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/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
32 #include <linux/vmacache.h>
33 #include <linux/nsproxy.h>
34 #include <linux/capability.h>
35 #include <linux/cpu.h>
36 #include <linux/cgroup.h>
37 #include <linux/security.h>
38 #include <linux/hugetlb.h>
39 #include <linux/seccomp.h>
40 #include <linux/swap.h>
41 #include <linux/syscalls.h>
42 #include <linux/jiffies.h>
43 #include <linux/futex.h>
44 #include <linux/compat.h>
45 #include <linux/kthread.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/rcupdate.h>
48 #include <linux/ptrace.h>
49 #include <linux/mount.h>
50 #include <linux/audit.h>
51 #include <linux/memcontrol.h>
52 #include <linux/ftrace.h>
53 #include <linux/proc_fs.h>
54 #include <linux/profile.h>
55 #include <linux/rmap.h>
56 #include <linux/ksm.h>
57 #include <linux/acct.h>
58 #include <linux/tsacct_kern.h>
59 #include <linux/cn_proc.h>
60 #include <linux/freezer.h>
61 #include <linux/delayacct.h>
62 #include <linux/taskstats_kern.h>
63 #include <linux/random.h>
64 #include <linux/tty.h>
65 #include <linux/blkdev.h>
66 #include <linux/fs_struct.h>
67 #include <linux/magic.h>
68 #include <linux/perf_event.h>
69 #include <linux/posix-timers.h>
70 #include <linux/user-return-notifier.h>
71 #include <linux/oom.h>
72 #include <linux/khugepaged.h>
73 #include <linux/signalfd.h>
74 #include <linux/uprobes.h>
75 #include <linux/aio.h>
76 #include <linux/compiler.h>
77 #include <linux/sysctl.h>
79 #include <asm/pgtable.h>
80 #include <asm/pgalloc.h>
81 #include <asm/uaccess.h>
82 #include <asm/mmu_context.h>
83 #include <asm/cacheflush.h>
84 #include <asm/tlbflush.h>
86 #include <trace/events/sched.h>
88 #define CREATE_TRACE_POINTS
89 #include <trace/events/task.h>
92 * Minimum number of threads to boot the kernel
94 #define MIN_THREADS 20
97 * Maximum number of threads
99 #define MAX_THREADS FUTEX_TID_MASK
102 * Protected counters by write_lock_irq(&tasklist_lock)
104 unsigned long total_forks; /* Handle normal Linux uptimes. */
105 int nr_threads; /* The idle threads do not count.. */
107 int max_threads; /* tunable limit on nr_threads */
109 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
111 DEFINE_RWLOCK(tasklist_lock); /* outer */
113 #ifdef CONFIG_PROVE_RCU
114 int lockdep_tasklist_lock_is_held(void)
116 return lockdep_is_held(&tasklist_lock);
118 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
119 #endif /* #ifdef CONFIG_PROVE_RCU */
121 int nr_processes(void)
126 for_each_possible_cpu(cpu)
127 total += per_cpu(process_counts, cpu);
132 void __weak arch_release_task_struct(struct task_struct *tsk)
136 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
137 static struct kmem_cache *task_struct_cachep;
139 static inline struct task_struct *alloc_task_struct_node(int node)
141 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
144 static inline void free_task_struct(struct task_struct *tsk)
146 kmem_cache_free(task_struct_cachep, tsk);
150 void __weak arch_release_thread_info(struct thread_info *ti)
154 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
157 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
158 * kmemcache based allocator.
160 # if THREAD_SIZE >= PAGE_SIZE
161 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
164 struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
167 return page ? page_address(page) : NULL;
170 static inline void free_thread_info(struct thread_info *ti)
172 free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
175 static struct kmem_cache *thread_info_cache;
177 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
180 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
183 static void free_thread_info(struct thread_info *ti)
185 kmem_cache_free(thread_info_cache, ti);
188 void thread_info_cache_init(void)
190 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
191 THREAD_SIZE, 0, NULL);
192 BUG_ON(thread_info_cache == NULL);
197 /* SLAB cache for signal_struct structures (tsk->signal) */
198 static struct kmem_cache *signal_cachep;
200 /* SLAB cache for sighand_struct structures (tsk->sighand) */
201 struct kmem_cache *sighand_cachep;
203 /* SLAB cache for files_struct structures (tsk->files) */
204 struct kmem_cache *files_cachep;
206 /* SLAB cache for fs_struct structures (tsk->fs) */
207 struct kmem_cache *fs_cachep;
209 /* SLAB cache for vm_area_struct structures */
210 struct kmem_cache *vm_area_cachep;
212 /* SLAB cache for mm_struct structures (tsk->mm) */
213 static struct kmem_cache *mm_cachep;
215 static void account_kernel_stack(struct thread_info *ti, int account)
217 struct zone *zone = page_zone(virt_to_page(ti));
219 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
222 void free_task(struct task_struct *tsk)
224 account_kernel_stack(tsk->stack, -1);
225 arch_release_thread_info(tsk->stack);
226 free_thread_info(tsk->stack);
227 rt_mutex_debug_task_free(tsk);
228 ftrace_graph_exit_task(tsk);
229 put_seccomp_filter(tsk);
230 arch_release_task_struct(tsk);
231 free_task_struct(tsk);
233 EXPORT_SYMBOL(free_task);
235 static inline void free_signal_struct(struct signal_struct *sig)
237 taskstats_tgid_free(sig);
238 sched_autogroup_exit(sig);
239 kmem_cache_free(signal_cachep, sig);
242 static inline void put_signal_struct(struct signal_struct *sig)
244 if (atomic_dec_and_test(&sig->sigcnt))
245 free_signal_struct(sig);
247 #ifdef CONFIG_PREEMPT_RT_BASE
250 void __put_task_struct(struct task_struct *tsk)
252 WARN_ON(!tsk->exit_state);
253 WARN_ON(atomic_read(&tsk->usage));
254 WARN_ON(tsk == current);
258 security_task_free(tsk);
260 delayacct_tsk_free(tsk);
261 put_signal_struct(tsk->signal);
263 if (!profile_handoff_task(tsk))
266 #ifndef CONFIG_PREEMPT_RT_BASE
267 EXPORT_SYMBOL_GPL(__put_task_struct);
269 void __put_task_struct_cb(struct rcu_head *rhp)
271 struct task_struct *tsk = container_of(rhp, struct task_struct, put_rcu);
273 __put_task_struct(tsk);
276 EXPORT_SYMBOL_GPL(__put_task_struct_cb);
279 void __init __weak arch_task_cache_init(void) { }
284 static void set_max_threads(unsigned int max_threads_suggested)
289 * The number of threads shall be limited such that the thread
290 * structures may only consume a small part of the available memory.
292 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
293 threads = MAX_THREADS;
295 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
296 (u64) THREAD_SIZE * 8UL);
298 if (threads > max_threads_suggested)
299 threads = max_threads_suggested;
301 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
304 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
305 /* Initialized by the architecture: */
306 int arch_task_struct_size __read_mostly;
309 void __init fork_init(void)
311 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
312 #ifndef ARCH_MIN_TASKALIGN
313 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
315 /* create a slab on which task_structs can be allocated */
317 kmem_cache_create("task_struct", arch_task_struct_size,
318 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
321 /* do the arch specific task caches init */
322 arch_task_cache_init();
324 set_max_threads(MAX_THREADS);
326 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
327 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
328 init_task.signal->rlim[RLIMIT_SIGPENDING] =
329 init_task.signal->rlim[RLIMIT_NPROC];
332 int __weak arch_dup_task_struct(struct task_struct *dst,
333 struct task_struct *src)
339 void set_task_stack_end_magic(struct task_struct *tsk)
341 unsigned long *stackend;
343 stackend = end_of_stack(tsk);
344 *stackend = STACK_END_MAGIC; /* for overflow detection */
347 static struct task_struct *dup_task_struct(struct task_struct *orig)
349 struct task_struct *tsk;
350 struct thread_info *ti;
351 int node = tsk_fork_get_node(orig);
354 tsk = alloc_task_struct_node(node);
358 ti = alloc_thread_info_node(tsk, node);
362 err = arch_dup_task_struct(tsk, orig);
367 #ifdef CONFIG_SECCOMP
369 * We must handle setting up seccomp filters once we're under
370 * the sighand lock in case orig has changed between now and
371 * then. Until then, filter must be NULL to avoid messing up
372 * the usage counts on the error path calling free_task.
374 tsk->seccomp.filter = NULL;
377 setup_thread_stack(tsk, orig);
378 clear_user_return_notifier(tsk);
379 clear_tsk_need_resched(tsk);
380 set_task_stack_end_magic(tsk);
382 #ifdef CONFIG_CC_STACKPROTECTOR
383 tsk->stack_canary = get_random_int();
387 * One for us, one for whoever does the "release_task()" (usually
390 atomic_set(&tsk->usage, 2);
391 #ifdef CONFIG_BLK_DEV_IO_TRACE
394 tsk->splice_pipe = NULL;
395 tsk->task_frag.page = NULL;
396 tsk->wake_q.next = NULL;
398 account_kernel_stack(ti, 1);
403 free_thread_info(ti);
405 free_task_struct(tsk);
410 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
412 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
413 struct rb_node **rb_link, *rb_parent;
415 unsigned long charge;
417 uprobe_start_dup_mmap();
418 down_write(&oldmm->mmap_sem);
419 flush_cache_dup_mm(oldmm);
420 uprobe_dup_mmap(oldmm, mm);
422 * Not linked in yet - no deadlock potential:
424 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
426 /* No ordering required: file already has been exposed. */
427 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
429 mm->total_vm = oldmm->total_vm;
430 mm->shared_vm = oldmm->shared_vm;
431 mm->exec_vm = oldmm->exec_vm;
432 mm->stack_vm = oldmm->stack_vm;
434 rb_link = &mm->mm_rb.rb_node;
437 retval = ksm_fork(mm, oldmm);
440 retval = khugepaged_fork(mm, oldmm);
445 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
448 if (mpnt->vm_flags & VM_DONTCOPY) {
449 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
454 if (mpnt->vm_flags & VM_ACCOUNT) {
455 unsigned long len = vma_pages(mpnt);
457 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
461 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
465 INIT_LIST_HEAD(&tmp->anon_vma_chain);
466 retval = vma_dup_policy(mpnt, tmp);
468 goto fail_nomem_policy;
470 if (anon_vma_fork(tmp, mpnt))
471 goto fail_nomem_anon_vma_fork;
473 ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
474 tmp->vm_next = tmp->vm_prev = NULL;
475 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
478 struct inode *inode = file_inode(file);
479 struct address_space *mapping = file->f_mapping;
482 if (tmp->vm_flags & VM_DENYWRITE)
483 atomic_dec(&inode->i_writecount);
484 i_mmap_lock_write(mapping);
485 if (tmp->vm_flags & VM_SHARED)
486 atomic_inc(&mapping->i_mmap_writable);
487 flush_dcache_mmap_lock(mapping);
488 /* insert tmp into the share list, just after mpnt */
489 vma_interval_tree_insert_after(tmp, mpnt,
491 flush_dcache_mmap_unlock(mapping);
492 i_mmap_unlock_write(mapping);
496 * Clear hugetlb-related page reserves for children. This only
497 * affects MAP_PRIVATE mappings. Faults generated by the child
498 * are not guaranteed to succeed, even if read-only
500 if (is_vm_hugetlb_page(tmp))
501 reset_vma_resv_huge_pages(tmp);
504 * Link in the new vma and copy the page table entries.
507 pprev = &tmp->vm_next;
511 __vma_link_rb(mm, tmp, rb_link, rb_parent);
512 rb_link = &tmp->vm_rb.rb_right;
513 rb_parent = &tmp->vm_rb;
516 retval = copy_page_range(mm, oldmm, mpnt);
518 if (tmp->vm_ops && tmp->vm_ops->open)
519 tmp->vm_ops->open(tmp);
524 /* a new mm has just been created */
525 arch_dup_mmap(oldmm, mm);
528 up_write(&mm->mmap_sem);
530 up_write(&oldmm->mmap_sem);
531 uprobe_end_dup_mmap();
533 fail_nomem_anon_vma_fork:
534 mpol_put(vma_policy(tmp));
536 kmem_cache_free(vm_area_cachep, tmp);
539 vm_unacct_memory(charge);
543 static inline int mm_alloc_pgd(struct mm_struct *mm)
545 mm->pgd = pgd_alloc(mm);
546 if (unlikely(!mm->pgd))
551 static inline void mm_free_pgd(struct mm_struct *mm)
553 pgd_free(mm, mm->pgd);
556 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
558 down_write(&oldmm->mmap_sem);
559 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
560 up_write(&oldmm->mmap_sem);
563 #define mm_alloc_pgd(mm) (0)
564 #define mm_free_pgd(mm)
565 #endif /* CONFIG_MMU */
567 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
569 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
570 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
572 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
574 static int __init coredump_filter_setup(char *s)
576 default_dump_filter =
577 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
578 MMF_DUMP_FILTER_MASK;
582 __setup("coredump_filter=", coredump_filter_setup);
584 #include <linux/init_task.h>
586 static void mm_init_aio(struct mm_struct *mm)
589 spin_lock_init(&mm->ioctx_lock);
590 mm->ioctx_table = NULL;
594 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
601 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
602 struct user_namespace *user_ns)
606 mm->vmacache_seqnum = 0;
607 atomic_set(&mm->mm_users, 1);
608 atomic_set(&mm->mm_count, 1);
609 init_rwsem(&mm->mmap_sem);
610 INIT_LIST_HEAD(&mm->mmlist);
611 mm->core_state = NULL;
612 atomic_long_set(&mm->nr_ptes, 0);
617 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
618 spin_lock_init(&mm->page_table_lock);
621 mm_init_owner(mm, p);
622 mmu_notifier_mm_init(mm);
623 clear_tlb_flush_pending(mm);
624 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
625 mm->pmd_huge_pte = NULL;
629 mm->flags = current->mm->flags & MMF_INIT_MASK;
630 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
632 mm->flags = default_dump_filter;
636 if (mm_alloc_pgd(mm))
639 if (init_new_context(p, mm))
642 mm->user_ns = get_user_ns(user_ns);
652 static void check_mm(struct mm_struct *mm)
656 for (i = 0; i < NR_MM_COUNTERS; i++) {
657 long x = atomic_long_read(&mm->rss_stat.count[i]);
660 printk(KERN_ALERT "BUG: Bad rss-counter state "
661 "mm:%p idx:%d val:%ld\n", mm, i, x);
664 if (atomic_long_read(&mm->nr_ptes))
665 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
666 atomic_long_read(&mm->nr_ptes));
668 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
671 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
672 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
677 * Allocate and initialize an mm_struct.
679 struct mm_struct *mm_alloc(void)
681 struct mm_struct *mm;
687 memset(mm, 0, sizeof(*mm));
688 return mm_init(mm, current, current_user_ns());
692 * Called when the last reference to the mm
693 * is dropped: either by a lazy thread or by
694 * mmput. Free the page directory and the mm.
696 void __mmdrop(struct mm_struct *mm)
698 BUG_ON(mm == &init_mm);
701 mmu_notifier_mm_destroy(mm);
703 put_user_ns(mm->user_ns);
706 EXPORT_SYMBOL_GPL(__mmdrop);
708 #ifdef CONFIG_PREEMPT_RT_BASE
710 * RCU callback for delayed mm drop. Not strictly rcu, but we don't
711 * want another facility to make this work.
713 void __mmdrop_delayed(struct rcu_head *rhp)
715 struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
722 * Decrement the use count and release all resources for an mm.
724 void mmput(struct mm_struct *mm)
728 if (atomic_dec_and_test(&mm->mm_users)) {
729 uprobe_clear_state(mm);
732 khugepaged_exit(mm); /* must run before exit_mmap */
734 set_mm_exe_file(mm, NULL);
735 if (!list_empty(&mm->mmlist)) {
736 spin_lock(&mmlist_lock);
737 list_del(&mm->mmlist);
738 spin_unlock(&mmlist_lock);
741 module_put(mm->binfmt->module);
745 EXPORT_SYMBOL_GPL(mmput);
748 * set_mm_exe_file - change a reference to the mm's executable file
750 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
752 * Main users are mmput() and sys_execve(). Callers prevent concurrent
753 * invocations: in mmput() nobody alive left, in execve task is single
754 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
755 * mm->exe_file, but does so without using set_mm_exe_file() in order
756 * to do avoid the need for any locks.
758 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
760 struct file *old_exe_file;
763 * It is safe to dereference the exe_file without RCU as
764 * this function is only called if nobody else can access
765 * this mm -- see comment above for justification.
767 old_exe_file = rcu_dereference_raw(mm->exe_file);
770 get_file(new_exe_file);
771 rcu_assign_pointer(mm->exe_file, new_exe_file);
777 * get_mm_exe_file - acquire a reference to the mm's executable file
779 * Returns %NULL if mm has no associated executable file.
780 * User must release file via fput().
782 struct file *get_mm_exe_file(struct mm_struct *mm)
784 struct file *exe_file;
787 exe_file = rcu_dereference(mm->exe_file);
788 if (exe_file && !get_file_rcu(exe_file))
793 EXPORT_SYMBOL(get_mm_exe_file);
796 * get_task_exe_file - acquire a reference to the task's executable file
798 * Returns %NULL if task's mm (if any) has no associated executable file or
799 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
800 * User must release file via fput().
802 struct file *get_task_exe_file(struct task_struct *task)
804 struct file *exe_file = NULL;
805 struct mm_struct *mm;
810 if (!(task->flags & PF_KTHREAD))
811 exe_file = get_mm_exe_file(mm);
816 EXPORT_SYMBOL(get_task_exe_file);
819 * get_task_mm - acquire a reference to the task's mm
821 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
822 * this kernel workthread has transiently adopted a user mm with use_mm,
823 * to do its AIO) is not set and if so returns a reference to it, after
824 * bumping up the use count. User must release the mm via mmput()
825 * after use. Typically used by /proc and ptrace.
827 struct mm_struct *get_task_mm(struct task_struct *task)
829 struct mm_struct *mm;
834 if (task->flags & PF_KTHREAD)
837 atomic_inc(&mm->mm_users);
842 EXPORT_SYMBOL_GPL(get_task_mm);
844 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
846 struct mm_struct *mm;
849 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
853 mm = get_task_mm(task);
854 if (mm && mm != current->mm &&
855 !ptrace_may_access(task, mode)) {
857 mm = ERR_PTR(-EACCES);
859 mutex_unlock(&task->signal->cred_guard_mutex);
864 static void complete_vfork_done(struct task_struct *tsk)
866 struct completion *vfork;
869 vfork = tsk->vfork_done;
871 tsk->vfork_done = NULL;
877 static int wait_for_vfork_done(struct task_struct *child,
878 struct completion *vfork)
882 freezer_do_not_count();
883 killed = wait_for_completion_killable(vfork);
888 child->vfork_done = NULL;
892 put_task_struct(child);
896 /* Please note the differences between mmput and mm_release.
897 * mmput is called whenever we stop holding onto a mm_struct,
898 * error success whatever.
900 * mm_release is called after a mm_struct has been removed
901 * from the current process.
903 * This difference is important for error handling, when we
904 * only half set up a mm_struct for a new process and need to restore
905 * the old one. Because we mmput the new mm_struct before
906 * restoring the old one. . .
907 * Eric Biederman 10 January 1998
909 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
911 /* Get rid of any futexes when releasing the mm */
913 if (unlikely(tsk->robust_list)) {
914 exit_robust_list(tsk);
915 tsk->robust_list = NULL;
918 if (unlikely(tsk->compat_robust_list)) {
919 compat_exit_robust_list(tsk);
920 tsk->compat_robust_list = NULL;
923 if (unlikely(!list_empty(&tsk->pi_state_list)))
924 exit_pi_state_list(tsk);
927 uprobe_free_utask(tsk);
929 /* Get rid of any cached register state */
930 deactivate_mm(tsk, mm);
933 * Signal userspace if we're not exiting with a core dump
934 * because we want to leave the value intact for debugging
937 if (tsk->clear_child_tid) {
938 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
939 atomic_read(&mm->mm_users) > 1) {
941 * We don't check the error code - if userspace has
942 * not set up a proper pointer then tough luck.
944 put_user(0, tsk->clear_child_tid);
945 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
948 tsk->clear_child_tid = NULL;
952 * All done, finally we can wake up parent and return this mm to him.
953 * Also kthread_stop() uses this completion for synchronization.
956 complete_vfork_done(tsk);
960 * Allocate a new mm structure and copy contents from the
961 * mm structure of the passed in task structure.
963 static struct mm_struct *dup_mm(struct task_struct *tsk)
965 struct mm_struct *mm, *oldmm = current->mm;
972 memcpy(mm, oldmm, sizeof(*mm));
974 if (!mm_init(mm, tsk, mm->user_ns))
977 err = dup_mmap(mm, oldmm);
981 mm->hiwater_rss = get_mm_rss(mm);
982 mm->hiwater_vm = mm->total_vm;
984 if (mm->binfmt && !try_module_get(mm->binfmt->module))
990 /* don't put binfmt in mmput, we haven't got module yet */
998 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1000 struct mm_struct *mm, *oldmm;
1003 tsk->min_flt = tsk->maj_flt = 0;
1004 tsk->nvcsw = tsk->nivcsw = 0;
1005 #ifdef CONFIG_DETECT_HUNG_TASK
1006 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1010 tsk->active_mm = NULL;
1013 * Are we cloning a kernel thread?
1015 * We need to steal a active VM for that..
1017 oldmm = current->mm;
1021 /* initialize the new vmacache entries */
1022 vmacache_flush(tsk);
1024 if (clone_flags & CLONE_VM) {
1025 atomic_inc(&oldmm->mm_users);
1037 tsk->active_mm = mm;
1044 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1046 struct fs_struct *fs = current->fs;
1047 if (clone_flags & CLONE_FS) {
1048 /* tsk->fs is already what we want */
1049 spin_lock(&fs->lock);
1051 spin_unlock(&fs->lock);
1055 spin_unlock(&fs->lock);
1058 tsk->fs = copy_fs_struct(fs);
1064 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1066 struct files_struct *oldf, *newf;
1070 * A background process may not have any files ...
1072 oldf = current->files;
1076 if (clone_flags & CLONE_FILES) {
1077 atomic_inc(&oldf->count);
1081 newf = dup_fd(oldf, &error);
1091 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1094 struct io_context *ioc = current->io_context;
1095 struct io_context *new_ioc;
1100 * Share io context with parent, if CLONE_IO is set
1102 if (clone_flags & CLONE_IO) {
1104 tsk->io_context = ioc;
1105 } else if (ioprio_valid(ioc->ioprio)) {
1106 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1107 if (unlikely(!new_ioc))
1110 new_ioc->ioprio = ioc->ioprio;
1111 put_io_context(new_ioc);
1117 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1119 struct sighand_struct *sig;
1121 if (clone_flags & CLONE_SIGHAND) {
1122 atomic_inc(¤t->sighand->count);
1125 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1126 rcu_assign_pointer(tsk->sighand, sig);
1130 atomic_set(&sig->count, 1);
1131 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1135 void __cleanup_sighand(struct sighand_struct *sighand)
1137 if (atomic_dec_and_test(&sighand->count)) {
1138 signalfd_cleanup(sighand);
1140 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1141 * without an RCU grace period, see __lock_task_sighand().
1143 kmem_cache_free(sighand_cachep, sighand);
1148 * Initialize POSIX timer handling for a thread group.
1150 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1152 unsigned long cpu_limit;
1154 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1155 if (cpu_limit != RLIM_INFINITY) {
1156 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1157 sig->cputimer.running = true;
1160 /* The timer lists. */
1161 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1162 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1163 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1166 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1168 struct signal_struct *sig;
1170 if (clone_flags & CLONE_THREAD)
1173 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1178 sig->nr_threads = 1;
1179 atomic_set(&sig->live, 1);
1180 atomic_set(&sig->sigcnt, 1);
1182 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1183 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1184 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1186 init_waitqueue_head(&sig->wait_chldexit);
1187 sig->curr_target = tsk;
1188 init_sigpending(&sig->shared_pending);
1189 INIT_LIST_HEAD(&sig->posix_timers);
1190 seqlock_init(&sig->stats_lock);
1191 prev_cputime_init(&sig->prev_cputime);
1193 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1194 sig->real_timer.function = it_real_fn;
1196 task_lock(current->group_leader);
1197 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1198 task_unlock(current->group_leader);
1200 posix_cpu_timers_init_group(sig);
1202 tty_audit_fork(sig);
1203 sched_autogroup_fork(sig);
1205 sig->oom_score_adj = current->signal->oom_score_adj;
1206 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1208 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1209 current->signal->is_child_subreaper;
1211 mutex_init(&sig->cred_guard_mutex);
1216 static void copy_seccomp(struct task_struct *p)
1218 #ifdef CONFIG_SECCOMP
1220 * Must be called with sighand->lock held, which is common to
1221 * all threads in the group. Holding cred_guard_mutex is not
1222 * needed because this new task is not yet running and cannot
1225 assert_spin_locked(¤t->sighand->siglock);
1227 /* Ref-count the new filter user, and assign it. */
1228 get_seccomp_filter(current);
1229 p->seccomp = current->seccomp;
1232 * Explicitly enable no_new_privs here in case it got set
1233 * between the task_struct being duplicated and holding the
1234 * sighand lock. The seccomp state and nnp must be in sync.
1236 if (task_no_new_privs(current))
1237 task_set_no_new_privs(p);
1240 * If the parent gained a seccomp mode after copying thread
1241 * flags and between before we held the sighand lock, we have
1242 * to manually enable the seccomp thread flag here.
1244 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1245 set_tsk_thread_flag(p, TIF_SECCOMP);
1249 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1251 current->clear_child_tid = tidptr;
1253 return task_pid_vnr(current);
1256 static void rt_mutex_init_task(struct task_struct *p)
1258 raw_spin_lock_init(&p->pi_lock);
1259 #ifdef CONFIG_RT_MUTEXES
1260 p->pi_waiters = RB_ROOT;
1261 p->pi_waiters_leftmost = NULL;
1262 p->pi_blocked_on = NULL;
1267 * Initialize POSIX timer handling for a single task.
1269 static void posix_cpu_timers_init(struct task_struct *tsk)
1271 #ifdef CONFIG_PREEMPT_RT_BASE
1272 tsk->posix_timer_list = NULL;
1274 tsk->cputime_expires.prof_exp = 0;
1275 tsk->cputime_expires.virt_exp = 0;
1276 tsk->cputime_expires.sched_exp = 0;
1277 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1278 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1279 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1283 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1285 task->pids[type].pid = pid;
1289 * This creates a new process as a copy of the old one,
1290 * but does not actually start it yet.
1292 * It copies the registers, and all the appropriate
1293 * parts of the process environment (as per the clone
1294 * flags). The actual kick-off is left to the caller.
1296 static struct task_struct *copy_process(unsigned long clone_flags,
1297 unsigned long stack_start,
1298 unsigned long stack_size,
1299 int __user *child_tidptr,
1305 struct task_struct *p;
1306 void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
1308 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1309 return ERR_PTR(-EINVAL);
1311 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1312 return ERR_PTR(-EINVAL);
1315 * Thread groups must share signals as well, and detached threads
1316 * can only be started up within the thread group.
1318 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1319 return ERR_PTR(-EINVAL);
1322 * Shared signal handlers imply shared VM. By way of the above,
1323 * thread groups also imply shared VM. Blocking this case allows
1324 * for various simplifications in other code.
1326 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1327 return ERR_PTR(-EINVAL);
1330 * Siblings of global init remain as zombies on exit since they are
1331 * not reaped by their parent (swapper). To solve this and to avoid
1332 * multi-rooted process trees, prevent global and container-inits
1333 * from creating siblings.
1335 if ((clone_flags & CLONE_PARENT) &&
1336 current->signal->flags & SIGNAL_UNKILLABLE)
1337 return ERR_PTR(-EINVAL);
1340 * If the new process will be in a different pid or user namespace
1341 * do not allow it to share a thread group with the forking task.
1343 if (clone_flags & CLONE_THREAD) {
1344 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1345 (task_active_pid_ns(current) !=
1346 current->nsproxy->pid_ns_for_children))
1347 return ERR_PTR(-EINVAL);
1350 retval = security_task_create(clone_flags);
1355 p = dup_task_struct(current);
1359 ftrace_graph_init_task(p);
1361 rt_mutex_init_task(p);
1363 #ifdef CONFIG_PROVE_LOCKING
1364 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1365 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1368 if (atomic_read(&p->real_cred->user->processes) >=
1369 task_rlimit(p, RLIMIT_NPROC)) {
1370 if (p->real_cred->user != INIT_USER &&
1371 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1374 current->flags &= ~PF_NPROC_EXCEEDED;
1376 retval = copy_creds(p, clone_flags);
1381 * If multiple threads are within copy_process(), then this check
1382 * triggers too late. This doesn't hurt, the check is only there
1383 * to stop root fork bombs.
1386 if (nr_threads >= max_threads)
1387 goto bad_fork_cleanup_count;
1389 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1390 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1391 p->flags |= PF_FORKNOEXEC;
1392 INIT_LIST_HEAD(&p->children);
1393 INIT_LIST_HEAD(&p->sibling);
1394 rcu_copy_process(p);
1395 p->vfork_done = NULL;
1396 spin_lock_init(&p->alloc_lock);
1398 init_sigpending(&p->pending);
1399 p->sigqueue_cache = NULL;
1401 p->utime = p->stime = p->gtime = 0;
1402 p->utimescaled = p->stimescaled = 0;
1403 prev_cputime_init(&p->prev_cputime);
1405 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1406 seqcount_init(&p->vtime_seqcount);
1408 p->vtime_snap_whence = VTIME_INACTIVE;
1411 #if defined(SPLIT_RSS_COUNTING)
1412 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1415 p->default_timer_slack_ns = current->timer_slack_ns;
1417 task_io_accounting_init(&p->ioac);
1418 acct_clear_integrals(p);
1420 posix_cpu_timers_init(p);
1422 p->start_time = ktime_get_ns();
1423 p->real_start_time = ktime_get_boot_ns();
1424 p->io_context = NULL;
1425 p->audit_context = NULL;
1428 p->mempolicy = mpol_dup(p->mempolicy);
1429 if (IS_ERR(p->mempolicy)) {
1430 retval = PTR_ERR(p->mempolicy);
1431 p->mempolicy = NULL;
1432 goto bad_fork_cleanup_threadgroup_lock;
1435 #ifdef CONFIG_CPUSETS
1436 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1437 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1438 seqcount_init(&p->mems_allowed_seq);
1440 #ifdef CONFIG_TRACE_IRQFLAGS
1442 p->hardirqs_enabled = 0;
1443 p->hardirq_enable_ip = 0;
1444 p->hardirq_enable_event = 0;
1445 p->hardirq_disable_ip = _THIS_IP_;
1446 p->hardirq_disable_event = 0;
1447 p->softirqs_enabled = 1;
1448 p->softirq_enable_ip = _THIS_IP_;
1449 p->softirq_enable_event = 0;
1450 p->softirq_disable_ip = 0;
1451 p->softirq_disable_event = 0;
1452 p->hardirq_context = 0;
1453 p->softirq_context = 0;
1456 p->pagefault_disabled = 0;
1458 #ifdef CONFIG_LOCKDEP
1459 p->lockdep_depth = 0; /* no locks held yet */
1460 p->curr_chain_key = 0;
1461 p->lockdep_recursion = 0;
1464 #ifdef CONFIG_DEBUG_MUTEXES
1465 p->blocked_on = NULL; /* not blocked yet */
1467 #ifdef CONFIG_BCACHE
1468 p->sequential_io = 0;
1469 p->sequential_io_avg = 0;
1472 /* Perform scheduler related setup. Assign this task to a CPU. */
1473 retval = sched_fork(clone_flags, p);
1475 goto bad_fork_cleanup_policy;
1477 retval = perf_event_init_task(p);
1479 goto bad_fork_cleanup_policy;
1480 retval = audit_alloc(p);
1482 goto bad_fork_cleanup_perf;
1483 /* copy all the process information */
1485 retval = copy_semundo(clone_flags, p);
1487 goto bad_fork_cleanup_audit;
1488 retval = copy_files(clone_flags, p);
1490 goto bad_fork_cleanup_semundo;
1491 retval = copy_fs(clone_flags, p);
1493 goto bad_fork_cleanup_files;
1494 retval = copy_sighand(clone_flags, p);
1496 goto bad_fork_cleanup_fs;
1497 retval = copy_signal(clone_flags, p);
1499 goto bad_fork_cleanup_sighand;
1500 retval = copy_mm(clone_flags, p);
1502 goto bad_fork_cleanup_signal;
1503 retval = copy_namespaces(clone_flags, p);
1505 goto bad_fork_cleanup_mm;
1506 retval = copy_io(clone_flags, p);
1508 goto bad_fork_cleanup_namespaces;
1509 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1511 goto bad_fork_cleanup_io;
1513 if (pid != &init_struct_pid) {
1514 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1516 retval = PTR_ERR(pid);
1517 goto bad_fork_cleanup_io;
1521 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1523 * Clear TID on mm_release()?
1525 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1530 p->robust_list = NULL;
1531 #ifdef CONFIG_COMPAT
1532 p->compat_robust_list = NULL;
1534 INIT_LIST_HEAD(&p->pi_state_list);
1535 p->pi_state_cache = NULL;
1538 * sigaltstack should be cleared when sharing the same VM
1540 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1541 p->sas_ss_sp = p->sas_ss_size = 0;
1544 * Syscall tracing and stepping should be turned off in the
1545 * child regardless of CLONE_PTRACE.
1547 user_disable_single_step(p);
1548 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1549 #ifdef TIF_SYSCALL_EMU
1550 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1552 clear_all_latency_tracing(p);
1554 /* ok, now we should be set up.. */
1555 p->pid = pid_nr(pid);
1556 if (clone_flags & CLONE_THREAD) {
1557 p->exit_signal = -1;
1558 p->group_leader = current->group_leader;
1559 p->tgid = current->tgid;
1561 if (clone_flags & CLONE_PARENT)
1562 p->exit_signal = current->group_leader->exit_signal;
1564 p->exit_signal = (clone_flags & CSIGNAL);
1565 p->group_leader = p;
1570 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1571 p->dirty_paused_when = 0;
1573 p->pdeath_signal = 0;
1574 INIT_LIST_HEAD(&p->thread_group);
1575 p->task_works = NULL;
1577 threadgroup_change_begin(current);
1579 * Ensure that the cgroup subsystem policies allow the new process to be
1580 * forked. It should be noted the the new process's css_set can be changed
1581 * between here and cgroup_post_fork() if an organisation operation is in
1584 retval = cgroup_can_fork(p, cgrp_ss_priv);
1586 goto bad_fork_free_pid;
1589 * Make it visible to the rest of the system, but dont wake it up yet.
1590 * Need tasklist lock for parent etc handling!
1592 write_lock_irq(&tasklist_lock);
1594 /* CLONE_PARENT re-uses the old parent */
1595 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1596 p->real_parent = current->real_parent;
1597 p->parent_exec_id = current->parent_exec_id;
1599 p->real_parent = current;
1600 p->parent_exec_id = current->self_exec_id;
1603 spin_lock(¤t->sighand->siglock);
1606 * Copy seccomp details explicitly here, in case they were changed
1607 * before holding sighand lock.
1612 * Process group and session signals need to be delivered to just the
1613 * parent before the fork or both the parent and the child after the
1614 * fork. Restart if a signal comes in before we add the new process to
1615 * it's process group.
1616 * A fatal signal pending means that current will exit, so the new
1617 * thread can't slip out of an OOM kill (or normal SIGKILL).
1619 recalc_sigpending();
1620 if (signal_pending(current)) {
1621 spin_unlock(¤t->sighand->siglock);
1622 write_unlock_irq(&tasklist_lock);
1623 retval = -ERESTARTNOINTR;
1624 goto bad_fork_cancel_cgroup;
1627 if (likely(p->pid)) {
1628 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1630 init_task_pid(p, PIDTYPE_PID, pid);
1631 if (thread_group_leader(p)) {
1632 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1633 init_task_pid(p, PIDTYPE_SID, task_session(current));
1635 if (is_child_reaper(pid)) {
1636 ns_of_pid(pid)->child_reaper = p;
1637 p->signal->flags |= SIGNAL_UNKILLABLE;
1640 p->signal->leader_pid = pid;
1641 p->signal->tty = tty_kref_get(current->signal->tty);
1642 list_add_tail(&p->sibling, &p->real_parent->children);
1643 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1644 attach_pid(p, PIDTYPE_PGID);
1645 attach_pid(p, PIDTYPE_SID);
1646 __this_cpu_inc(process_counts);
1648 current->signal->nr_threads++;
1649 atomic_inc(¤t->signal->live);
1650 atomic_inc(¤t->signal->sigcnt);
1651 list_add_tail_rcu(&p->thread_group,
1652 &p->group_leader->thread_group);
1653 list_add_tail_rcu(&p->thread_node,
1654 &p->signal->thread_head);
1656 attach_pid(p, PIDTYPE_PID);
1661 spin_unlock(¤t->sighand->siglock);
1662 syscall_tracepoint_update(p);
1663 write_unlock_irq(&tasklist_lock);
1665 proc_fork_connector(p);
1666 cgroup_post_fork(p, cgrp_ss_priv);
1667 threadgroup_change_end(current);
1670 trace_task_newtask(p, clone_flags);
1671 uprobe_copy_process(p, clone_flags);
1675 bad_fork_cancel_cgroup:
1676 cgroup_cancel_fork(p, cgrp_ss_priv);
1678 threadgroup_change_end(current);
1679 if (pid != &init_struct_pid)
1681 bad_fork_cleanup_io:
1684 bad_fork_cleanup_namespaces:
1685 exit_task_namespaces(p);
1686 bad_fork_cleanup_mm:
1689 bad_fork_cleanup_signal:
1690 if (!(clone_flags & CLONE_THREAD))
1691 free_signal_struct(p->signal);
1692 bad_fork_cleanup_sighand:
1693 __cleanup_sighand(p->sighand);
1694 bad_fork_cleanup_fs:
1695 exit_fs(p); /* blocking */
1696 bad_fork_cleanup_files:
1697 exit_files(p); /* blocking */
1698 bad_fork_cleanup_semundo:
1700 bad_fork_cleanup_audit:
1702 bad_fork_cleanup_perf:
1703 perf_event_free_task(p);
1704 bad_fork_cleanup_policy:
1706 mpol_put(p->mempolicy);
1707 bad_fork_cleanup_threadgroup_lock:
1709 delayacct_tsk_free(p);
1710 bad_fork_cleanup_count:
1711 atomic_dec(&p->cred->user->processes);
1716 return ERR_PTR(retval);
1719 static inline void init_idle_pids(struct pid_link *links)
1723 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1724 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1725 links[type].pid = &init_struct_pid;
1729 struct task_struct *fork_idle(int cpu)
1731 struct task_struct *task;
1732 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1733 if (!IS_ERR(task)) {
1734 init_idle_pids(task->pids);
1735 init_idle(task, cpu);
1742 * Ok, this is the main fork-routine.
1744 * It copies the process, and if successful kick-starts
1745 * it and waits for it to finish using the VM if required.
1747 long _do_fork(unsigned long clone_flags,
1748 unsigned long stack_start,
1749 unsigned long stack_size,
1750 int __user *parent_tidptr,
1751 int __user *child_tidptr,
1754 struct task_struct *p;
1759 * Determine whether and which event to report to ptracer. When
1760 * called from kernel_thread or CLONE_UNTRACED is explicitly
1761 * requested, no event is reported; otherwise, report if the event
1762 * for the type of forking is enabled.
1764 if (!(clone_flags & CLONE_UNTRACED)) {
1765 if (clone_flags & CLONE_VFORK)
1766 trace = PTRACE_EVENT_VFORK;
1767 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1768 trace = PTRACE_EVENT_CLONE;
1770 trace = PTRACE_EVENT_FORK;
1772 if (likely(!ptrace_event_enabled(current, trace)))
1776 p = copy_process(clone_flags, stack_start, stack_size,
1777 child_tidptr, NULL, trace, tls);
1779 * Do this prior waking up the new thread - the thread pointer
1780 * might get invalid after that point, if the thread exits quickly.
1783 struct completion vfork;
1786 trace_sched_process_fork(current, p);
1788 pid = get_task_pid(p, PIDTYPE_PID);
1791 if (clone_flags & CLONE_PARENT_SETTID)
1792 put_user(nr, parent_tidptr);
1794 if (clone_flags & CLONE_VFORK) {
1795 p->vfork_done = &vfork;
1796 init_completion(&vfork);
1800 wake_up_new_task(p);
1802 /* forking complete and child started to run, tell ptracer */
1803 if (unlikely(trace))
1804 ptrace_event_pid(trace, pid);
1806 if (clone_flags & CLONE_VFORK) {
1807 if (!wait_for_vfork_done(p, &vfork))
1808 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1818 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1819 /* For compatibility with architectures that call do_fork directly rather than
1820 * using the syscall entry points below. */
1821 long do_fork(unsigned long clone_flags,
1822 unsigned long stack_start,
1823 unsigned long stack_size,
1824 int __user *parent_tidptr,
1825 int __user *child_tidptr)
1827 return _do_fork(clone_flags, stack_start, stack_size,
1828 parent_tidptr, child_tidptr, 0);
1833 * Create a kernel thread.
1835 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1837 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1838 (unsigned long)arg, NULL, NULL, 0);
1841 #ifdef __ARCH_WANT_SYS_FORK
1842 SYSCALL_DEFINE0(fork)
1845 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1847 /* can not support in nommu mode */
1853 #ifdef __ARCH_WANT_SYS_VFORK
1854 SYSCALL_DEFINE0(vfork)
1856 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1861 #ifdef __ARCH_WANT_SYS_CLONE
1862 #ifdef CONFIG_CLONE_BACKWARDS
1863 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1864 int __user *, parent_tidptr,
1866 int __user *, child_tidptr)
1867 #elif defined(CONFIG_CLONE_BACKWARDS2)
1868 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1869 int __user *, parent_tidptr,
1870 int __user *, child_tidptr,
1872 #elif defined(CONFIG_CLONE_BACKWARDS3)
1873 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1875 int __user *, parent_tidptr,
1876 int __user *, child_tidptr,
1879 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1880 int __user *, parent_tidptr,
1881 int __user *, child_tidptr,
1885 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1889 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1890 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1893 static void sighand_ctor(void *data)
1895 struct sighand_struct *sighand = data;
1897 spin_lock_init(&sighand->siglock);
1898 init_waitqueue_head(&sighand->signalfd_wqh);
1901 void __init proc_caches_init(void)
1903 sighand_cachep = kmem_cache_create("sighand_cache",
1904 sizeof(struct sighand_struct), 0,
1905 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1906 SLAB_NOTRACK, sighand_ctor);
1907 signal_cachep = kmem_cache_create("signal_cache",
1908 sizeof(struct signal_struct), 0,
1909 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1910 files_cachep = kmem_cache_create("files_cache",
1911 sizeof(struct files_struct), 0,
1912 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1913 fs_cachep = kmem_cache_create("fs_cache",
1914 sizeof(struct fs_struct), 0,
1915 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1917 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1918 * whole struct cpumask for the OFFSTACK case. We could change
1919 * this to *only* allocate as much of it as required by the
1920 * maximum number of CPU's we can ever have. The cpumask_allocation
1921 * is at the end of the structure, exactly for that reason.
1923 mm_cachep = kmem_cache_create("mm_struct",
1924 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1925 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1926 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1928 nsproxy_cache_init();
1932 * Check constraints on flags passed to the unshare system call.
1934 static int check_unshare_flags(unsigned long unshare_flags)
1936 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1937 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1938 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1939 CLONE_NEWUSER|CLONE_NEWPID))
1942 * Not implemented, but pretend it works if there is nothing
1943 * to unshare. Note that unsharing the address space or the
1944 * signal handlers also need to unshare the signal queues (aka
1947 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1948 if (!thread_group_empty(current))
1951 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1952 if (atomic_read(¤t->sighand->count) > 1)
1955 if (unshare_flags & CLONE_VM) {
1956 if (!current_is_single_threaded())
1964 * Unshare the filesystem structure if it is being shared
1966 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1968 struct fs_struct *fs = current->fs;
1970 if (!(unshare_flags & CLONE_FS) || !fs)
1973 /* don't need lock here; in the worst case we'll do useless copy */
1977 *new_fsp = copy_fs_struct(fs);
1985 * Unshare file descriptor table if it is being shared
1987 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1989 struct files_struct *fd = current->files;
1992 if ((unshare_flags & CLONE_FILES) &&
1993 (fd && atomic_read(&fd->count) > 1)) {
1994 *new_fdp = dup_fd(fd, &error);
2003 * unshare allows a process to 'unshare' part of the process
2004 * context which was originally shared using clone. copy_*
2005 * functions used by do_fork() cannot be used here directly
2006 * because they modify an inactive task_struct that is being
2007 * constructed. Here we are modifying the current, active,
2010 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2012 struct fs_struct *fs, *new_fs = NULL;
2013 struct files_struct *fd, *new_fd = NULL;
2014 struct cred *new_cred = NULL;
2015 struct nsproxy *new_nsproxy = NULL;
2020 * If unsharing a user namespace must also unshare the thread group
2021 * and unshare the filesystem root and working directories.
2023 if (unshare_flags & CLONE_NEWUSER)
2024 unshare_flags |= CLONE_THREAD | CLONE_FS;
2026 * If unsharing vm, must also unshare signal handlers.
2028 if (unshare_flags & CLONE_VM)
2029 unshare_flags |= CLONE_SIGHAND;
2031 * If unsharing a signal handlers, must also unshare the signal queues.
2033 if (unshare_flags & CLONE_SIGHAND)
2034 unshare_flags |= CLONE_THREAD;
2036 * If unsharing namespace, must also unshare filesystem information.
2038 if (unshare_flags & CLONE_NEWNS)
2039 unshare_flags |= CLONE_FS;
2041 err = check_unshare_flags(unshare_flags);
2043 goto bad_unshare_out;
2045 * CLONE_NEWIPC must also detach from the undolist: after switching
2046 * to a new ipc namespace, the semaphore arrays from the old
2047 * namespace are unreachable.
2049 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2051 err = unshare_fs(unshare_flags, &new_fs);
2053 goto bad_unshare_out;
2054 err = unshare_fd(unshare_flags, &new_fd);
2056 goto bad_unshare_cleanup_fs;
2057 err = unshare_userns(unshare_flags, &new_cred);
2059 goto bad_unshare_cleanup_fd;
2060 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2063 goto bad_unshare_cleanup_cred;
2065 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2068 * CLONE_SYSVSEM is equivalent to sys_exit().
2072 if (unshare_flags & CLONE_NEWIPC) {
2073 /* Orphan segments in old ns (see sem above). */
2075 shm_init_task(current);
2079 switch_task_namespaces(current, new_nsproxy);
2085 spin_lock(&fs->lock);
2086 current->fs = new_fs;
2091 spin_unlock(&fs->lock);
2095 fd = current->files;
2096 current->files = new_fd;
2100 task_unlock(current);
2103 /* Install the new user namespace */
2104 commit_creds(new_cred);
2109 bad_unshare_cleanup_cred:
2112 bad_unshare_cleanup_fd:
2114 put_files_struct(new_fd);
2116 bad_unshare_cleanup_fs:
2118 free_fs_struct(new_fs);
2125 * Helper to unshare the files of the current task.
2126 * We don't want to expose copy_files internals to
2127 * the exec layer of the kernel.
2130 int unshare_files(struct files_struct **displaced)
2132 struct task_struct *task = current;
2133 struct files_struct *copy = NULL;
2136 error = unshare_fd(CLONE_FILES, ©);
2137 if (error || !copy) {
2141 *displaced = task->files;
2148 int sysctl_max_threads(struct ctl_table *table, int write,
2149 void __user *buffer, size_t *lenp, loff_t *ppos)
2153 int threads = max_threads;
2154 int min = MIN_THREADS;
2155 int max = MAX_THREADS;
2162 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2166 set_max_threads(threads);