4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 'fork.c' contains the help-routines for the 'fork' system call
9 * (see also entry.S and others).
10 * Fork is rather simple, once you get the hang of it, but the memory
11 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
14 #include <linux/anon_inodes.h>
15 #include <linux/slab.h>
16 #include <linux/sched/autogroup.h>
17 #include <linux/sched/mm.h>
18 #include <linux/sched/coredump.h>
19 #include <linux/sched/user.h>
20 #include <linux/sched/numa_balancing.h>
21 #include <linux/sched/stat.h>
22 #include <linux/sched/task.h>
23 #include <linux/sched/task_stack.h>
24 #include <linux/sched/cputime.h>
25 #include <linux/seq_file.h>
26 #include <linux/rtmutex.h>
27 #include <linux/init.h>
28 #include <linux/unistd.h>
29 #include <linux/module.h>
30 #include <linux/vmalloc.h>
31 #include <linux/completion.h>
32 #include <linux/personality.h>
33 #include <linux/mempolicy.h>
34 #include <linux/sem.h>
35 #include <linux/file.h>
36 #include <linux/fdtable.h>
37 #include <linux/iocontext.h>
38 #include <linux/key.h>
39 #include <linux/binfmts.h>
40 #include <linux/mman.h>
41 #include <linux/mmu_notifier.h>
42 #include <linux/hmm.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>
97 #include <asm/pgtable.h>
98 #include <asm/pgalloc.h>
99 #include <linux/uaccess.h>
100 #include <asm/mmu_context.h>
101 #include <asm/cacheflush.h>
102 #include <asm/tlbflush.h>
104 #include <trace/events/sched.h>
106 #define CREATE_TRACE_POINTS
107 #include <trace/events/task.h>
110 * Minimum number of threads to boot the kernel
112 #define MIN_THREADS 20
115 * Maximum number of threads
117 #define MAX_THREADS FUTEX_TID_MASK
120 * Protected counters by write_lock_irq(&tasklist_lock)
122 unsigned long total_forks; /* Handle normal Linux uptimes. */
123 int nr_threads; /* The idle threads do not count.. */
125 int max_threads; /* tunable limit on nr_threads */
127 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
129 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
131 #ifdef CONFIG_PROVE_RCU
132 int lockdep_tasklist_lock_is_held(void)
134 return lockdep_is_held(&tasklist_lock);
136 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
137 #endif /* #ifdef CONFIG_PROVE_RCU */
139 int nr_processes(void)
144 for_each_possible_cpu(cpu)
145 total += per_cpu(process_counts, cpu);
150 void __weak arch_release_task_struct(struct task_struct *tsk)
154 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
155 static struct kmem_cache *task_struct_cachep;
157 static inline struct task_struct *alloc_task_struct_node(int node)
159 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
162 static inline void free_task_struct(struct task_struct *tsk)
164 kmem_cache_free(task_struct_cachep, tsk);
168 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
171 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
172 * kmemcache based allocator.
174 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
176 #ifdef CONFIG_VMAP_STACK
178 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
179 * flush. Try to minimize the number of calls by caching stacks.
181 #define NR_CACHED_STACKS 2
182 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
184 static int free_vm_stack_cache(unsigned int cpu)
186 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
189 for (i = 0; i < NR_CACHED_STACKS; i++) {
190 struct vm_struct *vm_stack = cached_vm_stacks[i];
195 vfree(vm_stack->addr);
196 cached_vm_stacks[i] = NULL;
203 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
205 #ifdef CONFIG_VMAP_STACK
209 for (i = 0; i < NR_CACHED_STACKS; i++) {
212 s = this_cpu_xchg(cached_stacks[i], NULL);
217 /* Clear stale pointers from reused stack. */
218 memset(s->addr, 0, THREAD_SIZE);
220 tsk->stack_vm_area = s;
221 tsk->stack = s->addr;
226 * Allocated stacks are cached and later reused by new threads,
227 * so memcg accounting is performed manually on assigning/releasing
228 * stacks to tasks. Drop __GFP_ACCOUNT.
230 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
231 VMALLOC_START, VMALLOC_END,
232 THREADINFO_GFP & ~__GFP_ACCOUNT,
234 0, node, __builtin_return_address(0));
237 * We can't call find_vm_area() in interrupt context, and
238 * free_thread_stack() can be called in interrupt context,
239 * so cache the vm_struct.
242 tsk->stack_vm_area = find_vm_area(stack);
247 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
250 return page ? page_address(page) : NULL;
254 static inline void free_thread_stack(struct task_struct *tsk)
256 #ifdef CONFIG_VMAP_STACK
257 struct vm_struct *vm = task_stack_vm_area(tsk);
262 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
263 mod_memcg_page_state(vm->pages[i],
264 MEMCG_KERNEL_STACK_KB,
265 -(int)(PAGE_SIZE / 1024));
267 memcg_kmem_uncharge(vm->pages[i], 0);
270 for (i = 0; i < NR_CACHED_STACKS; i++) {
271 if (this_cpu_cmpxchg(cached_stacks[i],
272 NULL, tsk->stack_vm_area) != NULL)
278 vfree_atomic(tsk->stack);
283 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
286 static struct kmem_cache *thread_stack_cache;
288 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
291 unsigned long *stack;
292 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
297 static void free_thread_stack(struct task_struct *tsk)
299 kmem_cache_free(thread_stack_cache, tsk->stack);
302 void thread_stack_cache_init(void)
304 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
305 THREAD_SIZE, THREAD_SIZE, 0, 0,
307 BUG_ON(thread_stack_cache == NULL);
312 /* SLAB cache for signal_struct structures (tsk->signal) */
313 static struct kmem_cache *signal_cachep;
315 /* SLAB cache for sighand_struct structures (tsk->sighand) */
316 struct kmem_cache *sighand_cachep;
318 /* SLAB cache for files_struct structures (tsk->files) */
319 struct kmem_cache *files_cachep;
321 /* SLAB cache for fs_struct structures (tsk->fs) */
322 struct kmem_cache *fs_cachep;
324 /* SLAB cache for vm_area_struct structures */
325 static struct kmem_cache *vm_area_cachep;
327 /* SLAB cache for mm_struct structures (tsk->mm) */
328 static struct kmem_cache *mm_cachep;
330 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
332 struct vm_area_struct *vma;
334 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
340 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
342 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
346 INIT_LIST_HEAD(&new->anon_vma_chain);
351 void vm_area_free(struct vm_area_struct *vma)
353 kmem_cache_free(vm_area_cachep, vma);
356 static void account_kernel_stack(struct task_struct *tsk, int account)
358 void *stack = task_stack_page(tsk);
359 struct vm_struct *vm = task_stack_vm_area(tsk);
361 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
366 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
368 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
369 mod_zone_page_state(page_zone(vm->pages[i]),
371 PAGE_SIZE / 1024 * account);
375 * All stack pages are in the same zone and belong to the
378 struct page *first_page = virt_to_page(stack);
380 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
381 THREAD_SIZE / 1024 * account);
383 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
384 account * (THREAD_SIZE / 1024));
388 static int memcg_charge_kernel_stack(struct task_struct *tsk)
390 #ifdef CONFIG_VMAP_STACK
391 struct vm_struct *vm = task_stack_vm_area(tsk);
397 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
399 * If memcg_kmem_charge() fails, page->mem_cgroup
400 * pointer is NULL, and both memcg_kmem_uncharge()
401 * and mod_memcg_page_state() in free_thread_stack()
402 * will ignore this page. So it's safe.
404 ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
408 mod_memcg_page_state(vm->pages[i],
409 MEMCG_KERNEL_STACK_KB,
417 static void release_task_stack(struct task_struct *tsk)
419 if (WARN_ON(tsk->state != TASK_DEAD))
420 return; /* Better to leak the stack than to free prematurely */
422 account_kernel_stack(tsk, -1);
423 free_thread_stack(tsk);
425 #ifdef CONFIG_VMAP_STACK
426 tsk->stack_vm_area = NULL;
430 #ifdef CONFIG_THREAD_INFO_IN_TASK
431 void put_task_stack(struct task_struct *tsk)
433 if (refcount_dec_and_test(&tsk->stack_refcount))
434 release_task_stack(tsk);
438 void free_task(struct task_struct *tsk)
440 #ifndef CONFIG_THREAD_INFO_IN_TASK
442 * The task is finally done with both the stack and thread_info,
445 release_task_stack(tsk);
448 * If the task had a separate stack allocation, it should be gone
451 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
453 rt_mutex_debug_task_free(tsk);
454 ftrace_graph_exit_task(tsk);
455 put_seccomp_filter(tsk);
456 arch_release_task_struct(tsk);
457 if (tsk->flags & PF_KTHREAD)
458 free_kthread_struct(tsk);
459 free_task_struct(tsk);
461 EXPORT_SYMBOL(free_task);
464 static __latent_entropy int dup_mmap(struct mm_struct *mm,
465 struct mm_struct *oldmm)
467 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
468 struct rb_node **rb_link, *rb_parent;
470 unsigned long charge;
473 uprobe_start_dup_mmap();
474 if (down_write_killable(&oldmm->mmap_sem)) {
476 goto fail_uprobe_end;
478 flush_cache_dup_mm(oldmm);
479 uprobe_dup_mmap(oldmm, mm);
481 * Not linked in yet - no deadlock potential:
483 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
485 /* No ordering required: file already has been exposed. */
486 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
488 mm->total_vm = oldmm->total_vm;
489 mm->data_vm = oldmm->data_vm;
490 mm->exec_vm = oldmm->exec_vm;
491 mm->stack_vm = oldmm->stack_vm;
493 rb_link = &mm->mm_rb.rb_node;
496 retval = ksm_fork(mm, oldmm);
499 retval = khugepaged_fork(mm, oldmm);
504 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
507 if (mpnt->vm_flags & VM_DONTCOPY) {
508 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
513 * Don't duplicate many vmas if we've been oom-killed (for
516 if (fatal_signal_pending(current)) {
520 if (mpnt->vm_flags & VM_ACCOUNT) {
521 unsigned long len = vma_pages(mpnt);
523 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
527 tmp = vm_area_dup(mpnt);
530 retval = vma_dup_policy(mpnt, tmp);
532 goto fail_nomem_policy;
534 retval = dup_userfaultfd(tmp, &uf);
536 goto fail_nomem_anon_vma_fork;
537 if (tmp->vm_flags & VM_WIPEONFORK) {
538 /* VM_WIPEONFORK gets a clean slate in the child. */
539 tmp->anon_vma = NULL;
540 if (anon_vma_prepare(tmp))
541 goto fail_nomem_anon_vma_fork;
542 } else if (anon_vma_fork(tmp, mpnt))
543 goto fail_nomem_anon_vma_fork;
544 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
545 tmp->vm_next = tmp->vm_prev = NULL;
548 struct inode *inode = file_inode(file);
549 struct address_space *mapping = file->f_mapping;
552 if (tmp->vm_flags & VM_DENYWRITE)
553 atomic_dec(&inode->i_writecount);
554 i_mmap_lock_write(mapping);
555 if (tmp->vm_flags & VM_SHARED)
556 atomic_inc(&mapping->i_mmap_writable);
557 flush_dcache_mmap_lock(mapping);
558 /* insert tmp into the share list, just after mpnt */
559 vma_interval_tree_insert_after(tmp, mpnt,
561 flush_dcache_mmap_unlock(mapping);
562 i_mmap_unlock_write(mapping);
566 * Clear hugetlb-related page reserves for children. This only
567 * affects MAP_PRIVATE mappings. Faults generated by the child
568 * are not guaranteed to succeed, even if read-only
570 if (is_vm_hugetlb_page(tmp))
571 reset_vma_resv_huge_pages(tmp);
574 * Link in the new vma and copy the page table entries.
577 pprev = &tmp->vm_next;
581 __vma_link_rb(mm, tmp, rb_link, rb_parent);
582 rb_link = &tmp->vm_rb.rb_right;
583 rb_parent = &tmp->vm_rb;
586 if (!(tmp->vm_flags & VM_WIPEONFORK))
587 retval = copy_page_range(mm, oldmm, mpnt);
589 if (tmp->vm_ops && tmp->vm_ops->open)
590 tmp->vm_ops->open(tmp);
595 /* a new mm has just been created */
596 retval = arch_dup_mmap(oldmm, mm);
598 up_write(&mm->mmap_sem);
600 up_write(&oldmm->mmap_sem);
601 dup_userfaultfd_complete(&uf);
603 uprobe_end_dup_mmap();
605 fail_nomem_anon_vma_fork:
606 mpol_put(vma_policy(tmp));
611 vm_unacct_memory(charge);
615 static inline int mm_alloc_pgd(struct mm_struct *mm)
617 mm->pgd = pgd_alloc(mm);
618 if (unlikely(!mm->pgd))
623 static inline void mm_free_pgd(struct mm_struct *mm)
625 pgd_free(mm, mm->pgd);
628 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
630 down_write(&oldmm->mmap_sem);
631 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
632 up_write(&oldmm->mmap_sem);
635 #define mm_alloc_pgd(mm) (0)
636 #define mm_free_pgd(mm)
637 #endif /* CONFIG_MMU */
639 static void check_mm(struct mm_struct *mm)
643 for (i = 0; i < NR_MM_COUNTERS; i++) {
644 long x = atomic_long_read(&mm->rss_stat.count[i]);
647 printk(KERN_ALERT "BUG: Bad rss-counter state "
648 "mm:%p idx:%d val:%ld\n", mm, i, x);
651 if (mm_pgtables_bytes(mm))
652 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
653 mm_pgtables_bytes(mm));
655 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
656 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
660 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
661 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
664 * Called when the last reference to the mm
665 * is dropped: either by a lazy thread or by
666 * mmput. Free the page directory and the mm.
668 void __mmdrop(struct mm_struct *mm)
670 BUG_ON(mm == &init_mm);
671 WARN_ON_ONCE(mm == current->mm);
672 WARN_ON_ONCE(mm == current->active_mm);
676 mmu_notifier_mm_destroy(mm);
678 put_user_ns(mm->user_ns);
681 EXPORT_SYMBOL_GPL(__mmdrop);
683 static void mmdrop_async_fn(struct work_struct *work)
685 struct mm_struct *mm;
687 mm = container_of(work, struct mm_struct, async_put_work);
691 static void mmdrop_async(struct mm_struct *mm)
693 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
694 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
695 schedule_work(&mm->async_put_work);
699 static inline void free_signal_struct(struct signal_struct *sig)
701 taskstats_tgid_free(sig);
702 sched_autogroup_exit(sig);
704 * __mmdrop is not safe to call from softirq context on x86 due to
705 * pgd_dtor so postpone it to the async context
708 mmdrop_async(sig->oom_mm);
709 kmem_cache_free(signal_cachep, sig);
712 static inline void put_signal_struct(struct signal_struct *sig)
714 if (refcount_dec_and_test(&sig->sigcnt))
715 free_signal_struct(sig);
718 void __put_task_struct(struct task_struct *tsk)
720 WARN_ON(!tsk->exit_state);
721 WARN_ON(refcount_read(&tsk->usage));
722 WARN_ON(tsk == current);
726 security_task_free(tsk);
728 delayacct_tsk_free(tsk);
729 put_signal_struct(tsk->signal);
731 if (!profile_handoff_task(tsk))
734 EXPORT_SYMBOL_GPL(__put_task_struct);
736 void __init __weak arch_task_cache_init(void) { }
741 static void set_max_threads(unsigned int max_threads_suggested)
744 unsigned long nr_pages = totalram_pages();
747 * The number of threads shall be limited such that the thread
748 * structures may only consume a small part of the available memory.
750 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
751 threads = MAX_THREADS;
753 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
754 (u64) THREAD_SIZE * 8UL);
756 if (threads > max_threads_suggested)
757 threads = max_threads_suggested;
759 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
762 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
763 /* Initialized by the architecture: */
764 int arch_task_struct_size __read_mostly;
767 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
769 /* Fetch thread_struct whitelist for the architecture. */
770 arch_thread_struct_whitelist(offset, size);
773 * Handle zero-sized whitelist or empty thread_struct, otherwise
774 * adjust offset to position of thread_struct in task_struct.
776 if (unlikely(*size == 0))
779 *offset += offsetof(struct task_struct, thread);
782 void __init fork_init(void)
785 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
786 #ifndef ARCH_MIN_TASKALIGN
787 #define ARCH_MIN_TASKALIGN 0
789 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
790 unsigned long useroffset, usersize;
792 /* create a slab on which task_structs can be allocated */
793 task_struct_whitelist(&useroffset, &usersize);
794 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
795 arch_task_struct_size, align,
796 SLAB_PANIC|SLAB_ACCOUNT,
797 useroffset, usersize, NULL);
800 /* do the arch specific task caches init */
801 arch_task_cache_init();
803 set_max_threads(MAX_THREADS);
805 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
806 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
807 init_task.signal->rlim[RLIMIT_SIGPENDING] =
808 init_task.signal->rlim[RLIMIT_NPROC];
810 for (i = 0; i < UCOUNT_COUNTS; i++) {
811 init_user_ns.ucount_max[i] = max_threads/2;
814 #ifdef CONFIG_VMAP_STACK
815 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
816 NULL, free_vm_stack_cache);
819 lockdep_init_task(&init_task);
823 int __weak arch_dup_task_struct(struct task_struct *dst,
824 struct task_struct *src)
830 void set_task_stack_end_magic(struct task_struct *tsk)
832 unsigned long *stackend;
834 stackend = end_of_stack(tsk);
835 *stackend = STACK_END_MAGIC; /* for overflow detection */
838 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
840 struct task_struct *tsk;
841 unsigned long *stack;
842 struct vm_struct *stack_vm_area __maybe_unused;
845 if (node == NUMA_NO_NODE)
846 node = tsk_fork_get_node(orig);
847 tsk = alloc_task_struct_node(node);
851 stack = alloc_thread_stack_node(tsk, node);
855 if (memcg_charge_kernel_stack(tsk))
858 stack_vm_area = task_stack_vm_area(tsk);
860 err = arch_dup_task_struct(tsk, orig);
863 * arch_dup_task_struct() clobbers the stack-related fields. Make
864 * sure they're properly initialized before using any stack-related
868 #ifdef CONFIG_VMAP_STACK
869 tsk->stack_vm_area = stack_vm_area;
871 #ifdef CONFIG_THREAD_INFO_IN_TASK
872 refcount_set(&tsk->stack_refcount, 1);
878 #ifdef CONFIG_SECCOMP
880 * We must handle setting up seccomp filters once we're under
881 * the sighand lock in case orig has changed between now and
882 * then. Until then, filter must be NULL to avoid messing up
883 * the usage counts on the error path calling free_task.
885 tsk->seccomp.filter = NULL;
888 setup_thread_stack(tsk, orig);
889 clear_user_return_notifier(tsk);
890 clear_tsk_need_resched(tsk);
891 set_task_stack_end_magic(tsk);
893 #ifdef CONFIG_STACKPROTECTOR
894 tsk->stack_canary = get_random_canary();
898 * One for us, one for whoever does the "release_task()" (usually
901 refcount_set(&tsk->usage, 2);
902 #ifdef CONFIG_BLK_DEV_IO_TRACE
905 tsk->splice_pipe = NULL;
906 tsk->task_frag.page = NULL;
907 tsk->wake_q.next = NULL;
909 account_kernel_stack(tsk, 1);
913 #ifdef CONFIG_FAULT_INJECTION
917 #ifdef CONFIG_BLK_CGROUP
918 tsk->throttle_queue = NULL;
919 tsk->use_memdelay = 0;
923 tsk->active_memcg = NULL;
928 free_thread_stack(tsk);
930 free_task_struct(tsk);
934 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
936 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
938 static int __init coredump_filter_setup(char *s)
940 default_dump_filter =
941 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
942 MMF_DUMP_FILTER_MASK;
946 __setup("coredump_filter=", coredump_filter_setup);
948 #include <linux/init_task.h>
950 static void mm_init_aio(struct mm_struct *mm)
953 spin_lock_init(&mm->ioctx_lock);
954 mm->ioctx_table = NULL;
958 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
965 static void mm_init_uprobes_state(struct mm_struct *mm)
967 #ifdef CONFIG_UPROBES
968 mm->uprobes_state.xol_area = NULL;
972 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
973 struct user_namespace *user_ns)
977 mm->vmacache_seqnum = 0;
978 atomic_set(&mm->mm_users, 1);
979 atomic_set(&mm->mm_count, 1);
980 init_rwsem(&mm->mmap_sem);
981 INIT_LIST_HEAD(&mm->mmlist);
982 mm->core_state = NULL;
983 mm_pgtables_bytes_init(mm);
986 atomic64_set(&mm->pinned_vm, 0);
987 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
988 spin_lock_init(&mm->page_table_lock);
989 spin_lock_init(&mm->arg_lock);
992 mm_init_owner(mm, p);
993 RCU_INIT_POINTER(mm->exe_file, NULL);
994 mmu_notifier_mm_init(mm);
996 init_tlb_flush_pending(mm);
997 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
998 mm->pmd_huge_pte = NULL;
1000 mm_init_uprobes_state(mm);
1003 mm->flags = current->mm->flags & MMF_INIT_MASK;
1004 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1006 mm->flags = default_dump_filter;
1010 if (mm_alloc_pgd(mm))
1013 if (init_new_context(p, mm))
1014 goto fail_nocontext;
1016 mm->user_ns = get_user_ns(user_ns);
1027 * Allocate and initialize an mm_struct.
1029 struct mm_struct *mm_alloc(void)
1031 struct mm_struct *mm;
1037 memset(mm, 0, sizeof(*mm));
1038 return mm_init(mm, current, current_user_ns());
1041 static inline void __mmput(struct mm_struct *mm)
1043 VM_BUG_ON(atomic_read(&mm->mm_users));
1045 uprobe_clear_state(mm);
1048 khugepaged_exit(mm); /* must run before exit_mmap */
1050 mm_put_huge_zero_page(mm);
1051 set_mm_exe_file(mm, NULL);
1052 if (!list_empty(&mm->mmlist)) {
1053 spin_lock(&mmlist_lock);
1054 list_del(&mm->mmlist);
1055 spin_unlock(&mmlist_lock);
1058 module_put(mm->binfmt->module);
1063 * Decrement the use count and release all resources for an mm.
1065 void mmput(struct mm_struct *mm)
1069 if (atomic_dec_and_test(&mm->mm_users))
1072 EXPORT_SYMBOL_GPL(mmput);
1075 static void mmput_async_fn(struct work_struct *work)
1077 struct mm_struct *mm = container_of(work, struct mm_struct,
1083 void mmput_async(struct mm_struct *mm)
1085 if (atomic_dec_and_test(&mm->mm_users)) {
1086 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1087 schedule_work(&mm->async_put_work);
1093 * set_mm_exe_file - change a reference to the mm's executable file
1095 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1097 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1098 * invocations: in mmput() nobody alive left, in execve task is single
1099 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1100 * mm->exe_file, but does so without using set_mm_exe_file() in order
1101 * to do avoid the need for any locks.
1103 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1105 struct file *old_exe_file;
1108 * It is safe to dereference the exe_file without RCU as
1109 * this function is only called if nobody else can access
1110 * this mm -- see comment above for justification.
1112 old_exe_file = rcu_dereference_raw(mm->exe_file);
1115 get_file(new_exe_file);
1116 rcu_assign_pointer(mm->exe_file, new_exe_file);
1122 * get_mm_exe_file - acquire a reference to the mm's executable file
1124 * Returns %NULL if mm has no associated executable file.
1125 * User must release file via fput().
1127 struct file *get_mm_exe_file(struct mm_struct *mm)
1129 struct file *exe_file;
1132 exe_file = rcu_dereference(mm->exe_file);
1133 if (exe_file && !get_file_rcu(exe_file))
1138 EXPORT_SYMBOL(get_mm_exe_file);
1141 * get_task_exe_file - acquire a reference to the task's executable file
1143 * Returns %NULL if task's mm (if any) has no associated executable file or
1144 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1145 * User must release file via fput().
1147 struct file *get_task_exe_file(struct task_struct *task)
1149 struct file *exe_file = NULL;
1150 struct mm_struct *mm;
1155 if (!(task->flags & PF_KTHREAD))
1156 exe_file = get_mm_exe_file(mm);
1161 EXPORT_SYMBOL(get_task_exe_file);
1164 * get_task_mm - acquire a reference to the task's mm
1166 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1167 * this kernel workthread has transiently adopted a user mm with use_mm,
1168 * to do its AIO) is not set and if so returns a reference to it, after
1169 * bumping up the use count. User must release the mm via mmput()
1170 * after use. Typically used by /proc and ptrace.
1172 struct mm_struct *get_task_mm(struct task_struct *task)
1174 struct mm_struct *mm;
1179 if (task->flags & PF_KTHREAD)
1187 EXPORT_SYMBOL_GPL(get_task_mm);
1189 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1191 struct mm_struct *mm;
1194 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1196 return ERR_PTR(err);
1198 mm = get_task_mm(task);
1199 if (mm && mm != current->mm &&
1200 !ptrace_may_access(task, mode)) {
1202 mm = ERR_PTR(-EACCES);
1204 mutex_unlock(&task->signal->cred_guard_mutex);
1209 static void complete_vfork_done(struct task_struct *tsk)
1211 struct completion *vfork;
1214 vfork = tsk->vfork_done;
1215 if (likely(vfork)) {
1216 tsk->vfork_done = NULL;
1222 static int wait_for_vfork_done(struct task_struct *child,
1223 struct completion *vfork)
1227 freezer_do_not_count();
1228 killed = wait_for_completion_killable(vfork);
1233 child->vfork_done = NULL;
1237 put_task_struct(child);
1241 /* Please note the differences between mmput and mm_release.
1242 * mmput is called whenever we stop holding onto a mm_struct,
1243 * error success whatever.
1245 * mm_release is called after a mm_struct has been removed
1246 * from the current process.
1248 * This difference is important for error handling, when we
1249 * only half set up a mm_struct for a new process and need to restore
1250 * the old one. Because we mmput the new mm_struct before
1251 * restoring the old one. . .
1252 * Eric Biederman 10 January 1998
1254 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1256 /* Get rid of any futexes when releasing the mm */
1258 if (unlikely(tsk->robust_list)) {
1259 exit_robust_list(tsk);
1260 tsk->robust_list = NULL;
1262 #ifdef CONFIG_COMPAT
1263 if (unlikely(tsk->compat_robust_list)) {
1264 compat_exit_robust_list(tsk);
1265 tsk->compat_robust_list = NULL;
1268 if (unlikely(!list_empty(&tsk->pi_state_list)))
1269 exit_pi_state_list(tsk);
1272 uprobe_free_utask(tsk);
1274 /* Get rid of any cached register state */
1275 deactivate_mm(tsk, mm);
1278 * Signal userspace if we're not exiting with a core dump
1279 * because we want to leave the value intact for debugging
1282 if (tsk->clear_child_tid) {
1283 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1284 atomic_read(&mm->mm_users) > 1) {
1286 * We don't check the error code - if userspace has
1287 * not set up a proper pointer then tough luck.
1289 put_user(0, tsk->clear_child_tid);
1290 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1291 1, NULL, NULL, 0, 0);
1293 tsk->clear_child_tid = NULL;
1297 * All done, finally we can wake up parent and return this mm to him.
1298 * Also kthread_stop() uses this completion for synchronization.
1300 if (tsk->vfork_done)
1301 complete_vfork_done(tsk);
1305 * dup_mm() - duplicates an existing mm structure
1306 * @tsk: the task_struct with which the new mm will be associated.
1307 * @oldmm: the mm to duplicate.
1309 * Allocates a new mm structure and duplicates the provided @oldmm structure
1312 * Return: the duplicated mm or NULL on failure.
1314 static struct mm_struct *dup_mm(struct task_struct *tsk,
1315 struct mm_struct *oldmm)
1317 struct mm_struct *mm;
1324 memcpy(mm, oldmm, sizeof(*mm));
1326 if (!mm_init(mm, tsk, mm->user_ns))
1329 err = dup_mmap(mm, oldmm);
1333 mm->hiwater_rss = get_mm_rss(mm);
1334 mm->hiwater_vm = mm->total_vm;
1336 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1342 /* don't put binfmt in mmput, we haven't got module yet */
1350 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1352 struct mm_struct *mm, *oldmm;
1355 tsk->min_flt = tsk->maj_flt = 0;
1356 tsk->nvcsw = tsk->nivcsw = 0;
1357 #ifdef CONFIG_DETECT_HUNG_TASK
1358 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1359 tsk->last_switch_time = 0;
1363 tsk->active_mm = NULL;
1366 * Are we cloning a kernel thread?
1368 * We need to steal a active VM for that..
1370 oldmm = current->mm;
1374 /* initialize the new vmacache entries */
1375 vmacache_flush(tsk);
1377 if (clone_flags & CLONE_VM) {
1384 mm = dup_mm(tsk, current->mm);
1390 tsk->active_mm = mm;
1397 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1399 struct fs_struct *fs = current->fs;
1400 if (clone_flags & CLONE_FS) {
1401 /* tsk->fs is already what we want */
1402 spin_lock(&fs->lock);
1404 spin_unlock(&fs->lock);
1408 spin_unlock(&fs->lock);
1411 tsk->fs = copy_fs_struct(fs);
1417 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1419 struct files_struct *oldf, *newf;
1423 * A background process may not have any files ...
1425 oldf = current->files;
1429 if (clone_flags & CLONE_FILES) {
1430 atomic_inc(&oldf->count);
1434 newf = dup_fd(oldf, &error);
1444 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1447 struct io_context *ioc = current->io_context;
1448 struct io_context *new_ioc;
1453 * Share io context with parent, if CLONE_IO is set
1455 if (clone_flags & CLONE_IO) {
1457 tsk->io_context = ioc;
1458 } else if (ioprio_valid(ioc->ioprio)) {
1459 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1460 if (unlikely(!new_ioc))
1463 new_ioc->ioprio = ioc->ioprio;
1464 put_io_context(new_ioc);
1470 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1472 struct sighand_struct *sig;
1474 if (clone_flags & CLONE_SIGHAND) {
1475 refcount_inc(¤t->sighand->count);
1478 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1479 rcu_assign_pointer(tsk->sighand, sig);
1483 refcount_set(&sig->count, 1);
1484 spin_lock_irq(¤t->sighand->siglock);
1485 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1486 spin_unlock_irq(¤t->sighand->siglock);
1490 void __cleanup_sighand(struct sighand_struct *sighand)
1492 if (refcount_dec_and_test(&sighand->count)) {
1493 signalfd_cleanup(sighand);
1495 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1496 * without an RCU grace period, see __lock_task_sighand().
1498 kmem_cache_free(sighand_cachep, sighand);
1502 #ifdef CONFIG_POSIX_TIMERS
1504 * Initialize POSIX timer handling for a thread group.
1506 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1508 unsigned long cpu_limit;
1510 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1511 if (cpu_limit != RLIM_INFINITY) {
1512 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1513 sig->cputimer.running = true;
1516 /* The timer lists. */
1517 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1518 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1519 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1522 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1525 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1527 struct signal_struct *sig;
1529 if (clone_flags & CLONE_THREAD)
1532 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1537 sig->nr_threads = 1;
1538 atomic_set(&sig->live, 1);
1539 refcount_set(&sig->sigcnt, 1);
1541 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1542 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1543 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1545 init_waitqueue_head(&sig->wait_chldexit);
1546 sig->curr_target = tsk;
1547 init_sigpending(&sig->shared_pending);
1548 INIT_HLIST_HEAD(&sig->multiprocess);
1549 seqlock_init(&sig->stats_lock);
1550 prev_cputime_init(&sig->prev_cputime);
1552 #ifdef CONFIG_POSIX_TIMERS
1553 INIT_LIST_HEAD(&sig->posix_timers);
1554 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1555 sig->real_timer.function = it_real_fn;
1558 task_lock(current->group_leader);
1559 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1560 task_unlock(current->group_leader);
1562 posix_cpu_timers_init_group(sig);
1564 tty_audit_fork(sig);
1565 sched_autogroup_fork(sig);
1567 sig->oom_score_adj = current->signal->oom_score_adj;
1568 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1570 mutex_init(&sig->cred_guard_mutex);
1575 static void copy_seccomp(struct task_struct *p)
1577 #ifdef CONFIG_SECCOMP
1579 * Must be called with sighand->lock held, which is common to
1580 * all threads in the group. Holding cred_guard_mutex is not
1581 * needed because this new task is not yet running and cannot
1584 assert_spin_locked(¤t->sighand->siglock);
1586 /* Ref-count the new filter user, and assign it. */
1587 get_seccomp_filter(current);
1588 p->seccomp = current->seccomp;
1591 * Explicitly enable no_new_privs here in case it got set
1592 * between the task_struct being duplicated and holding the
1593 * sighand lock. The seccomp state and nnp must be in sync.
1595 if (task_no_new_privs(current))
1596 task_set_no_new_privs(p);
1599 * If the parent gained a seccomp mode after copying thread
1600 * flags and between before we held the sighand lock, we have
1601 * to manually enable the seccomp thread flag here.
1603 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1604 set_tsk_thread_flag(p, TIF_SECCOMP);
1608 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1610 current->clear_child_tid = tidptr;
1612 return task_pid_vnr(current);
1615 static void rt_mutex_init_task(struct task_struct *p)
1617 raw_spin_lock_init(&p->pi_lock);
1618 #ifdef CONFIG_RT_MUTEXES
1619 p->pi_waiters = RB_ROOT_CACHED;
1620 p->pi_top_task = NULL;
1621 p->pi_blocked_on = NULL;
1625 #ifdef CONFIG_POSIX_TIMERS
1627 * Initialize POSIX timer handling for a single task.
1629 static void posix_cpu_timers_init(struct task_struct *tsk)
1631 tsk->cputime_expires.prof_exp = 0;
1632 tsk->cputime_expires.virt_exp = 0;
1633 tsk->cputime_expires.sched_exp = 0;
1634 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1635 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1636 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1639 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1642 static inline void init_task_pid_links(struct task_struct *task)
1646 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1647 INIT_HLIST_NODE(&task->pid_links[type]);
1652 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1654 if (type == PIDTYPE_PID)
1655 task->thread_pid = pid;
1657 task->signal->pids[type] = pid;
1660 static inline void rcu_copy_process(struct task_struct *p)
1662 #ifdef CONFIG_PREEMPT_RCU
1663 p->rcu_read_lock_nesting = 0;
1664 p->rcu_read_unlock_special.s = 0;
1665 p->rcu_blocked_node = NULL;
1666 INIT_LIST_HEAD(&p->rcu_node_entry);
1667 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1668 #ifdef CONFIG_TASKS_RCU
1669 p->rcu_tasks_holdout = false;
1670 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1671 p->rcu_tasks_idle_cpu = -1;
1672 #endif /* #ifdef CONFIG_TASKS_RCU */
1675 static int pidfd_release(struct inode *inode, struct file *file)
1677 struct pid *pid = file->private_data;
1679 file->private_data = NULL;
1684 #ifdef CONFIG_PROC_FS
1685 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1687 struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1688 struct pid *pid = f->private_data;
1690 seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1695 const struct file_operations pidfd_fops = {
1696 .release = pidfd_release,
1697 #ifdef CONFIG_PROC_FS
1698 .show_fdinfo = pidfd_show_fdinfo,
1703 * pidfd_create() - Create a new pid file descriptor.
1705 * @pid: struct pid that the pidfd will reference
1707 * This creates a new pid file descriptor with the O_CLOEXEC flag set.
1709 * Note, that this function can only be called after the fd table has
1710 * been unshared to avoid leaking the pidfd to the new process.
1712 * Return: On success, a cloexec pidfd is returned.
1713 * On error, a negative errno number will be returned.
1715 static int pidfd_create(struct pid *pid)
1719 fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
1720 O_RDWR | O_CLOEXEC);
1728 * This creates a new process as a copy of the old one,
1729 * but does not actually start it yet.
1731 * It copies the registers, and all the appropriate
1732 * parts of the process environment (as per the clone
1733 * flags). The actual kick-off is left to the caller.
1735 static __latent_entropy struct task_struct *copy_process(
1736 unsigned long clone_flags,
1737 unsigned long stack_start,
1738 unsigned long stack_size,
1739 int __user *parent_tidptr,
1740 int __user *child_tidptr,
1746 int pidfd = -1, retval;
1747 struct task_struct *p;
1748 struct multiprocess_signals delayed;
1751 * Don't allow sharing the root directory with processes in a different
1754 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1755 return ERR_PTR(-EINVAL);
1757 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1758 return ERR_PTR(-EINVAL);
1761 * Thread groups must share signals as well, and detached threads
1762 * can only be started up within the thread group.
1764 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1765 return ERR_PTR(-EINVAL);
1768 * Shared signal handlers imply shared VM. By way of the above,
1769 * thread groups also imply shared VM. Blocking this case allows
1770 * for various simplifications in other code.
1772 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1773 return ERR_PTR(-EINVAL);
1776 * Siblings of global init remain as zombies on exit since they are
1777 * not reaped by their parent (swapper). To solve this and to avoid
1778 * multi-rooted process trees, prevent global and container-inits
1779 * from creating siblings.
1781 if ((clone_flags & CLONE_PARENT) &&
1782 current->signal->flags & SIGNAL_UNKILLABLE)
1783 return ERR_PTR(-EINVAL);
1786 * If the new process will be in a different pid or user namespace
1787 * do not allow it to share a thread group with the forking task.
1789 if (clone_flags & CLONE_THREAD) {
1790 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1791 (task_active_pid_ns(current) !=
1792 current->nsproxy->pid_ns_for_children))
1793 return ERR_PTR(-EINVAL);
1796 if (clone_flags & CLONE_PIDFD) {
1800 * - CLONE_PARENT_SETTID is useless for pidfds and also
1801 * parent_tidptr is used to return pidfds.
1802 * - CLONE_DETACHED is blocked so that we can potentially
1803 * reuse it later for CLONE_PIDFD.
1804 * - CLONE_THREAD is blocked until someone really needs it.
1807 (CLONE_DETACHED | CLONE_PARENT_SETTID | CLONE_THREAD))
1808 return ERR_PTR(-EINVAL);
1811 * Verify that parent_tidptr is sane so we can potentially
1814 if (get_user(reserved, parent_tidptr))
1815 return ERR_PTR(-EFAULT);
1818 return ERR_PTR(-EINVAL);
1822 * Force any signals received before this point to be delivered
1823 * before the fork happens. Collect up signals sent to multiple
1824 * processes that happen during the fork and delay them so that
1825 * they appear to happen after the fork.
1827 sigemptyset(&delayed.signal);
1828 INIT_HLIST_NODE(&delayed.node);
1830 spin_lock_irq(¤t->sighand->siglock);
1831 if (!(clone_flags & CLONE_THREAD))
1832 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1833 recalc_sigpending();
1834 spin_unlock_irq(¤t->sighand->siglock);
1835 retval = -ERESTARTNOINTR;
1836 if (signal_pending(current))
1840 p = dup_task_struct(current, node);
1845 * This _must_ happen before we call free_task(), i.e. before we jump
1846 * to any of the bad_fork_* labels. This is to avoid freeing
1847 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1848 * kernel threads (PF_KTHREAD).
1850 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1852 * Clear TID on mm_release()?
1854 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1856 ftrace_graph_init_task(p);
1858 rt_mutex_init_task(p);
1860 #ifdef CONFIG_PROVE_LOCKING
1861 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1862 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1865 if (atomic_read(&p->real_cred->user->processes) >=
1866 task_rlimit(p, RLIMIT_NPROC)) {
1867 if (p->real_cred->user != INIT_USER &&
1868 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1871 current->flags &= ~PF_NPROC_EXCEEDED;
1873 retval = copy_creds(p, clone_flags);
1878 * If multiple threads are within copy_process(), then this check
1879 * triggers too late. This doesn't hurt, the check is only there
1880 * to stop root fork bombs.
1883 if (nr_threads >= max_threads)
1884 goto bad_fork_cleanup_count;
1886 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1887 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1888 p->flags |= PF_FORKNOEXEC;
1889 INIT_LIST_HEAD(&p->children);
1890 INIT_LIST_HEAD(&p->sibling);
1891 rcu_copy_process(p);
1892 p->vfork_done = NULL;
1893 spin_lock_init(&p->alloc_lock);
1895 init_sigpending(&p->pending);
1897 p->utime = p->stime = p->gtime = 0;
1898 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1899 p->utimescaled = p->stimescaled = 0;
1901 prev_cputime_init(&p->prev_cputime);
1903 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1904 seqcount_init(&p->vtime.seqcount);
1905 p->vtime.starttime = 0;
1906 p->vtime.state = VTIME_INACTIVE;
1909 #if defined(SPLIT_RSS_COUNTING)
1910 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1913 p->default_timer_slack_ns = current->timer_slack_ns;
1919 task_io_accounting_init(&p->ioac);
1920 acct_clear_integrals(p);
1922 posix_cpu_timers_init(p);
1924 p->io_context = NULL;
1925 audit_set_context(p, NULL);
1928 p->mempolicy = mpol_dup(p->mempolicy);
1929 if (IS_ERR(p->mempolicy)) {
1930 retval = PTR_ERR(p->mempolicy);
1931 p->mempolicy = NULL;
1932 goto bad_fork_cleanup_threadgroup_lock;
1935 #ifdef CONFIG_CPUSETS
1936 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1937 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1938 seqcount_init(&p->mems_allowed_seq);
1940 #ifdef CONFIG_TRACE_IRQFLAGS
1942 p->hardirqs_enabled = 0;
1943 p->hardirq_enable_ip = 0;
1944 p->hardirq_enable_event = 0;
1945 p->hardirq_disable_ip = _THIS_IP_;
1946 p->hardirq_disable_event = 0;
1947 p->softirqs_enabled = 1;
1948 p->softirq_enable_ip = _THIS_IP_;
1949 p->softirq_enable_event = 0;
1950 p->softirq_disable_ip = 0;
1951 p->softirq_disable_event = 0;
1952 p->hardirq_context = 0;
1953 p->softirq_context = 0;
1956 p->pagefault_disabled = 0;
1958 #ifdef CONFIG_LOCKDEP
1959 p->lockdep_depth = 0; /* no locks held yet */
1960 p->curr_chain_key = 0;
1961 p->lockdep_recursion = 0;
1962 lockdep_init_task(p);
1965 #ifdef CONFIG_DEBUG_MUTEXES
1966 p->blocked_on = NULL; /* not blocked yet */
1968 #ifdef CONFIG_BCACHE
1969 p->sequential_io = 0;
1970 p->sequential_io_avg = 0;
1973 /* Perform scheduler related setup. Assign this task to a CPU. */
1974 retval = sched_fork(clone_flags, p);
1976 goto bad_fork_cleanup_policy;
1978 retval = perf_event_init_task(p);
1980 goto bad_fork_cleanup_policy;
1981 retval = audit_alloc(p);
1983 goto bad_fork_cleanup_perf;
1984 /* copy all the process information */
1986 retval = security_task_alloc(p, clone_flags);
1988 goto bad_fork_cleanup_audit;
1989 retval = copy_semundo(clone_flags, p);
1991 goto bad_fork_cleanup_security;
1992 retval = copy_files(clone_flags, p);
1994 goto bad_fork_cleanup_semundo;
1995 retval = copy_fs(clone_flags, p);
1997 goto bad_fork_cleanup_files;
1998 retval = copy_sighand(clone_flags, p);
2000 goto bad_fork_cleanup_fs;
2001 retval = copy_signal(clone_flags, p);
2003 goto bad_fork_cleanup_sighand;
2004 retval = copy_mm(clone_flags, p);
2006 goto bad_fork_cleanup_signal;
2007 retval = copy_namespaces(clone_flags, p);
2009 goto bad_fork_cleanup_mm;
2010 retval = copy_io(clone_flags, p);
2012 goto bad_fork_cleanup_namespaces;
2013 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
2015 goto bad_fork_cleanup_io;
2017 stackleak_task_init(p);
2019 if (pid != &init_struct_pid) {
2020 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
2022 retval = PTR_ERR(pid);
2023 goto bad_fork_cleanup_thread;
2028 * This has to happen after we've potentially unshared the file
2029 * descriptor table (so that the pidfd doesn't leak into the child
2030 * if the fd table isn't shared).
2032 if (clone_flags & CLONE_PIDFD) {
2033 retval = pidfd_create(pid);
2035 goto bad_fork_free_pid;
2038 retval = put_user(pidfd, parent_tidptr);
2040 goto bad_fork_put_pidfd;
2047 p->robust_list = NULL;
2048 #ifdef CONFIG_COMPAT
2049 p->compat_robust_list = NULL;
2051 INIT_LIST_HEAD(&p->pi_state_list);
2052 p->pi_state_cache = NULL;
2055 * sigaltstack should be cleared when sharing the same VM
2057 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2061 * Syscall tracing and stepping should be turned off in the
2062 * child regardless of CLONE_PTRACE.
2064 user_disable_single_step(p);
2065 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2066 #ifdef TIF_SYSCALL_EMU
2067 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2069 clear_all_latency_tracing(p);
2071 /* ok, now we should be set up.. */
2072 p->pid = pid_nr(pid);
2073 if (clone_flags & CLONE_THREAD) {
2074 p->exit_signal = -1;
2075 p->group_leader = current->group_leader;
2076 p->tgid = current->tgid;
2078 if (clone_flags & CLONE_PARENT)
2079 p->exit_signal = current->group_leader->exit_signal;
2081 p->exit_signal = (clone_flags & CSIGNAL);
2082 p->group_leader = p;
2087 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2088 p->dirty_paused_when = 0;
2090 p->pdeath_signal = 0;
2091 INIT_LIST_HEAD(&p->thread_group);
2092 p->task_works = NULL;
2094 cgroup_threadgroup_change_begin(current);
2096 * Ensure that the cgroup subsystem policies allow the new process to be
2097 * forked. It should be noted the the new process's css_set can be changed
2098 * between here and cgroup_post_fork() if an organisation operation is in
2101 retval = cgroup_can_fork(p);
2103 goto bad_fork_put_pidfd;
2106 * From this point on we must avoid any synchronous user-space
2107 * communication until we take the tasklist-lock. In particular, we do
2108 * not want user-space to be able to predict the process start-time by
2109 * stalling fork(2) after we recorded the start_time but before it is
2110 * visible to the system.
2113 p->start_time = ktime_get_ns();
2114 p->real_start_time = ktime_get_boot_ns();
2117 * Make it visible to the rest of the system, but dont wake it up yet.
2118 * Need tasklist lock for parent etc handling!
2120 write_lock_irq(&tasklist_lock);
2122 /* CLONE_PARENT re-uses the old parent */
2123 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2124 p->real_parent = current->real_parent;
2125 p->parent_exec_id = current->parent_exec_id;
2127 p->real_parent = current;
2128 p->parent_exec_id = current->self_exec_id;
2131 klp_copy_process(p);
2133 spin_lock(¤t->sighand->siglock);
2136 * Copy seccomp details explicitly here, in case they were changed
2137 * before holding sighand lock.
2141 rseq_fork(p, clone_flags);
2143 /* Don't start children in a dying pid namespace */
2144 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2146 goto bad_fork_cancel_cgroup;
2149 /* Let kill terminate clone/fork in the middle */
2150 if (fatal_signal_pending(current)) {
2152 goto bad_fork_cancel_cgroup;
2156 init_task_pid_links(p);
2157 if (likely(p->pid)) {
2158 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2160 init_task_pid(p, PIDTYPE_PID, pid);
2161 if (thread_group_leader(p)) {
2162 init_task_pid(p, PIDTYPE_TGID, pid);
2163 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2164 init_task_pid(p, PIDTYPE_SID, task_session(current));
2166 if (is_child_reaper(pid)) {
2167 ns_of_pid(pid)->child_reaper = p;
2168 p->signal->flags |= SIGNAL_UNKILLABLE;
2170 p->signal->shared_pending.signal = delayed.signal;
2171 p->signal->tty = tty_kref_get(current->signal->tty);
2173 * Inherit has_child_subreaper flag under the same
2174 * tasklist_lock with adding child to the process tree
2175 * for propagate_has_child_subreaper optimization.
2177 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2178 p->real_parent->signal->is_child_subreaper;
2179 list_add_tail(&p->sibling, &p->real_parent->children);
2180 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2181 attach_pid(p, PIDTYPE_TGID);
2182 attach_pid(p, PIDTYPE_PGID);
2183 attach_pid(p, PIDTYPE_SID);
2184 __this_cpu_inc(process_counts);
2186 current->signal->nr_threads++;
2187 atomic_inc(¤t->signal->live);
2188 refcount_inc(¤t->signal->sigcnt);
2189 task_join_group_stop(p);
2190 list_add_tail_rcu(&p->thread_group,
2191 &p->group_leader->thread_group);
2192 list_add_tail_rcu(&p->thread_node,
2193 &p->signal->thread_head);
2195 attach_pid(p, PIDTYPE_PID);
2199 hlist_del_init(&delayed.node);
2200 spin_unlock(¤t->sighand->siglock);
2201 syscall_tracepoint_update(p);
2202 write_unlock_irq(&tasklist_lock);
2204 proc_fork_connector(p);
2205 cgroup_post_fork(p);
2206 cgroup_threadgroup_change_end(current);
2209 trace_task_newtask(p, clone_flags);
2210 uprobe_copy_process(p, clone_flags);
2214 bad_fork_cancel_cgroup:
2215 spin_unlock(¤t->sighand->siglock);
2216 write_unlock_irq(&tasklist_lock);
2217 cgroup_cancel_fork(p);
2219 if (clone_flags & CLONE_PIDFD)
2222 cgroup_threadgroup_change_end(current);
2223 if (pid != &init_struct_pid)
2225 bad_fork_cleanup_thread:
2227 bad_fork_cleanup_io:
2230 bad_fork_cleanup_namespaces:
2231 exit_task_namespaces(p);
2232 bad_fork_cleanup_mm:
2235 bad_fork_cleanup_signal:
2236 if (!(clone_flags & CLONE_THREAD))
2237 free_signal_struct(p->signal);
2238 bad_fork_cleanup_sighand:
2239 __cleanup_sighand(p->sighand);
2240 bad_fork_cleanup_fs:
2241 exit_fs(p); /* blocking */
2242 bad_fork_cleanup_files:
2243 exit_files(p); /* blocking */
2244 bad_fork_cleanup_semundo:
2246 bad_fork_cleanup_security:
2247 security_task_free(p);
2248 bad_fork_cleanup_audit:
2250 bad_fork_cleanup_perf:
2251 perf_event_free_task(p);
2252 bad_fork_cleanup_policy:
2253 lockdep_free_task(p);
2255 mpol_put(p->mempolicy);
2256 bad_fork_cleanup_threadgroup_lock:
2258 delayacct_tsk_free(p);
2259 bad_fork_cleanup_count:
2260 atomic_dec(&p->cred->user->processes);
2263 p->state = TASK_DEAD;
2267 spin_lock_irq(¤t->sighand->siglock);
2268 hlist_del_init(&delayed.node);
2269 spin_unlock_irq(¤t->sighand->siglock);
2270 return ERR_PTR(retval);
2273 static inline void init_idle_pids(struct task_struct *idle)
2277 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2278 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2279 init_task_pid(idle, type, &init_struct_pid);
2283 struct task_struct *fork_idle(int cpu)
2285 struct task_struct *task;
2286 task = copy_process(CLONE_VM, 0, 0, NULL, NULL, &init_struct_pid, 0, 0,
2288 if (!IS_ERR(task)) {
2289 init_idle_pids(task);
2290 init_idle(task, cpu);
2296 struct mm_struct *copy_init_mm(void)
2298 return dup_mm(NULL, &init_mm);
2302 * Ok, this is the main fork-routine.
2304 * It copies the process, and if successful kick-starts
2305 * it and waits for it to finish using the VM if required.
2307 long _do_fork(unsigned long clone_flags,
2308 unsigned long stack_start,
2309 unsigned long stack_size,
2310 int __user *parent_tidptr,
2311 int __user *child_tidptr,
2314 struct completion vfork;
2316 struct task_struct *p;
2321 * Determine whether and which event to report to ptracer. When
2322 * called from kernel_thread or CLONE_UNTRACED is explicitly
2323 * requested, no event is reported; otherwise, report if the event
2324 * for the type of forking is enabled.
2326 if (!(clone_flags & CLONE_UNTRACED)) {
2327 if (clone_flags & CLONE_VFORK)
2328 trace = PTRACE_EVENT_VFORK;
2329 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2330 trace = PTRACE_EVENT_CLONE;
2332 trace = PTRACE_EVENT_FORK;
2334 if (likely(!ptrace_event_enabled(current, trace)))
2338 p = copy_process(clone_flags, stack_start, stack_size, parent_tidptr,
2339 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2340 add_latent_entropy();
2346 * Do this prior waking up the new thread - the thread pointer
2347 * might get invalid after that point, if the thread exits quickly.
2349 trace_sched_process_fork(current, p);
2351 pid = get_task_pid(p, PIDTYPE_PID);
2354 if (clone_flags & CLONE_PARENT_SETTID)
2355 put_user(nr, parent_tidptr);
2357 if (clone_flags & CLONE_VFORK) {
2358 p->vfork_done = &vfork;
2359 init_completion(&vfork);
2363 wake_up_new_task(p);
2365 /* forking complete and child started to run, tell ptracer */
2366 if (unlikely(trace))
2367 ptrace_event_pid(trace, pid);
2369 if (clone_flags & CLONE_VFORK) {
2370 if (!wait_for_vfork_done(p, &vfork))
2371 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2378 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2379 /* For compatibility with architectures that call do_fork directly rather than
2380 * using the syscall entry points below. */
2381 long do_fork(unsigned long clone_flags,
2382 unsigned long stack_start,
2383 unsigned long stack_size,
2384 int __user *parent_tidptr,
2385 int __user *child_tidptr)
2387 return _do_fork(clone_flags, stack_start, stack_size,
2388 parent_tidptr, child_tidptr, 0);
2393 * Create a kernel thread.
2395 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2397 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2398 (unsigned long)arg, NULL, NULL, 0);
2401 #ifdef __ARCH_WANT_SYS_FORK
2402 SYSCALL_DEFINE0(fork)
2405 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2407 /* can not support in nommu mode */
2413 #ifdef __ARCH_WANT_SYS_VFORK
2414 SYSCALL_DEFINE0(vfork)
2416 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2421 #ifdef __ARCH_WANT_SYS_CLONE
2422 #ifdef CONFIG_CLONE_BACKWARDS
2423 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2424 int __user *, parent_tidptr,
2426 int __user *, child_tidptr)
2427 #elif defined(CONFIG_CLONE_BACKWARDS2)
2428 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2429 int __user *, parent_tidptr,
2430 int __user *, child_tidptr,
2432 #elif defined(CONFIG_CLONE_BACKWARDS3)
2433 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2435 int __user *, parent_tidptr,
2436 int __user *, child_tidptr,
2439 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2440 int __user *, parent_tidptr,
2441 int __user *, child_tidptr,
2445 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2449 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2451 struct task_struct *leader, *parent, *child;
2454 read_lock(&tasklist_lock);
2455 leader = top = top->group_leader;
2457 for_each_thread(leader, parent) {
2458 list_for_each_entry(child, &parent->children, sibling) {
2459 res = visitor(child, data);
2471 if (leader != top) {
2473 parent = child->real_parent;
2474 leader = parent->group_leader;
2478 read_unlock(&tasklist_lock);
2481 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2482 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2485 static void sighand_ctor(void *data)
2487 struct sighand_struct *sighand = data;
2489 spin_lock_init(&sighand->siglock);
2490 init_waitqueue_head(&sighand->signalfd_wqh);
2493 void __init proc_caches_init(void)
2495 unsigned int mm_size;
2497 sighand_cachep = kmem_cache_create("sighand_cache",
2498 sizeof(struct sighand_struct), 0,
2499 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2500 SLAB_ACCOUNT, sighand_ctor);
2501 signal_cachep = kmem_cache_create("signal_cache",
2502 sizeof(struct signal_struct), 0,
2503 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2505 files_cachep = kmem_cache_create("files_cache",
2506 sizeof(struct files_struct), 0,
2507 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2509 fs_cachep = kmem_cache_create("fs_cache",
2510 sizeof(struct fs_struct), 0,
2511 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2515 * The mm_cpumask is located at the end of mm_struct, and is
2516 * dynamically sized based on the maximum CPU number this system
2517 * can have, taking hotplug into account (nr_cpu_ids).
2519 mm_size = sizeof(struct mm_struct) + cpumask_size();
2521 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2522 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2523 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2524 offsetof(struct mm_struct, saved_auxv),
2525 sizeof_field(struct mm_struct, saved_auxv),
2527 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2529 nsproxy_cache_init();
2533 * Check constraints on flags passed to the unshare system call.
2535 static int check_unshare_flags(unsigned long unshare_flags)
2537 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2538 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2539 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2540 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2543 * Not implemented, but pretend it works if there is nothing
2544 * to unshare. Note that unsharing the address space or the
2545 * signal handlers also need to unshare the signal queues (aka
2548 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2549 if (!thread_group_empty(current))
2552 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2553 if (refcount_read(¤t->sighand->count) > 1)
2556 if (unshare_flags & CLONE_VM) {
2557 if (!current_is_single_threaded())
2565 * Unshare the filesystem structure if it is being shared
2567 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2569 struct fs_struct *fs = current->fs;
2571 if (!(unshare_flags & CLONE_FS) || !fs)
2574 /* don't need lock here; in the worst case we'll do useless copy */
2578 *new_fsp = copy_fs_struct(fs);
2586 * Unshare file descriptor table if it is being shared
2588 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2590 struct files_struct *fd = current->files;
2593 if ((unshare_flags & CLONE_FILES) &&
2594 (fd && atomic_read(&fd->count) > 1)) {
2595 *new_fdp = dup_fd(fd, &error);
2604 * unshare allows a process to 'unshare' part of the process
2605 * context which was originally shared using clone. copy_*
2606 * functions used by do_fork() cannot be used here directly
2607 * because they modify an inactive task_struct that is being
2608 * constructed. Here we are modifying the current, active,
2611 int ksys_unshare(unsigned long unshare_flags)
2613 struct fs_struct *fs, *new_fs = NULL;
2614 struct files_struct *fd, *new_fd = NULL;
2615 struct cred *new_cred = NULL;
2616 struct nsproxy *new_nsproxy = NULL;
2621 * If unsharing a user namespace must also unshare the thread group
2622 * and unshare the filesystem root and working directories.
2624 if (unshare_flags & CLONE_NEWUSER)
2625 unshare_flags |= CLONE_THREAD | CLONE_FS;
2627 * If unsharing vm, must also unshare signal handlers.
2629 if (unshare_flags & CLONE_VM)
2630 unshare_flags |= CLONE_SIGHAND;
2632 * If unsharing a signal handlers, must also unshare the signal queues.
2634 if (unshare_flags & CLONE_SIGHAND)
2635 unshare_flags |= CLONE_THREAD;
2637 * If unsharing namespace, must also unshare filesystem information.
2639 if (unshare_flags & CLONE_NEWNS)
2640 unshare_flags |= CLONE_FS;
2642 err = check_unshare_flags(unshare_flags);
2644 goto bad_unshare_out;
2646 * CLONE_NEWIPC must also detach from the undolist: after switching
2647 * to a new ipc namespace, the semaphore arrays from the old
2648 * namespace are unreachable.
2650 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2652 err = unshare_fs(unshare_flags, &new_fs);
2654 goto bad_unshare_out;
2655 err = unshare_fd(unshare_flags, &new_fd);
2657 goto bad_unshare_cleanup_fs;
2658 err = unshare_userns(unshare_flags, &new_cred);
2660 goto bad_unshare_cleanup_fd;
2661 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2664 goto bad_unshare_cleanup_cred;
2666 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2669 * CLONE_SYSVSEM is equivalent to sys_exit().
2673 if (unshare_flags & CLONE_NEWIPC) {
2674 /* Orphan segments in old ns (see sem above). */
2676 shm_init_task(current);
2680 switch_task_namespaces(current, new_nsproxy);
2686 spin_lock(&fs->lock);
2687 current->fs = new_fs;
2692 spin_unlock(&fs->lock);
2696 fd = current->files;
2697 current->files = new_fd;
2701 task_unlock(current);
2704 /* Install the new user namespace */
2705 commit_creds(new_cred);
2710 perf_event_namespaces(current);
2712 bad_unshare_cleanup_cred:
2715 bad_unshare_cleanup_fd:
2717 put_files_struct(new_fd);
2719 bad_unshare_cleanup_fs:
2721 free_fs_struct(new_fs);
2727 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2729 return ksys_unshare(unshare_flags);
2733 * Helper to unshare the files of the current task.
2734 * We don't want to expose copy_files internals to
2735 * the exec layer of the kernel.
2738 int unshare_files(struct files_struct **displaced)
2740 struct task_struct *task = current;
2741 struct files_struct *copy = NULL;
2744 error = unshare_fd(CLONE_FILES, ©);
2745 if (error || !copy) {
2749 *displaced = task->files;
2756 int sysctl_max_threads(struct ctl_table *table, int write,
2757 void __user *buffer, size_t *lenp, loff_t *ppos)
2761 int threads = max_threads;
2762 int min = MIN_THREADS;
2763 int max = MAX_THREADS;
2770 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2774 set_max_threads(threads);