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/init.h>
21 #include <linux/unistd.h>
22 #include <linux/module.h>
23 #include <linux/vmalloc.h>
24 #include <linux/completion.h>
25 #include <linux/personality.h>
26 #include <linux/mempolicy.h>
27 #include <linux/sem.h>
28 #include <linux/file.h>
29 #include <linux/fdtable.h>
30 #include <linux/iocontext.h>
31 #include <linux/key.h>
32 #include <linux/binfmts.h>
33 #include <linux/mman.h>
34 #include <linux/mmu_notifier.h>
37 #include <linux/vmacache.h>
38 #include <linux/nsproxy.h>
39 #include <linux/capability.h>
40 #include <linux/cpu.h>
41 #include <linux/cgroup.h>
42 #include <linux/security.h>
43 #include <linux/hugetlb.h>
44 #include <linux/seccomp.h>
45 #include <linux/swap.h>
46 #include <linux/syscalls.h>
47 #include <linux/jiffies.h>
48 #include <linux/futex.h>
49 #include <linux/compat.h>
50 #include <linux/kthread.h>
51 #include <linux/task_io_accounting_ops.h>
52 #include <linux/rcupdate.h>
53 #include <linux/ptrace.h>
54 #include <linux/mount.h>
55 #include <linux/audit.h>
56 #include <linux/memcontrol.h>
57 #include <linux/ftrace.h>
58 #include <linux/proc_fs.h>
59 #include <linux/profile.h>
60 #include <linux/rmap.h>
61 #include <linux/ksm.h>
62 #include <linux/acct.h>
63 #include <linux/userfaultfd_k.h>
64 #include <linux/tsacct_kern.h>
65 #include <linux/cn_proc.h>
66 #include <linux/freezer.h>
67 #include <linux/delayacct.h>
68 #include <linux/taskstats_kern.h>
69 #include <linux/random.h>
70 #include <linux/tty.h>
71 #include <linux/blkdev.h>
72 #include <linux/fs_struct.h>
73 #include <linux/magic.h>
74 #include <linux/perf_event.h>
75 #include <linux/posix-timers.h>
76 #include <linux/user-return-notifier.h>
77 #include <linux/oom.h>
78 #include <linux/khugepaged.h>
79 #include <linux/signalfd.h>
80 #include <linux/uprobes.h>
81 #include <linux/aio.h>
82 #include <linux/compiler.h>
83 #include <linux/sysctl.h>
84 #include <linux/kcov.h>
86 #include <asm/pgtable.h>
87 #include <asm/pgalloc.h>
88 #include <linux/uaccess.h>
89 #include <asm/mmu_context.h>
90 #include <asm/cacheflush.h>
91 #include <asm/tlbflush.h>
93 #include <trace/events/sched.h>
95 #define CREATE_TRACE_POINTS
96 #include <trace/events/task.h>
99 * Minimum number of threads to boot the kernel
101 #define MIN_THREADS 20
104 * Maximum number of threads
106 #define MAX_THREADS FUTEX_TID_MASK
109 * Protected counters by write_lock_irq(&tasklist_lock)
111 unsigned long total_forks; /* Handle normal Linux uptimes. */
112 int nr_threads; /* The idle threads do not count.. */
114 int max_threads; /* tunable limit on nr_threads */
116 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
118 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
120 #ifdef CONFIG_PROVE_RCU
121 int lockdep_tasklist_lock_is_held(void)
123 return lockdep_is_held(&tasklist_lock);
125 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
126 #endif /* #ifdef CONFIG_PROVE_RCU */
128 int nr_processes(void)
133 for_each_possible_cpu(cpu)
134 total += per_cpu(process_counts, cpu);
139 void __weak arch_release_task_struct(struct task_struct *tsk)
143 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
144 static struct kmem_cache *task_struct_cachep;
146 static inline struct task_struct *alloc_task_struct_node(int node)
148 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
151 static inline void free_task_struct(struct task_struct *tsk)
153 kmem_cache_free(task_struct_cachep, tsk);
157 void __weak arch_release_thread_stack(unsigned long *stack)
161 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
164 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
165 * kmemcache based allocator.
167 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
169 #ifdef CONFIG_VMAP_STACK
171 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
172 * flush. Try to minimize the number of calls by caching stacks.
174 #define NR_CACHED_STACKS 2
175 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
178 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
180 #ifdef CONFIG_VMAP_STACK
185 for (i = 0; i < NR_CACHED_STACKS; i++) {
186 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
190 this_cpu_write(cached_stacks[i], NULL);
192 tsk->stack_vm_area = s;
198 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
199 VMALLOC_START, VMALLOC_END,
200 THREADINFO_GFP | __GFP_HIGHMEM,
202 0, node, __builtin_return_address(0));
205 * We can't call find_vm_area() in interrupt context, and
206 * free_thread_stack() can be called in interrupt context,
207 * so cache the vm_struct.
210 tsk->stack_vm_area = find_vm_area(stack);
213 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
216 return page ? page_address(page) : NULL;
220 static inline void free_thread_stack(struct task_struct *tsk)
222 #ifdef CONFIG_VMAP_STACK
223 if (task_stack_vm_area(tsk)) {
227 local_irq_save(flags);
228 for (i = 0; i < NR_CACHED_STACKS; i++) {
229 if (this_cpu_read(cached_stacks[i]))
232 this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
233 local_irq_restore(flags);
236 local_irq_restore(flags);
238 vfree_atomic(tsk->stack);
243 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
246 static struct kmem_cache *thread_stack_cache;
248 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
251 return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
254 static void free_thread_stack(struct task_struct *tsk)
256 kmem_cache_free(thread_stack_cache, tsk->stack);
259 void thread_stack_cache_init(void)
261 thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
262 THREAD_SIZE, 0, NULL);
263 BUG_ON(thread_stack_cache == NULL);
268 /* SLAB cache for signal_struct structures (tsk->signal) */
269 static struct kmem_cache *signal_cachep;
271 /* SLAB cache for sighand_struct structures (tsk->sighand) */
272 struct kmem_cache *sighand_cachep;
274 /* SLAB cache for files_struct structures (tsk->files) */
275 struct kmem_cache *files_cachep;
277 /* SLAB cache for fs_struct structures (tsk->fs) */
278 struct kmem_cache *fs_cachep;
280 /* SLAB cache for vm_area_struct structures */
281 struct kmem_cache *vm_area_cachep;
283 /* SLAB cache for mm_struct structures (tsk->mm) */
284 static struct kmem_cache *mm_cachep;
286 static void account_kernel_stack(struct task_struct *tsk, int account)
288 void *stack = task_stack_page(tsk);
289 struct vm_struct *vm = task_stack_vm_area(tsk);
291 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
296 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
298 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
299 mod_zone_page_state(page_zone(vm->pages[i]),
301 PAGE_SIZE / 1024 * account);
304 /* All stack pages belong to the same memcg. */
305 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
306 account * (THREAD_SIZE / 1024));
309 * All stack pages are in the same zone and belong to the
312 struct page *first_page = virt_to_page(stack);
314 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
315 THREAD_SIZE / 1024 * account);
317 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
318 account * (THREAD_SIZE / 1024));
322 static void release_task_stack(struct task_struct *tsk)
324 if (WARN_ON(tsk->state != TASK_DEAD))
325 return; /* Better to leak the stack than to free prematurely */
327 account_kernel_stack(tsk, -1);
328 arch_release_thread_stack(tsk->stack);
329 free_thread_stack(tsk);
331 #ifdef CONFIG_VMAP_STACK
332 tsk->stack_vm_area = NULL;
336 #ifdef CONFIG_THREAD_INFO_IN_TASK
337 void put_task_stack(struct task_struct *tsk)
339 if (atomic_dec_and_test(&tsk->stack_refcount))
340 release_task_stack(tsk);
344 void free_task(struct task_struct *tsk)
346 #ifndef CONFIG_THREAD_INFO_IN_TASK
348 * The task is finally done with both the stack and thread_info,
351 release_task_stack(tsk);
354 * If the task had a separate stack allocation, it should be gone
357 WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
359 rt_mutex_debug_task_free(tsk);
360 ftrace_graph_exit_task(tsk);
361 put_seccomp_filter(tsk);
362 arch_release_task_struct(tsk);
363 if (tsk->flags & PF_KTHREAD)
364 free_kthread_struct(tsk);
365 free_task_struct(tsk);
367 EXPORT_SYMBOL(free_task);
369 static inline void free_signal_struct(struct signal_struct *sig)
371 taskstats_tgid_free(sig);
372 sched_autogroup_exit(sig);
374 * __mmdrop is not safe to call from softirq context on x86 due to
375 * pgd_dtor so postpone it to the async context
378 mmdrop_async(sig->oom_mm);
379 kmem_cache_free(signal_cachep, sig);
382 static inline void put_signal_struct(struct signal_struct *sig)
384 if (atomic_dec_and_test(&sig->sigcnt))
385 free_signal_struct(sig);
388 void __put_task_struct(struct task_struct *tsk)
390 WARN_ON(!tsk->exit_state);
391 WARN_ON(atomic_read(&tsk->usage));
392 WARN_ON(tsk == current);
396 security_task_free(tsk);
398 delayacct_tsk_free(tsk);
399 put_signal_struct(tsk->signal);
401 if (!profile_handoff_task(tsk))
404 EXPORT_SYMBOL_GPL(__put_task_struct);
406 void __init __weak arch_task_cache_init(void) { }
411 static void set_max_threads(unsigned int max_threads_suggested)
416 * The number of threads shall be limited such that the thread
417 * structures may only consume a small part of the available memory.
419 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
420 threads = MAX_THREADS;
422 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
423 (u64) THREAD_SIZE * 8UL);
425 if (threads > max_threads_suggested)
426 threads = max_threads_suggested;
428 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
431 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
432 /* Initialized by the architecture: */
433 int arch_task_struct_size __read_mostly;
436 void __init fork_init(void)
439 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
440 #ifndef ARCH_MIN_TASKALIGN
441 #define ARCH_MIN_TASKALIGN 0
443 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
445 /* create a slab on which task_structs can be allocated */
446 task_struct_cachep = kmem_cache_create("task_struct",
447 arch_task_struct_size, align,
448 SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
451 /* do the arch specific task caches init */
452 arch_task_cache_init();
454 set_max_threads(MAX_THREADS);
456 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
457 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
458 init_task.signal->rlim[RLIMIT_SIGPENDING] =
459 init_task.signal->rlim[RLIMIT_NPROC];
461 for (i = 0; i < UCOUNT_COUNTS; i++) {
462 init_user_ns.ucount_max[i] = max_threads/2;
466 int __weak arch_dup_task_struct(struct task_struct *dst,
467 struct task_struct *src)
473 void set_task_stack_end_magic(struct task_struct *tsk)
475 unsigned long *stackend;
477 stackend = end_of_stack(tsk);
478 *stackend = STACK_END_MAGIC; /* for overflow detection */
481 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
483 struct task_struct *tsk;
484 unsigned long *stack;
485 struct vm_struct *stack_vm_area;
488 if (node == NUMA_NO_NODE)
489 node = tsk_fork_get_node(orig);
490 tsk = alloc_task_struct_node(node);
494 stack = alloc_thread_stack_node(tsk, node);
498 stack_vm_area = task_stack_vm_area(tsk);
500 err = arch_dup_task_struct(tsk, orig);
503 * arch_dup_task_struct() clobbers the stack-related fields. Make
504 * sure they're properly initialized before using any stack-related
508 #ifdef CONFIG_VMAP_STACK
509 tsk->stack_vm_area = stack_vm_area;
511 #ifdef CONFIG_THREAD_INFO_IN_TASK
512 atomic_set(&tsk->stack_refcount, 1);
518 #ifdef CONFIG_SECCOMP
520 * We must handle setting up seccomp filters once we're under
521 * the sighand lock in case orig has changed between now and
522 * then. Until then, filter must be NULL to avoid messing up
523 * the usage counts on the error path calling free_task.
525 tsk->seccomp.filter = NULL;
528 setup_thread_stack(tsk, orig);
529 clear_user_return_notifier(tsk);
530 clear_tsk_need_resched(tsk);
531 set_task_stack_end_magic(tsk);
533 #ifdef CONFIG_CC_STACKPROTECTOR
534 tsk->stack_canary = get_random_int();
538 * One for us, one for whoever does the "release_task()" (usually
541 atomic_set(&tsk->usage, 2);
542 #ifdef CONFIG_BLK_DEV_IO_TRACE
545 tsk->splice_pipe = NULL;
546 tsk->task_frag.page = NULL;
547 tsk->wake_q.next = NULL;
549 account_kernel_stack(tsk, 1);
556 free_thread_stack(tsk);
558 free_task_struct(tsk);
563 static __latent_entropy int dup_mmap(struct mm_struct *mm,
564 struct mm_struct *oldmm)
566 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
567 struct rb_node **rb_link, *rb_parent;
569 unsigned long charge;
572 uprobe_start_dup_mmap();
573 if (down_write_killable(&oldmm->mmap_sem)) {
575 goto fail_uprobe_end;
577 flush_cache_dup_mm(oldmm);
578 uprobe_dup_mmap(oldmm, mm);
580 * Not linked in yet - no deadlock potential:
582 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
584 /* No ordering required: file already has been exposed. */
585 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
587 mm->total_vm = oldmm->total_vm;
588 mm->data_vm = oldmm->data_vm;
589 mm->exec_vm = oldmm->exec_vm;
590 mm->stack_vm = oldmm->stack_vm;
592 rb_link = &mm->mm_rb.rb_node;
595 retval = ksm_fork(mm, oldmm);
598 retval = khugepaged_fork(mm, oldmm);
603 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
606 if (mpnt->vm_flags & VM_DONTCOPY) {
607 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
611 if (mpnt->vm_flags & VM_ACCOUNT) {
612 unsigned long len = vma_pages(mpnt);
614 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
618 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
622 INIT_LIST_HEAD(&tmp->anon_vma_chain);
623 retval = vma_dup_policy(mpnt, tmp);
625 goto fail_nomem_policy;
627 retval = dup_userfaultfd(tmp, &uf);
629 goto fail_nomem_anon_vma_fork;
630 if (anon_vma_fork(tmp, mpnt))
631 goto fail_nomem_anon_vma_fork;
632 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
633 tmp->vm_next = tmp->vm_prev = NULL;
636 struct inode *inode = file_inode(file);
637 struct address_space *mapping = file->f_mapping;
640 if (tmp->vm_flags & VM_DENYWRITE)
641 atomic_dec(&inode->i_writecount);
642 i_mmap_lock_write(mapping);
643 if (tmp->vm_flags & VM_SHARED)
644 atomic_inc(&mapping->i_mmap_writable);
645 flush_dcache_mmap_lock(mapping);
646 /* insert tmp into the share list, just after mpnt */
647 vma_interval_tree_insert_after(tmp, mpnt,
649 flush_dcache_mmap_unlock(mapping);
650 i_mmap_unlock_write(mapping);
654 * Clear hugetlb-related page reserves for children. This only
655 * affects MAP_PRIVATE mappings. Faults generated by the child
656 * are not guaranteed to succeed, even if read-only
658 if (is_vm_hugetlb_page(tmp))
659 reset_vma_resv_huge_pages(tmp);
662 * Link in the new vma and copy the page table entries.
665 pprev = &tmp->vm_next;
669 __vma_link_rb(mm, tmp, rb_link, rb_parent);
670 rb_link = &tmp->vm_rb.rb_right;
671 rb_parent = &tmp->vm_rb;
674 retval = copy_page_range(mm, oldmm, mpnt);
676 if (tmp->vm_ops && tmp->vm_ops->open)
677 tmp->vm_ops->open(tmp);
682 /* a new mm has just been created */
683 arch_dup_mmap(oldmm, mm);
686 up_write(&mm->mmap_sem);
688 up_write(&oldmm->mmap_sem);
689 dup_userfaultfd_complete(&uf);
691 uprobe_end_dup_mmap();
693 fail_nomem_anon_vma_fork:
694 mpol_put(vma_policy(tmp));
696 kmem_cache_free(vm_area_cachep, tmp);
699 vm_unacct_memory(charge);
703 static inline int mm_alloc_pgd(struct mm_struct *mm)
705 mm->pgd = pgd_alloc(mm);
706 if (unlikely(!mm->pgd))
711 static inline void mm_free_pgd(struct mm_struct *mm)
713 pgd_free(mm, mm->pgd);
716 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
718 down_write(&oldmm->mmap_sem);
719 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
720 up_write(&oldmm->mmap_sem);
723 #define mm_alloc_pgd(mm) (0)
724 #define mm_free_pgd(mm)
725 #endif /* CONFIG_MMU */
727 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
729 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
730 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
732 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
734 static int __init coredump_filter_setup(char *s)
736 default_dump_filter =
737 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
738 MMF_DUMP_FILTER_MASK;
742 __setup("coredump_filter=", coredump_filter_setup);
744 #include <linux/init_task.h>
746 static void mm_init_aio(struct mm_struct *mm)
749 spin_lock_init(&mm->ioctx_lock);
750 mm->ioctx_table = NULL;
754 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
761 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
762 struct user_namespace *user_ns)
766 mm->vmacache_seqnum = 0;
767 atomic_set(&mm->mm_users, 1);
768 atomic_set(&mm->mm_count, 1);
769 init_rwsem(&mm->mmap_sem);
770 INIT_LIST_HEAD(&mm->mmlist);
771 mm->core_state = NULL;
772 atomic_long_set(&mm->nr_ptes, 0);
777 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
778 spin_lock_init(&mm->page_table_lock);
781 mm_init_owner(mm, p);
782 mmu_notifier_mm_init(mm);
783 clear_tlb_flush_pending(mm);
784 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
785 mm->pmd_huge_pte = NULL;
789 mm->flags = current->mm->flags & MMF_INIT_MASK;
790 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
792 mm->flags = default_dump_filter;
796 if (mm_alloc_pgd(mm))
799 if (init_new_context(p, mm))
802 mm->user_ns = get_user_ns(user_ns);
812 static void check_mm(struct mm_struct *mm)
816 for (i = 0; i < NR_MM_COUNTERS; i++) {
817 long x = atomic_long_read(&mm->rss_stat.count[i]);
820 printk(KERN_ALERT "BUG: Bad rss-counter state "
821 "mm:%p idx:%d val:%ld\n", mm, i, x);
824 if (atomic_long_read(&mm->nr_ptes))
825 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
826 atomic_long_read(&mm->nr_ptes));
828 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
831 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
832 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
837 * Allocate and initialize an mm_struct.
839 struct mm_struct *mm_alloc(void)
841 struct mm_struct *mm;
847 memset(mm, 0, sizeof(*mm));
848 return mm_init(mm, current, current_user_ns());
852 * Called when the last reference to the mm
853 * is dropped: either by a lazy thread or by
854 * mmput. Free the page directory and the mm.
856 void __mmdrop(struct mm_struct *mm)
858 BUG_ON(mm == &init_mm);
861 mmu_notifier_mm_destroy(mm);
863 put_user_ns(mm->user_ns);
866 EXPORT_SYMBOL_GPL(__mmdrop);
868 static inline void __mmput(struct mm_struct *mm)
870 VM_BUG_ON(atomic_read(&mm->mm_users));
872 uprobe_clear_state(mm);
875 khugepaged_exit(mm); /* must run before exit_mmap */
877 mm_put_huge_zero_page(mm);
878 set_mm_exe_file(mm, NULL);
879 if (!list_empty(&mm->mmlist)) {
880 spin_lock(&mmlist_lock);
881 list_del(&mm->mmlist);
882 spin_unlock(&mmlist_lock);
885 module_put(mm->binfmt->module);
886 set_bit(MMF_OOM_SKIP, &mm->flags);
891 * Decrement the use count and release all resources for an mm.
893 void mmput(struct mm_struct *mm)
897 if (atomic_dec_and_test(&mm->mm_users))
900 EXPORT_SYMBOL_GPL(mmput);
903 static void mmput_async_fn(struct work_struct *work)
905 struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
909 void mmput_async(struct mm_struct *mm)
911 if (atomic_dec_and_test(&mm->mm_users)) {
912 INIT_WORK(&mm->async_put_work, mmput_async_fn);
913 schedule_work(&mm->async_put_work);
919 * set_mm_exe_file - change a reference to the mm's executable file
921 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
923 * Main users are mmput() and sys_execve(). Callers prevent concurrent
924 * invocations: in mmput() nobody alive left, in execve task is single
925 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
926 * mm->exe_file, but does so without using set_mm_exe_file() in order
927 * to do avoid the need for any locks.
929 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
931 struct file *old_exe_file;
934 * It is safe to dereference the exe_file without RCU as
935 * this function is only called if nobody else can access
936 * this mm -- see comment above for justification.
938 old_exe_file = rcu_dereference_raw(mm->exe_file);
941 get_file(new_exe_file);
942 rcu_assign_pointer(mm->exe_file, new_exe_file);
948 * get_mm_exe_file - acquire a reference to the mm's executable file
950 * Returns %NULL if mm has no associated executable file.
951 * User must release file via fput().
953 struct file *get_mm_exe_file(struct mm_struct *mm)
955 struct file *exe_file;
958 exe_file = rcu_dereference(mm->exe_file);
959 if (exe_file && !get_file_rcu(exe_file))
964 EXPORT_SYMBOL(get_mm_exe_file);
967 * get_task_exe_file - acquire a reference to the task's executable file
969 * Returns %NULL if task's mm (if any) has no associated executable file or
970 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
971 * User must release file via fput().
973 struct file *get_task_exe_file(struct task_struct *task)
975 struct file *exe_file = NULL;
976 struct mm_struct *mm;
981 if (!(task->flags & PF_KTHREAD))
982 exe_file = get_mm_exe_file(mm);
987 EXPORT_SYMBOL(get_task_exe_file);
990 * get_task_mm - acquire a reference to the task's mm
992 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
993 * this kernel workthread has transiently adopted a user mm with use_mm,
994 * to do its AIO) is not set and if so returns a reference to it, after
995 * bumping up the use count. User must release the mm via mmput()
996 * after use. Typically used by /proc and ptrace.
998 struct mm_struct *get_task_mm(struct task_struct *task)
1000 struct mm_struct *mm;
1005 if (task->flags & PF_KTHREAD)
1013 EXPORT_SYMBOL_GPL(get_task_mm);
1015 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1017 struct mm_struct *mm;
1020 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1022 return ERR_PTR(err);
1024 mm = get_task_mm(task);
1025 if (mm && mm != current->mm &&
1026 !ptrace_may_access(task, mode)) {
1028 mm = ERR_PTR(-EACCES);
1030 mutex_unlock(&task->signal->cred_guard_mutex);
1035 static void complete_vfork_done(struct task_struct *tsk)
1037 struct completion *vfork;
1040 vfork = tsk->vfork_done;
1041 if (likely(vfork)) {
1042 tsk->vfork_done = NULL;
1048 static int wait_for_vfork_done(struct task_struct *child,
1049 struct completion *vfork)
1053 freezer_do_not_count();
1054 killed = wait_for_completion_killable(vfork);
1059 child->vfork_done = NULL;
1063 put_task_struct(child);
1067 /* Please note the differences between mmput and mm_release.
1068 * mmput is called whenever we stop holding onto a mm_struct,
1069 * error success whatever.
1071 * mm_release is called after a mm_struct has been removed
1072 * from the current process.
1074 * This difference is important for error handling, when we
1075 * only half set up a mm_struct for a new process and need to restore
1076 * the old one. Because we mmput the new mm_struct before
1077 * restoring the old one. . .
1078 * Eric Biederman 10 January 1998
1080 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1082 /* Get rid of any futexes when releasing the mm */
1084 if (unlikely(tsk->robust_list)) {
1085 exit_robust_list(tsk);
1086 tsk->robust_list = NULL;
1088 #ifdef CONFIG_COMPAT
1089 if (unlikely(tsk->compat_robust_list)) {
1090 compat_exit_robust_list(tsk);
1091 tsk->compat_robust_list = NULL;
1094 if (unlikely(!list_empty(&tsk->pi_state_list)))
1095 exit_pi_state_list(tsk);
1098 uprobe_free_utask(tsk);
1100 /* Get rid of any cached register state */
1101 deactivate_mm(tsk, mm);
1104 * Signal userspace if we're not exiting with a core dump
1105 * because we want to leave the value intact for debugging
1108 if (tsk->clear_child_tid) {
1109 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1110 atomic_read(&mm->mm_users) > 1) {
1112 * We don't check the error code - if userspace has
1113 * not set up a proper pointer then tough luck.
1115 put_user(0, tsk->clear_child_tid);
1116 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1119 tsk->clear_child_tid = NULL;
1123 * All done, finally we can wake up parent and return this mm to him.
1124 * Also kthread_stop() uses this completion for synchronization.
1126 if (tsk->vfork_done)
1127 complete_vfork_done(tsk);
1131 * Allocate a new mm structure and copy contents from the
1132 * mm structure of the passed in task structure.
1134 static struct mm_struct *dup_mm(struct task_struct *tsk)
1136 struct mm_struct *mm, *oldmm = current->mm;
1143 memcpy(mm, oldmm, sizeof(*mm));
1145 if (!mm_init(mm, tsk, mm->user_ns))
1148 err = dup_mmap(mm, oldmm);
1152 mm->hiwater_rss = get_mm_rss(mm);
1153 mm->hiwater_vm = mm->total_vm;
1155 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1161 /* don't put binfmt in mmput, we haven't got module yet */
1169 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1171 struct mm_struct *mm, *oldmm;
1174 tsk->min_flt = tsk->maj_flt = 0;
1175 tsk->nvcsw = tsk->nivcsw = 0;
1176 #ifdef CONFIG_DETECT_HUNG_TASK
1177 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1181 tsk->active_mm = NULL;
1184 * Are we cloning a kernel thread?
1186 * We need to steal a active VM for that..
1188 oldmm = current->mm;
1192 /* initialize the new vmacache entries */
1193 vmacache_flush(tsk);
1195 if (clone_flags & CLONE_VM) {
1208 tsk->active_mm = mm;
1215 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1217 struct fs_struct *fs = current->fs;
1218 if (clone_flags & CLONE_FS) {
1219 /* tsk->fs is already what we want */
1220 spin_lock(&fs->lock);
1222 spin_unlock(&fs->lock);
1226 spin_unlock(&fs->lock);
1229 tsk->fs = copy_fs_struct(fs);
1235 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1237 struct files_struct *oldf, *newf;
1241 * A background process may not have any files ...
1243 oldf = current->files;
1247 if (clone_flags & CLONE_FILES) {
1248 atomic_inc(&oldf->count);
1252 newf = dup_fd(oldf, &error);
1262 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1265 struct io_context *ioc = current->io_context;
1266 struct io_context *new_ioc;
1271 * Share io context with parent, if CLONE_IO is set
1273 if (clone_flags & CLONE_IO) {
1275 tsk->io_context = ioc;
1276 } else if (ioprio_valid(ioc->ioprio)) {
1277 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1278 if (unlikely(!new_ioc))
1281 new_ioc->ioprio = ioc->ioprio;
1282 put_io_context(new_ioc);
1288 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1290 struct sighand_struct *sig;
1292 if (clone_flags & CLONE_SIGHAND) {
1293 atomic_inc(¤t->sighand->count);
1296 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1297 rcu_assign_pointer(tsk->sighand, sig);
1301 atomic_set(&sig->count, 1);
1302 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1306 void __cleanup_sighand(struct sighand_struct *sighand)
1308 if (atomic_dec_and_test(&sighand->count)) {
1309 signalfd_cleanup(sighand);
1311 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1312 * without an RCU grace period, see __lock_task_sighand().
1314 kmem_cache_free(sighand_cachep, sighand);
1318 #ifdef CONFIG_POSIX_TIMERS
1320 * Initialize POSIX timer handling for a thread group.
1322 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1324 unsigned long cpu_limit;
1326 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1327 if (cpu_limit != RLIM_INFINITY) {
1328 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1329 sig->cputimer.running = true;
1332 /* The timer lists. */
1333 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1334 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1335 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1338 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1341 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1343 struct signal_struct *sig;
1345 if (clone_flags & CLONE_THREAD)
1348 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1353 sig->nr_threads = 1;
1354 atomic_set(&sig->live, 1);
1355 atomic_set(&sig->sigcnt, 1);
1357 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1358 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1359 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1361 init_waitqueue_head(&sig->wait_chldexit);
1362 sig->curr_target = tsk;
1363 init_sigpending(&sig->shared_pending);
1364 seqlock_init(&sig->stats_lock);
1365 prev_cputime_init(&sig->prev_cputime);
1367 #ifdef CONFIG_POSIX_TIMERS
1368 INIT_LIST_HEAD(&sig->posix_timers);
1369 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1370 sig->real_timer.function = it_real_fn;
1373 task_lock(current->group_leader);
1374 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1375 task_unlock(current->group_leader);
1377 posix_cpu_timers_init_group(sig);
1379 tty_audit_fork(sig);
1380 sched_autogroup_fork(sig);
1382 sig->oom_score_adj = current->signal->oom_score_adj;
1383 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1385 mutex_init(&sig->cred_guard_mutex);
1390 static void copy_seccomp(struct task_struct *p)
1392 #ifdef CONFIG_SECCOMP
1394 * Must be called with sighand->lock held, which is common to
1395 * all threads in the group. Holding cred_guard_mutex is not
1396 * needed because this new task is not yet running and cannot
1399 assert_spin_locked(¤t->sighand->siglock);
1401 /* Ref-count the new filter user, and assign it. */
1402 get_seccomp_filter(current);
1403 p->seccomp = current->seccomp;
1406 * Explicitly enable no_new_privs here in case it got set
1407 * between the task_struct being duplicated and holding the
1408 * sighand lock. The seccomp state and nnp must be in sync.
1410 if (task_no_new_privs(current))
1411 task_set_no_new_privs(p);
1414 * If the parent gained a seccomp mode after copying thread
1415 * flags and between before we held the sighand lock, we have
1416 * to manually enable the seccomp thread flag here.
1418 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1419 set_tsk_thread_flag(p, TIF_SECCOMP);
1423 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1425 current->clear_child_tid = tidptr;
1427 return task_pid_vnr(current);
1430 static void rt_mutex_init_task(struct task_struct *p)
1432 raw_spin_lock_init(&p->pi_lock);
1433 #ifdef CONFIG_RT_MUTEXES
1434 p->pi_waiters = RB_ROOT;
1435 p->pi_waiters_leftmost = NULL;
1436 p->pi_blocked_on = NULL;
1440 #ifdef CONFIG_POSIX_TIMERS
1442 * Initialize POSIX timer handling for a single task.
1444 static void posix_cpu_timers_init(struct task_struct *tsk)
1446 tsk->cputime_expires.prof_exp = 0;
1447 tsk->cputime_expires.virt_exp = 0;
1448 tsk->cputime_expires.sched_exp = 0;
1449 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1450 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1451 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1454 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1458 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1460 task->pids[type].pid = pid;
1464 * This creates a new process as a copy of the old one,
1465 * but does not actually start it yet.
1467 * It copies the registers, and all the appropriate
1468 * parts of the process environment (as per the clone
1469 * flags). The actual kick-off is left to the caller.
1471 static __latent_entropy struct task_struct *copy_process(
1472 unsigned long clone_flags,
1473 unsigned long stack_start,
1474 unsigned long stack_size,
1475 int __user *child_tidptr,
1482 struct task_struct *p;
1484 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1485 return ERR_PTR(-EINVAL);
1487 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1488 return ERR_PTR(-EINVAL);
1491 * Thread groups must share signals as well, and detached threads
1492 * can only be started up within the thread group.
1494 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1495 return ERR_PTR(-EINVAL);
1498 * Shared signal handlers imply shared VM. By way of the above,
1499 * thread groups also imply shared VM. Blocking this case allows
1500 * for various simplifications in other code.
1502 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1503 return ERR_PTR(-EINVAL);
1506 * Siblings of global init remain as zombies on exit since they are
1507 * not reaped by their parent (swapper). To solve this and to avoid
1508 * multi-rooted process trees, prevent global and container-inits
1509 * from creating siblings.
1511 if ((clone_flags & CLONE_PARENT) &&
1512 current->signal->flags & SIGNAL_UNKILLABLE)
1513 return ERR_PTR(-EINVAL);
1516 * If the new process will be in a different pid or user namespace
1517 * do not allow it to share a thread group with the forking task.
1519 if (clone_flags & CLONE_THREAD) {
1520 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1521 (task_active_pid_ns(current) !=
1522 current->nsproxy->pid_ns_for_children))
1523 return ERR_PTR(-EINVAL);
1526 retval = security_task_create(clone_flags);
1531 p = dup_task_struct(current, node);
1535 ftrace_graph_init_task(p);
1537 rt_mutex_init_task(p);
1539 #ifdef CONFIG_PROVE_LOCKING
1540 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1541 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1544 if (atomic_read(&p->real_cred->user->processes) >=
1545 task_rlimit(p, RLIMIT_NPROC)) {
1546 if (p->real_cred->user != INIT_USER &&
1547 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1550 current->flags &= ~PF_NPROC_EXCEEDED;
1552 retval = copy_creds(p, clone_flags);
1557 * If multiple threads are within copy_process(), then this check
1558 * triggers too late. This doesn't hurt, the check is only there
1559 * to stop root fork bombs.
1562 if (nr_threads >= max_threads)
1563 goto bad_fork_cleanup_count;
1565 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1566 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1567 p->flags |= PF_FORKNOEXEC;
1568 INIT_LIST_HEAD(&p->children);
1569 INIT_LIST_HEAD(&p->sibling);
1570 rcu_copy_process(p);
1571 p->vfork_done = NULL;
1572 spin_lock_init(&p->alloc_lock);
1574 init_sigpending(&p->pending);
1576 p->utime = p->stime = p->gtime = 0;
1577 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1578 p->utimescaled = p->stimescaled = 0;
1580 prev_cputime_init(&p->prev_cputime);
1582 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1583 seqcount_init(&p->vtime_seqcount);
1585 p->vtime_snap_whence = VTIME_INACTIVE;
1588 #if defined(SPLIT_RSS_COUNTING)
1589 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1592 p->default_timer_slack_ns = current->timer_slack_ns;
1594 task_io_accounting_init(&p->ioac);
1595 acct_clear_integrals(p);
1597 posix_cpu_timers_init(p);
1599 p->start_time = ktime_get_ns();
1600 p->real_start_time = ktime_get_boot_ns();
1601 p->io_context = NULL;
1602 p->audit_context = NULL;
1605 p->mempolicy = mpol_dup(p->mempolicy);
1606 if (IS_ERR(p->mempolicy)) {
1607 retval = PTR_ERR(p->mempolicy);
1608 p->mempolicy = NULL;
1609 goto bad_fork_cleanup_threadgroup_lock;
1612 #ifdef CONFIG_CPUSETS
1613 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1614 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1615 seqcount_init(&p->mems_allowed_seq);
1617 #ifdef CONFIG_TRACE_IRQFLAGS
1619 p->hardirqs_enabled = 0;
1620 p->hardirq_enable_ip = 0;
1621 p->hardirq_enable_event = 0;
1622 p->hardirq_disable_ip = _THIS_IP_;
1623 p->hardirq_disable_event = 0;
1624 p->softirqs_enabled = 1;
1625 p->softirq_enable_ip = _THIS_IP_;
1626 p->softirq_enable_event = 0;
1627 p->softirq_disable_ip = 0;
1628 p->softirq_disable_event = 0;
1629 p->hardirq_context = 0;
1630 p->softirq_context = 0;
1633 p->pagefault_disabled = 0;
1635 #ifdef CONFIG_LOCKDEP
1636 p->lockdep_depth = 0; /* no locks held yet */
1637 p->curr_chain_key = 0;
1638 p->lockdep_recursion = 0;
1641 #ifdef CONFIG_DEBUG_MUTEXES
1642 p->blocked_on = NULL; /* not blocked yet */
1644 #ifdef CONFIG_BCACHE
1645 p->sequential_io = 0;
1646 p->sequential_io_avg = 0;
1649 /* Perform scheduler related setup. Assign this task to a CPU. */
1650 retval = sched_fork(clone_flags, p);
1652 goto bad_fork_cleanup_policy;
1654 retval = perf_event_init_task(p);
1656 goto bad_fork_cleanup_policy;
1657 retval = audit_alloc(p);
1659 goto bad_fork_cleanup_perf;
1660 /* copy all the process information */
1662 retval = copy_semundo(clone_flags, p);
1664 goto bad_fork_cleanup_audit;
1665 retval = copy_files(clone_flags, p);
1667 goto bad_fork_cleanup_semundo;
1668 retval = copy_fs(clone_flags, p);
1670 goto bad_fork_cleanup_files;
1671 retval = copy_sighand(clone_flags, p);
1673 goto bad_fork_cleanup_fs;
1674 retval = copy_signal(clone_flags, p);
1676 goto bad_fork_cleanup_sighand;
1677 retval = copy_mm(clone_flags, p);
1679 goto bad_fork_cleanup_signal;
1680 retval = copy_namespaces(clone_flags, p);
1682 goto bad_fork_cleanup_mm;
1683 retval = copy_io(clone_flags, p);
1685 goto bad_fork_cleanup_namespaces;
1686 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1688 goto bad_fork_cleanup_io;
1690 if (pid != &init_struct_pid) {
1691 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1693 retval = PTR_ERR(pid);
1694 goto bad_fork_cleanup_thread;
1698 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1700 * Clear TID on mm_release()?
1702 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1707 p->robust_list = NULL;
1708 #ifdef CONFIG_COMPAT
1709 p->compat_robust_list = NULL;
1711 INIT_LIST_HEAD(&p->pi_state_list);
1712 p->pi_state_cache = NULL;
1715 * sigaltstack should be cleared when sharing the same VM
1717 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1721 * Syscall tracing and stepping should be turned off in the
1722 * child regardless of CLONE_PTRACE.
1724 user_disable_single_step(p);
1725 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1726 #ifdef TIF_SYSCALL_EMU
1727 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1729 clear_all_latency_tracing(p);
1731 /* ok, now we should be set up.. */
1732 p->pid = pid_nr(pid);
1733 if (clone_flags & CLONE_THREAD) {
1734 p->exit_signal = -1;
1735 p->group_leader = current->group_leader;
1736 p->tgid = current->tgid;
1738 if (clone_flags & CLONE_PARENT)
1739 p->exit_signal = current->group_leader->exit_signal;
1741 p->exit_signal = (clone_flags & CSIGNAL);
1742 p->group_leader = p;
1747 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1748 p->dirty_paused_when = 0;
1750 p->pdeath_signal = 0;
1751 INIT_LIST_HEAD(&p->thread_group);
1752 p->task_works = NULL;
1754 cgroup_threadgroup_change_begin(current);
1756 * Ensure that the cgroup subsystem policies allow the new process to be
1757 * forked. It should be noted the the new process's css_set can be changed
1758 * between here and cgroup_post_fork() if an organisation operation is in
1761 retval = cgroup_can_fork(p);
1763 goto bad_fork_free_pid;
1766 * Make it visible to the rest of the system, but dont wake it up yet.
1767 * Need tasklist lock for parent etc handling!
1769 write_lock_irq(&tasklist_lock);
1771 /* CLONE_PARENT re-uses the old parent */
1772 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1773 p->real_parent = current->real_parent;
1774 p->parent_exec_id = current->parent_exec_id;
1776 p->real_parent = current;
1777 p->parent_exec_id = current->self_exec_id;
1780 spin_lock(¤t->sighand->siglock);
1783 * Copy seccomp details explicitly here, in case they were changed
1784 * before holding sighand lock.
1789 * Process group and session signals need to be delivered to just the
1790 * parent before the fork or both the parent and the child after the
1791 * fork. Restart if a signal comes in before we add the new process to
1792 * it's process group.
1793 * A fatal signal pending means that current will exit, so the new
1794 * thread can't slip out of an OOM kill (or normal SIGKILL).
1796 recalc_sigpending();
1797 if (signal_pending(current)) {
1798 spin_unlock(¤t->sighand->siglock);
1799 write_unlock_irq(&tasklist_lock);
1800 retval = -ERESTARTNOINTR;
1801 goto bad_fork_cancel_cgroup;
1804 if (likely(p->pid)) {
1805 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1807 init_task_pid(p, PIDTYPE_PID, pid);
1808 if (thread_group_leader(p)) {
1809 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1810 init_task_pid(p, PIDTYPE_SID, task_session(current));
1812 if (is_child_reaper(pid)) {
1813 ns_of_pid(pid)->child_reaper = p;
1814 p->signal->flags |= SIGNAL_UNKILLABLE;
1817 p->signal->leader_pid = pid;
1818 p->signal->tty = tty_kref_get(current->signal->tty);
1820 * Inherit has_child_subreaper flag under the same
1821 * tasklist_lock with adding child to the process tree
1822 * for propagate_has_child_subreaper optimization.
1824 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1825 p->real_parent->signal->is_child_subreaper;
1826 list_add_tail(&p->sibling, &p->real_parent->children);
1827 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1828 attach_pid(p, PIDTYPE_PGID);
1829 attach_pid(p, PIDTYPE_SID);
1830 __this_cpu_inc(process_counts);
1832 current->signal->nr_threads++;
1833 atomic_inc(¤t->signal->live);
1834 atomic_inc(¤t->signal->sigcnt);
1835 list_add_tail_rcu(&p->thread_group,
1836 &p->group_leader->thread_group);
1837 list_add_tail_rcu(&p->thread_node,
1838 &p->signal->thread_head);
1840 attach_pid(p, PIDTYPE_PID);
1845 spin_unlock(¤t->sighand->siglock);
1846 syscall_tracepoint_update(p);
1847 write_unlock_irq(&tasklist_lock);
1849 proc_fork_connector(p);
1850 cgroup_post_fork(p);
1851 cgroup_threadgroup_change_end(current);
1854 trace_task_newtask(p, clone_flags);
1855 uprobe_copy_process(p, clone_flags);
1859 bad_fork_cancel_cgroup:
1860 cgroup_cancel_fork(p);
1862 cgroup_threadgroup_change_end(current);
1863 if (pid != &init_struct_pid)
1865 bad_fork_cleanup_thread:
1867 bad_fork_cleanup_io:
1870 bad_fork_cleanup_namespaces:
1871 exit_task_namespaces(p);
1872 bad_fork_cleanup_mm:
1875 bad_fork_cleanup_signal:
1876 if (!(clone_flags & CLONE_THREAD))
1877 free_signal_struct(p->signal);
1878 bad_fork_cleanup_sighand:
1879 __cleanup_sighand(p->sighand);
1880 bad_fork_cleanup_fs:
1881 exit_fs(p); /* blocking */
1882 bad_fork_cleanup_files:
1883 exit_files(p); /* blocking */
1884 bad_fork_cleanup_semundo:
1886 bad_fork_cleanup_audit:
1888 bad_fork_cleanup_perf:
1889 perf_event_free_task(p);
1890 bad_fork_cleanup_policy:
1892 mpol_put(p->mempolicy);
1893 bad_fork_cleanup_threadgroup_lock:
1895 delayacct_tsk_free(p);
1896 bad_fork_cleanup_count:
1897 atomic_dec(&p->cred->user->processes);
1900 p->state = TASK_DEAD;
1904 return ERR_PTR(retval);
1907 static inline void init_idle_pids(struct pid_link *links)
1911 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1912 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1913 links[type].pid = &init_struct_pid;
1917 struct task_struct *fork_idle(int cpu)
1919 struct task_struct *task;
1920 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1922 if (!IS_ERR(task)) {
1923 init_idle_pids(task->pids);
1924 init_idle(task, cpu);
1931 * Ok, this is the main fork-routine.
1933 * It copies the process, and if successful kick-starts
1934 * it and waits for it to finish using the VM if required.
1936 long _do_fork(unsigned long clone_flags,
1937 unsigned long stack_start,
1938 unsigned long stack_size,
1939 int __user *parent_tidptr,
1940 int __user *child_tidptr,
1943 struct task_struct *p;
1948 * Determine whether and which event to report to ptracer. When
1949 * called from kernel_thread or CLONE_UNTRACED is explicitly
1950 * requested, no event is reported; otherwise, report if the event
1951 * for the type of forking is enabled.
1953 if (!(clone_flags & CLONE_UNTRACED)) {
1954 if (clone_flags & CLONE_VFORK)
1955 trace = PTRACE_EVENT_VFORK;
1956 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1957 trace = PTRACE_EVENT_CLONE;
1959 trace = PTRACE_EVENT_FORK;
1961 if (likely(!ptrace_event_enabled(current, trace)))
1965 p = copy_process(clone_flags, stack_start, stack_size,
1966 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1967 add_latent_entropy();
1969 * Do this prior waking up the new thread - the thread pointer
1970 * might get invalid after that point, if the thread exits quickly.
1973 struct completion vfork;
1976 trace_sched_process_fork(current, p);
1978 pid = get_task_pid(p, PIDTYPE_PID);
1981 if (clone_flags & CLONE_PARENT_SETTID)
1982 put_user(nr, parent_tidptr);
1984 if (clone_flags & CLONE_VFORK) {
1985 p->vfork_done = &vfork;
1986 init_completion(&vfork);
1990 wake_up_new_task(p);
1992 /* forking complete and child started to run, tell ptracer */
1993 if (unlikely(trace))
1994 ptrace_event_pid(trace, pid);
1996 if (clone_flags & CLONE_VFORK) {
1997 if (!wait_for_vfork_done(p, &vfork))
1998 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2008 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2009 /* For compatibility with architectures that call do_fork directly rather than
2010 * using the syscall entry points below. */
2011 long do_fork(unsigned long clone_flags,
2012 unsigned long stack_start,
2013 unsigned long stack_size,
2014 int __user *parent_tidptr,
2015 int __user *child_tidptr)
2017 return _do_fork(clone_flags, stack_start, stack_size,
2018 parent_tidptr, child_tidptr, 0);
2023 * Create a kernel thread.
2025 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2027 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2028 (unsigned long)arg, NULL, NULL, 0);
2031 #ifdef __ARCH_WANT_SYS_FORK
2032 SYSCALL_DEFINE0(fork)
2035 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2037 /* can not support in nommu mode */
2043 #ifdef __ARCH_WANT_SYS_VFORK
2044 SYSCALL_DEFINE0(vfork)
2046 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2051 #ifdef __ARCH_WANT_SYS_CLONE
2052 #ifdef CONFIG_CLONE_BACKWARDS
2053 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2054 int __user *, parent_tidptr,
2056 int __user *, child_tidptr)
2057 #elif defined(CONFIG_CLONE_BACKWARDS2)
2058 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2059 int __user *, parent_tidptr,
2060 int __user *, child_tidptr,
2062 #elif defined(CONFIG_CLONE_BACKWARDS3)
2063 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2065 int __user *, parent_tidptr,
2066 int __user *, child_tidptr,
2069 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2070 int __user *, parent_tidptr,
2071 int __user *, child_tidptr,
2075 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2079 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2081 struct task_struct *leader, *parent, *child;
2084 read_lock(&tasklist_lock);
2085 leader = top = top->group_leader;
2087 for_each_thread(leader, parent) {
2088 list_for_each_entry(child, &parent->children, sibling) {
2089 res = visitor(child, data);
2101 if (leader != top) {
2103 parent = child->real_parent;
2104 leader = parent->group_leader;
2108 read_unlock(&tasklist_lock);
2111 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2112 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2115 static void sighand_ctor(void *data)
2117 struct sighand_struct *sighand = data;
2119 spin_lock_init(&sighand->siglock);
2120 init_waitqueue_head(&sighand->signalfd_wqh);
2123 void __init proc_caches_init(void)
2125 sighand_cachep = kmem_cache_create("sighand_cache",
2126 sizeof(struct sighand_struct), 0,
2127 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2128 SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2129 signal_cachep = kmem_cache_create("signal_cache",
2130 sizeof(struct signal_struct), 0,
2131 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2133 files_cachep = kmem_cache_create("files_cache",
2134 sizeof(struct files_struct), 0,
2135 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2137 fs_cachep = kmem_cache_create("fs_cache",
2138 sizeof(struct fs_struct), 0,
2139 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2142 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2143 * whole struct cpumask for the OFFSTACK case. We could change
2144 * this to *only* allocate as much of it as required by the
2145 * maximum number of CPU's we can ever have. The cpumask_allocation
2146 * is at the end of the structure, exactly for that reason.
2148 mm_cachep = kmem_cache_create("mm_struct",
2149 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2150 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2152 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2154 nsproxy_cache_init();
2158 * Check constraints on flags passed to the unshare system call.
2160 static int check_unshare_flags(unsigned long unshare_flags)
2162 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2163 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2164 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2165 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2168 * Not implemented, but pretend it works if there is nothing
2169 * to unshare. Note that unsharing the address space or the
2170 * signal handlers also need to unshare the signal queues (aka
2173 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2174 if (!thread_group_empty(current))
2177 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2178 if (atomic_read(¤t->sighand->count) > 1)
2181 if (unshare_flags & CLONE_VM) {
2182 if (!current_is_single_threaded())
2190 * Unshare the filesystem structure if it is being shared
2192 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2194 struct fs_struct *fs = current->fs;
2196 if (!(unshare_flags & CLONE_FS) || !fs)
2199 /* don't need lock here; in the worst case we'll do useless copy */
2203 *new_fsp = copy_fs_struct(fs);
2211 * Unshare file descriptor table if it is being shared
2213 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2215 struct files_struct *fd = current->files;
2218 if ((unshare_flags & CLONE_FILES) &&
2219 (fd && atomic_read(&fd->count) > 1)) {
2220 *new_fdp = dup_fd(fd, &error);
2229 * unshare allows a process to 'unshare' part of the process
2230 * context which was originally shared using clone. copy_*
2231 * functions used by do_fork() cannot be used here directly
2232 * because they modify an inactive task_struct that is being
2233 * constructed. Here we are modifying the current, active,
2236 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2238 struct fs_struct *fs, *new_fs = NULL;
2239 struct files_struct *fd, *new_fd = NULL;
2240 struct cred *new_cred = NULL;
2241 struct nsproxy *new_nsproxy = NULL;
2246 * If unsharing a user namespace must also unshare the thread group
2247 * and unshare the filesystem root and working directories.
2249 if (unshare_flags & CLONE_NEWUSER)
2250 unshare_flags |= CLONE_THREAD | CLONE_FS;
2252 * If unsharing vm, must also unshare signal handlers.
2254 if (unshare_flags & CLONE_VM)
2255 unshare_flags |= CLONE_SIGHAND;
2257 * If unsharing a signal handlers, must also unshare the signal queues.
2259 if (unshare_flags & CLONE_SIGHAND)
2260 unshare_flags |= CLONE_THREAD;
2262 * If unsharing namespace, must also unshare filesystem information.
2264 if (unshare_flags & CLONE_NEWNS)
2265 unshare_flags |= CLONE_FS;
2267 err = check_unshare_flags(unshare_flags);
2269 goto bad_unshare_out;
2271 * CLONE_NEWIPC must also detach from the undolist: after switching
2272 * to a new ipc namespace, the semaphore arrays from the old
2273 * namespace are unreachable.
2275 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2277 err = unshare_fs(unshare_flags, &new_fs);
2279 goto bad_unshare_out;
2280 err = unshare_fd(unshare_flags, &new_fd);
2282 goto bad_unshare_cleanup_fs;
2283 err = unshare_userns(unshare_flags, &new_cred);
2285 goto bad_unshare_cleanup_fd;
2286 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2289 goto bad_unshare_cleanup_cred;
2291 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2294 * CLONE_SYSVSEM is equivalent to sys_exit().
2298 if (unshare_flags & CLONE_NEWIPC) {
2299 /* Orphan segments in old ns (see sem above). */
2301 shm_init_task(current);
2305 switch_task_namespaces(current, new_nsproxy);
2311 spin_lock(&fs->lock);
2312 current->fs = new_fs;
2317 spin_unlock(&fs->lock);
2321 fd = current->files;
2322 current->files = new_fd;
2326 task_unlock(current);
2329 /* Install the new user namespace */
2330 commit_creds(new_cred);
2335 bad_unshare_cleanup_cred:
2338 bad_unshare_cleanup_fd:
2340 put_files_struct(new_fd);
2342 bad_unshare_cleanup_fs:
2344 free_fs_struct(new_fs);
2351 * Helper to unshare the files of the current task.
2352 * We don't want to expose copy_files internals to
2353 * the exec layer of the kernel.
2356 int unshare_files(struct files_struct **displaced)
2358 struct task_struct *task = current;
2359 struct files_struct *copy = NULL;
2362 error = unshare_fd(CLONE_FILES, ©);
2363 if (error || !copy) {
2367 *displaced = task->files;
2374 int sysctl_max_threads(struct ctl_table *table, int write,
2375 void __user *buffer, size_t *lenp, loff_t *ppos)
2379 int threads = max_threads;
2380 int min = MIN_THREADS;
2381 int max = MAX_THREADS;
2388 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2392 set_max_threads(threads);