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_canary();
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);
576 #ifdef CONFIG_FAULT_INJECTION
583 free_thread_stack(tsk);
585 free_task_struct(tsk);
590 static __latent_entropy int dup_mmap(struct mm_struct *mm,
591 struct mm_struct *oldmm)
593 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
594 struct rb_node **rb_link, *rb_parent;
596 unsigned long charge;
599 uprobe_start_dup_mmap();
600 if (down_write_killable(&oldmm->mmap_sem)) {
602 goto fail_uprobe_end;
604 flush_cache_dup_mm(oldmm);
605 uprobe_dup_mmap(oldmm, mm);
607 * Not linked in yet - no deadlock potential:
609 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
611 /* No ordering required: file already has been exposed. */
612 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
614 mm->total_vm = oldmm->total_vm;
615 mm->data_vm = oldmm->data_vm;
616 mm->exec_vm = oldmm->exec_vm;
617 mm->stack_vm = oldmm->stack_vm;
619 rb_link = &mm->mm_rb.rb_node;
622 retval = ksm_fork(mm, oldmm);
625 retval = khugepaged_fork(mm, oldmm);
630 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
633 if (mpnt->vm_flags & VM_DONTCOPY) {
634 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
638 if (mpnt->vm_flags & VM_ACCOUNT) {
639 unsigned long len = vma_pages(mpnt);
641 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
645 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
649 INIT_LIST_HEAD(&tmp->anon_vma_chain);
650 retval = vma_dup_policy(mpnt, tmp);
652 goto fail_nomem_policy;
654 retval = dup_userfaultfd(tmp, &uf);
656 goto fail_nomem_anon_vma_fork;
657 if (anon_vma_fork(tmp, mpnt))
658 goto fail_nomem_anon_vma_fork;
659 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
660 tmp->vm_next = tmp->vm_prev = NULL;
663 struct inode *inode = file_inode(file);
664 struct address_space *mapping = file->f_mapping;
667 if (tmp->vm_flags & VM_DENYWRITE)
668 atomic_dec(&inode->i_writecount);
669 i_mmap_lock_write(mapping);
670 if (tmp->vm_flags & VM_SHARED)
671 atomic_inc(&mapping->i_mmap_writable);
672 flush_dcache_mmap_lock(mapping);
673 /* insert tmp into the share list, just after mpnt */
674 vma_interval_tree_insert_after(tmp, mpnt,
676 flush_dcache_mmap_unlock(mapping);
677 i_mmap_unlock_write(mapping);
681 * Clear hugetlb-related page reserves for children. This only
682 * affects MAP_PRIVATE mappings. Faults generated by the child
683 * are not guaranteed to succeed, even if read-only
685 if (is_vm_hugetlb_page(tmp))
686 reset_vma_resv_huge_pages(tmp);
689 * Link in the new vma and copy the page table entries.
692 pprev = &tmp->vm_next;
696 __vma_link_rb(mm, tmp, rb_link, rb_parent);
697 rb_link = &tmp->vm_rb.rb_right;
698 rb_parent = &tmp->vm_rb;
701 retval = copy_page_range(mm, oldmm, mpnt);
703 if (tmp->vm_ops && tmp->vm_ops->open)
704 tmp->vm_ops->open(tmp);
709 /* a new mm has just been created */
710 arch_dup_mmap(oldmm, mm);
713 up_write(&mm->mmap_sem);
715 up_write(&oldmm->mmap_sem);
716 dup_userfaultfd_complete(&uf);
718 uprobe_end_dup_mmap();
720 fail_nomem_anon_vma_fork:
721 mpol_put(vma_policy(tmp));
723 kmem_cache_free(vm_area_cachep, tmp);
726 vm_unacct_memory(charge);
730 static inline int mm_alloc_pgd(struct mm_struct *mm)
732 mm->pgd = pgd_alloc(mm);
733 if (unlikely(!mm->pgd))
738 static inline void mm_free_pgd(struct mm_struct *mm)
740 pgd_free(mm, mm->pgd);
743 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
745 down_write(&oldmm->mmap_sem);
746 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
747 up_write(&oldmm->mmap_sem);
750 #define mm_alloc_pgd(mm) (0)
751 #define mm_free_pgd(mm)
752 #endif /* CONFIG_MMU */
754 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
756 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
757 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
759 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
761 static int __init coredump_filter_setup(char *s)
763 default_dump_filter =
764 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
765 MMF_DUMP_FILTER_MASK;
769 __setup("coredump_filter=", coredump_filter_setup);
771 #include <linux/init_task.h>
773 static void mm_init_aio(struct mm_struct *mm)
776 spin_lock_init(&mm->ioctx_lock);
777 mm->ioctx_table = NULL;
781 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
788 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
789 struct user_namespace *user_ns)
793 mm->vmacache_seqnum = 0;
794 atomic_set(&mm->mm_users, 1);
795 atomic_set(&mm->mm_count, 1);
796 init_rwsem(&mm->mmap_sem);
797 INIT_LIST_HEAD(&mm->mmlist);
798 mm->core_state = NULL;
799 atomic_long_set(&mm->nr_ptes, 0);
804 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
805 spin_lock_init(&mm->page_table_lock);
808 mm_init_owner(mm, p);
809 RCU_INIT_POINTER(mm->exe_file, NULL);
810 mmu_notifier_mm_init(mm);
811 init_tlb_flush_pending(mm);
812 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
813 mm->pmd_huge_pte = NULL;
817 mm->flags = current->mm->flags & MMF_INIT_MASK;
818 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
820 mm->flags = default_dump_filter;
824 if (mm_alloc_pgd(mm))
827 if (init_new_context(p, mm))
830 mm->user_ns = get_user_ns(user_ns);
840 static void check_mm(struct mm_struct *mm)
844 for (i = 0; i < NR_MM_COUNTERS; i++) {
845 long x = atomic_long_read(&mm->rss_stat.count[i]);
848 printk(KERN_ALERT "BUG: Bad rss-counter state "
849 "mm:%p idx:%d val:%ld\n", mm, i, x);
852 if (atomic_long_read(&mm->nr_ptes))
853 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
854 atomic_long_read(&mm->nr_ptes));
856 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
859 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
860 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
865 * Allocate and initialize an mm_struct.
867 struct mm_struct *mm_alloc(void)
869 struct mm_struct *mm;
875 memset(mm, 0, sizeof(*mm));
876 return mm_init(mm, current, current_user_ns());
880 * Called when the last reference to the mm
881 * is dropped: either by a lazy thread or by
882 * mmput. Free the page directory and the mm.
884 void __mmdrop(struct mm_struct *mm)
886 BUG_ON(mm == &init_mm);
889 mmu_notifier_mm_destroy(mm);
891 put_user_ns(mm->user_ns);
894 EXPORT_SYMBOL_GPL(__mmdrop);
896 static inline void __mmput(struct mm_struct *mm)
898 VM_BUG_ON(atomic_read(&mm->mm_users));
900 uprobe_clear_state(mm);
903 khugepaged_exit(mm); /* must run before exit_mmap */
905 mm_put_huge_zero_page(mm);
906 set_mm_exe_file(mm, NULL);
907 if (!list_empty(&mm->mmlist)) {
908 spin_lock(&mmlist_lock);
909 list_del(&mm->mmlist);
910 spin_unlock(&mmlist_lock);
913 module_put(mm->binfmt->module);
914 set_bit(MMF_OOM_SKIP, &mm->flags);
919 * Decrement the use count and release all resources for an mm.
921 void mmput(struct mm_struct *mm)
925 if (atomic_dec_and_test(&mm->mm_users))
928 EXPORT_SYMBOL_GPL(mmput);
931 static void mmput_async_fn(struct work_struct *work)
933 struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
937 void mmput_async(struct mm_struct *mm)
939 if (atomic_dec_and_test(&mm->mm_users)) {
940 INIT_WORK(&mm->async_put_work, mmput_async_fn);
941 schedule_work(&mm->async_put_work);
947 * set_mm_exe_file - change a reference to the mm's executable file
949 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
951 * Main users are mmput() and sys_execve(). Callers prevent concurrent
952 * invocations: in mmput() nobody alive left, in execve task is single
953 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
954 * mm->exe_file, but does so without using set_mm_exe_file() in order
955 * to do avoid the need for any locks.
957 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
959 struct file *old_exe_file;
962 * It is safe to dereference the exe_file without RCU as
963 * this function is only called if nobody else can access
964 * this mm -- see comment above for justification.
966 old_exe_file = rcu_dereference_raw(mm->exe_file);
969 get_file(new_exe_file);
970 rcu_assign_pointer(mm->exe_file, new_exe_file);
976 * get_mm_exe_file - acquire a reference to the mm's executable file
978 * Returns %NULL if mm has no associated executable file.
979 * User must release file via fput().
981 struct file *get_mm_exe_file(struct mm_struct *mm)
983 struct file *exe_file;
986 exe_file = rcu_dereference(mm->exe_file);
987 if (exe_file && !get_file_rcu(exe_file))
992 EXPORT_SYMBOL(get_mm_exe_file);
995 * get_task_exe_file - acquire a reference to the task's executable file
997 * Returns %NULL if task's mm (if any) has no associated executable file or
998 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
999 * User must release file via fput().
1001 struct file *get_task_exe_file(struct task_struct *task)
1003 struct file *exe_file = NULL;
1004 struct mm_struct *mm;
1009 if (!(task->flags & PF_KTHREAD))
1010 exe_file = get_mm_exe_file(mm);
1015 EXPORT_SYMBOL(get_task_exe_file);
1018 * get_task_mm - acquire a reference to the task's mm
1020 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1021 * this kernel workthread has transiently adopted a user mm with use_mm,
1022 * to do its AIO) is not set and if so returns a reference to it, after
1023 * bumping up the use count. User must release the mm via mmput()
1024 * after use. Typically used by /proc and ptrace.
1026 struct mm_struct *get_task_mm(struct task_struct *task)
1028 struct mm_struct *mm;
1033 if (task->flags & PF_KTHREAD)
1041 EXPORT_SYMBOL_GPL(get_task_mm);
1043 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1045 struct mm_struct *mm;
1048 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1050 return ERR_PTR(err);
1052 mm = get_task_mm(task);
1053 if (mm && mm != current->mm &&
1054 !ptrace_may_access(task, mode)) {
1056 mm = ERR_PTR(-EACCES);
1058 mutex_unlock(&task->signal->cred_guard_mutex);
1063 static void complete_vfork_done(struct task_struct *tsk)
1065 struct completion *vfork;
1068 vfork = tsk->vfork_done;
1069 if (likely(vfork)) {
1070 tsk->vfork_done = NULL;
1076 static int wait_for_vfork_done(struct task_struct *child,
1077 struct completion *vfork)
1081 freezer_do_not_count();
1082 killed = wait_for_completion_killable(vfork);
1087 child->vfork_done = NULL;
1091 put_task_struct(child);
1095 /* Please note the differences between mmput and mm_release.
1096 * mmput is called whenever we stop holding onto a mm_struct,
1097 * error success whatever.
1099 * mm_release is called after a mm_struct has been removed
1100 * from the current process.
1102 * This difference is important for error handling, when we
1103 * only half set up a mm_struct for a new process and need to restore
1104 * the old one. Because we mmput the new mm_struct before
1105 * restoring the old one. . .
1106 * Eric Biederman 10 January 1998
1108 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1110 /* Get rid of any futexes when releasing the mm */
1112 if (unlikely(tsk->robust_list)) {
1113 exit_robust_list(tsk);
1114 tsk->robust_list = NULL;
1116 #ifdef CONFIG_COMPAT
1117 if (unlikely(tsk->compat_robust_list)) {
1118 compat_exit_robust_list(tsk);
1119 tsk->compat_robust_list = NULL;
1122 if (unlikely(!list_empty(&tsk->pi_state_list)))
1123 exit_pi_state_list(tsk);
1126 uprobe_free_utask(tsk);
1128 /* Get rid of any cached register state */
1129 deactivate_mm(tsk, mm);
1132 * Signal userspace if we're not exiting with a core dump
1133 * because we want to leave the value intact for debugging
1136 if (tsk->clear_child_tid) {
1137 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1138 atomic_read(&mm->mm_users) > 1) {
1140 * We don't check the error code - if userspace has
1141 * not set up a proper pointer then tough luck.
1143 put_user(0, tsk->clear_child_tid);
1144 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1147 tsk->clear_child_tid = NULL;
1151 * All done, finally we can wake up parent and return this mm to him.
1152 * Also kthread_stop() uses this completion for synchronization.
1154 if (tsk->vfork_done)
1155 complete_vfork_done(tsk);
1159 * Allocate a new mm structure and copy contents from the
1160 * mm structure of the passed in task structure.
1162 static struct mm_struct *dup_mm(struct task_struct *tsk)
1164 struct mm_struct *mm, *oldmm = current->mm;
1171 memcpy(mm, oldmm, sizeof(*mm));
1173 if (!mm_init(mm, tsk, mm->user_ns))
1176 err = dup_mmap(mm, oldmm);
1180 mm->hiwater_rss = get_mm_rss(mm);
1181 mm->hiwater_vm = mm->total_vm;
1183 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1189 /* don't put binfmt in mmput, we haven't got module yet */
1197 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1199 struct mm_struct *mm, *oldmm;
1202 tsk->min_flt = tsk->maj_flt = 0;
1203 tsk->nvcsw = tsk->nivcsw = 0;
1204 #ifdef CONFIG_DETECT_HUNG_TASK
1205 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1209 tsk->active_mm = NULL;
1212 * Are we cloning a kernel thread?
1214 * We need to steal a active VM for that..
1216 oldmm = current->mm;
1220 /* initialize the new vmacache entries */
1221 vmacache_flush(tsk);
1223 if (clone_flags & CLONE_VM) {
1236 tsk->active_mm = mm;
1243 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1245 struct fs_struct *fs = current->fs;
1246 if (clone_flags & CLONE_FS) {
1247 /* tsk->fs is already what we want */
1248 spin_lock(&fs->lock);
1250 spin_unlock(&fs->lock);
1254 spin_unlock(&fs->lock);
1257 tsk->fs = copy_fs_struct(fs);
1263 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1265 struct files_struct *oldf, *newf;
1269 * A background process may not have any files ...
1271 oldf = current->files;
1275 if (clone_flags & CLONE_FILES) {
1276 atomic_inc(&oldf->count);
1280 newf = dup_fd(oldf, &error);
1290 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1293 struct io_context *ioc = current->io_context;
1294 struct io_context *new_ioc;
1299 * Share io context with parent, if CLONE_IO is set
1301 if (clone_flags & CLONE_IO) {
1303 tsk->io_context = ioc;
1304 } else if (ioprio_valid(ioc->ioprio)) {
1305 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1306 if (unlikely(!new_ioc))
1309 new_ioc->ioprio = ioc->ioprio;
1310 put_io_context(new_ioc);
1316 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1318 struct sighand_struct *sig;
1320 if (clone_flags & CLONE_SIGHAND) {
1321 atomic_inc(¤t->sighand->count);
1324 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1325 rcu_assign_pointer(tsk->sighand, sig);
1329 atomic_set(&sig->count, 1);
1330 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1334 void __cleanup_sighand(struct sighand_struct *sighand)
1336 if (atomic_dec_and_test(&sighand->count)) {
1337 signalfd_cleanup(sighand);
1339 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1340 * without an RCU grace period, see __lock_task_sighand().
1342 kmem_cache_free(sighand_cachep, sighand);
1346 #ifdef CONFIG_POSIX_TIMERS
1348 * Initialize POSIX timer handling for a thread group.
1350 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1352 unsigned long cpu_limit;
1354 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1355 if (cpu_limit != RLIM_INFINITY) {
1356 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1357 sig->cputimer.running = true;
1360 /* The timer lists. */
1361 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1362 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1363 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1366 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1369 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1371 struct signal_struct *sig;
1373 if (clone_flags & CLONE_THREAD)
1376 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1381 sig->nr_threads = 1;
1382 atomic_set(&sig->live, 1);
1383 atomic_set(&sig->sigcnt, 1);
1385 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1386 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1387 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1389 init_waitqueue_head(&sig->wait_chldexit);
1390 sig->curr_target = tsk;
1391 init_sigpending(&sig->shared_pending);
1392 seqlock_init(&sig->stats_lock);
1393 prev_cputime_init(&sig->prev_cputime);
1395 #ifdef CONFIG_POSIX_TIMERS
1396 INIT_LIST_HEAD(&sig->posix_timers);
1397 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1398 sig->real_timer.function = it_real_fn;
1401 task_lock(current->group_leader);
1402 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1403 task_unlock(current->group_leader);
1405 posix_cpu_timers_init_group(sig);
1407 tty_audit_fork(sig);
1408 sched_autogroup_fork(sig);
1410 sig->oom_score_adj = current->signal->oom_score_adj;
1411 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1413 mutex_init(&sig->cred_guard_mutex);
1418 static void copy_seccomp(struct task_struct *p)
1420 #ifdef CONFIG_SECCOMP
1422 * Must be called with sighand->lock held, which is common to
1423 * all threads in the group. Holding cred_guard_mutex is not
1424 * needed because this new task is not yet running and cannot
1427 assert_spin_locked(¤t->sighand->siglock);
1429 /* Ref-count the new filter user, and assign it. */
1430 get_seccomp_filter(current);
1431 p->seccomp = current->seccomp;
1434 * Explicitly enable no_new_privs here in case it got set
1435 * between the task_struct being duplicated and holding the
1436 * sighand lock. The seccomp state and nnp must be in sync.
1438 if (task_no_new_privs(current))
1439 task_set_no_new_privs(p);
1442 * If the parent gained a seccomp mode after copying thread
1443 * flags and between before we held the sighand lock, we have
1444 * to manually enable the seccomp thread flag here.
1446 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1447 set_tsk_thread_flag(p, TIF_SECCOMP);
1451 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1453 current->clear_child_tid = tidptr;
1455 return task_pid_vnr(current);
1458 static void rt_mutex_init_task(struct task_struct *p)
1460 raw_spin_lock_init(&p->pi_lock);
1461 #ifdef CONFIG_RT_MUTEXES
1462 p->pi_waiters = RB_ROOT;
1463 p->pi_waiters_leftmost = NULL;
1464 p->pi_top_task = NULL;
1465 p->pi_blocked_on = NULL;
1469 #ifdef CONFIG_POSIX_TIMERS
1471 * Initialize POSIX timer handling for a single task.
1473 static void posix_cpu_timers_init(struct task_struct *tsk)
1475 tsk->cputime_expires.prof_exp = 0;
1476 tsk->cputime_expires.virt_exp = 0;
1477 tsk->cputime_expires.sched_exp = 0;
1478 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1479 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1480 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1483 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1487 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1489 task->pids[type].pid = pid;
1492 static inline void rcu_copy_process(struct task_struct *p)
1494 #ifdef CONFIG_PREEMPT_RCU
1495 p->rcu_read_lock_nesting = 0;
1496 p->rcu_read_unlock_special.s = 0;
1497 p->rcu_blocked_node = NULL;
1498 INIT_LIST_HEAD(&p->rcu_node_entry);
1499 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1500 #ifdef CONFIG_TASKS_RCU
1501 p->rcu_tasks_holdout = false;
1502 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1503 p->rcu_tasks_idle_cpu = -1;
1504 #endif /* #ifdef CONFIG_TASKS_RCU */
1508 * This creates a new process as a copy of the old one,
1509 * but does not actually start it yet.
1511 * It copies the registers, and all the appropriate
1512 * parts of the process environment (as per the clone
1513 * flags). The actual kick-off is left to the caller.
1515 static __latent_entropy struct task_struct *copy_process(
1516 unsigned long clone_flags,
1517 unsigned long stack_start,
1518 unsigned long stack_size,
1519 int __user *child_tidptr,
1526 struct task_struct *p;
1528 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1529 return ERR_PTR(-EINVAL);
1531 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1532 return ERR_PTR(-EINVAL);
1535 * Thread groups must share signals as well, and detached threads
1536 * can only be started up within the thread group.
1538 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1539 return ERR_PTR(-EINVAL);
1542 * Shared signal handlers imply shared VM. By way of the above,
1543 * thread groups also imply shared VM. Blocking this case allows
1544 * for various simplifications in other code.
1546 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1547 return ERR_PTR(-EINVAL);
1550 * Siblings of global init remain as zombies on exit since they are
1551 * not reaped by their parent (swapper). To solve this and to avoid
1552 * multi-rooted process trees, prevent global and container-inits
1553 * from creating siblings.
1555 if ((clone_flags & CLONE_PARENT) &&
1556 current->signal->flags & SIGNAL_UNKILLABLE)
1557 return ERR_PTR(-EINVAL);
1560 * If the new process will be in a different pid or user namespace
1561 * do not allow it to share a thread group with the forking task.
1563 if (clone_flags & CLONE_THREAD) {
1564 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1565 (task_active_pid_ns(current) !=
1566 current->nsproxy->pid_ns_for_children))
1567 return ERR_PTR(-EINVAL);
1570 retval = security_task_create(clone_flags);
1575 p = dup_task_struct(current, node);
1580 * This _must_ happen before we call free_task(), i.e. before we jump
1581 * to any of the bad_fork_* labels. This is to avoid freeing
1582 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1583 * kernel threads (PF_KTHREAD).
1585 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1587 * Clear TID on mm_release()?
1589 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1591 ftrace_graph_init_task(p);
1593 rt_mutex_init_task(p);
1595 #ifdef CONFIG_PROVE_LOCKING
1596 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1597 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1600 if (atomic_read(&p->real_cred->user->processes) >=
1601 task_rlimit(p, RLIMIT_NPROC)) {
1602 if (p->real_cred->user != INIT_USER &&
1603 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1606 current->flags &= ~PF_NPROC_EXCEEDED;
1608 retval = copy_creds(p, clone_flags);
1613 * If multiple threads are within copy_process(), then this check
1614 * triggers too late. This doesn't hurt, the check is only there
1615 * to stop root fork bombs.
1618 if (nr_threads >= max_threads)
1619 goto bad_fork_cleanup_count;
1621 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1622 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1623 p->flags |= PF_FORKNOEXEC;
1624 INIT_LIST_HEAD(&p->children);
1625 INIT_LIST_HEAD(&p->sibling);
1626 rcu_copy_process(p);
1627 p->vfork_done = NULL;
1628 spin_lock_init(&p->alloc_lock);
1630 init_sigpending(&p->pending);
1632 p->utime = p->stime = p->gtime = 0;
1633 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1634 p->utimescaled = p->stimescaled = 0;
1636 prev_cputime_init(&p->prev_cputime);
1638 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1639 seqcount_init(&p->vtime.seqcount);
1640 p->vtime.starttime = 0;
1641 p->vtime.state = VTIME_INACTIVE;
1644 #if defined(SPLIT_RSS_COUNTING)
1645 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1648 p->default_timer_slack_ns = current->timer_slack_ns;
1650 task_io_accounting_init(&p->ioac);
1651 acct_clear_integrals(p);
1653 posix_cpu_timers_init(p);
1655 p->start_time = ktime_get_ns();
1656 p->real_start_time = ktime_get_boot_ns();
1657 p->io_context = NULL;
1658 p->audit_context = NULL;
1661 p->mempolicy = mpol_dup(p->mempolicy);
1662 if (IS_ERR(p->mempolicy)) {
1663 retval = PTR_ERR(p->mempolicy);
1664 p->mempolicy = NULL;
1665 goto bad_fork_cleanup_threadgroup_lock;
1668 #ifdef CONFIG_CPUSETS
1669 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1670 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1671 seqcount_init(&p->mems_allowed_seq);
1673 #ifdef CONFIG_TRACE_IRQFLAGS
1675 p->hardirqs_enabled = 0;
1676 p->hardirq_enable_ip = 0;
1677 p->hardirq_enable_event = 0;
1678 p->hardirq_disable_ip = _THIS_IP_;
1679 p->hardirq_disable_event = 0;
1680 p->softirqs_enabled = 1;
1681 p->softirq_enable_ip = _THIS_IP_;
1682 p->softirq_enable_event = 0;
1683 p->softirq_disable_ip = 0;
1684 p->softirq_disable_event = 0;
1685 p->hardirq_context = 0;
1686 p->softirq_context = 0;
1689 p->pagefault_disabled = 0;
1691 #ifdef CONFIG_LOCKDEP
1692 p->lockdep_depth = 0; /* no locks held yet */
1693 p->curr_chain_key = 0;
1694 p->lockdep_recursion = 0;
1697 #ifdef CONFIG_DEBUG_MUTEXES
1698 p->blocked_on = NULL; /* not blocked yet */
1700 #ifdef CONFIG_BCACHE
1701 p->sequential_io = 0;
1702 p->sequential_io_avg = 0;
1705 /* Perform scheduler related setup. Assign this task to a CPU. */
1706 retval = sched_fork(clone_flags, p);
1708 goto bad_fork_cleanup_policy;
1710 retval = perf_event_init_task(p);
1712 goto bad_fork_cleanup_policy;
1713 retval = audit_alloc(p);
1715 goto bad_fork_cleanup_perf;
1716 /* copy all the process information */
1718 retval = security_task_alloc(p, clone_flags);
1720 goto bad_fork_cleanup_audit;
1721 retval = copy_semundo(clone_flags, p);
1723 goto bad_fork_cleanup_security;
1724 retval = copy_files(clone_flags, p);
1726 goto bad_fork_cleanup_semundo;
1727 retval = copy_fs(clone_flags, p);
1729 goto bad_fork_cleanup_files;
1730 retval = copy_sighand(clone_flags, p);
1732 goto bad_fork_cleanup_fs;
1733 retval = copy_signal(clone_flags, p);
1735 goto bad_fork_cleanup_sighand;
1736 retval = copy_mm(clone_flags, p);
1738 goto bad_fork_cleanup_signal;
1739 retval = copy_namespaces(clone_flags, p);
1741 goto bad_fork_cleanup_mm;
1742 retval = copy_io(clone_flags, p);
1744 goto bad_fork_cleanup_namespaces;
1745 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1747 goto bad_fork_cleanup_io;
1749 if (pid != &init_struct_pid) {
1750 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1752 retval = PTR_ERR(pid);
1753 goto bad_fork_cleanup_thread;
1761 p->robust_list = NULL;
1762 #ifdef CONFIG_COMPAT
1763 p->compat_robust_list = NULL;
1765 INIT_LIST_HEAD(&p->pi_state_list);
1766 p->pi_state_cache = NULL;
1769 * sigaltstack should be cleared when sharing the same VM
1771 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1775 * Syscall tracing and stepping should be turned off in the
1776 * child regardless of CLONE_PTRACE.
1778 user_disable_single_step(p);
1779 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1780 #ifdef TIF_SYSCALL_EMU
1781 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1783 clear_all_latency_tracing(p);
1785 /* ok, now we should be set up.. */
1786 p->pid = pid_nr(pid);
1787 if (clone_flags & CLONE_THREAD) {
1788 p->exit_signal = -1;
1789 p->group_leader = current->group_leader;
1790 p->tgid = current->tgid;
1792 if (clone_flags & CLONE_PARENT)
1793 p->exit_signal = current->group_leader->exit_signal;
1795 p->exit_signal = (clone_flags & CSIGNAL);
1796 p->group_leader = p;
1801 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1802 p->dirty_paused_when = 0;
1804 p->pdeath_signal = 0;
1805 INIT_LIST_HEAD(&p->thread_group);
1806 p->task_works = NULL;
1808 cgroup_threadgroup_change_begin(current);
1810 * Ensure that the cgroup subsystem policies allow the new process to be
1811 * forked. It should be noted the the new process's css_set can be changed
1812 * between here and cgroup_post_fork() if an organisation operation is in
1815 retval = cgroup_can_fork(p);
1817 goto bad_fork_free_pid;
1820 * Make it visible to the rest of the system, but dont wake it up yet.
1821 * Need tasklist lock for parent etc handling!
1823 write_lock_irq(&tasklist_lock);
1825 /* CLONE_PARENT re-uses the old parent */
1826 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1827 p->real_parent = current->real_parent;
1828 p->parent_exec_id = current->parent_exec_id;
1830 p->real_parent = current;
1831 p->parent_exec_id = current->self_exec_id;
1834 klp_copy_process(p);
1836 spin_lock(¤t->sighand->siglock);
1839 * Copy seccomp details explicitly here, in case they were changed
1840 * before holding sighand lock.
1845 * Process group and session signals need to be delivered to just the
1846 * parent before the fork or both the parent and the child after the
1847 * fork. Restart if a signal comes in before we add the new process to
1848 * it's process group.
1849 * A fatal signal pending means that current will exit, so the new
1850 * thread can't slip out of an OOM kill (or normal SIGKILL).
1852 recalc_sigpending();
1853 if (signal_pending(current)) {
1854 retval = -ERESTARTNOINTR;
1855 goto bad_fork_cancel_cgroup;
1857 if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1859 goto bad_fork_cancel_cgroup;
1862 if (likely(p->pid)) {
1863 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1865 init_task_pid(p, PIDTYPE_PID, pid);
1866 if (thread_group_leader(p)) {
1867 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1868 init_task_pid(p, PIDTYPE_SID, task_session(current));
1870 if (is_child_reaper(pid)) {
1871 ns_of_pid(pid)->child_reaper = p;
1872 p->signal->flags |= SIGNAL_UNKILLABLE;
1875 p->signal->leader_pid = pid;
1876 p->signal->tty = tty_kref_get(current->signal->tty);
1878 * Inherit has_child_subreaper flag under the same
1879 * tasklist_lock with adding child to the process tree
1880 * for propagate_has_child_subreaper optimization.
1882 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1883 p->real_parent->signal->is_child_subreaper;
1884 list_add_tail(&p->sibling, &p->real_parent->children);
1885 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1886 attach_pid(p, PIDTYPE_PGID);
1887 attach_pid(p, PIDTYPE_SID);
1888 __this_cpu_inc(process_counts);
1890 current->signal->nr_threads++;
1891 atomic_inc(¤t->signal->live);
1892 atomic_inc(¤t->signal->sigcnt);
1893 list_add_tail_rcu(&p->thread_group,
1894 &p->group_leader->thread_group);
1895 list_add_tail_rcu(&p->thread_node,
1896 &p->signal->thread_head);
1898 attach_pid(p, PIDTYPE_PID);
1903 spin_unlock(¤t->sighand->siglock);
1904 syscall_tracepoint_update(p);
1905 write_unlock_irq(&tasklist_lock);
1907 proc_fork_connector(p);
1908 cgroup_post_fork(p);
1909 cgroup_threadgroup_change_end(current);
1912 trace_task_newtask(p, clone_flags);
1913 uprobe_copy_process(p, clone_flags);
1917 bad_fork_cancel_cgroup:
1918 spin_unlock(¤t->sighand->siglock);
1919 write_unlock_irq(&tasklist_lock);
1920 cgroup_cancel_fork(p);
1922 cgroup_threadgroup_change_end(current);
1923 if (pid != &init_struct_pid)
1925 bad_fork_cleanup_thread:
1927 bad_fork_cleanup_io:
1930 bad_fork_cleanup_namespaces:
1931 exit_task_namespaces(p);
1932 bad_fork_cleanup_mm:
1935 bad_fork_cleanup_signal:
1936 if (!(clone_flags & CLONE_THREAD))
1937 free_signal_struct(p->signal);
1938 bad_fork_cleanup_sighand:
1939 __cleanup_sighand(p->sighand);
1940 bad_fork_cleanup_fs:
1941 exit_fs(p); /* blocking */
1942 bad_fork_cleanup_files:
1943 exit_files(p); /* blocking */
1944 bad_fork_cleanup_semundo:
1946 bad_fork_cleanup_security:
1947 security_task_free(p);
1948 bad_fork_cleanup_audit:
1950 bad_fork_cleanup_perf:
1951 perf_event_free_task(p);
1952 bad_fork_cleanup_policy:
1954 mpol_put(p->mempolicy);
1955 bad_fork_cleanup_threadgroup_lock:
1957 delayacct_tsk_free(p);
1958 bad_fork_cleanup_count:
1959 atomic_dec(&p->cred->user->processes);
1962 p->state = TASK_DEAD;
1966 return ERR_PTR(retval);
1969 static inline void init_idle_pids(struct pid_link *links)
1973 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1974 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1975 links[type].pid = &init_struct_pid;
1979 struct task_struct *fork_idle(int cpu)
1981 struct task_struct *task;
1982 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1984 if (!IS_ERR(task)) {
1985 init_idle_pids(task->pids);
1986 init_idle(task, cpu);
1993 * Ok, this is the main fork-routine.
1995 * It copies the process, and if successful kick-starts
1996 * it and waits for it to finish using the VM if required.
1998 long _do_fork(unsigned long clone_flags,
1999 unsigned long stack_start,
2000 unsigned long stack_size,
2001 int __user *parent_tidptr,
2002 int __user *child_tidptr,
2005 struct task_struct *p;
2010 * Determine whether and which event to report to ptracer. When
2011 * called from kernel_thread or CLONE_UNTRACED is explicitly
2012 * requested, no event is reported; otherwise, report if the event
2013 * for the type of forking is enabled.
2015 if (!(clone_flags & CLONE_UNTRACED)) {
2016 if (clone_flags & CLONE_VFORK)
2017 trace = PTRACE_EVENT_VFORK;
2018 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2019 trace = PTRACE_EVENT_CLONE;
2021 trace = PTRACE_EVENT_FORK;
2023 if (likely(!ptrace_event_enabled(current, trace)))
2027 p = copy_process(clone_flags, stack_start, stack_size,
2028 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2029 add_latent_entropy();
2031 * Do this prior waking up the new thread - the thread pointer
2032 * might get invalid after that point, if the thread exits quickly.
2035 struct completion vfork;
2038 trace_sched_process_fork(current, p);
2040 pid = get_task_pid(p, PIDTYPE_PID);
2043 if (clone_flags & CLONE_PARENT_SETTID)
2044 put_user(nr, parent_tidptr);
2046 if (clone_flags & CLONE_VFORK) {
2047 p->vfork_done = &vfork;
2048 init_completion(&vfork);
2052 wake_up_new_task(p);
2054 /* forking complete and child started to run, tell ptracer */
2055 if (unlikely(trace))
2056 ptrace_event_pid(trace, pid);
2058 if (clone_flags & CLONE_VFORK) {
2059 if (!wait_for_vfork_done(p, &vfork))
2060 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2070 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2071 /* For compatibility with architectures that call do_fork directly rather than
2072 * using the syscall entry points below. */
2073 long do_fork(unsigned long clone_flags,
2074 unsigned long stack_start,
2075 unsigned long stack_size,
2076 int __user *parent_tidptr,
2077 int __user *child_tidptr)
2079 return _do_fork(clone_flags, stack_start, stack_size,
2080 parent_tidptr, child_tidptr, 0);
2085 * Create a kernel thread.
2087 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2089 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2090 (unsigned long)arg, NULL, NULL, 0);
2093 #ifdef __ARCH_WANT_SYS_FORK
2094 SYSCALL_DEFINE0(fork)
2097 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2099 /* can not support in nommu mode */
2105 #ifdef __ARCH_WANT_SYS_VFORK
2106 SYSCALL_DEFINE0(vfork)
2108 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2113 #ifdef __ARCH_WANT_SYS_CLONE
2114 #ifdef CONFIG_CLONE_BACKWARDS
2115 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2116 int __user *, parent_tidptr,
2118 int __user *, child_tidptr)
2119 #elif defined(CONFIG_CLONE_BACKWARDS2)
2120 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2121 int __user *, parent_tidptr,
2122 int __user *, child_tidptr,
2124 #elif defined(CONFIG_CLONE_BACKWARDS3)
2125 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2127 int __user *, parent_tidptr,
2128 int __user *, child_tidptr,
2131 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2132 int __user *, parent_tidptr,
2133 int __user *, child_tidptr,
2137 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2141 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2143 struct task_struct *leader, *parent, *child;
2146 read_lock(&tasklist_lock);
2147 leader = top = top->group_leader;
2149 for_each_thread(leader, parent) {
2150 list_for_each_entry(child, &parent->children, sibling) {
2151 res = visitor(child, data);
2163 if (leader != top) {
2165 parent = child->real_parent;
2166 leader = parent->group_leader;
2170 read_unlock(&tasklist_lock);
2173 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2174 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2177 static void sighand_ctor(void *data)
2179 struct sighand_struct *sighand = data;
2181 spin_lock_init(&sighand->siglock);
2182 init_waitqueue_head(&sighand->signalfd_wqh);
2185 void __init proc_caches_init(void)
2187 sighand_cachep = kmem_cache_create("sighand_cache",
2188 sizeof(struct sighand_struct), 0,
2189 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2190 SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2191 signal_cachep = kmem_cache_create("signal_cache",
2192 sizeof(struct signal_struct), 0,
2193 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2195 files_cachep = kmem_cache_create("files_cache",
2196 sizeof(struct files_struct), 0,
2197 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2199 fs_cachep = kmem_cache_create("fs_cache",
2200 sizeof(struct fs_struct), 0,
2201 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2204 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2205 * whole struct cpumask for the OFFSTACK case. We could change
2206 * this to *only* allocate as much of it as required by the
2207 * maximum number of CPU's we can ever have. The cpumask_allocation
2208 * is at the end of the structure, exactly for that reason.
2210 mm_cachep = kmem_cache_create("mm_struct",
2211 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2212 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2214 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2216 nsproxy_cache_init();
2220 * Check constraints on flags passed to the unshare system call.
2222 static int check_unshare_flags(unsigned long unshare_flags)
2224 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2225 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2226 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2227 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2230 * Not implemented, but pretend it works if there is nothing
2231 * to unshare. Note that unsharing the address space or the
2232 * signal handlers also need to unshare the signal queues (aka
2235 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2236 if (!thread_group_empty(current))
2239 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2240 if (atomic_read(¤t->sighand->count) > 1)
2243 if (unshare_flags & CLONE_VM) {
2244 if (!current_is_single_threaded())
2252 * Unshare the filesystem structure if it is being shared
2254 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2256 struct fs_struct *fs = current->fs;
2258 if (!(unshare_flags & CLONE_FS) || !fs)
2261 /* don't need lock here; in the worst case we'll do useless copy */
2265 *new_fsp = copy_fs_struct(fs);
2273 * Unshare file descriptor table if it is being shared
2275 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2277 struct files_struct *fd = current->files;
2280 if ((unshare_flags & CLONE_FILES) &&
2281 (fd && atomic_read(&fd->count) > 1)) {
2282 *new_fdp = dup_fd(fd, &error);
2291 * unshare allows a process to 'unshare' part of the process
2292 * context which was originally shared using clone. copy_*
2293 * functions used by do_fork() cannot be used here directly
2294 * because they modify an inactive task_struct that is being
2295 * constructed. Here we are modifying the current, active,
2298 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2300 struct fs_struct *fs, *new_fs = NULL;
2301 struct files_struct *fd, *new_fd = NULL;
2302 struct cred *new_cred = NULL;
2303 struct nsproxy *new_nsproxy = NULL;
2308 * If unsharing a user namespace must also unshare the thread group
2309 * and unshare the filesystem root and working directories.
2311 if (unshare_flags & CLONE_NEWUSER)
2312 unshare_flags |= CLONE_THREAD | CLONE_FS;
2314 * If unsharing vm, must also unshare signal handlers.
2316 if (unshare_flags & CLONE_VM)
2317 unshare_flags |= CLONE_SIGHAND;
2319 * If unsharing a signal handlers, must also unshare the signal queues.
2321 if (unshare_flags & CLONE_SIGHAND)
2322 unshare_flags |= CLONE_THREAD;
2324 * If unsharing namespace, must also unshare filesystem information.
2326 if (unshare_flags & CLONE_NEWNS)
2327 unshare_flags |= CLONE_FS;
2329 err = check_unshare_flags(unshare_flags);
2331 goto bad_unshare_out;
2333 * CLONE_NEWIPC must also detach from the undolist: after switching
2334 * to a new ipc namespace, the semaphore arrays from the old
2335 * namespace are unreachable.
2337 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2339 err = unshare_fs(unshare_flags, &new_fs);
2341 goto bad_unshare_out;
2342 err = unshare_fd(unshare_flags, &new_fd);
2344 goto bad_unshare_cleanup_fs;
2345 err = unshare_userns(unshare_flags, &new_cred);
2347 goto bad_unshare_cleanup_fd;
2348 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2351 goto bad_unshare_cleanup_cred;
2353 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2356 * CLONE_SYSVSEM is equivalent to sys_exit().
2360 if (unshare_flags & CLONE_NEWIPC) {
2361 /* Orphan segments in old ns (see sem above). */
2363 shm_init_task(current);
2367 switch_task_namespaces(current, new_nsproxy);
2373 spin_lock(&fs->lock);
2374 current->fs = new_fs;
2379 spin_unlock(&fs->lock);
2383 fd = current->files;
2384 current->files = new_fd;
2388 task_unlock(current);
2391 /* Install the new user namespace */
2392 commit_creds(new_cred);
2397 perf_event_namespaces(current);
2399 bad_unshare_cleanup_cred:
2402 bad_unshare_cleanup_fd:
2404 put_files_struct(new_fd);
2406 bad_unshare_cleanup_fs:
2408 free_fs_struct(new_fs);
2415 * Helper to unshare the files of the current task.
2416 * We don't want to expose copy_files internals to
2417 * the exec layer of the kernel.
2420 int unshare_files(struct files_struct **displaced)
2422 struct task_struct *task = current;
2423 struct files_struct *copy = NULL;
2426 error = unshare_fd(CLONE_FILES, ©);
2427 if (error || !copy) {
2431 *displaced = task->files;
2438 int sysctl_max_threads(struct ctl_table *table, int write,
2439 void __user *buffer, size_t *lenp, loff_t *ppos)
2443 int threads = max_threads;
2444 int min = MIN_THREADS;
2445 int max = MAX_THREADS;
2452 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2456 set_max_threads(threads);