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/sched/numa_balancing.h>
20 #include <linux/sched/stat.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/task_stack.h>
23 #include <linux/sched/cputime.h>
24 #include <linux/rtmutex.h>
25 #include <linux/init.h>
26 #include <linux/unistd.h>
27 #include <linux/module.h>
28 #include <linux/vmalloc.h>
29 #include <linux/completion.h>
30 #include <linux/personality.h>
31 #include <linux/mempolicy.h>
32 #include <linux/sem.h>
33 #include <linux/file.h>
34 #include <linux/fdtable.h>
35 #include <linux/iocontext.h>
36 #include <linux/key.h>
37 #include <linux/binfmts.h>
38 #include <linux/mman.h>
39 #include <linux/mmu_notifier.h>
42 #include <linux/vmacache.h>
43 #include <linux/nsproxy.h>
44 #include <linux/capability.h>
45 #include <linux/cpu.h>
46 #include <linux/cgroup.h>
47 #include <linux/security.h>
48 #include <linux/hugetlb.h>
49 #include <linux/seccomp.h>
50 #include <linux/swap.h>
51 #include <linux/syscalls.h>
52 #include <linux/jiffies.h>
53 #include <linux/futex.h>
54 #include <linux/compat.h>
55 #include <linux/kthread.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/rcupdate.h>
58 #include <linux/ptrace.h>
59 #include <linux/mount.h>
60 #include <linux/audit.h>
61 #include <linux/memcontrol.h>
62 #include <linux/ftrace.h>
63 #include <linux/proc_fs.h>
64 #include <linux/profile.h>
65 #include <linux/rmap.h>
66 #include <linux/ksm.h>
67 #include <linux/acct.h>
68 #include <linux/userfaultfd_k.h>
69 #include <linux/tsacct_kern.h>
70 #include <linux/cn_proc.h>
71 #include <linux/freezer.h>
72 #include <linux/delayacct.h>
73 #include <linux/taskstats_kern.h>
74 #include <linux/random.h>
75 #include <linux/tty.h>
76 #include <linux/blkdev.h>
77 #include <linux/fs_struct.h>
78 #include <linux/magic.h>
79 #include <linux/perf_event.h>
80 #include <linux/posix-timers.h>
81 #include <linux/user-return-notifier.h>
82 #include <linux/oom.h>
83 #include <linux/khugepaged.h>
84 #include <linux/signalfd.h>
85 #include <linux/uprobes.h>
86 #include <linux/aio.h>
87 #include <linux/compiler.h>
88 #include <linux/sysctl.h>
89 #include <linux/kcov.h>
90 #include <linux/livepatch.h>
92 #include <asm/pgtable.h>
93 #include <asm/pgalloc.h>
94 #include <linux/uaccess.h>
95 #include <asm/mmu_context.h>
96 #include <asm/cacheflush.h>
97 #include <asm/tlbflush.h>
99 #include <trace/events/sched.h>
101 #define CREATE_TRACE_POINTS
102 #include <trace/events/task.h>
105 * Minimum number of threads to boot the kernel
107 #define MIN_THREADS 20
110 * Maximum number of threads
112 #define MAX_THREADS FUTEX_TID_MASK
115 * Protected counters by write_lock_irq(&tasklist_lock)
117 unsigned long total_forks; /* Handle normal Linux uptimes. */
118 int nr_threads; /* The idle threads do not count.. */
120 int max_threads; /* tunable limit on nr_threads */
122 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
124 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
126 #ifdef CONFIG_PROVE_RCU
127 int lockdep_tasklist_lock_is_held(void)
129 return lockdep_is_held(&tasklist_lock);
131 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
132 #endif /* #ifdef CONFIG_PROVE_RCU */
134 int nr_processes(void)
139 for_each_possible_cpu(cpu)
140 total += per_cpu(process_counts, cpu);
145 void __weak arch_release_task_struct(struct task_struct *tsk)
149 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
150 static struct kmem_cache *task_struct_cachep;
152 static inline struct task_struct *alloc_task_struct_node(int node)
154 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
157 static inline void free_task_struct(struct task_struct *tsk)
159 kmem_cache_free(task_struct_cachep, tsk);
163 void __weak arch_release_thread_stack(unsigned long *stack)
167 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
170 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
171 * kmemcache based allocator.
173 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
175 #ifdef CONFIG_VMAP_STACK
177 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
178 * flush. Try to minimize the number of calls by caching stacks.
180 #define NR_CACHED_STACKS 2
181 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
183 static int free_vm_stack_cache(unsigned int cpu)
185 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
188 for (i = 0; i < NR_CACHED_STACKS; i++) {
189 struct vm_struct *vm_stack = cached_vm_stacks[i];
194 vfree(vm_stack->addr);
195 cached_vm_stacks[i] = NULL;
202 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
204 #ifdef CONFIG_VMAP_STACK
208 for (i = 0; i < NR_CACHED_STACKS; i++) {
211 s = this_cpu_xchg(cached_stacks[i], NULL);
216 tsk->stack_vm_area = s;
220 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
221 VMALLOC_START, VMALLOC_END,
224 0, node, __builtin_return_address(0));
227 * We can't call find_vm_area() in interrupt context, and
228 * free_thread_stack() can be called in interrupt context,
229 * so cache the vm_struct.
232 tsk->stack_vm_area = find_vm_area(stack);
235 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
238 return page ? page_address(page) : NULL;
242 static inline void free_thread_stack(struct task_struct *tsk)
244 #ifdef CONFIG_VMAP_STACK
245 if (task_stack_vm_area(tsk)) {
248 for (i = 0; i < NR_CACHED_STACKS; i++) {
249 if (this_cpu_cmpxchg(cached_stacks[i],
250 NULL, tsk->stack_vm_area) != NULL)
256 vfree_atomic(tsk->stack);
261 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
264 static struct kmem_cache *thread_stack_cache;
266 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
269 return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
272 static void free_thread_stack(struct task_struct *tsk)
274 kmem_cache_free(thread_stack_cache, tsk->stack);
277 void thread_stack_cache_init(void)
279 thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
280 THREAD_SIZE, 0, NULL);
281 BUG_ON(thread_stack_cache == NULL);
286 /* SLAB cache for signal_struct structures (tsk->signal) */
287 static struct kmem_cache *signal_cachep;
289 /* SLAB cache for sighand_struct structures (tsk->sighand) */
290 struct kmem_cache *sighand_cachep;
292 /* SLAB cache for files_struct structures (tsk->files) */
293 struct kmem_cache *files_cachep;
295 /* SLAB cache for fs_struct structures (tsk->fs) */
296 struct kmem_cache *fs_cachep;
298 /* SLAB cache for vm_area_struct structures */
299 struct kmem_cache *vm_area_cachep;
301 /* SLAB cache for mm_struct structures (tsk->mm) */
302 static struct kmem_cache *mm_cachep;
304 static void account_kernel_stack(struct task_struct *tsk, int account)
306 void *stack = task_stack_page(tsk);
307 struct vm_struct *vm = task_stack_vm_area(tsk);
309 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
314 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
316 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
317 mod_zone_page_state(page_zone(vm->pages[i]),
319 PAGE_SIZE / 1024 * account);
322 /* All stack pages belong to the same memcg. */
323 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
324 account * (THREAD_SIZE / 1024));
327 * All stack pages are in the same zone and belong to the
330 struct page *first_page = virt_to_page(stack);
332 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
333 THREAD_SIZE / 1024 * account);
335 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
336 account * (THREAD_SIZE / 1024));
340 static void release_task_stack(struct task_struct *tsk)
342 if (WARN_ON(tsk->state != TASK_DEAD))
343 return; /* Better to leak the stack than to free prematurely */
345 account_kernel_stack(tsk, -1);
346 arch_release_thread_stack(tsk->stack);
347 free_thread_stack(tsk);
349 #ifdef CONFIG_VMAP_STACK
350 tsk->stack_vm_area = NULL;
354 #ifdef CONFIG_THREAD_INFO_IN_TASK
355 void put_task_stack(struct task_struct *tsk)
357 if (atomic_dec_and_test(&tsk->stack_refcount))
358 release_task_stack(tsk);
362 void free_task(struct task_struct *tsk)
364 #ifndef CONFIG_THREAD_INFO_IN_TASK
366 * The task is finally done with both the stack and thread_info,
369 release_task_stack(tsk);
372 * If the task had a separate stack allocation, it should be gone
375 WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
377 rt_mutex_debug_task_free(tsk);
378 ftrace_graph_exit_task(tsk);
379 put_seccomp_filter(tsk);
380 arch_release_task_struct(tsk);
381 if (tsk->flags & PF_KTHREAD)
382 free_kthread_struct(tsk);
383 free_task_struct(tsk);
385 EXPORT_SYMBOL(free_task);
387 static inline void free_signal_struct(struct signal_struct *sig)
389 taskstats_tgid_free(sig);
390 sched_autogroup_exit(sig);
392 * __mmdrop is not safe to call from softirq context on x86 due to
393 * pgd_dtor so postpone it to the async context
396 mmdrop_async(sig->oom_mm);
397 kmem_cache_free(signal_cachep, sig);
400 static inline void put_signal_struct(struct signal_struct *sig)
402 if (atomic_dec_and_test(&sig->sigcnt))
403 free_signal_struct(sig);
406 void __put_task_struct(struct task_struct *tsk)
408 WARN_ON(!tsk->exit_state);
409 WARN_ON(atomic_read(&tsk->usage));
410 WARN_ON(tsk == current);
414 security_task_free(tsk);
416 delayacct_tsk_free(tsk);
417 put_signal_struct(tsk->signal);
419 if (!profile_handoff_task(tsk))
422 EXPORT_SYMBOL_GPL(__put_task_struct);
424 void __init __weak arch_task_cache_init(void) { }
429 static void set_max_threads(unsigned int max_threads_suggested)
434 * The number of threads shall be limited such that the thread
435 * structures may only consume a small part of the available memory.
437 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
438 threads = MAX_THREADS;
440 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
441 (u64) THREAD_SIZE * 8UL);
443 if (threads > max_threads_suggested)
444 threads = max_threads_suggested;
446 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
449 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
450 /* Initialized by the architecture: */
451 int arch_task_struct_size __read_mostly;
454 void __init fork_init(void)
457 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
458 #ifndef ARCH_MIN_TASKALIGN
459 #define ARCH_MIN_TASKALIGN 0
461 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
463 /* create a slab on which task_structs can be allocated */
464 task_struct_cachep = kmem_cache_create("task_struct",
465 arch_task_struct_size, align,
466 SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
469 /* do the arch specific task caches init */
470 arch_task_cache_init();
472 set_max_threads(MAX_THREADS);
474 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
475 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
476 init_task.signal->rlim[RLIMIT_SIGPENDING] =
477 init_task.signal->rlim[RLIMIT_NPROC];
479 for (i = 0; i < UCOUNT_COUNTS; i++) {
480 init_user_ns.ucount_max[i] = max_threads/2;
483 #ifdef CONFIG_VMAP_STACK
484 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
485 NULL, free_vm_stack_cache);
489 int __weak arch_dup_task_struct(struct task_struct *dst,
490 struct task_struct *src)
496 void set_task_stack_end_magic(struct task_struct *tsk)
498 unsigned long *stackend;
500 stackend = end_of_stack(tsk);
501 *stackend = STACK_END_MAGIC; /* for overflow detection */
504 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
506 struct task_struct *tsk;
507 unsigned long *stack;
508 struct vm_struct *stack_vm_area;
511 if (node == NUMA_NO_NODE)
512 node = tsk_fork_get_node(orig);
513 tsk = alloc_task_struct_node(node);
517 stack = alloc_thread_stack_node(tsk, node);
521 stack_vm_area = task_stack_vm_area(tsk);
523 err = arch_dup_task_struct(tsk, orig);
526 * arch_dup_task_struct() clobbers the stack-related fields. Make
527 * sure they're properly initialized before using any stack-related
531 #ifdef CONFIG_VMAP_STACK
532 tsk->stack_vm_area = stack_vm_area;
534 #ifdef CONFIG_THREAD_INFO_IN_TASK
535 atomic_set(&tsk->stack_refcount, 1);
541 #ifdef CONFIG_SECCOMP
543 * We must handle setting up seccomp filters once we're under
544 * the sighand lock in case orig has changed between now and
545 * then. Until then, filter must be NULL to avoid messing up
546 * the usage counts on the error path calling free_task.
548 tsk->seccomp.filter = NULL;
551 setup_thread_stack(tsk, orig);
552 clear_user_return_notifier(tsk);
553 clear_tsk_need_resched(tsk);
554 set_task_stack_end_magic(tsk);
556 #ifdef CONFIG_CC_STACKPROTECTOR
557 tsk->stack_canary = get_random_long();
561 * One for us, one for whoever does the "release_task()" (usually
564 atomic_set(&tsk->usage, 2);
565 #ifdef CONFIG_BLK_DEV_IO_TRACE
568 tsk->splice_pipe = NULL;
569 tsk->task_frag.page = NULL;
570 tsk->wake_q.next = NULL;
572 account_kernel_stack(tsk, 1);
579 free_thread_stack(tsk);
581 free_task_struct(tsk);
586 static __latent_entropy int dup_mmap(struct mm_struct *mm,
587 struct mm_struct *oldmm)
589 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
590 struct rb_node **rb_link, *rb_parent;
592 unsigned long charge;
595 uprobe_start_dup_mmap();
596 if (down_write_killable(&oldmm->mmap_sem)) {
598 goto fail_uprobe_end;
600 flush_cache_dup_mm(oldmm);
601 uprobe_dup_mmap(oldmm, mm);
603 * Not linked in yet - no deadlock potential:
605 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
607 /* No ordering required: file already has been exposed. */
608 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
610 mm->total_vm = oldmm->total_vm;
611 mm->data_vm = oldmm->data_vm;
612 mm->exec_vm = oldmm->exec_vm;
613 mm->stack_vm = oldmm->stack_vm;
615 rb_link = &mm->mm_rb.rb_node;
618 retval = ksm_fork(mm, oldmm);
621 retval = khugepaged_fork(mm, oldmm);
626 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
629 if (mpnt->vm_flags & VM_DONTCOPY) {
630 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
634 if (mpnt->vm_flags & VM_ACCOUNT) {
635 unsigned long len = vma_pages(mpnt);
637 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
641 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
645 INIT_LIST_HEAD(&tmp->anon_vma_chain);
646 retval = vma_dup_policy(mpnt, tmp);
648 goto fail_nomem_policy;
650 retval = dup_userfaultfd(tmp, &uf);
652 goto fail_nomem_anon_vma_fork;
653 if (anon_vma_fork(tmp, mpnt))
654 goto fail_nomem_anon_vma_fork;
655 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
656 tmp->vm_next = tmp->vm_prev = NULL;
659 struct inode *inode = file_inode(file);
660 struct address_space *mapping = file->f_mapping;
663 if (tmp->vm_flags & VM_DENYWRITE)
664 atomic_dec(&inode->i_writecount);
665 i_mmap_lock_write(mapping);
666 if (tmp->vm_flags & VM_SHARED)
667 atomic_inc(&mapping->i_mmap_writable);
668 flush_dcache_mmap_lock(mapping);
669 /* insert tmp into the share list, just after mpnt */
670 vma_interval_tree_insert_after(tmp, mpnt,
672 flush_dcache_mmap_unlock(mapping);
673 i_mmap_unlock_write(mapping);
677 * Clear hugetlb-related page reserves for children. This only
678 * affects MAP_PRIVATE mappings. Faults generated by the child
679 * are not guaranteed to succeed, even if read-only
681 if (is_vm_hugetlb_page(tmp))
682 reset_vma_resv_huge_pages(tmp);
685 * Link in the new vma and copy the page table entries.
688 pprev = &tmp->vm_next;
692 __vma_link_rb(mm, tmp, rb_link, rb_parent);
693 rb_link = &tmp->vm_rb.rb_right;
694 rb_parent = &tmp->vm_rb;
697 retval = copy_page_range(mm, oldmm, mpnt);
699 if (tmp->vm_ops && tmp->vm_ops->open)
700 tmp->vm_ops->open(tmp);
705 /* a new mm has just been created */
706 arch_dup_mmap(oldmm, mm);
709 up_write(&mm->mmap_sem);
711 up_write(&oldmm->mmap_sem);
712 dup_userfaultfd_complete(&uf);
714 uprobe_end_dup_mmap();
716 fail_nomem_anon_vma_fork:
717 mpol_put(vma_policy(tmp));
719 kmem_cache_free(vm_area_cachep, tmp);
722 vm_unacct_memory(charge);
726 static inline int mm_alloc_pgd(struct mm_struct *mm)
728 mm->pgd = pgd_alloc(mm);
729 if (unlikely(!mm->pgd))
734 static inline void mm_free_pgd(struct mm_struct *mm)
736 pgd_free(mm, mm->pgd);
739 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
741 down_write(&oldmm->mmap_sem);
742 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
743 up_write(&oldmm->mmap_sem);
746 #define mm_alloc_pgd(mm) (0)
747 #define mm_free_pgd(mm)
748 #endif /* CONFIG_MMU */
750 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
752 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
753 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
755 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
757 static int __init coredump_filter_setup(char *s)
759 default_dump_filter =
760 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
761 MMF_DUMP_FILTER_MASK;
765 __setup("coredump_filter=", coredump_filter_setup);
767 #include <linux/init_task.h>
769 static void mm_init_aio(struct mm_struct *mm)
772 spin_lock_init(&mm->ioctx_lock);
773 mm->ioctx_table = NULL;
777 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
784 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
785 struct user_namespace *user_ns)
789 mm->vmacache_seqnum = 0;
790 atomic_set(&mm->mm_users, 1);
791 atomic_set(&mm->mm_count, 1);
792 init_rwsem(&mm->mmap_sem);
793 INIT_LIST_HEAD(&mm->mmlist);
794 mm->core_state = NULL;
795 atomic_long_set(&mm->nr_ptes, 0);
800 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
801 spin_lock_init(&mm->page_table_lock);
804 mm_init_owner(mm, p);
805 mmu_notifier_mm_init(mm);
806 clear_tlb_flush_pending(mm);
807 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
808 mm->pmd_huge_pte = NULL;
812 mm->flags = current->mm->flags & MMF_INIT_MASK;
813 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
815 mm->flags = default_dump_filter;
819 if (mm_alloc_pgd(mm))
822 if (init_new_context(p, mm))
825 mm->user_ns = get_user_ns(user_ns);
835 static void check_mm(struct mm_struct *mm)
839 for (i = 0; i < NR_MM_COUNTERS; i++) {
840 long x = atomic_long_read(&mm->rss_stat.count[i]);
843 printk(KERN_ALERT "BUG: Bad rss-counter state "
844 "mm:%p idx:%d val:%ld\n", mm, i, x);
847 if (atomic_long_read(&mm->nr_ptes))
848 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
849 atomic_long_read(&mm->nr_ptes));
851 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
854 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
855 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
860 * Allocate and initialize an mm_struct.
862 struct mm_struct *mm_alloc(void)
864 struct mm_struct *mm;
870 memset(mm, 0, sizeof(*mm));
871 return mm_init(mm, current, current_user_ns());
875 * Called when the last reference to the mm
876 * is dropped: either by a lazy thread or by
877 * mmput. Free the page directory and the mm.
879 void __mmdrop(struct mm_struct *mm)
881 BUG_ON(mm == &init_mm);
884 mmu_notifier_mm_destroy(mm);
886 put_user_ns(mm->user_ns);
889 EXPORT_SYMBOL_GPL(__mmdrop);
891 static inline void __mmput(struct mm_struct *mm)
893 VM_BUG_ON(atomic_read(&mm->mm_users));
895 uprobe_clear_state(mm);
898 khugepaged_exit(mm); /* must run before exit_mmap */
900 mm_put_huge_zero_page(mm);
901 set_mm_exe_file(mm, NULL);
902 if (!list_empty(&mm->mmlist)) {
903 spin_lock(&mmlist_lock);
904 list_del(&mm->mmlist);
905 spin_unlock(&mmlist_lock);
908 module_put(mm->binfmt->module);
909 set_bit(MMF_OOM_SKIP, &mm->flags);
914 * Decrement the use count and release all resources for an mm.
916 void mmput(struct mm_struct *mm)
920 if (atomic_dec_and_test(&mm->mm_users))
923 EXPORT_SYMBOL_GPL(mmput);
926 static void mmput_async_fn(struct work_struct *work)
928 struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
932 void mmput_async(struct mm_struct *mm)
934 if (atomic_dec_and_test(&mm->mm_users)) {
935 INIT_WORK(&mm->async_put_work, mmput_async_fn);
936 schedule_work(&mm->async_put_work);
942 * set_mm_exe_file - change a reference to the mm's executable file
944 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
946 * Main users are mmput() and sys_execve(). Callers prevent concurrent
947 * invocations: in mmput() nobody alive left, in execve task is single
948 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
949 * mm->exe_file, but does so without using set_mm_exe_file() in order
950 * to do avoid the need for any locks.
952 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
954 struct file *old_exe_file;
957 * It is safe to dereference the exe_file without RCU as
958 * this function is only called if nobody else can access
959 * this mm -- see comment above for justification.
961 old_exe_file = rcu_dereference_raw(mm->exe_file);
964 get_file(new_exe_file);
965 rcu_assign_pointer(mm->exe_file, new_exe_file);
971 * get_mm_exe_file - acquire a reference to the mm's executable file
973 * Returns %NULL if mm has no associated executable file.
974 * User must release file via fput().
976 struct file *get_mm_exe_file(struct mm_struct *mm)
978 struct file *exe_file;
981 exe_file = rcu_dereference(mm->exe_file);
982 if (exe_file && !get_file_rcu(exe_file))
987 EXPORT_SYMBOL(get_mm_exe_file);
990 * get_task_exe_file - acquire a reference to the task's executable file
992 * Returns %NULL if task's mm (if any) has no associated executable file or
993 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
994 * User must release file via fput().
996 struct file *get_task_exe_file(struct task_struct *task)
998 struct file *exe_file = NULL;
999 struct mm_struct *mm;
1004 if (!(task->flags & PF_KTHREAD))
1005 exe_file = get_mm_exe_file(mm);
1010 EXPORT_SYMBOL(get_task_exe_file);
1013 * get_task_mm - acquire a reference to the task's mm
1015 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1016 * this kernel workthread has transiently adopted a user mm with use_mm,
1017 * to do its AIO) is not set and if so returns a reference to it, after
1018 * bumping up the use count. User must release the mm via mmput()
1019 * after use. Typically used by /proc and ptrace.
1021 struct mm_struct *get_task_mm(struct task_struct *task)
1023 struct mm_struct *mm;
1028 if (task->flags & PF_KTHREAD)
1036 EXPORT_SYMBOL_GPL(get_task_mm);
1038 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1040 struct mm_struct *mm;
1043 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1045 return ERR_PTR(err);
1047 mm = get_task_mm(task);
1048 if (mm && mm != current->mm &&
1049 !ptrace_may_access(task, mode)) {
1051 mm = ERR_PTR(-EACCES);
1053 mutex_unlock(&task->signal->cred_guard_mutex);
1058 static void complete_vfork_done(struct task_struct *tsk)
1060 struct completion *vfork;
1063 vfork = tsk->vfork_done;
1064 if (likely(vfork)) {
1065 tsk->vfork_done = NULL;
1071 static int wait_for_vfork_done(struct task_struct *child,
1072 struct completion *vfork)
1076 freezer_do_not_count();
1077 killed = wait_for_completion_killable(vfork);
1082 child->vfork_done = NULL;
1086 put_task_struct(child);
1090 /* Please note the differences between mmput and mm_release.
1091 * mmput is called whenever we stop holding onto a mm_struct,
1092 * error success whatever.
1094 * mm_release is called after a mm_struct has been removed
1095 * from the current process.
1097 * This difference is important for error handling, when we
1098 * only half set up a mm_struct for a new process and need to restore
1099 * the old one. Because we mmput the new mm_struct before
1100 * restoring the old one. . .
1101 * Eric Biederman 10 January 1998
1103 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1105 /* Get rid of any futexes when releasing the mm */
1107 if (unlikely(tsk->robust_list)) {
1108 exit_robust_list(tsk);
1109 tsk->robust_list = NULL;
1111 #ifdef CONFIG_COMPAT
1112 if (unlikely(tsk->compat_robust_list)) {
1113 compat_exit_robust_list(tsk);
1114 tsk->compat_robust_list = NULL;
1117 if (unlikely(!list_empty(&tsk->pi_state_list)))
1118 exit_pi_state_list(tsk);
1121 uprobe_free_utask(tsk);
1123 /* Get rid of any cached register state */
1124 deactivate_mm(tsk, mm);
1127 * Signal userspace if we're not exiting with a core dump
1128 * because we want to leave the value intact for debugging
1131 if (tsk->clear_child_tid) {
1132 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1133 atomic_read(&mm->mm_users) > 1) {
1135 * We don't check the error code - if userspace has
1136 * not set up a proper pointer then tough luck.
1138 put_user(0, tsk->clear_child_tid);
1139 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1142 tsk->clear_child_tid = NULL;
1146 * All done, finally we can wake up parent and return this mm to him.
1147 * Also kthread_stop() uses this completion for synchronization.
1149 if (tsk->vfork_done)
1150 complete_vfork_done(tsk);
1154 * Allocate a new mm structure and copy contents from the
1155 * mm structure of the passed in task structure.
1157 static struct mm_struct *dup_mm(struct task_struct *tsk)
1159 struct mm_struct *mm, *oldmm = current->mm;
1166 memcpy(mm, oldmm, sizeof(*mm));
1168 if (!mm_init(mm, tsk, mm->user_ns))
1171 err = dup_mmap(mm, oldmm);
1175 mm->hiwater_rss = get_mm_rss(mm);
1176 mm->hiwater_vm = mm->total_vm;
1178 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1184 /* don't put binfmt in mmput, we haven't got module yet */
1192 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1194 struct mm_struct *mm, *oldmm;
1197 tsk->min_flt = tsk->maj_flt = 0;
1198 tsk->nvcsw = tsk->nivcsw = 0;
1199 #ifdef CONFIG_DETECT_HUNG_TASK
1200 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1204 tsk->active_mm = NULL;
1207 * Are we cloning a kernel thread?
1209 * We need to steal a active VM for that..
1211 oldmm = current->mm;
1215 /* initialize the new vmacache entries */
1216 vmacache_flush(tsk);
1218 if (clone_flags & CLONE_VM) {
1231 tsk->active_mm = mm;
1238 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1240 struct fs_struct *fs = current->fs;
1241 if (clone_flags & CLONE_FS) {
1242 /* tsk->fs is already what we want */
1243 spin_lock(&fs->lock);
1245 spin_unlock(&fs->lock);
1249 spin_unlock(&fs->lock);
1252 tsk->fs = copy_fs_struct(fs);
1258 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1260 struct files_struct *oldf, *newf;
1264 * A background process may not have any files ...
1266 oldf = current->files;
1270 if (clone_flags & CLONE_FILES) {
1271 atomic_inc(&oldf->count);
1275 newf = dup_fd(oldf, &error);
1285 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1288 struct io_context *ioc = current->io_context;
1289 struct io_context *new_ioc;
1294 * Share io context with parent, if CLONE_IO is set
1296 if (clone_flags & CLONE_IO) {
1298 tsk->io_context = ioc;
1299 } else if (ioprio_valid(ioc->ioprio)) {
1300 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1301 if (unlikely(!new_ioc))
1304 new_ioc->ioprio = ioc->ioprio;
1305 put_io_context(new_ioc);
1311 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1313 struct sighand_struct *sig;
1315 if (clone_flags & CLONE_SIGHAND) {
1316 atomic_inc(¤t->sighand->count);
1319 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1320 rcu_assign_pointer(tsk->sighand, sig);
1324 atomic_set(&sig->count, 1);
1325 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1329 void __cleanup_sighand(struct sighand_struct *sighand)
1331 if (atomic_dec_and_test(&sighand->count)) {
1332 signalfd_cleanup(sighand);
1334 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1335 * without an RCU grace period, see __lock_task_sighand().
1337 kmem_cache_free(sighand_cachep, sighand);
1341 #ifdef CONFIG_POSIX_TIMERS
1343 * Initialize POSIX timer handling for a thread group.
1345 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1347 unsigned long cpu_limit;
1349 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1350 if (cpu_limit != RLIM_INFINITY) {
1351 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1352 sig->cputimer.running = true;
1355 /* The timer lists. */
1356 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1357 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1358 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1361 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1364 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1366 struct signal_struct *sig;
1368 if (clone_flags & CLONE_THREAD)
1371 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1376 sig->nr_threads = 1;
1377 atomic_set(&sig->live, 1);
1378 atomic_set(&sig->sigcnt, 1);
1380 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1381 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1382 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1384 init_waitqueue_head(&sig->wait_chldexit);
1385 sig->curr_target = tsk;
1386 init_sigpending(&sig->shared_pending);
1387 seqlock_init(&sig->stats_lock);
1388 prev_cputime_init(&sig->prev_cputime);
1390 #ifdef CONFIG_POSIX_TIMERS
1391 INIT_LIST_HEAD(&sig->posix_timers);
1392 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1393 sig->real_timer.function = it_real_fn;
1396 task_lock(current->group_leader);
1397 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1398 task_unlock(current->group_leader);
1400 posix_cpu_timers_init_group(sig);
1402 tty_audit_fork(sig);
1403 sched_autogroup_fork(sig);
1405 sig->oom_score_adj = current->signal->oom_score_adj;
1406 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1408 mutex_init(&sig->cred_guard_mutex);
1413 static void copy_seccomp(struct task_struct *p)
1415 #ifdef CONFIG_SECCOMP
1417 * Must be called with sighand->lock held, which is common to
1418 * all threads in the group. Holding cred_guard_mutex is not
1419 * needed because this new task is not yet running and cannot
1422 assert_spin_locked(¤t->sighand->siglock);
1424 /* Ref-count the new filter user, and assign it. */
1425 get_seccomp_filter(current);
1426 p->seccomp = current->seccomp;
1429 * Explicitly enable no_new_privs here in case it got set
1430 * between the task_struct being duplicated and holding the
1431 * sighand lock. The seccomp state and nnp must be in sync.
1433 if (task_no_new_privs(current))
1434 task_set_no_new_privs(p);
1437 * If the parent gained a seccomp mode after copying thread
1438 * flags and between before we held the sighand lock, we have
1439 * to manually enable the seccomp thread flag here.
1441 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1442 set_tsk_thread_flag(p, TIF_SECCOMP);
1446 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1448 current->clear_child_tid = tidptr;
1450 return task_pid_vnr(current);
1453 static void rt_mutex_init_task(struct task_struct *p)
1455 raw_spin_lock_init(&p->pi_lock);
1456 #ifdef CONFIG_RT_MUTEXES
1457 p->pi_waiters = RB_ROOT;
1458 p->pi_waiters_leftmost = NULL;
1459 p->pi_top_task = NULL;
1460 p->pi_blocked_on = NULL;
1464 #ifdef CONFIG_POSIX_TIMERS
1466 * Initialize POSIX timer handling for a single task.
1468 static void posix_cpu_timers_init(struct task_struct *tsk)
1470 tsk->cputime_expires.prof_exp = 0;
1471 tsk->cputime_expires.virt_exp = 0;
1472 tsk->cputime_expires.sched_exp = 0;
1473 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1474 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1475 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1478 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1482 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1484 task->pids[type].pid = pid;
1487 static inline void rcu_copy_process(struct task_struct *p)
1489 #ifdef CONFIG_PREEMPT_RCU
1490 p->rcu_read_lock_nesting = 0;
1491 p->rcu_read_unlock_special.s = 0;
1492 p->rcu_blocked_node = NULL;
1493 INIT_LIST_HEAD(&p->rcu_node_entry);
1494 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1495 #ifdef CONFIG_TASKS_RCU
1496 p->rcu_tasks_holdout = false;
1497 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1498 p->rcu_tasks_idle_cpu = -1;
1499 #endif /* #ifdef CONFIG_TASKS_RCU */
1503 * This creates a new process as a copy of the old one,
1504 * but does not actually start it yet.
1506 * It copies the registers, and all the appropriate
1507 * parts of the process environment (as per the clone
1508 * flags). The actual kick-off is left to the caller.
1510 static __latent_entropy struct task_struct *copy_process(
1511 unsigned long clone_flags,
1512 unsigned long stack_start,
1513 unsigned long stack_size,
1514 int __user *child_tidptr,
1521 struct task_struct *p;
1523 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1524 return ERR_PTR(-EINVAL);
1526 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1527 return ERR_PTR(-EINVAL);
1530 * Thread groups must share signals as well, and detached threads
1531 * can only be started up within the thread group.
1533 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1534 return ERR_PTR(-EINVAL);
1537 * Shared signal handlers imply shared VM. By way of the above,
1538 * thread groups also imply shared VM. Blocking this case allows
1539 * for various simplifications in other code.
1541 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1542 return ERR_PTR(-EINVAL);
1545 * Siblings of global init remain as zombies on exit since they are
1546 * not reaped by their parent (swapper). To solve this and to avoid
1547 * multi-rooted process trees, prevent global and container-inits
1548 * from creating siblings.
1550 if ((clone_flags & CLONE_PARENT) &&
1551 current->signal->flags & SIGNAL_UNKILLABLE)
1552 return ERR_PTR(-EINVAL);
1555 * If the new process will be in a different pid or user namespace
1556 * do not allow it to share a thread group with the forking task.
1558 if (clone_flags & CLONE_THREAD) {
1559 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1560 (task_active_pid_ns(current) !=
1561 current->nsproxy->pid_ns_for_children))
1562 return ERR_PTR(-EINVAL);
1565 retval = security_task_create(clone_flags);
1570 p = dup_task_struct(current, node);
1575 * This _must_ happen before we call free_task(), i.e. before we jump
1576 * to any of the bad_fork_* labels. This is to avoid freeing
1577 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1578 * kernel threads (PF_KTHREAD).
1580 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1582 * Clear TID on mm_release()?
1584 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1586 ftrace_graph_init_task(p);
1588 rt_mutex_init_task(p);
1590 #ifdef CONFIG_PROVE_LOCKING
1591 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1592 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1595 if (atomic_read(&p->real_cred->user->processes) >=
1596 task_rlimit(p, RLIMIT_NPROC)) {
1597 if (p->real_cred->user != INIT_USER &&
1598 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1601 current->flags &= ~PF_NPROC_EXCEEDED;
1603 retval = copy_creds(p, clone_flags);
1608 * If multiple threads are within copy_process(), then this check
1609 * triggers too late. This doesn't hurt, the check is only there
1610 * to stop root fork bombs.
1613 if (nr_threads >= max_threads)
1614 goto bad_fork_cleanup_count;
1616 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1617 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1618 p->flags |= PF_FORKNOEXEC;
1619 INIT_LIST_HEAD(&p->children);
1620 INIT_LIST_HEAD(&p->sibling);
1621 rcu_copy_process(p);
1622 p->vfork_done = NULL;
1623 spin_lock_init(&p->alloc_lock);
1625 init_sigpending(&p->pending);
1627 p->utime = p->stime = p->gtime = 0;
1628 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1629 p->utimescaled = p->stimescaled = 0;
1631 prev_cputime_init(&p->prev_cputime);
1633 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1634 seqcount_init(&p->vtime.seqcount);
1635 p->vtime.starttime = 0;
1636 p->vtime.state = VTIME_INACTIVE;
1639 #if defined(SPLIT_RSS_COUNTING)
1640 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1643 p->default_timer_slack_ns = current->timer_slack_ns;
1645 task_io_accounting_init(&p->ioac);
1646 acct_clear_integrals(p);
1648 posix_cpu_timers_init(p);
1650 p->start_time = ktime_get_ns();
1651 p->real_start_time = ktime_get_boot_ns();
1652 p->io_context = NULL;
1653 p->audit_context = NULL;
1656 p->mempolicy = mpol_dup(p->mempolicy);
1657 if (IS_ERR(p->mempolicy)) {
1658 retval = PTR_ERR(p->mempolicy);
1659 p->mempolicy = NULL;
1660 goto bad_fork_cleanup_threadgroup_lock;
1663 #ifdef CONFIG_CPUSETS
1664 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1665 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1666 seqcount_init(&p->mems_allowed_seq);
1668 #ifdef CONFIG_TRACE_IRQFLAGS
1670 p->hardirqs_enabled = 0;
1671 p->hardirq_enable_ip = 0;
1672 p->hardirq_enable_event = 0;
1673 p->hardirq_disable_ip = _THIS_IP_;
1674 p->hardirq_disable_event = 0;
1675 p->softirqs_enabled = 1;
1676 p->softirq_enable_ip = _THIS_IP_;
1677 p->softirq_enable_event = 0;
1678 p->softirq_disable_ip = 0;
1679 p->softirq_disable_event = 0;
1680 p->hardirq_context = 0;
1681 p->softirq_context = 0;
1684 p->pagefault_disabled = 0;
1686 #ifdef CONFIG_LOCKDEP
1687 p->lockdep_depth = 0; /* no locks held yet */
1688 p->curr_chain_key = 0;
1689 p->lockdep_recursion = 0;
1692 #ifdef CONFIG_DEBUG_MUTEXES
1693 p->blocked_on = NULL; /* not blocked yet */
1695 #ifdef CONFIG_BCACHE
1696 p->sequential_io = 0;
1697 p->sequential_io_avg = 0;
1700 /* Perform scheduler related setup. Assign this task to a CPU. */
1701 retval = sched_fork(clone_flags, p);
1703 goto bad_fork_cleanup_policy;
1705 retval = perf_event_init_task(p);
1707 goto bad_fork_cleanup_policy;
1708 retval = audit_alloc(p);
1710 goto bad_fork_cleanup_perf;
1711 /* copy all the process information */
1713 retval = security_task_alloc(p, clone_flags);
1715 goto bad_fork_cleanup_audit;
1716 retval = copy_semundo(clone_flags, p);
1718 goto bad_fork_cleanup_security;
1719 retval = copy_files(clone_flags, p);
1721 goto bad_fork_cleanup_semundo;
1722 retval = copy_fs(clone_flags, p);
1724 goto bad_fork_cleanup_files;
1725 retval = copy_sighand(clone_flags, p);
1727 goto bad_fork_cleanup_fs;
1728 retval = copy_signal(clone_flags, p);
1730 goto bad_fork_cleanup_sighand;
1731 retval = copy_mm(clone_flags, p);
1733 goto bad_fork_cleanup_signal;
1734 retval = copy_namespaces(clone_flags, p);
1736 goto bad_fork_cleanup_mm;
1737 retval = copy_io(clone_flags, p);
1739 goto bad_fork_cleanup_namespaces;
1740 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1742 goto bad_fork_cleanup_io;
1744 if (pid != &init_struct_pid) {
1745 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1747 retval = PTR_ERR(pid);
1748 goto bad_fork_cleanup_thread;
1756 p->robust_list = NULL;
1757 #ifdef CONFIG_COMPAT
1758 p->compat_robust_list = NULL;
1760 INIT_LIST_HEAD(&p->pi_state_list);
1761 p->pi_state_cache = NULL;
1764 * sigaltstack should be cleared when sharing the same VM
1766 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1770 * Syscall tracing and stepping should be turned off in the
1771 * child regardless of CLONE_PTRACE.
1773 user_disable_single_step(p);
1774 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1775 #ifdef TIF_SYSCALL_EMU
1776 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1778 clear_all_latency_tracing(p);
1780 /* ok, now we should be set up.. */
1781 p->pid = pid_nr(pid);
1782 if (clone_flags & CLONE_THREAD) {
1783 p->exit_signal = -1;
1784 p->group_leader = current->group_leader;
1785 p->tgid = current->tgid;
1787 if (clone_flags & CLONE_PARENT)
1788 p->exit_signal = current->group_leader->exit_signal;
1790 p->exit_signal = (clone_flags & CSIGNAL);
1791 p->group_leader = p;
1796 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1797 p->dirty_paused_when = 0;
1799 p->pdeath_signal = 0;
1800 INIT_LIST_HEAD(&p->thread_group);
1801 p->task_works = NULL;
1803 cgroup_threadgroup_change_begin(current);
1805 * Ensure that the cgroup subsystem policies allow the new process to be
1806 * forked. It should be noted the the new process's css_set can be changed
1807 * between here and cgroup_post_fork() if an organisation operation is in
1810 retval = cgroup_can_fork(p);
1812 goto bad_fork_free_pid;
1815 * Make it visible to the rest of the system, but dont wake it up yet.
1816 * Need tasklist lock for parent etc handling!
1818 write_lock_irq(&tasklist_lock);
1820 /* CLONE_PARENT re-uses the old parent */
1821 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1822 p->real_parent = current->real_parent;
1823 p->parent_exec_id = current->parent_exec_id;
1825 p->real_parent = current;
1826 p->parent_exec_id = current->self_exec_id;
1829 klp_copy_process(p);
1831 spin_lock(¤t->sighand->siglock);
1834 * Copy seccomp details explicitly here, in case they were changed
1835 * before holding sighand lock.
1840 * Process group and session signals need to be delivered to just the
1841 * parent before the fork or both the parent and the child after the
1842 * fork. Restart if a signal comes in before we add the new process to
1843 * it's process group.
1844 * A fatal signal pending means that current will exit, so the new
1845 * thread can't slip out of an OOM kill (or normal SIGKILL).
1847 recalc_sigpending();
1848 if (signal_pending(current)) {
1849 retval = -ERESTARTNOINTR;
1850 goto bad_fork_cancel_cgroup;
1852 if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1854 goto bad_fork_cancel_cgroup;
1857 if (likely(p->pid)) {
1858 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1860 init_task_pid(p, PIDTYPE_PID, pid);
1861 if (thread_group_leader(p)) {
1862 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1863 init_task_pid(p, PIDTYPE_SID, task_session(current));
1865 if (is_child_reaper(pid)) {
1866 ns_of_pid(pid)->child_reaper = p;
1867 p->signal->flags |= SIGNAL_UNKILLABLE;
1870 p->signal->leader_pid = pid;
1871 p->signal->tty = tty_kref_get(current->signal->tty);
1873 * Inherit has_child_subreaper flag under the same
1874 * tasklist_lock with adding child to the process tree
1875 * for propagate_has_child_subreaper optimization.
1877 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1878 p->real_parent->signal->is_child_subreaper;
1879 list_add_tail(&p->sibling, &p->real_parent->children);
1880 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1881 attach_pid(p, PIDTYPE_PGID);
1882 attach_pid(p, PIDTYPE_SID);
1883 __this_cpu_inc(process_counts);
1885 current->signal->nr_threads++;
1886 atomic_inc(¤t->signal->live);
1887 atomic_inc(¤t->signal->sigcnt);
1888 list_add_tail_rcu(&p->thread_group,
1889 &p->group_leader->thread_group);
1890 list_add_tail_rcu(&p->thread_node,
1891 &p->signal->thread_head);
1893 attach_pid(p, PIDTYPE_PID);
1898 spin_unlock(¤t->sighand->siglock);
1899 syscall_tracepoint_update(p);
1900 write_unlock_irq(&tasklist_lock);
1902 proc_fork_connector(p);
1903 cgroup_post_fork(p);
1904 cgroup_threadgroup_change_end(current);
1907 trace_task_newtask(p, clone_flags);
1908 uprobe_copy_process(p, clone_flags);
1912 bad_fork_cancel_cgroup:
1913 spin_unlock(¤t->sighand->siglock);
1914 write_unlock_irq(&tasklist_lock);
1915 cgroup_cancel_fork(p);
1917 cgroup_threadgroup_change_end(current);
1918 if (pid != &init_struct_pid)
1920 bad_fork_cleanup_thread:
1922 bad_fork_cleanup_io:
1925 bad_fork_cleanup_namespaces:
1926 exit_task_namespaces(p);
1927 bad_fork_cleanup_mm:
1930 bad_fork_cleanup_signal:
1931 if (!(clone_flags & CLONE_THREAD))
1932 free_signal_struct(p->signal);
1933 bad_fork_cleanup_sighand:
1934 __cleanup_sighand(p->sighand);
1935 bad_fork_cleanup_fs:
1936 exit_fs(p); /* blocking */
1937 bad_fork_cleanup_files:
1938 exit_files(p); /* blocking */
1939 bad_fork_cleanup_semundo:
1941 bad_fork_cleanup_security:
1942 security_task_free(p);
1943 bad_fork_cleanup_audit:
1945 bad_fork_cleanup_perf:
1946 perf_event_free_task(p);
1947 bad_fork_cleanup_policy:
1949 mpol_put(p->mempolicy);
1950 bad_fork_cleanup_threadgroup_lock:
1952 delayacct_tsk_free(p);
1953 bad_fork_cleanup_count:
1954 atomic_dec(&p->cred->user->processes);
1957 p->state = TASK_DEAD;
1961 return ERR_PTR(retval);
1964 static inline void init_idle_pids(struct pid_link *links)
1968 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1969 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1970 links[type].pid = &init_struct_pid;
1974 struct task_struct *fork_idle(int cpu)
1976 struct task_struct *task;
1977 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1979 if (!IS_ERR(task)) {
1980 init_idle_pids(task->pids);
1981 init_idle(task, cpu);
1988 * Ok, this is the main fork-routine.
1990 * It copies the process, and if successful kick-starts
1991 * it and waits for it to finish using the VM if required.
1993 long _do_fork(unsigned long clone_flags,
1994 unsigned long stack_start,
1995 unsigned long stack_size,
1996 int __user *parent_tidptr,
1997 int __user *child_tidptr,
2000 struct task_struct *p;
2005 * Determine whether and which event to report to ptracer. When
2006 * called from kernel_thread or CLONE_UNTRACED is explicitly
2007 * requested, no event is reported; otherwise, report if the event
2008 * for the type of forking is enabled.
2010 if (!(clone_flags & CLONE_UNTRACED)) {
2011 if (clone_flags & CLONE_VFORK)
2012 trace = PTRACE_EVENT_VFORK;
2013 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2014 trace = PTRACE_EVENT_CLONE;
2016 trace = PTRACE_EVENT_FORK;
2018 if (likely(!ptrace_event_enabled(current, trace)))
2022 p = copy_process(clone_flags, stack_start, stack_size,
2023 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2024 add_latent_entropy();
2026 * Do this prior waking up the new thread - the thread pointer
2027 * might get invalid after that point, if the thread exits quickly.
2030 struct completion vfork;
2033 trace_sched_process_fork(current, p);
2035 pid = get_task_pid(p, PIDTYPE_PID);
2038 if (clone_flags & CLONE_PARENT_SETTID)
2039 put_user(nr, parent_tidptr);
2041 if (clone_flags & CLONE_VFORK) {
2042 p->vfork_done = &vfork;
2043 init_completion(&vfork);
2047 wake_up_new_task(p);
2049 /* forking complete and child started to run, tell ptracer */
2050 if (unlikely(trace))
2051 ptrace_event_pid(trace, pid);
2053 if (clone_flags & CLONE_VFORK) {
2054 if (!wait_for_vfork_done(p, &vfork))
2055 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2065 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2066 /* For compatibility with architectures that call do_fork directly rather than
2067 * using the syscall entry points below. */
2068 long do_fork(unsigned long clone_flags,
2069 unsigned long stack_start,
2070 unsigned long stack_size,
2071 int __user *parent_tidptr,
2072 int __user *child_tidptr)
2074 return _do_fork(clone_flags, stack_start, stack_size,
2075 parent_tidptr, child_tidptr, 0);
2080 * Create a kernel thread.
2082 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2084 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2085 (unsigned long)arg, NULL, NULL, 0);
2088 #ifdef __ARCH_WANT_SYS_FORK
2089 SYSCALL_DEFINE0(fork)
2092 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2094 /* can not support in nommu mode */
2100 #ifdef __ARCH_WANT_SYS_VFORK
2101 SYSCALL_DEFINE0(vfork)
2103 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2108 #ifdef __ARCH_WANT_SYS_CLONE
2109 #ifdef CONFIG_CLONE_BACKWARDS
2110 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2111 int __user *, parent_tidptr,
2113 int __user *, child_tidptr)
2114 #elif defined(CONFIG_CLONE_BACKWARDS2)
2115 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2116 int __user *, parent_tidptr,
2117 int __user *, child_tidptr,
2119 #elif defined(CONFIG_CLONE_BACKWARDS3)
2120 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2122 int __user *, parent_tidptr,
2123 int __user *, child_tidptr,
2126 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2127 int __user *, parent_tidptr,
2128 int __user *, child_tidptr,
2132 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2136 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2138 struct task_struct *leader, *parent, *child;
2141 read_lock(&tasklist_lock);
2142 leader = top = top->group_leader;
2144 for_each_thread(leader, parent) {
2145 list_for_each_entry(child, &parent->children, sibling) {
2146 res = visitor(child, data);
2158 if (leader != top) {
2160 parent = child->real_parent;
2161 leader = parent->group_leader;
2165 read_unlock(&tasklist_lock);
2168 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2169 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2172 static void sighand_ctor(void *data)
2174 struct sighand_struct *sighand = data;
2176 spin_lock_init(&sighand->siglock);
2177 init_waitqueue_head(&sighand->signalfd_wqh);
2180 void __init proc_caches_init(void)
2182 sighand_cachep = kmem_cache_create("sighand_cache",
2183 sizeof(struct sighand_struct), 0,
2184 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2185 SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2186 signal_cachep = kmem_cache_create("signal_cache",
2187 sizeof(struct signal_struct), 0,
2188 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2190 files_cachep = kmem_cache_create("files_cache",
2191 sizeof(struct files_struct), 0,
2192 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2194 fs_cachep = kmem_cache_create("fs_cache",
2195 sizeof(struct fs_struct), 0,
2196 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2199 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2200 * whole struct cpumask for the OFFSTACK case. We could change
2201 * this to *only* allocate as much of it as required by the
2202 * maximum number of CPU's we can ever have. The cpumask_allocation
2203 * is at the end of the structure, exactly for that reason.
2205 mm_cachep = kmem_cache_create("mm_struct",
2206 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2207 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2209 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2211 nsproxy_cache_init();
2215 * Check constraints on flags passed to the unshare system call.
2217 static int check_unshare_flags(unsigned long unshare_flags)
2219 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2220 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2221 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2222 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2225 * Not implemented, but pretend it works if there is nothing
2226 * to unshare. Note that unsharing the address space or the
2227 * signal handlers also need to unshare the signal queues (aka
2230 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2231 if (!thread_group_empty(current))
2234 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2235 if (atomic_read(¤t->sighand->count) > 1)
2238 if (unshare_flags & CLONE_VM) {
2239 if (!current_is_single_threaded())
2247 * Unshare the filesystem structure if it is being shared
2249 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2251 struct fs_struct *fs = current->fs;
2253 if (!(unshare_flags & CLONE_FS) || !fs)
2256 /* don't need lock here; in the worst case we'll do useless copy */
2260 *new_fsp = copy_fs_struct(fs);
2268 * Unshare file descriptor table if it is being shared
2270 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2272 struct files_struct *fd = current->files;
2275 if ((unshare_flags & CLONE_FILES) &&
2276 (fd && atomic_read(&fd->count) > 1)) {
2277 *new_fdp = dup_fd(fd, &error);
2286 * unshare allows a process to 'unshare' part of the process
2287 * context which was originally shared using clone. copy_*
2288 * functions used by do_fork() cannot be used here directly
2289 * because they modify an inactive task_struct that is being
2290 * constructed. Here we are modifying the current, active,
2293 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2295 struct fs_struct *fs, *new_fs = NULL;
2296 struct files_struct *fd, *new_fd = NULL;
2297 struct cred *new_cred = NULL;
2298 struct nsproxy *new_nsproxy = NULL;
2303 * If unsharing a user namespace must also unshare the thread group
2304 * and unshare the filesystem root and working directories.
2306 if (unshare_flags & CLONE_NEWUSER)
2307 unshare_flags |= CLONE_THREAD | CLONE_FS;
2309 * If unsharing vm, must also unshare signal handlers.
2311 if (unshare_flags & CLONE_VM)
2312 unshare_flags |= CLONE_SIGHAND;
2314 * If unsharing a signal handlers, must also unshare the signal queues.
2316 if (unshare_flags & CLONE_SIGHAND)
2317 unshare_flags |= CLONE_THREAD;
2319 * If unsharing namespace, must also unshare filesystem information.
2321 if (unshare_flags & CLONE_NEWNS)
2322 unshare_flags |= CLONE_FS;
2324 err = check_unshare_flags(unshare_flags);
2326 goto bad_unshare_out;
2328 * CLONE_NEWIPC must also detach from the undolist: after switching
2329 * to a new ipc namespace, the semaphore arrays from the old
2330 * namespace are unreachable.
2332 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2334 err = unshare_fs(unshare_flags, &new_fs);
2336 goto bad_unshare_out;
2337 err = unshare_fd(unshare_flags, &new_fd);
2339 goto bad_unshare_cleanup_fs;
2340 err = unshare_userns(unshare_flags, &new_cred);
2342 goto bad_unshare_cleanup_fd;
2343 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2346 goto bad_unshare_cleanup_cred;
2348 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2351 * CLONE_SYSVSEM is equivalent to sys_exit().
2355 if (unshare_flags & CLONE_NEWIPC) {
2356 /* Orphan segments in old ns (see sem above). */
2358 shm_init_task(current);
2362 switch_task_namespaces(current, new_nsproxy);
2368 spin_lock(&fs->lock);
2369 current->fs = new_fs;
2374 spin_unlock(&fs->lock);
2378 fd = current->files;
2379 current->files = new_fd;
2383 task_unlock(current);
2386 /* Install the new user namespace */
2387 commit_creds(new_cred);
2392 perf_event_namespaces(current);
2394 bad_unshare_cleanup_cred:
2397 bad_unshare_cleanup_fd:
2399 put_files_struct(new_fd);
2401 bad_unshare_cleanup_fs:
2403 free_fs_struct(new_fs);
2410 * Helper to unshare the files of the current task.
2411 * We don't want to expose copy_files internals to
2412 * the exec layer of the kernel.
2415 int unshare_files(struct files_struct **displaced)
2417 struct task_struct *task = current;
2418 struct files_struct *copy = NULL;
2421 error = unshare_fd(CLONE_FILES, ©);
2422 if (error || !copy) {
2426 *displaced = task->files;
2433 int sysctl_max_threads(struct ctl_table *table, int write,
2434 void __user *buffer, size_t *lenp, loff_t *ppos)
2438 int threads = max_threads;
2439 int min = MIN_THREADS;
2440 int max = MAX_THREADS;
2447 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2451 set_max_threads(threads);