR(-EINVAL);
1904 * Siblings of global init remain as zombies on exit since they are
1905 * not reaped by their parent (swapper). To solve this and to avoid
1906 * multi-rooted process trees, prevent global and container-inits
1907 * from creating siblings.
1909 if ((clone_flags & CLONE_PARENT) &&
1910 current->signal->flags & SIGNAL_UNKILLABLE)
1911 return ERR_PTR(-EINVAL);
1914 * If the new process will be in a different pid or user namespace
1915 * do not allow it to share a thread group with the forking task.
1917 if (clone_flags & CLONE_THREAD) {
1918 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1919 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1920 return ERR_PTR(-EINVAL);
1924 * If the new process will be in a different time namespace
1925 * do not allow it to share VM or a thread group with the forking task.
1927 if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1928 if (nsp->time_ns != nsp->time_ns_for_children)
1929 return ERR_PTR(-EINVAL);
1932 if (clone_flags & CLONE_PIDFD) {
1934 * - CLONE_DETACHED is blocked so that we can potentially
1935 * reuse it later for CLONE_PIDFD.
1936 * - CLONE_THREAD is blocked until someone really needs it.
1938 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1939 return ERR_PTR(-EINVAL);
1943 * Force any signals received before this point to be delivered
1944 * before the fork happens. Collect up signals sent to multiple
1945 * processes that happen during the fork and delay them so that
1946 * they appear to happen after the fork.
1948 sigemptyset(&delayed.signal);
1949 INIT_HLIST_NODE(&delayed.node);
1951 spin_lock_irq(¤t->sighand->siglock);
1952 if (!(clone_flags & CLONE_THREAD))
1953 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1954 recalc_sigpending();
1955 spin_unlock_irq(¤t->sighand->siglock);
1956 retval = -ERESTARTNOINTR;
1957 if (task_sigpending(current))
1961 p = dup_task_struct(current, node);
1964 if (args->io_thread) {
1966 * Mark us an IO worker, and block any signal that isn't
1969 p->flags |= PF_IO_WORKER;
1970 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
1974 * This _must_ happen before we call free_task(), i.e. before we jump
1975 * to any of the bad_fork_* labels. This is to avoid freeing
1976 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1977 * kernel threads (PF_KTHREAD).
1979 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1981 * Clear TID on mm_release()?
1983 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1985 ftrace_graph_init_task(p);
1987 rt_mutex_init_task(p);
1989 lockdep_assert_irqs_enabled();
1990 #ifdef CONFIG_PROVE_LOCKING
1991 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1994 if (atomic_read(&p->real_cred->user->processes) >=
1995 task_rlimit(p, RLIMIT_NPROC)) {
1996 if (p->real_cred->user != INIT_USER &&
1997 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2000 current->flags &= ~PF_NPROC_EXCEEDED;
2002 retval = copy_creds(p, clone_flags);
2007 * If multiple threads are within copy_process(), then this check
2008 * triggers too late. This doesn't hurt, the check is only there
2009 * to stop root fork bombs.
2012 if (data_race(nr_threads >= max_threads))
2013 goto bad_fork_cleanup_count;
2015 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2016 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
2017 p->flags |= PF_FORKNOEXEC;
2018 INIT_LIST_HEAD(&p->children);
2019 INIT_LIST_HEAD(&p->sibling);
2020 rcu_copy_process(p);
2021 p->vfork_done = NULL;
2022 spin_lock_init(&p->alloc_lock);
2024 init_sigpending(&p->pending);
2026 p->utime = p->stime = p->gtime = 0;
2027 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2028 p->utimescaled = p->stimescaled = 0;
2030 prev_cputime_init(&p->prev_cputime);
2032 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2033 seqcount_init(&p->vtime.seqcount);
2034 p->vtime.starttime = 0;
2035 p->vtime.state = VTIME_INACTIVE;
2038 #ifdef CONFIG_IO_URING
2042 #if defined(SPLIT_RSS_COUNTING)
2043 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2046 p->default_timer_slack_ns = current->timer_slack_ns;
2052 task_io_accounting_init(&p->ioac);
2053 acct_clear_integrals(p);
2055 posix_cputimers_init(&p->posix_cputimers);
2057 p->io_context = NULL;
2058 audit_set_context(p, NULL);
2061 p->mempolicy = mpol_dup(p->mempolicy);
2062 if (IS_ERR(p->mempolicy)) {
2063 retval = PTR_ERR(p->mempolicy);
2064 p->mempolicy = NULL;
2065 goto bad_fork_cleanup_threadgroup_lock;
2068 #ifdef CONFIG_CPUSETS
2069 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2070 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2071 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2073 #ifdef CONFIG_TRACE_IRQFLAGS
2074 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2075 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2076 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2077 p->softirqs_enabled = 1;
2078 p->softirq_context = 0;
2081 p->pagefault_disabled = 0;
2083 #ifdef CONFIG_LOCKDEP
2084 lockdep_init_task(p);
2087 #ifdef CONFIG_DEBUG_MUTEXES
2088 p->blocked_on = NULL; /* not blocked yet */
2090 #ifdef CONFIG_BCACHE
2091 p->sequential_io = 0;
2092 p->sequential_io_avg = 0;
2095 /* Perform scheduler related setup. Assign this task to a CPU. */
2096 retval = sched_fork(clone_flags, p);
2098 goto bad_fork_cleanup_policy;
2100 retval = perf_event_init_task(p);
2102 goto bad_fork_cleanup_policy;
2103 retval = audit_alloc(p);
2105 goto bad_fork_cleanup_perf;
2106 /* copy all the process information */
2108 retval = security_task_alloc(p, clone_flags);
2110 goto bad_fork_cleanup_audit;
2111 retval = copy_semundo(clone_flags, p);
2113 goto bad_fork_cleanup_security;
2114 retval = copy_files(clone_flags, p);
2116 goto bad_fork_cleanup_semundo;
2117 retval = copy_fs(clone_flags, p);
2119 goto bad_fork_cleanup_files;
2120 retval = copy_sighand(clone_flags, p);
2122 goto bad_fork_cleanup_fs;
2123 retval = copy_signal(clone_flags, p);
2125 goto bad_fork_cleanup_sighand;
2126 retval = copy_mm(clone_flags, p);
2128 goto bad_fork_cleanup_signal;
2129 retval = copy_namespaces(clone_flags, p);
2131 goto bad_fork_cleanup_mm;
2132 retval = copy_io(clone_flags, p);
2134 goto bad_fork_cleanup_namespaces;
2135 retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2137 goto bad_fork_cleanup_io;
2139 stackleak_task_init(p);
2141 if (pid != &init_struct_pid) {
2142 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2143 args->set_tid_size);
2145 retval = PTR_ERR(pid);
2146 goto bad_fork_cleanup_thread;
2151 * This has to happen after we've potentially unshared the file
2152 * descriptor table (so that the pidfd doesn't leak into the child
2153 * if the fd table isn't shared).
2155 if (clone_flags & CLONE_PIDFD) {
2156 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2158 goto bad_fork_free_pid;
2162 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2163 O_RDWR | O_CLOEXEC);
2164 if (IS_ERR(pidfile)) {
2165 put_unused_fd(pidfd);
2166 retval = PTR_ERR(pidfile);
2167 goto bad_fork_free_pid;
2169 get_pid(pid); /* held by pidfile now */
2171 retval = put_user(pidfd, args->pidfd);
2173 goto bad_fork_put_pidfd;
2182 * sigaltstack should be cleared when sharing the same VM
2184 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2188 * Syscall tracing and stepping should be turned off in the
2189 * child regardless of CLONE_PTRACE.
2191 user_disable_single_step(p);
2192 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2193 #ifdef TIF_SYSCALL_EMU
2194 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2196 clear_tsk_latency_tracing(p);
2198 /* ok, now we should be set up.. */
2199 p->pid = pid_nr(pid);
2200 if (clone_flags & CLONE_THREAD) {
2201 p->group_leader = current->group_leader;
2202 p->tgid = current->tgid;
2204 p->group_leader = p;
2209 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2210 p->dirty_paused_when = 0;
2212 p->pdeath_signal = 0;
2213 INIT_LIST_HEAD(&p->thread_group);
2214 p->task_works = NULL;
2215 clear_posix_cputimers_work(p);
2218 * Ensure that the cgroup subsystem policies allow the new process to be
2219 * forked. It should be noted that the new process's css_set can be changed
2220 * between here and cgroup_post_fork() if an organisation operation is in
2223 retval = cgroup_can_fork(p, args);
2225 goto bad_fork_put_pidfd;
2228 * From this point on we must avoid any synchronous user-space
2229 * communication until we take the tasklist-lock. In particular, we do
2230 * not want user-space to be able to predict the process start-time by
2231 * stalling fork(2) after we recorded the start_time but before it is
2232 * visible to the system.
2235 p->start_time = ktime_get_ns();
2236 p->start_boottime = ktime_get_boottime_ns();
2239 * Make it visible to the rest of the system, but dont wake it up yet.
2240 * Need tasklist lock for parent etc handling!
2242 write_lock_irq(&tasklist_lock);
2244 /* CLONE_PARENT re-uses the old parent */
2245 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2246 p->real_parent = current->real_parent;
2247 p->parent_exec_id = current->parent_exec_id;
2248 if (clone_flags & CLONE_THREAD)
2249 p->exit_signal = -1;
2251 p->exit_signal = current->group_leader->exit_signal;
2253 p->real_parent = current;
2254 p->parent_exec_id = current->self_exec_id;
2255 p->exit_signal = args->exit_signal;
2258 klp_copy_process(p);
2260 spin_lock(¤t->sighand->siglock);
2262 rseq_fork(p, clone_flags);
2264 /* Don't start children in a dying pid namespace */
2265 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2267 goto bad_fork_cancel_cgroup;
2270 /* Let kill terminate clone/fork in the middle */
2271 if (fatal_signal_pending(current)) {
2273 goto bad_fork_cancel_cgroup;
2276 /* No more failure paths after this point. */
2279 * Copy seccomp details explicitly here, in case they were changed
2280 * before holding sighand lock.
2284 init_task_pid_links(p);
2285 if (likely(p->pid)) {
2286 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2288 init_task_pid(p, PIDTYPE_PID, pid);
2289 if (thread_group_leader(p)) {
2290 init_task_pid(p, PIDTYPE_TGID, pid);
2291 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2292 init_task_pid(p, PIDTYPE_SID, task_session(current));
2294 if (is_child_reaper(pid)) {
2295 ns_of_pid(pid)->child_reaper = p;
2296 p->signal->flags |= SIGNAL_UNKILLABLE;
2298 p->signal->shared_pending.signal = delayed.signal;
2299 p->signal->tty = tty_kref_get(current->signal->tty);
2301 * Inherit has_child_subreaper flag under the same
2302 * tasklist_lock with adding child to the process tree
2303 * for propagate_has_child_subreaper optimization.
2305 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2306 p->real_parent->signal->is_child_subreaper;
2307 list_add_tail(&p->sibling, &p->real_parent->children);
2308 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2309 attach_pid(p, PIDTYPE_TGID);
2310 attach_pid(p, PIDTYPE_PGID);
2311 attach_pid(p, PIDTYPE_SID);
2312 __this_cpu_inc(process_counts);
2314 current->signal->nr_threads++;
2315 atomic_inc(¤t->signal->live);
2316 refcount_inc(¤t->signal->sigcnt);
2317 task_join_group_stop(p);
2318 list_add_tail_rcu(&p->thread_group,
2319 &p->group_leader->thread_group);
2320 list_add_tail_rcu(&p->thread_node,
2321 &p->signal->thread_head);
2323 attach_pid(p, PIDTYPE_PID);
2327 hlist_del_init(&delayed.node);
2328 spin_unlock(¤t->sighand->siglock);
2329 syscall_tracepoint_update(p);
2330 write_unlock_irq(&tasklist_lock);
2333 fd_install(pidfd, pidfile);
2335 proc_fork_connector(p);
2336 sched_post_fork(p, args);
2337 cgroup_post_fork(p, args);
2340 trace_task_newtask(p, clone_flags);
2341 uprobe_copy_process(p, clone_flags);
2343 copy_oom_score_adj(clone_flags, p);
2347 bad_fork_cancel_cgroup:
2348 spin_unlock(¤t->sighand->siglock);
2349 write_unlock_irq(&tasklist_lock);
2350 cgroup_cancel_fork(p, args);
2352 if (clone_flags & CLONE_PIDFD) {
2354 put_unused_fd(pidfd);
2357 if (pid != &init_struct_pid)
2359 bad_fork_cleanup_thread:
2361 bad_fork_cleanup_io:
2364 bad_fork_cleanup_namespaces:
2365 exit_task_namespaces(p);
2366 bad_fork_cleanup_mm:
2368 mm_clear_owner(p->mm, p);
2371 bad_fork_cleanup_signal:
2372 if (!(clone_flags & CLONE_THREAD))
2373 free_signal_struct(p->signal);
2374 bad_fork_cleanup_sighand:
2375 __cleanup_sighand(p->sighand);
2376 bad_fork_cleanup_fs:
2377 exit_fs(p); /* blocking */
2378 bad_fork_cleanup_files:
2379 exit_files(p); /* blocking */
2380 bad_fork_cleanup_semundo:
2382 bad_fork_cleanup_security:
2383 security_task_free(p);
2384 bad_fork_cleanup_audit:
2386 bad_fork_cleanup_perf:
2387 perf_event_free_task(p);
2388 bad_fork_cleanup_policy:
2389 lockdep_free_task(p);
2391 mpol_put(p->mempolicy);
2392 bad_fork_cleanup_threadgroup_lock:
2394 delayacct_tsk_free(p);
2395 bad_fork_cleanup_count:
2396 atomic_dec(&p->cred->user->processes);
2399 p->state = TASK_DEAD;
2401 delayed_free_task(p);
2403 spin_lock_irq(¤t->sighand->siglock);
2404 hlist_del_init(&delayed.node);
2405 spin_unlock_irq(¤t->sighand->siglock);
2406 return ERR_PTR(retval);
2409 static inline void init_idle_pids(struct task_struct *idle)
2413 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2414 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2415 init_task_pid(idle, type, &init_struct_pid);
2419 struct task_struct * __init fork_idle(int cpu)
2421 struct task_struct *task;
2422 struct kernel_clone_args args = {
2426 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2427 if (!IS_ERR(task)) {
2428 init_idle_pids(task);
2429 init_idle(task, cpu);
2436 * This is like kernel_clone(), but shaved down and tailored to just
2437 * creating io_uring workers. It returns a created task, or an error pointer.
2438 * The returned task is inactive, and the caller must fire it up through
2439 * wake_up_new_task(p). All signals are blocked in the created task.
2441 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2443 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2445 struct kernel_clone_args args = {
2446 .flags = ((lower_32_bits(flags) | CLONE_VM |
2447 CLONE_UNTRACED) & ~CSIGNAL),
2448 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2449 .stack = (unsigned long)fn,
2450 .stack_size = (unsigned long)arg,
2454 return copy_process(NULL, 0, node, &args);
2458 * Ok, this is the main fork-routine.
2460 * It copies the process, and if successful kick-starts
2461 * it and waits for it to finish using the VM if required.
2463 * args->exit_signal is expected to be checked for sanity by the caller.
2465 pid_t kernel_clone(struct kernel_clone_args *args)
2467 u64 clone_flags = args->flags;
2468 struct completion vfork;
2470 struct task_struct *p;
2475 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2476 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2477 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2478 * field in struct clone_args and it still doesn't make sense to have
2479 * them both point at the same memory location. Performing this check
2480 * here has the advantage that we don't need to have a separate helper
2481 * to check for legacy clone().
2483 if ((args->flags & CLONE_PIDFD) &&
2484 (args->flags & CLONE_PARENT_SETTID) &&
2485 (args->pidfd == args->parent_tid))
2489 * Determine whether and which event to report to ptracer. When
2490 * called from kernel_thread or CLONE_UNTRACED is explicitly
2491 * requested, no event is reported; otherwise, report if the event
2492 * for the type of forking is enabled.
2494 if (!(clone_flags & CLONE_UNTRACED)) {
2495 if (clone_flags & CLONE_VFORK)
2496 trace = PTRACE_EVENT_VFORK;
2497 else if (args->exit_signal != SIGCHLD)
2498 trace = PTRACE_EVENT_CLONE;
2500 trace = PTRACE_EVENT_FORK;
2502 if (likely(!ptrace_event_enabled(current, trace)))
2506 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2507 add_latent_entropy();
2513 * Do this prior waking up the new thread - the thread pointer
2514 * might get invalid after that point, if the thread exits quickly.
2516 trace_sched_process_fork(current, p);
2518 pid = get_task_pid(p, PIDTYPE_PID);
2521 if (clone_flags & CLONE_PARENT_SETTID)
2522 put_user(nr, args->parent_tid);
2524 if (clone_flags & CLONE_VFORK) {
2525 p->vfork_done = &vfork;
2526 init_completion(&vfork);
2530 wake_up_new_task(p);
2532 /* forking complete and child started to run, tell ptracer */
2533 if (unlikely(trace))
2534 ptrace_event_pid(trace, pid);
2536 if (clone_flags & CLONE_VFORK) {
2537 if (!wait_for_vfork_done(p, &vfork))
2538 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2546 * Create a kernel thread.
2548 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2550 struct kernel_clone_args args = {
2551 .flags = ((lower_32_bits(flags) | CLONE_VM |
2552 CLONE_UNTRACED) & ~CSIGNAL),
2553 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2554 .stack = (unsigned long)fn,
2555 .stack_size = (unsigned long)arg,
2558 return kernel_clone(&args);
2561 #ifdef __ARCH_WANT_SYS_FORK
2562 SYSCALL_DEFINE0(fork)
2565 struct kernel_clone_args args = {
2566 .exit_signal = SIGCHLD,
2569 return kernel_clone(&args);
2571 /* can not support in nommu mode */
2577 #ifdef __ARCH_WANT_SYS_VFORK
2578 SYSCALL_DEFINE0(vfork)
2580 struct kernel_clone_args args = {
2581 .flags = CLONE_VFORK | CLONE_VM,
2582 .exit_signal = SIGCHLD,
2585 return kernel_clone(&args);
2589 #ifdef __ARCH_WANT_SYS_CLONE
2590 #ifdef CONFIG_CLONE_BACKWARDS
2591 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2592 int __user *, parent_tidptr,
2594 int __user *, child_tidptr)
2595 #elif defined(CONFIG_CLONE_BACKWARDS2)
2596 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2597 int __user *, parent_tidptr,
2598 int __user *, child_tidptr,
2600 #elif defined(CONFIG_CLONE_BACKWARDS3)
2601 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2603 int __user *, parent_tidptr,
2604 int __user *, child_tidptr,
2607 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2608 int __user *, parent_tidptr,
2609 int __user *, child_tidptr,
2613 struct kernel_clone_args args = {
2614 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2615 .pidfd = parent_tidptr,
2616 .child_tid = child_tidptr,
2617 .parent_tid = parent_tidptr,
2618 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2623 return kernel_clone(&args);
2627 #ifdef __ARCH_WANT_SYS_CLONE3
2629 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2630 struct clone_args __user *uargs,
2634 struct clone_args args;
2635 pid_t *kset_tid = kargs->set_tid;
2637 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2638 CLONE_ARGS_SIZE_VER0);
2639 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2640 CLONE_ARGS_SIZE_VER1);
2641 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2642 CLONE_ARGS_SIZE_VER2);
2643 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2645 if (unlikely(usize > PAGE_SIZE))
2647 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2650 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2654 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2657 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2660 if (unlikely(args.set_tid && args.set_tid_size == 0))
2664 * Verify that higher 32bits of exit_signal are unset and that
2665 * it is a valid signal
2667 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2668 !valid_signal(args.exit_signal)))
2671 if ((args.flags & CLONE_INTO_CGROUP) &&
2672 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2675 *kargs = (struct kernel_clone_args){
2676 .flags = args.flags,
2677 .pidfd = u64_to_user_ptr(args.pidfd),
2678 .child_tid = u64_to_user_ptr(args.child_tid),
2679 .parent_tid = u64_to_user_ptr(args.parent_tid),
2680 .exit_signal = args.exit_signal,
2681 .stack = args.stack,
2682 .stack_size = args.stack_size,
2684 .set_tid_size = args.set_tid_size,
2685 .cgroup = args.cgroup,
2689 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2690 (kargs->set_tid_size * sizeof(pid_t))))
2693 kargs->set_tid = kset_tid;
2699 * clone3_stack_valid - check and prepare stack
2700 * @kargs: kernel clone args
2702 * Verify that the stack arguments userspace gave us are sane.
2703 * In addition, set the stack direction for userspace since it's easy for us to
2706 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2708 if (kargs->stack == 0) {
2709 if (kargs->stack_size > 0)
2712 if (kargs->stack_size == 0)
2715 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2718 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2719 kargs->stack += kargs->stack_size;
2726 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2728 /* Verify that no unknown flags are passed along. */
2730 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2734 * - make the CLONE_DETACHED bit reuseable for clone3
2735 * - make the CSIGNAL bits reuseable for clone3
2737 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
2740 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2741 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2744 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2748 if (!clone3_stack_valid(kargs))
2755 * clone3 - create a new process with specific properties
2756 * @uargs: argument structure
2757 * @size: size of @uargs
2759 * clone3() is the extensible successor to clone()/clone2().
2760 * It takes a struct as argument that is versioned by its size.
2762 * Return: On success, a positive PID for the child process.
2763 * On error, a negative errno number.
2765 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2769 struct kernel_clone_args kargs;
2770 pid_t set_tid[MAX_PID_NS_LEVEL];
2772 kargs.set_tid = set_tid;
2774 err = copy_clone_args_from_user(&kargs, uargs, size);
2778 if (!clone3_args_valid(&kargs))
2781 return kernel_clone(&kargs);
2785 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2787 struct task_struct *leader, *parent, *child;
2790 read_lock(&tasklist_lock);
2791 leader = top = top->group_leader;
2793 for_each_thread(leader, parent) {
2794 list_for_each_entry(child, &parent->children, sibling) {
2795 res = visitor(child, data);
2807 if (leader != top) {
2809 parent = child->real_parent;
2810 leader = parent->group_leader;
2814 read_unlock(&tasklist_lock);
2817 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2818 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2821 static void sighand_ctor(void *data)
2823 struct sighand_struct *sighand = data;
2825 spin_lock_init(&sighand->siglock);
2826 init_waitqueue_head(&sighand->signalfd_wqh);
2829 void __init mm_cache_init(void)
2831 unsigned int mm_size;
2834 * The mm_cpumask is located at the end of mm_struct, and is
2835 * dynamically sized based on the maximum CPU number this system
2836 * can have, taking hotplug into account (nr_cpu_ids).
2838 mm_size = sizeof(struct mm_struct) + cpumask_size();
2840 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2841 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2842 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2843 offsetof(struct mm_struct, saved_auxv),
2844 sizeof_field(struct mm_struct, saved_auxv),
2848 void __init proc_caches_init(void)
2850 sighand_cachep = kmem_cache_create("sighand_cache",
2851 sizeof(struct sighand_struct), 0,
2852 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2853 SLAB_ACCOUNT, sighand_ctor);
2854 signal_cachep = kmem_cache_create("signal_cache",
2855 sizeof(struct signal_struct), 0,
2856 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2858 files_cachep = kmem_cache_create("files_cache",
2859 sizeof(struct files_struct), 0,
2860 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2862 fs_cachep = kmem_cache_create("fs_cache",
2863 sizeof(struct fs_struct), 0,
2864 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2867 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2869 nsproxy_cache_init();
2873 * Check constraints on flags passed to the unshare system call.
2875 static int check_unshare_flags(unsigned long unshare_flags)
2877 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2878 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2879 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2880 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2884 * Not implemented, but pretend it works if there is nothing
2885 * to unshare. Note that unsharing the address space or the
2886 * signal handlers also need to unshare the signal queues (aka
2889 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2890 if (!thread_group_empty(current))
2893 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2894 if (refcount_read(¤t->sighand->count) > 1)
2897 if (unshare_flags & CLONE_VM) {
2898 if (!current_is_single_threaded())
2906 * Unshare the filesystem structure if it is being shared
2908 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2910 struct fs_struct *fs = current->fs;
2912 if (!(unshare_flags & CLONE_FS) || !fs)
2915 /* don't need lock here; in the worst case we'll do useless copy */
2919 *new_fsp = copy_fs_struct(fs);
2927 * Unshare file descriptor table if it is being shared
2929 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2930 struct files_struct **new_fdp)
2932 struct files_struct *fd = current->files;
2935 if ((unshare_flags & CLONE_FILES) &&
2936 (fd && atomic_read(&fd->count) > 1)) {
2937 *new_fdp = dup_fd(fd, max_fds, &error);
2946 * unshare allows a process to 'unshare' part of the process
2947 * context which was originally shared using clone. copy_*
2948 * functions used by kernel_clone() cannot be used here directly
2949 * because they modify an inactive task_struct that is being
2950 * constructed. Here we are modifying the current, active,
2953 int ksys_unshare(unsigned long unshare_flags)
2955 struct fs_struct *fs, *new_fs = NULL;
2956 struct files_struct *fd, *new_fd = NULL;
2957 struct cred *new_cred = NULL;
2958 struct nsproxy *new_nsproxy = NULL;
2963 * If unsharing a user namespace must also unshare the thread group
2964 * and unshare the filesystem root and working directories.
2966 if (unshare_flags & CLONE_NEWUSER)
2967 unshare_flags |= CLONE_THREAD | CLONE_FS;
2969 * If unsharing vm, must also unshare signal handlers.
2971 if (unshare_flags & CLONE_VM)
2972 unshare_flags |= CLONE_SIGHAND;
2974 * If unsharing a signal handlers, must also unshare the signal queues.
2976 if (unshare_flags & CLONE_SIGHAND)
2977 unshare_flags |= CLONE_THREAD;
2979 * If unsharing namespace, must also unshare filesystem information.
2981 if (unshare_flags & CLONE_NEWNS)
2982 unshare_flags |= CLONE_FS;
2984 err = check_unshare_flags(unshare_flags);
2986 goto bad_unshare_out;
2988 * CLONE_NEWIPC must also detach from the undolist: after switching
2989 * to a new ipc namespace, the semaphore arrays from the old
2990 * namespace are unreachable.
2992 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2994 err = unshare_fs(unshare_flags, &new_fs);
2996 goto bad_unshare_out;
2997 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2999 goto bad_unshare_cleanup_fs;
3000 err = unshare_userns(unshare_flags, &new_cred);
3002 goto bad_unshare_cleanup_fd;
3003 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3006 goto bad_unshare_cleanup_cred;
3008 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3011 * CLONE_SYSVSEM is equivalent to sys_exit().
3015 if (unshare_flags & CLONE_NEWIPC) {
3016 /* Orphan segments in old ns (see sem above). */
3018 shm_init_task(current);
3022 switch_task_namespaces(current, new_nsproxy);
3028 spin_lock(&fs->lock);
3029 current->fs = new_fs;
3034 spin_unlock(&fs->lock);
3038 fd = current->files;
3039 current->files = new_fd;
3043 task_unlock(current);
3046 /* Install the new user namespace */
3047 commit_creds(new_cred);
3052 perf_event_namespaces(current);
3054 bad_unshare_cleanup_cred:
3057 bad_unshare_cleanup_fd:
3059 put_files_struct(new_fd);
3061 bad_unshare_cleanup_fs:
3063 free_fs_struct(new_fs);
3069 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3071 return ksys_unshare(unshare_flags);
3075 * Helper to unshare the files of the current task.
3076 * We don't want to expose copy_files internals to
3077 * the exec layer of the kernel.
3080 int unshare_files(struct files_struct **displaced)
3082 struct task_struct *task = current;
3083 struct files_struct *copy = NULL;
3086 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3087 if (error || !copy) {
3091 *displaced = task->files;
3098 int sysctl_max_threads(struct ctl_table *table, int write,
3099 void *buffer, size_t *lenp, loff_t *ppos)
3103 int threads = max_threads;
3105 int max = MAX_THREADS;
3112 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3116 max_threads = threads;