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/slab.h>
15 #include <linux/sched/autogroup.h>
16 #include <linux/sched/mm.h>
17 #include <linux/sched/coredump.h>
18 #include <linux/sched/user.h>
19 #include <linux/init.h>
20 #include <linux/unistd.h>
21 #include <linux/module.h>
22 #include <linux/vmalloc.h>
23 #include <linux/completion.h>
24 #include <linux/personality.h>
25 #include <linux/mempolicy.h>
26 #include <linux/sem.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/iocontext.h>
30 #include <linux/key.h>
31 #include <linux/binfmts.h>
32 #include <linux/mman.h>
33 #include <linux/mmu_notifier.h>
36 #include <linux/vmacache.h>
37 #include <linux/nsproxy.h>
38 #include <linux/capability.h>
39 #include <linux/cpu.h>
40 #include <linux/cgroup.h>
41 #include <linux/security.h>
42 #include <linux/hugetlb.h>
43 #include <linux/seccomp.h>
44 #include <linux/swap.h>
45 #include <linux/syscalls.h>
46 #include <linux/jiffies.h>
47 #include <linux/futex.h>
48 #include <linux/compat.h>
49 #include <linux/kthread.h>
50 #include <linux/task_io_accounting_ops.h>
51 #include <linux/rcupdate.h>
52 #include <linux/ptrace.h>
53 #include <linux/mount.h>
54 #include <linux/audit.h>
55 #include <linux/memcontrol.h>
56 #include <linux/ftrace.h>
57 #include <linux/proc_fs.h>
58 #include <linux/profile.h>
59 #include <linux/rmap.h>
60 #include <linux/ksm.h>
61 #include <linux/acct.h>
62 #include <linux/userfaultfd_k.h>
63 #include <linux/tsacct_kern.h>
64 #include <linux/cn_proc.h>
65 #include <linux/freezer.h>
66 #include <linux/delayacct.h>
67 #include <linux/taskstats_kern.h>
68 #include <linux/random.h>
69 #include <linux/tty.h>
70 #include <linux/blkdev.h>
71 #include <linux/fs_struct.h>
72 #include <linux/magic.h>
73 #include <linux/perf_event.h>
74 #include <linux/posix-timers.h>
75 #include <linux/user-return-notifier.h>
76 #include <linux/oom.h>
77 #include <linux/khugepaged.h>
78 #include <linux/signalfd.h>
79 #include <linux/uprobes.h>
80 #include <linux/aio.h>
81 #include <linux/compiler.h>
82 #include <linux/sysctl.h>
83 #include <linux/kcov.h>
85 #include <asm/pgtable.h>
86 #include <asm/pgalloc.h>
87 #include <linux/uaccess.h>
88 #include <asm/mmu_context.h>
89 #include <asm/cacheflush.h>
90 #include <asm/tlbflush.h>
92 #include <trace/events/sched.h>
94 #define CREATE_TRACE_POINTS
95 #include <trace/events/task.h>
98 * Minimum number of threads to boot the kernel
100 #define MIN_THREADS 20
103 * Maximum number of threads
105 #define MAX_THREADS FUTEX_TID_MASK
108 * Protected counters by write_lock_irq(&tasklist_lock)
110 unsigned long total_forks; /* Handle normal Linux uptimes. */
111 int nr_threads; /* The idle threads do not count.. */
113 int max_threads; /* tunable limit on nr_threads */
115 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
117 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
119 #ifdef CONFIG_PROVE_RCU
120 int lockdep_tasklist_lock_is_held(void)
122 return lockdep_is_held(&tasklist_lock);
124 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
125 #endif /* #ifdef CONFIG_PROVE_RCU */
127 int nr_processes(void)
132 for_each_possible_cpu(cpu)
133 total += per_cpu(process_counts, cpu);
138 void __weak arch_release_task_struct(struct task_struct *tsk)
142 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
143 static struct kmem_cache *task_struct_cachep;
145 static inline struct task_struct *alloc_task_struct_node(int node)
147 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
150 static inline void free_task_struct(struct task_struct *tsk)
152 kmem_cache_free(task_struct_cachep, tsk);
156 void __weak arch_release_thread_stack(unsigned long *stack)
160 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
163 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
164 * kmemcache based allocator.
166 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
168 #ifdef CONFIG_VMAP_STACK
170 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
171 * flush. Try to minimize the number of calls by caching stacks.
173 #define NR_CACHED_STACKS 2
174 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
177 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
179 #ifdef CONFIG_VMAP_STACK
184 for (i = 0; i < NR_CACHED_STACKS; i++) {
185 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
189 this_cpu_write(cached_stacks[i], NULL);
191 tsk->stack_vm_area = s;
197 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
198 VMALLOC_START, VMALLOC_END,
199 THREADINFO_GFP | __GFP_HIGHMEM,
201 0, node, __builtin_return_address(0));
204 * We can't call find_vm_area() in interrupt context, and
205 * free_thread_stack() can be called in interrupt context,
206 * so cache the vm_struct.
209 tsk->stack_vm_area = find_vm_area(stack);
212 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
215 return page ? page_address(page) : NULL;
219 static inline void free_thread_stack(struct task_struct *tsk)
221 #ifdef CONFIG_VMAP_STACK
222 if (task_stack_vm_area(tsk)) {
226 local_irq_save(flags);
227 for (i = 0; i < NR_CACHED_STACKS; i++) {
228 if (this_cpu_read(cached_stacks[i]))
231 this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
232 local_irq_restore(flags);
235 local_irq_restore(flags);
237 vfree_atomic(tsk->stack);
242 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
245 static struct kmem_cache *thread_stack_cache;
247 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
250 return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
253 static void free_thread_stack(struct task_struct *tsk)
255 kmem_cache_free(thread_stack_cache, tsk->stack);
258 void thread_stack_cache_init(void)
260 thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
261 THREAD_SIZE, 0, NULL);
262 BUG_ON(thread_stack_cache == NULL);
267 /* SLAB cache for signal_struct structures (tsk->signal) */
268 static struct kmem_cache *signal_cachep;
270 /* SLAB cache for sighand_struct structures (tsk->sighand) */
271 struct kmem_cache *sighand_cachep;
273 /* SLAB cache for files_struct structures (tsk->files) */
274 struct kmem_cache *files_cachep;
276 /* SLAB cache for fs_struct structures (tsk->fs) */
277 struct kmem_cache *fs_cachep;
279 /* SLAB cache for vm_area_struct structures */
280 struct kmem_cache *vm_area_cachep;
282 /* SLAB cache for mm_struct structures (tsk->mm) */
283 static struct kmem_cache *mm_cachep;
285 static void account_kernel_stack(struct task_struct *tsk, int account)
287 void *stack = task_stack_page(tsk);
288 struct vm_struct *vm = task_stack_vm_area(tsk);
290 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
295 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
297 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
298 mod_zone_page_state(page_zone(vm->pages[i]),
300 PAGE_SIZE / 1024 * account);
303 /* All stack pages belong to the same memcg. */
304 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
305 account * (THREAD_SIZE / 1024));
308 * All stack pages are in the same zone and belong to the
311 struct page *first_page = virt_to_page(stack);
313 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
314 THREAD_SIZE / 1024 * account);
316 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
317 account * (THREAD_SIZE / 1024));
321 static void release_task_stack(struct task_struct *tsk)
323 if (WARN_ON(tsk->state != TASK_DEAD))
324 return; /* Better to leak the stack than to free prematurely */
326 account_kernel_stack(tsk, -1);
327 arch_release_thread_stack(tsk->stack);
328 free_thread_stack(tsk);
330 #ifdef CONFIG_VMAP_STACK
331 tsk->stack_vm_area = NULL;
335 #ifdef CONFIG_THREAD_INFO_IN_TASK
336 void put_task_stack(struct task_struct *tsk)
338 if (atomic_dec_and_test(&tsk->stack_refcount))
339 release_task_stack(tsk);
343 void free_task(struct task_struct *tsk)
345 #ifndef CONFIG_THREAD_INFO_IN_TASK
347 * The task is finally done with both the stack and thread_info,
350 release_task_stack(tsk);
353 * If the task had a separate stack allocation, it should be gone
356 WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
358 rt_mutex_debug_task_free(tsk);
359 ftrace_graph_exit_task(tsk);
360 put_seccomp_filter(tsk);
361 arch_release_task_struct(tsk);
362 if (tsk->flags & PF_KTHREAD)
363 free_kthread_struct(tsk);
364 free_task_struct(tsk);
366 EXPORT_SYMBOL(free_task);
368 static inline void free_signal_struct(struct signal_struct *sig)
370 taskstats_tgid_free(sig);
371 sched_autogroup_exit(sig);
373 * __mmdrop is not safe to call from softirq context on x86 due to
374 * pgd_dtor so postpone it to the async context
377 mmdrop_async(sig->oom_mm);
378 kmem_cache_free(signal_cachep, sig);
381 static inline void put_signal_struct(struct signal_struct *sig)
383 if (atomic_dec_and_test(&sig->sigcnt))
384 free_signal_struct(sig);
387 void __put_task_struct(struct task_struct *tsk)
389 WARN_ON(!tsk->exit_state);
390 WARN_ON(atomic_read(&tsk->usage));
391 WARN_ON(tsk == current);
395 security_task_free(tsk);
397 delayacct_tsk_free(tsk);
398 put_signal_struct(tsk->signal);
400 if (!profile_handoff_task(tsk))
403 EXPORT_SYMBOL_GPL(__put_task_struct);
405 void __init __weak arch_task_cache_init(void) { }
410 static void set_max_threads(unsigned int max_threads_suggested)
415 * The number of threads shall be limited such that the thread
416 * structures may only consume a small part of the available memory.
418 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
419 threads = MAX_THREADS;
421 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
422 (u64) THREAD_SIZE * 8UL);
424 if (threads > max_threads_suggested)
425 threads = max_threads_suggested;
427 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
430 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
431 /* Initialized by the architecture: */
432 int arch_task_struct_size __read_mostly;
435 void __init fork_init(void)
438 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
439 #ifndef ARCH_MIN_TASKALIGN
440 #define ARCH_MIN_TASKALIGN 0
442 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
444 /* create a slab on which task_structs can be allocated */
445 task_struct_cachep = kmem_cache_create("task_struct",
446 arch_task_struct_size, align,
447 SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
450 /* do the arch specific task caches init */
451 arch_task_cache_init();
453 set_max_threads(MAX_THREADS);
455 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
456 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
457 init_task.signal->rlim[RLIMIT_SIGPENDING] =
458 init_task.signal->rlim[RLIMIT_NPROC];
460 for (i = 0; i < UCOUNT_COUNTS; i++) {
461 init_user_ns.ucount_max[i] = max_threads/2;
465 int __weak arch_dup_task_struct(struct task_struct *dst,
466 struct task_struct *src)
472 void set_task_stack_end_magic(struct task_struct *tsk)
474 unsigned long *stackend;
476 stackend = end_of_stack(tsk);
477 *stackend = STACK_END_MAGIC; /* for overflow detection */
480 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
482 struct task_struct *tsk;
483 unsigned long *stack;
484 struct vm_struct *stack_vm_area;
487 if (node == NUMA_NO_NODE)
488 node = tsk_fork_get_node(orig);
489 tsk = alloc_task_struct_node(node);
493 stack = alloc_thread_stack_node(tsk, node);
497 stack_vm_area = task_stack_vm_area(tsk);
499 err = arch_dup_task_struct(tsk, orig);
502 * arch_dup_task_struct() clobbers the stack-related fields. Make
503 * sure they're properly initialized before using any stack-related
507 #ifdef CONFIG_VMAP_STACK
508 tsk->stack_vm_area = stack_vm_area;
510 #ifdef CONFIG_THREAD_INFO_IN_TASK
511 atomic_set(&tsk->stack_refcount, 1);
517 #ifdef CONFIG_SECCOMP
519 * We must handle setting up seccomp filters once we're under
520 * the sighand lock in case orig has changed between now and
521 * then. Until then, filter must be NULL to avoid messing up
522 * the usage counts on the error path calling free_task.
524 tsk->seccomp.filter = NULL;
527 setup_thread_stack(tsk, orig);
528 clear_user_return_notifier(tsk);
529 clear_tsk_need_resched(tsk);
530 set_task_stack_end_magic(tsk);
532 #ifdef CONFIG_CC_STACKPROTECTOR
533 tsk->stack_canary = get_random_int();
537 * One for us, one for whoever does the "release_task()" (usually
540 atomic_set(&tsk->usage, 2);
541 #ifdef CONFIG_BLK_DEV_IO_TRACE
544 tsk->splice_pipe = NULL;
545 tsk->task_frag.page = NULL;
546 tsk->wake_q.next = NULL;
548 account_kernel_stack(tsk, 1);
555 free_thread_stack(tsk);
557 free_task_struct(tsk);
562 static __latent_entropy int dup_mmap(struct mm_struct *mm,
563 struct mm_struct *oldmm)
565 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
566 struct rb_node **rb_link, *rb_parent;
568 unsigned long charge;
571 uprobe_start_dup_mmap();
572 if (down_write_killable(&oldmm->mmap_sem)) {
574 goto fail_uprobe_end;
576 flush_cache_dup_mm(oldmm);
577 uprobe_dup_mmap(oldmm, mm);
579 * Not linked in yet - no deadlock potential:
581 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
583 /* No ordering required: file already has been exposed. */
584 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
586 mm->total_vm = oldmm->total_vm;
587 mm->data_vm = oldmm->data_vm;
588 mm->exec_vm = oldmm->exec_vm;
589 mm->stack_vm = oldmm->stack_vm;
591 rb_link = &mm->mm_rb.rb_node;
594 retval = ksm_fork(mm, oldmm);
597 retval = khugepaged_fork(mm, oldmm);
602 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
605 if (mpnt->vm_flags & VM_DONTCOPY) {
606 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
610 if (mpnt->vm_flags & VM_ACCOUNT) {
611 unsigned long len = vma_pages(mpnt);
613 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
617 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
621 INIT_LIST_HEAD(&tmp->anon_vma_chain);
622 retval = vma_dup_policy(mpnt, tmp);
624 goto fail_nomem_policy;
626 retval = dup_userfaultfd(tmp, &uf);
628 goto fail_nomem_anon_vma_fork;
629 if (anon_vma_fork(tmp, mpnt))
630 goto fail_nomem_anon_vma_fork;
631 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
632 tmp->vm_next = tmp->vm_prev = NULL;
635 struct inode *inode = file_inode(file);
636 struct address_space *mapping = file->f_mapping;
639 if (tmp->vm_flags & VM_DENYWRITE)
640 atomic_dec(&inode->i_writecount);
641 i_mmap_lock_write(mapping);
642 if (tmp->vm_flags & VM_SHARED)
643 atomic_inc(&mapping->i_mmap_writable);
644 flush_dcache_mmap_lock(mapping);
645 /* insert tmp into the share list, just after mpnt */
646 vma_interval_tree_insert_after(tmp, mpnt,
648 flush_dcache_mmap_unlock(mapping);
649 i_mmap_unlock_write(mapping);
653 * Clear hugetlb-related page reserves for children. This only
654 * affects MAP_PRIVATE mappings. Faults generated by the child
655 * are not guaranteed to succeed, even if read-only
657 if (is_vm_hugetlb_page(tmp))
658 reset_vma_resv_huge_pages(tmp);
661 * Link in the new vma and copy the page table entries.
664 pprev = &tmp->vm_next;
668 __vma_link_rb(mm, tmp, rb_link, rb_parent);
669 rb_link = &tmp->vm_rb.rb_right;
670 rb_parent = &tmp->vm_rb;
673 retval = copy_page_range(mm, oldmm, mpnt);
675 if (tmp->vm_ops && tmp->vm_ops->open)
676 tmp->vm_ops->open(tmp);
681 /* a new mm has just been created */
682 arch_dup_mmap(oldmm, mm);
685 up_write(&mm->mmap_sem);
687 up_write(&oldmm->mmap_sem);
688 dup_userfaultfd_complete(&uf);
690 uprobe_end_dup_mmap();
692 fail_nomem_anon_vma_fork:
693 mpol_put(vma_policy(tmp));
695 kmem_cache_free(vm_area_cachep, tmp);
698 vm_unacct_memory(charge);
702 static inline int mm_alloc_pgd(struct mm_struct *mm)
704 mm->pgd = pgd_alloc(mm);
705 if (unlikely(!mm->pgd))
710 static inline void mm_free_pgd(struct mm_struct *mm)
712 pgd_free(mm, mm->pgd);
715 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
717 down_write(&oldmm->mmap_sem);
718 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
719 up_write(&oldmm->mmap_sem);
722 #define mm_alloc_pgd(mm) (0)
723 #define mm_free_pgd(mm)
724 #endif /* CONFIG_MMU */
726 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
728 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
729 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
731 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
733 static int __init coredump_filter_setup(char *s)
735 default_dump_filter =
736 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
737 MMF_DUMP_FILTER_MASK;
741 __setup("coredump_filter=", coredump_filter_setup);
743 #include <linux/init_task.h>
745 static void mm_init_aio(struct mm_struct *mm)
748 spin_lock_init(&mm->ioctx_lock);
749 mm->ioctx_table = NULL;
753 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
760 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
761 struct user_namespace *user_ns)
765 mm->vmacache_seqnum = 0;
766 atomic_set(&mm->mm_users, 1);
767 atomic_set(&mm->mm_count, 1);
768 init_rwsem(&mm->mmap_sem);
769 INIT_LIST_HEAD(&mm->mmlist);
770 mm->core_state = NULL;
771 atomic_long_set(&mm->nr_ptes, 0);
776 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
777 spin_lock_init(&mm->page_table_lock);
780 mm_init_owner(mm, p);
781 mmu_notifier_mm_init(mm);
782 clear_tlb_flush_pending(mm);
783 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
784 mm->pmd_huge_pte = NULL;
788 mm->flags = current->mm->flags & MMF_INIT_MASK;
789 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
791 mm->flags = default_dump_filter;
795 if (mm_alloc_pgd(mm))
798 if (init_new_context(p, mm))
801 mm->user_ns = get_user_ns(user_ns);
811 static void check_mm(struct mm_struct *mm)
815 for (i = 0; i < NR_MM_COUNTERS; i++) {
816 long x = atomic_long_read(&mm->rss_stat.count[i]);
819 printk(KERN_ALERT "BUG: Bad rss-counter state "
820 "mm:%p idx:%d val:%ld\n", mm, i, x);
823 if (atomic_long_read(&mm->nr_ptes))
824 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
825 atomic_long_read(&mm->nr_ptes));
827 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
830 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
831 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
836 * Allocate and initialize an mm_struct.
838 struct mm_struct *mm_alloc(void)
840 struct mm_struct *mm;
846 memset(mm, 0, sizeof(*mm));
847 return mm_init(mm, current, current_user_ns());
851 * Called when the last reference to the mm
852 * is dropped: either by a lazy thread or by
853 * mmput. Free the page directory and the mm.
855 void __mmdrop(struct mm_struct *mm)
857 BUG_ON(mm == &init_mm);
860 mmu_notifier_mm_destroy(mm);
862 put_user_ns(mm->user_ns);
865 EXPORT_SYMBOL_GPL(__mmdrop);
867 static inline void __mmput(struct mm_struct *mm)
869 VM_BUG_ON(atomic_read(&mm->mm_users));
871 uprobe_clear_state(mm);
874 khugepaged_exit(mm); /* must run before exit_mmap */
876 mm_put_huge_zero_page(mm);
877 set_mm_exe_file(mm, NULL);
878 if (!list_empty(&mm->mmlist)) {
879 spin_lock(&mmlist_lock);
880 list_del(&mm->mmlist);
881 spin_unlock(&mmlist_lock);
884 module_put(mm->binfmt->module);
885 set_bit(MMF_OOM_SKIP, &mm->flags);
890 * Decrement the use count and release all resources for an mm.
892 void mmput(struct mm_struct *mm)
896 if (atomic_dec_and_test(&mm->mm_users))
899 EXPORT_SYMBOL_GPL(mmput);
902 static void mmput_async_fn(struct work_struct *work)
904 struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
908 void mmput_async(struct mm_struct *mm)
910 if (atomic_dec_and_test(&mm->mm_users)) {
911 INIT_WORK(&mm->async_put_work, mmput_async_fn);
912 schedule_work(&mm->async_put_work);
918 * set_mm_exe_file - change a reference to the mm's executable file
920 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
922 * Main users are mmput() and sys_execve(). Callers prevent concurrent
923 * invocations: in mmput() nobody alive left, in execve task is single
924 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
925 * mm->exe_file, but does so without using set_mm_exe_file() in order
926 * to do avoid the need for any locks.
928 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
930 struct file *old_exe_file;
933 * It is safe to dereference the exe_file without RCU as
934 * this function is only called if nobody else can access
935 * this mm -- see comment above for justification.
937 old_exe_file = rcu_dereference_raw(mm->exe_file);
940 get_file(new_exe_file);
941 rcu_assign_pointer(mm->exe_file, new_exe_file);
947 * get_mm_exe_file - acquire a reference to the mm's executable file
949 * Returns %NULL if mm has no associated executable file.
950 * User must release file via fput().
952 struct file *get_mm_exe_file(struct mm_struct *mm)
954 struct file *exe_file;
957 exe_file = rcu_dereference(mm->exe_file);
958 if (exe_file && !get_file_rcu(exe_file))
963 EXPORT_SYMBOL(get_mm_exe_file);
966 * get_task_exe_file - acquire a reference to the task's executable file
968 * Returns %NULL if task's mm (if any) has no associated executable file or
969 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
970 * User must release file via fput().
972 struct file *get_task_exe_file(struct task_struct *task)
974 struct file *exe_file = NULL;
975 struct mm_struct *mm;
980 if (!(task->flags & PF_KTHREAD))
981 exe_file = get_mm_exe_file(mm);
986 EXPORT_SYMBOL(get_task_exe_file);
989 * get_task_mm - acquire a reference to the task's mm
991 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
992 * this kernel workthread has transiently adopted a user mm with use_mm,
993 * to do its AIO) is not set and if so returns a reference to it, after
994 * bumping up the use count. User must release the mm via mmput()
995 * after use. Typically used by /proc and ptrace.
997 struct mm_struct *get_task_mm(struct task_struct *task)
999 struct mm_struct *mm;
1004 if (task->flags & PF_KTHREAD)
1012 EXPORT_SYMBOL_GPL(get_task_mm);
1014 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1016 struct mm_struct *mm;
1019 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1021 return ERR_PTR(err);
1023 mm = get_task_mm(task);
1024 if (mm && mm != current->mm &&
1025 !ptrace_may_access(task, mode)) {
1027 mm = ERR_PTR(-EACCES);
1029 mutex_unlock(&task->signal->cred_guard_mutex);
1034 static void complete_vfork_done(struct task_struct *tsk)
1036 struct completion *vfork;
1039 vfork = tsk->vfork_done;
1040 if (likely(vfork)) {
1041 tsk->vfork_done = NULL;
1047 static int wait_for_vfork_done(struct task_struct *child,
1048 struct completion *vfork)
1052 freezer_do_not_count();
1053 killed = wait_for_completion_killable(vfork);
1058 child->vfork_done = NULL;
1062 put_task_struct(child);
1066 /* Please note the differences between mmput and mm_release.
1067 * mmput is called whenever we stop holding onto a mm_struct,
1068 * error success whatever.
1070 * mm_release is called after a mm_struct has been removed
1071 * from the current process.
1073 * This difference is important for error handling, when we
1074 * only half set up a mm_struct for a new process and need to restore
1075 * the old one. Because we mmput the new mm_struct before
1076 * restoring the old one. . .
1077 * Eric Biederman 10 January 1998
1079 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1081 /* Get rid of any futexes when releasing the mm */
1083 if (unlikely(tsk->robust_list)) {
1084 exit_robust_list(tsk);
1085 tsk->robust_list = NULL;
1087 #ifdef CONFIG_COMPAT
1088 if (unlikely(tsk->compat_robust_list)) {
1089 compat_exit_robust_list(tsk);
1090 tsk->compat_robust_list = NULL;
1093 if (unlikely(!list_empty(&tsk->pi_state_list)))
1094 exit_pi_state_list(tsk);
1097 uprobe_free_utask(tsk);
1099 /* Get rid of any cached register state */
1100 deactivate_mm(tsk, mm);
1103 * Signal userspace if we're not exiting with a core dump
1104 * because we want to leave the value intact for debugging
1107 if (tsk->clear_child_tid) {
1108 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1109 atomic_read(&mm->mm_users) > 1) {
1111 * We don't check the error code - if userspace has
1112 * not set up a proper pointer then tough luck.
1114 put_user(0, tsk->clear_child_tid);
1115 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1118 tsk->clear_child_tid = NULL;
1122 * All done, finally we can wake up parent and return this mm to him.
1123 * Also kthread_stop() uses this completion for synchronization.
1125 if (tsk->vfork_done)
1126 complete_vfork_done(tsk);
1130 * Allocate a new mm structure and copy contents from the
1131 * mm structure of the passed in task structure.
1133 static struct mm_struct *dup_mm(struct task_struct *tsk)
1135 struct mm_struct *mm, *oldmm = current->mm;
1142 memcpy(mm, oldmm, sizeof(*mm));
1144 if (!mm_init(mm, tsk, mm->user_ns))
1147 err = dup_mmap(mm, oldmm);
1151 mm->hiwater_rss = get_mm_rss(mm);
1152 mm->hiwater_vm = mm->total_vm;
1154 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1160 /* don't put binfmt in mmput, we haven't got module yet */
1168 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1170 struct mm_struct *mm, *oldmm;
1173 tsk->min_flt = tsk->maj_flt = 0;
1174 tsk->nvcsw = tsk->nivcsw = 0;
1175 #ifdef CONFIG_DETECT_HUNG_TASK
1176 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1180 tsk->active_mm = NULL;
1183 * Are we cloning a kernel thread?
1185 * We need to steal a active VM for that..
1187 oldmm = current->mm;
1191 /* initialize the new vmacache entries */
1192 vmacache_flush(tsk);
1194 if (clone_flags & CLONE_VM) {
1207 tsk->active_mm = mm;
1214 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1216 struct fs_struct *fs = current->fs;
1217 if (clone_flags & CLONE_FS) {
1218 /* tsk->fs is already what we want */
1219 spin_lock(&fs->lock);
1221 spin_unlock(&fs->lock);
1225 spin_unlock(&fs->lock);
1228 tsk->fs = copy_fs_struct(fs);
1234 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1236 struct files_struct *oldf, *newf;
1240 * A background process may not have any files ...
1242 oldf = current->files;
1246 if (clone_flags & CLONE_FILES) {
1247 atomic_inc(&oldf->count);
1251 newf = dup_fd(oldf, &error);
1261 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1264 struct io_context *ioc = current->io_context;
1265 struct io_context *new_ioc;
1270 * Share io context with parent, if CLONE_IO is set
1272 if (clone_flags & CLONE_IO) {
1274 tsk->io_context = ioc;
1275 } else if (ioprio_valid(ioc->ioprio)) {
1276 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1277 if (unlikely(!new_ioc))
1280 new_ioc->ioprio = ioc->ioprio;
1281 put_io_context(new_ioc);
1287 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1289 struct sighand_struct *sig;
1291 if (clone_flags & CLONE_SIGHAND) {
1292 atomic_inc(¤t->sighand->count);
1295 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1296 rcu_assign_pointer(tsk->sighand, sig);
1300 atomic_set(&sig->count, 1);
1301 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1305 void __cleanup_sighand(struct sighand_struct *sighand)
1307 if (atomic_dec_and_test(&sighand->count)) {
1308 signalfd_cleanup(sighand);
1310 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1311 * without an RCU grace period, see __lock_task_sighand().
1313 kmem_cache_free(sighand_cachep, sighand);
1317 #ifdef CONFIG_POSIX_TIMERS
1319 * Initialize POSIX timer handling for a thread group.
1321 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1323 unsigned long cpu_limit;
1325 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1326 if (cpu_limit != RLIM_INFINITY) {
1327 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1328 sig->cputimer.running = true;
1331 /* The timer lists. */
1332 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1333 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1334 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1337 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1340 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1342 struct signal_struct *sig;
1344 if (clone_flags & CLONE_THREAD)
1347 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1352 sig->nr_threads = 1;
1353 atomic_set(&sig->live, 1);
1354 atomic_set(&sig->sigcnt, 1);
1356 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1357 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1358 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1360 init_waitqueue_head(&sig->wait_chldexit);
1361 sig->curr_target = tsk;
1362 init_sigpending(&sig->shared_pending);
1363 seqlock_init(&sig->stats_lock);
1364 prev_cputime_init(&sig->prev_cputime);
1366 #ifdef CONFIG_POSIX_TIMERS
1367 INIT_LIST_HEAD(&sig->posix_timers);
1368 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1369 sig->real_timer.function = it_real_fn;
1372 task_lock(current->group_leader);
1373 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1374 task_unlock(current->group_leader);
1376 posix_cpu_timers_init_group(sig);
1378 tty_audit_fork(sig);
1379 sched_autogroup_fork(sig);
1381 sig->oom_score_adj = current->signal->oom_score_adj;
1382 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1384 mutex_init(&sig->cred_guard_mutex);
1389 static void copy_seccomp(struct task_struct *p)
1391 #ifdef CONFIG_SECCOMP
1393 * Must be called with sighand->lock held, which is common to
1394 * all threads in the group. Holding cred_guard_mutex is not
1395 * needed because this new task is not yet running and cannot
1398 assert_spin_locked(¤t->sighand->siglock);
1400 /* Ref-count the new filter user, and assign it. */
1401 get_seccomp_filter(current);
1402 p->seccomp = current->seccomp;
1405 * Explicitly enable no_new_privs here in case it got set
1406 * between the task_struct being duplicated and holding the
1407 * sighand lock. The seccomp state and nnp must be in sync.
1409 if (task_no_new_privs(current))
1410 task_set_no_new_privs(p);
1413 * If the parent gained a seccomp mode after copying thread
1414 * flags and between before we held the sighand lock, we have
1415 * to manually enable the seccomp thread flag here.
1417 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1418 set_tsk_thread_flag(p, TIF_SECCOMP);
1422 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1424 current->clear_child_tid = tidptr;
1426 return task_pid_vnr(current);
1429 static void rt_mutex_init_task(struct task_struct *p)
1431 raw_spin_lock_init(&p->pi_lock);
1432 #ifdef CONFIG_RT_MUTEXES
1433 p->pi_waiters = RB_ROOT;
1434 p->pi_waiters_leftmost = NULL;
1435 p->pi_blocked_on = NULL;
1439 #ifdef CONFIG_POSIX_TIMERS
1441 * Initialize POSIX timer handling for a single task.
1443 static void posix_cpu_timers_init(struct task_struct *tsk)
1445 tsk->cputime_expires.prof_exp = 0;
1446 tsk->cputime_expires.virt_exp = 0;
1447 tsk->cputime_expires.sched_exp = 0;
1448 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1449 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1450 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1453 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1457 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1459 task->pids[type].pid = pid;
1463 * This creates a new process as a copy of the old one,
1464 * but does not actually start it yet.
1466 * It copies the registers, and all the appropriate
1467 * parts of the process environment (as per the clone
1468 * flags). The actual kick-off is left to the caller.
1470 static __latent_entropy struct task_struct *copy_process(
1471 unsigned long clone_flags,
1472 unsigned long stack_start,
1473 unsigned long stack_size,
1474 int __user *child_tidptr,
1481 struct task_struct *p;
1483 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1484 return ERR_PTR(-EINVAL);
1486 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1487 return ERR_PTR(-EINVAL);
1490 * Thread groups must share signals as well, and detached threads
1491 * can only be started up within the thread group.
1493 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1494 return ERR_PTR(-EINVAL);
1497 * Shared signal handlers imply shared VM. By way of the above,
1498 * thread groups also imply shared VM. Blocking this case allows
1499 * for various simplifications in other code.
1501 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1502 return ERR_PTR(-EINVAL);
1505 * Siblings of global init remain as zombies on exit since they are
1506 * not reaped by their parent (swapper). To solve this and to avoid
1507 * multi-rooted process trees, prevent global and container-inits
1508 * from creating siblings.
1510 if ((clone_flags & CLONE_PARENT) &&
1511 current->signal->flags & SIGNAL_UNKILLABLE)
1512 return ERR_PTR(-EINVAL);
1515 * If the new process will be in a different pid or user namespace
1516 * do not allow it to share a thread group with the forking task.
1518 if (clone_flags & CLONE_THREAD) {
1519 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1520 (task_active_pid_ns(current) !=
1521 current->nsproxy->pid_ns_for_children))
1522 return ERR_PTR(-EINVAL);
1525 retval = security_task_create(clone_flags);
1530 p = dup_task_struct(current, node);
1534 ftrace_graph_init_task(p);
1536 rt_mutex_init_task(p);
1538 #ifdef CONFIG_PROVE_LOCKING
1539 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1540 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1543 if (atomic_read(&p->real_cred->user->processes) >=
1544 task_rlimit(p, RLIMIT_NPROC)) {
1545 if (p->real_cred->user != INIT_USER &&
1546 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1549 current->flags &= ~PF_NPROC_EXCEEDED;
1551 retval = copy_creds(p, clone_flags);
1556 * If multiple threads are within copy_process(), then this check
1557 * triggers too late. This doesn't hurt, the check is only there
1558 * to stop root fork bombs.
1561 if (nr_threads >= max_threads)
1562 goto bad_fork_cleanup_count;
1564 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1565 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1566 p->flags |= PF_FORKNOEXEC;
1567 INIT_LIST_HEAD(&p->children);
1568 INIT_LIST_HEAD(&p->sibling);
1569 rcu_copy_process(p);
1570 p->vfork_done = NULL;
1571 spin_lock_init(&p->alloc_lock);
1573 init_sigpending(&p->pending);
1575 p->utime = p->stime = p->gtime = 0;
1576 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1577 p->utimescaled = p->stimescaled = 0;
1579 prev_cputime_init(&p->prev_cputime);
1581 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1582 seqcount_init(&p->vtime_seqcount);
1584 p->vtime_snap_whence = VTIME_INACTIVE;
1587 #if defined(SPLIT_RSS_COUNTING)
1588 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1591 p->default_timer_slack_ns = current->timer_slack_ns;
1593 task_io_accounting_init(&p->ioac);
1594 acct_clear_integrals(p);
1596 posix_cpu_timers_init(p);
1598 p->start_time = ktime_get_ns();
1599 p->real_start_time = ktime_get_boot_ns();
1600 p->io_context = NULL;
1601 p->audit_context = NULL;
1604 p->mempolicy = mpol_dup(p->mempolicy);
1605 if (IS_ERR(p->mempolicy)) {
1606 retval = PTR_ERR(p->mempolicy);
1607 p->mempolicy = NULL;
1608 goto bad_fork_cleanup_threadgroup_lock;
1611 #ifdef CONFIG_CPUSETS
1612 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1613 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1614 seqcount_init(&p->mems_allowed_seq);
1616 #ifdef CONFIG_TRACE_IRQFLAGS
1618 p->hardirqs_enabled = 0;
1619 p->hardirq_enable_ip = 0;
1620 p->hardirq_enable_event = 0;
1621 p->hardirq_disable_ip = _THIS_IP_;
1622 p->hardirq_disable_event = 0;
1623 p->softirqs_enabled = 1;
1624 p->softirq_enable_ip = _THIS_IP_;
1625 p->softirq_enable_event = 0;
1626 p->softirq_disable_ip = 0;
1627 p->softirq_disable_event = 0;
1628 p->hardirq_context = 0;
1629 p->softirq_context = 0;
1632 p->pagefault_disabled = 0;
1634 #ifdef CONFIG_LOCKDEP
1635 p->lockdep_depth = 0; /* no locks held yet */
1636 p->curr_chain_key = 0;
1637 p->lockdep_recursion = 0;
1640 #ifdef CONFIG_DEBUG_MUTEXES
1641 p->blocked_on = NULL; /* not blocked yet */
1643 #ifdef CONFIG_BCACHE
1644 p->sequential_io = 0;
1645 p->sequential_io_avg = 0;
1648 /* Perform scheduler related setup. Assign this task to a CPU. */
1649 retval = sched_fork(clone_flags, p);
1651 goto bad_fork_cleanup_policy;
1653 retval = perf_event_init_task(p);
1655 goto bad_fork_cleanup_policy;
1656 retval = audit_alloc(p);
1658 goto bad_fork_cleanup_perf;
1659 /* copy all the process information */
1661 retval = copy_semundo(clone_flags, p);
1663 goto bad_fork_cleanup_audit;
1664 retval = copy_files(clone_flags, p);
1666 goto bad_fork_cleanup_semundo;
1667 retval = copy_fs(clone_flags, p);
1669 goto bad_fork_cleanup_files;
1670 retval = copy_sighand(clone_flags, p);
1672 goto bad_fork_cleanup_fs;
1673 retval = copy_signal(clone_flags, p);
1675 goto bad_fork_cleanup_sighand;
1676 retval = copy_mm(clone_flags, p);
1678 goto bad_fork_cleanup_signal;
1679 retval = copy_namespaces(clone_flags, p);
1681 goto bad_fork_cleanup_mm;
1682 retval = copy_io(clone_flags, p);
1684 goto bad_fork_cleanup_namespaces;
1685 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1687 goto bad_fork_cleanup_io;
1689 if (pid != &init_struct_pid) {
1690 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1692 retval = PTR_ERR(pid);
1693 goto bad_fork_cleanup_thread;
1697 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1699 * Clear TID on mm_release()?
1701 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1706 p->robust_list = NULL;
1707 #ifdef CONFIG_COMPAT
1708 p->compat_robust_list = NULL;
1710 INIT_LIST_HEAD(&p->pi_state_list);
1711 p->pi_state_cache = NULL;
1714 * sigaltstack should be cleared when sharing the same VM
1716 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1720 * Syscall tracing and stepping should be turned off in the
1721 * child regardless of CLONE_PTRACE.
1723 user_disable_single_step(p);
1724 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1725 #ifdef TIF_SYSCALL_EMU
1726 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1728 clear_all_latency_tracing(p);
1730 /* ok, now we should be set up.. */
1731 p->pid = pid_nr(pid);
1732 if (clone_flags & CLONE_THREAD) {
1733 p->exit_signal = -1;
1734 p->group_leader = current->group_leader;
1735 p->tgid = current->tgid;
1737 if (clone_flags & CLONE_PARENT)
1738 p->exit_signal = current->group_leader->exit_signal;
1740 p->exit_signal = (clone_flags & CSIGNAL);
1741 p->group_leader = p;
1746 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1747 p->dirty_paused_when = 0;
1749 p->pdeath_signal = 0;
1750 INIT_LIST_HEAD(&p->thread_group);
1751 p->task_works = NULL;
1753 cgroup_threadgroup_change_begin(current);
1755 * Ensure that the cgroup subsystem policies allow the new process to be
1756 * forked. It should be noted the the new process's css_set can be changed
1757 * between here and cgroup_post_fork() if an organisation operation is in
1760 retval = cgroup_can_fork(p);
1762 goto bad_fork_free_pid;
1765 * Make it visible to the rest of the system, but dont wake it up yet.
1766 * Need tasklist lock for parent etc handling!
1768 write_lock_irq(&tasklist_lock);
1770 /* CLONE_PARENT re-uses the old parent */
1771 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1772 p->real_parent = current->real_parent;
1773 p->parent_exec_id = current->parent_exec_id;
1775 p->real_parent = current;
1776 p->parent_exec_id = current->self_exec_id;
1779 spin_lock(¤t->sighand->siglock);
1782 * Copy seccomp details explicitly here, in case they were changed
1783 * before holding sighand lock.
1788 * Process group and session signals need to be delivered to just the
1789 * parent before the fork or both the parent and the child after the
1790 * fork. Restart if a signal comes in before we add the new process to
1791 * it's process group.
1792 * A fatal signal pending means that current will exit, so the new
1793 * thread can't slip out of an OOM kill (or normal SIGKILL).
1795 recalc_sigpending();
1796 if (signal_pending(current)) {
1797 spin_unlock(¤t->sighand->siglock);
1798 write_unlock_irq(&tasklist_lock);
1799 retval = -ERESTARTNOINTR;
1800 goto bad_fork_cancel_cgroup;
1803 if (likely(p->pid)) {
1804 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1806 init_task_pid(p, PIDTYPE_PID, pid);
1807 if (thread_group_leader(p)) {
1808 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1809 init_task_pid(p, PIDTYPE_SID, task_session(current));
1811 if (is_child_reaper(pid)) {
1812 ns_of_pid(pid)->child_reaper = p;
1813 p->signal->flags |= SIGNAL_UNKILLABLE;
1816 p->signal->leader_pid = pid;
1817 p->signal->tty = tty_kref_get(current->signal->tty);
1819 * Inherit has_child_subreaper flag under the same
1820 * tasklist_lock with adding child to the process tree
1821 * for propagate_has_child_subreaper optimization.
1823 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1824 p->real_parent->signal->is_child_subreaper;
1825 list_add_tail(&p->sibling, &p->real_parent->children);
1826 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1827 attach_pid(p, PIDTYPE_PGID);
1828 attach_pid(p, PIDTYPE_SID);
1829 __this_cpu_inc(process_counts);
1831 current->signal->nr_threads++;
1832 atomic_inc(¤t->signal->live);
1833 atomic_inc(¤t->signal->sigcnt);
1834 list_add_tail_rcu(&p->thread_group,
1835 &p->group_leader->thread_group);
1836 list_add_tail_rcu(&p->thread_node,
1837 &p->signal->thread_head);
1839 attach_pid(p, PIDTYPE_PID);
1844 spin_unlock(¤t->sighand->siglock);
1845 syscall_tracepoint_update(p);
1846 write_unlock_irq(&tasklist_lock);
1848 proc_fork_connector(p);
1849 cgroup_post_fork(p);
1850 cgroup_threadgroup_change_end(current);
1853 trace_task_newtask(p, clone_flags);
1854 uprobe_copy_process(p, clone_flags);
1858 bad_fork_cancel_cgroup:
1859 cgroup_cancel_fork(p);
1861 cgroup_threadgroup_change_end(current);
1862 if (pid != &init_struct_pid)
1864 bad_fork_cleanup_thread:
1866 bad_fork_cleanup_io:
1869 bad_fork_cleanup_namespaces:
1870 exit_task_namespaces(p);
1871 bad_fork_cleanup_mm:
1874 bad_fork_cleanup_signal:
1875 if (!(clone_flags & CLONE_THREAD))
1876 free_signal_struct(p->signal);
1877 bad_fork_cleanup_sighand:
1878 __cleanup_sighand(p->sighand);
1879 bad_fork_cleanup_fs:
1880 exit_fs(p); /* blocking */
1881 bad_fork_cleanup_files:
1882 exit_files(p); /* blocking */
1883 bad_fork_cleanup_semundo:
1885 bad_fork_cleanup_audit:
1887 bad_fork_cleanup_perf:
1888 perf_event_free_task(p);
1889 bad_fork_cleanup_policy:
1891 mpol_put(p->mempolicy);
1892 bad_fork_cleanup_threadgroup_lock:
1894 delayacct_tsk_free(p);
1895 bad_fork_cleanup_count:
1896 atomic_dec(&p->cred->user->processes);
1899 p->state = TASK_DEAD;
1903 return ERR_PTR(retval);
1906 static inline void init_idle_pids(struct pid_link *links)
1910 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1911 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1912 links[type].pid = &init_struct_pid;
1916 struct task_struct *fork_idle(int cpu)
1918 struct task_struct *task;
1919 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1921 if (!IS_ERR(task)) {
1922 init_idle_pids(task->pids);
1923 init_idle(task, cpu);
1930 * Ok, this is the main fork-routine.
1932 * It copies the process, and if successful kick-starts
1933 * it and waits for it to finish using the VM if required.
1935 long _do_fork(unsigned long clone_flags,
1936 unsigned long stack_start,
1937 unsigned long stack_size,
1938 int __user *parent_tidptr,
1939 int __user *child_tidptr,
1942 struct task_struct *p;
1947 * Determine whether and which event to report to ptracer. When
1948 * called from kernel_thread or CLONE_UNTRACED is explicitly
1949 * requested, no event is reported; otherwise, report if the event
1950 * for the type of forking is enabled.
1952 if (!(clone_flags & CLONE_UNTRACED)) {
1953 if (clone_flags & CLONE_VFORK)
1954 trace = PTRACE_EVENT_VFORK;
1955 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1956 trace = PTRACE_EVENT_CLONE;
1958 trace = PTRACE_EVENT_FORK;
1960 if (likely(!ptrace_event_enabled(current, trace)))
1964 p = copy_process(clone_flags, stack_start, stack_size,
1965 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1966 add_latent_entropy();
1968 * Do this prior waking up the new thread - the thread pointer
1969 * might get invalid after that point, if the thread exits quickly.
1972 struct completion vfork;
1975 trace_sched_process_fork(current, p);
1977 pid = get_task_pid(p, PIDTYPE_PID);
1980 if (clone_flags & CLONE_PARENT_SETTID)
1981 put_user(nr, parent_tidptr);
1983 if (clone_flags & CLONE_VFORK) {
1984 p->vfork_done = &vfork;
1985 init_completion(&vfork);
1989 wake_up_new_task(p);
1991 /* forking complete and child started to run, tell ptracer */
1992 if (unlikely(trace))
1993 ptrace_event_pid(trace, pid);
1995 if (clone_flags & CLONE_VFORK) {
1996 if (!wait_for_vfork_done(p, &vfork))
1997 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2007 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2008 /* For compatibility with architectures that call do_fork directly rather than
2009 * using the syscall entry points below. */
2010 long do_fork(unsigned long clone_flags,
2011 unsigned long stack_start,
2012 unsigned long stack_size,
2013 int __user *parent_tidptr,
2014 int __user *child_tidptr)
2016 return _do_fork(clone_flags, stack_start, stack_size,
2017 parent_tidptr, child_tidptr, 0);
2022 * Create a kernel thread.
2024 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2026 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2027 (unsigned long)arg, NULL, NULL, 0);
2030 #ifdef __ARCH_WANT_SYS_FORK
2031 SYSCALL_DEFINE0(fork)
2034 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2036 /* can not support in nommu mode */
2042 #ifdef __ARCH_WANT_SYS_VFORK
2043 SYSCALL_DEFINE0(vfork)
2045 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2050 #ifdef __ARCH_WANT_SYS_CLONE
2051 #ifdef CONFIG_CLONE_BACKWARDS
2052 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2053 int __user *, parent_tidptr,
2055 int __user *, child_tidptr)
2056 #elif defined(CONFIG_CLONE_BACKWARDS2)
2057 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2058 int __user *, parent_tidptr,
2059 int __user *, child_tidptr,
2061 #elif defined(CONFIG_CLONE_BACKWARDS3)
2062 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2064 int __user *, parent_tidptr,
2065 int __user *, child_tidptr,
2068 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2069 int __user *, parent_tidptr,
2070 int __user *, child_tidptr,
2074 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2078 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2080 struct task_struct *leader, *parent, *child;
2083 read_lock(&tasklist_lock);
2084 leader = top = top->group_leader;
2086 for_each_thread(leader, parent) {
2087 list_for_each_entry(child, &parent->children, sibling) {
2088 res = visitor(child, data);
2100 if (leader != top) {
2102 parent = child->real_parent;
2103 leader = parent->group_leader;
2107 read_unlock(&tasklist_lock);
2110 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2111 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2114 static void sighand_ctor(void *data)
2116 struct sighand_struct *sighand = data;
2118 spin_lock_init(&sighand->siglock);
2119 init_waitqueue_head(&sighand->signalfd_wqh);
2122 void __init proc_caches_init(void)
2124 sighand_cachep = kmem_cache_create("sighand_cache",
2125 sizeof(struct sighand_struct), 0,
2126 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2127 SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2128 signal_cachep = kmem_cache_create("signal_cache",
2129 sizeof(struct signal_struct), 0,
2130 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2132 files_cachep = kmem_cache_create("files_cache",
2133 sizeof(struct files_struct), 0,
2134 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2136 fs_cachep = kmem_cache_create("fs_cache",
2137 sizeof(struct fs_struct), 0,
2138 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2141 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2142 * whole struct cpumask for the OFFSTACK case. We could change
2143 * this to *only* allocate as much of it as required by the
2144 * maximum number of CPU's we can ever have. The cpumask_allocation
2145 * is at the end of the structure, exactly for that reason.
2147 mm_cachep = kmem_cache_create("mm_struct",
2148 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2149 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2151 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2153 nsproxy_cache_init();
2157 * Check constraints on flags passed to the unshare system call.
2159 static int check_unshare_flags(unsigned long unshare_flags)
2161 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2162 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2163 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2164 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2167 * Not implemented, but pretend it works if there is nothing
2168 * to unshare. Note that unsharing the address space or the
2169 * signal handlers also need to unshare the signal queues (aka
2172 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2173 if (!thread_group_empty(current))
2176 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2177 if (atomic_read(¤t->sighand->count) > 1)
2180 if (unshare_flags & CLONE_VM) {
2181 if (!current_is_single_threaded())
2189 * Unshare the filesystem structure if it is being shared
2191 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2193 struct fs_struct *fs = current->fs;
2195 if (!(unshare_flags & CLONE_FS) || !fs)
2198 /* don't need lock here; in the worst case we'll do useless copy */
2202 *new_fsp = copy_fs_struct(fs);
2210 * Unshare file descriptor table if it is being shared
2212 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2214 struct files_struct *fd = current->files;
2217 if ((unshare_flags & CLONE_FILES) &&
2218 (fd && atomic_read(&fd->count) > 1)) {
2219 *new_fdp = dup_fd(fd, &error);
2228 * unshare allows a process to 'unshare' part of the process
2229 * context which was originally shared using clone. copy_*
2230 * functions used by do_fork() cannot be used here directly
2231 * because they modify an inactive task_struct that is being
2232 * constructed. Here we are modifying the current, active,
2235 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2237 struct fs_struct *fs, *new_fs = NULL;
2238 struct files_struct *fd, *new_fd = NULL;
2239 struct cred *new_cred = NULL;
2240 struct nsproxy *new_nsproxy = NULL;
2245 * If unsharing a user namespace must also unshare the thread group
2246 * and unshare the filesystem root and working directories.
2248 if (unshare_flags & CLONE_NEWUSER)
2249 unshare_flags |= CLONE_THREAD | CLONE_FS;
2251 * If unsharing vm, must also unshare signal handlers.
2253 if (unshare_flags & CLONE_VM)
2254 unshare_flags |= CLONE_SIGHAND;
2256 * If unsharing a signal handlers, must also unshare the signal queues.
2258 if (unshare_flags & CLONE_SIGHAND)
2259 unshare_flags |= CLONE_THREAD;
2261 * If unsharing namespace, must also unshare filesystem information.
2263 if (unshare_flags & CLONE_NEWNS)
2264 unshare_flags |= CLONE_FS;
2266 err = check_unshare_flags(unshare_flags);
2268 goto bad_unshare_out;
2270 * CLONE_NEWIPC must also detach from the undolist: after switching
2271 * to a new ipc namespace, the semaphore arrays from the old
2272 * namespace are unreachable.
2274 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2276 err = unshare_fs(unshare_flags, &new_fs);
2278 goto bad_unshare_out;
2279 err = unshare_fd(unshare_flags, &new_fd);
2281 goto bad_unshare_cleanup_fs;
2282 err = unshare_userns(unshare_flags, &new_cred);
2284 goto bad_unshare_cleanup_fd;
2285 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2288 goto bad_unshare_cleanup_cred;
2290 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2293 * CLONE_SYSVSEM is equivalent to sys_exit().
2297 if (unshare_flags & CLONE_NEWIPC) {
2298 /* Orphan segments in old ns (see sem above). */
2300 shm_init_task(current);
2304 switch_task_namespaces(current, new_nsproxy);
2310 spin_lock(&fs->lock);
2311 current->fs = new_fs;
2316 spin_unlock(&fs->lock);
2320 fd = current->files;
2321 current->files = new_fd;
2325 task_unlock(current);
2328 /* Install the new user namespace */
2329 commit_creds(new_cred);
2334 bad_unshare_cleanup_cred:
2337 bad_unshare_cleanup_fd:
2339 put_files_struct(new_fd);
2341 bad_unshare_cleanup_fs:
2343 free_fs_struct(new_fs);
2350 * Helper to unshare the files of the current task.
2351 * We don't want to expose copy_files internals to
2352 * the exec layer of the kernel.
2355 int unshare_files(struct files_struct **displaced)
2357 struct task_struct *task = current;
2358 struct files_struct *copy = NULL;
2361 error = unshare_fd(CLONE_FILES, ©);
2362 if (error || !copy) {
2366 *displaced = task->files;
2373 int sysctl_max_threads(struct ctl_table *table, int write,
2374 void __user *buffer, size_t *lenp, loff_t *ppos)
2378 int threads = max_threads;
2379 int min = MIN_THREADS;
2380 int max = MAX_THREADS;
2387 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2391 set_max_threads(threads);