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 __cacheline_aligned 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,
168 memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
169 1 << THREAD_SIZE_ORDER);
171 return page ? page_address(page) : NULL;
174 static inline void free_thread_info(struct thread_info *ti)
176 struct page *page = virt_to_page(ti);
178 memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
179 -(1 << THREAD_SIZE_ORDER));
180 __free_kmem_pages(page, THREAD_SIZE_ORDER);
183 static struct kmem_cache *thread_info_cache;
185 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
188 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
191 static void free_thread_info(struct thread_info *ti)
193 kmem_cache_free(thread_info_cache, ti);
196 void thread_info_cache_init(void)
198 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
199 THREAD_SIZE, 0, NULL);
200 BUG_ON(thread_info_cache == NULL);
205 /* SLAB cache for signal_struct structures (tsk->signal) */
206 static struct kmem_cache *signal_cachep;
208 /* SLAB cache for sighand_struct structures (tsk->sighand) */
209 struct kmem_cache *sighand_cachep;
211 /* SLAB cache for files_struct structures (tsk->files) */
212 struct kmem_cache *files_cachep;
214 /* SLAB cache for fs_struct structures (tsk->fs) */
215 struct kmem_cache *fs_cachep;
217 /* SLAB cache for vm_area_struct structures */
218 struct kmem_cache *vm_area_cachep;
220 /* SLAB cache for mm_struct structures (tsk->mm) */
221 static struct kmem_cache *mm_cachep;
223 static void account_kernel_stack(struct thread_info *ti, int account)
225 struct zone *zone = page_zone(virt_to_page(ti));
227 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
230 void free_task(struct task_struct *tsk)
232 account_kernel_stack(tsk->stack, -1);
233 arch_release_thread_info(tsk->stack);
234 free_thread_info(tsk->stack);
235 rt_mutex_debug_task_free(tsk);
236 ftrace_graph_exit_task(tsk);
237 put_seccomp_filter(tsk);
238 arch_release_task_struct(tsk);
239 free_task_struct(tsk);
241 EXPORT_SYMBOL(free_task);
243 static inline void free_signal_struct(struct signal_struct *sig)
245 taskstats_tgid_free(sig);
246 sched_autogroup_exit(sig);
247 kmem_cache_free(signal_cachep, sig);
250 static inline void put_signal_struct(struct signal_struct *sig)
252 if (atomic_dec_and_test(&sig->sigcnt))
253 free_signal_struct(sig);
256 void __put_task_struct(struct task_struct *tsk)
258 WARN_ON(!tsk->exit_state);
259 WARN_ON(atomic_read(&tsk->usage));
260 WARN_ON(tsk == current);
264 security_task_free(tsk);
266 delayacct_tsk_free(tsk);
267 put_signal_struct(tsk->signal);
269 if (!profile_handoff_task(tsk))
272 EXPORT_SYMBOL_GPL(__put_task_struct);
274 void __init __weak arch_task_cache_init(void) { }
279 static void set_max_threads(unsigned int max_threads_suggested)
284 * The number of threads shall be limited such that the thread
285 * structures may only consume a small part of the available memory.
287 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
288 threads = MAX_THREADS;
290 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
291 (u64) THREAD_SIZE * 8UL);
293 if (threads > max_threads_suggested)
294 threads = max_threads_suggested;
296 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
299 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
300 /* Initialized by the architecture: */
301 int arch_task_struct_size __read_mostly;
304 void __init fork_init(void)
306 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
307 #ifndef ARCH_MIN_TASKALIGN
308 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
310 /* create a slab on which task_structs can be allocated */
311 task_struct_cachep = kmem_cache_create("task_struct",
312 arch_task_struct_size, ARCH_MIN_TASKALIGN,
313 SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
316 /* do the arch specific task caches init */
317 arch_task_cache_init();
319 set_max_threads(MAX_THREADS);
321 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
322 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
323 init_task.signal->rlim[RLIMIT_SIGPENDING] =
324 init_task.signal->rlim[RLIMIT_NPROC];
327 int __weak arch_dup_task_struct(struct task_struct *dst,
328 struct task_struct *src)
334 void set_task_stack_end_magic(struct task_struct *tsk)
336 unsigned long *stackend;
338 stackend = end_of_stack(tsk);
339 *stackend = STACK_END_MAGIC; /* for overflow detection */
342 static struct task_struct *dup_task_struct(struct task_struct *orig)
344 struct task_struct *tsk;
345 struct thread_info *ti;
346 int node = tsk_fork_get_node(orig);
349 tsk = alloc_task_struct_node(node);
353 ti = alloc_thread_info_node(tsk, node);
357 err = arch_dup_task_struct(tsk, orig);
362 #ifdef CONFIG_SECCOMP
364 * We must handle setting up seccomp filters once we're under
365 * the sighand lock in case orig has changed between now and
366 * then. Until then, filter must be NULL to avoid messing up
367 * the usage counts on the error path calling free_task.
369 tsk->seccomp.filter = NULL;
372 setup_thread_stack(tsk, orig);
373 clear_user_return_notifier(tsk);
374 clear_tsk_need_resched(tsk);
375 set_task_stack_end_magic(tsk);
377 #ifdef CONFIG_CC_STACKPROTECTOR
378 tsk->stack_canary = get_random_int();
382 * One for us, one for whoever does the "release_task()" (usually
385 atomic_set(&tsk->usage, 2);
386 #ifdef CONFIG_BLK_DEV_IO_TRACE
389 tsk->splice_pipe = NULL;
390 tsk->task_frag.page = NULL;
391 tsk->wake_q.next = NULL;
393 account_kernel_stack(ti, 1);
398 free_thread_info(ti);
400 free_task_struct(tsk);
405 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
407 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
408 struct rb_node **rb_link, *rb_parent;
410 unsigned long charge;
412 uprobe_start_dup_mmap();
413 down_write(&oldmm->mmap_sem);
414 flush_cache_dup_mm(oldmm);
415 uprobe_dup_mmap(oldmm, mm);
417 * Not linked in yet - no deadlock potential:
419 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
421 /* No ordering required: file already has been exposed. */
422 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
424 mm->total_vm = oldmm->total_vm;
425 mm->data_vm = oldmm->data_vm;
426 mm->exec_vm = oldmm->exec_vm;
427 mm->stack_vm = oldmm->stack_vm;
429 rb_link = &mm->mm_rb.rb_node;
432 retval = ksm_fork(mm, oldmm);
435 retval = khugepaged_fork(mm, oldmm);
440 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
443 if (mpnt->vm_flags & VM_DONTCOPY) {
444 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
448 if (mpnt->vm_flags & VM_ACCOUNT) {
449 unsigned long len = vma_pages(mpnt);
451 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
455 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
459 INIT_LIST_HEAD(&tmp->anon_vma_chain);
460 retval = vma_dup_policy(mpnt, tmp);
462 goto fail_nomem_policy;
464 if (anon_vma_fork(tmp, mpnt))
465 goto fail_nomem_anon_vma_fork;
467 ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
468 tmp->vm_next = tmp->vm_prev = NULL;
469 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
472 struct inode *inode = file_inode(file);
473 struct address_space *mapping = file->f_mapping;
476 if (tmp->vm_flags & VM_DENYWRITE)
477 atomic_dec(&inode->i_writecount);
478 i_mmap_lock_write(mapping);
479 if (tmp->vm_flags & VM_SHARED)
480 atomic_inc(&mapping->i_mmap_writable);
481 flush_dcache_mmap_lock(mapping);
482 /* insert tmp into the share list, just after mpnt */
483 vma_interval_tree_insert_after(tmp, mpnt,
485 flush_dcache_mmap_unlock(mapping);
486 i_mmap_unlock_write(mapping);
490 * Clear hugetlb-related page reserves for children. This only
491 * affects MAP_PRIVATE mappings. Faults generated by the child
492 * are not guaranteed to succeed, even if read-only
494 if (is_vm_hugetlb_page(tmp))
495 reset_vma_resv_huge_pages(tmp);
498 * Link in the new vma and copy the page table entries.
501 pprev = &tmp->vm_next;
505 __vma_link_rb(mm, tmp, rb_link, rb_parent);
506 rb_link = &tmp->vm_rb.rb_right;
507 rb_parent = &tmp->vm_rb;
510 retval = copy_page_range(mm, oldmm, mpnt);
512 if (tmp->vm_ops && tmp->vm_ops->open)
513 tmp->vm_ops->open(tmp);
518 /* a new mm has just been created */
519 arch_dup_mmap(oldmm, mm);
522 up_write(&mm->mmap_sem);
524 up_write(&oldmm->mmap_sem);
525 uprobe_end_dup_mmap();
527 fail_nomem_anon_vma_fork:
528 mpol_put(vma_policy(tmp));
530 kmem_cache_free(vm_area_cachep, tmp);
533 vm_unacct_memory(charge);
537 static inline int mm_alloc_pgd(struct mm_struct *mm)
539 mm->pgd = pgd_alloc(mm);
540 if (unlikely(!mm->pgd))
545 static inline void mm_free_pgd(struct mm_struct *mm)
547 pgd_free(mm, mm->pgd);
550 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
552 down_write(&oldmm->mmap_sem);
553 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
554 up_write(&oldmm->mmap_sem);
557 #define mm_alloc_pgd(mm) (0)
558 #define mm_free_pgd(mm)
559 #endif /* CONFIG_MMU */
561 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
563 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
564 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
566 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
568 static int __init coredump_filter_setup(char *s)
570 default_dump_filter =
571 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
572 MMF_DUMP_FILTER_MASK;
576 __setup("coredump_filter=", coredump_filter_setup);
578 #include <linux/init_task.h>
580 static void mm_init_aio(struct mm_struct *mm)
583 spin_lock_init(&mm->ioctx_lock);
584 mm->ioctx_table = NULL;
588 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
595 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
599 mm->vmacache_seqnum = 0;
600 atomic_set(&mm->mm_users, 1);
601 atomic_set(&mm->mm_count, 1);
602 init_rwsem(&mm->mmap_sem);
603 INIT_LIST_HEAD(&mm->mmlist);
604 mm->core_state = NULL;
605 atomic_long_set(&mm->nr_ptes, 0);
610 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
611 spin_lock_init(&mm->page_table_lock);
614 mm_init_owner(mm, p);
615 mmu_notifier_mm_init(mm);
616 clear_tlb_flush_pending(mm);
617 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
618 mm->pmd_huge_pte = NULL;
622 mm->flags = current->mm->flags & MMF_INIT_MASK;
623 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
625 mm->flags = default_dump_filter;
629 if (mm_alloc_pgd(mm))
632 if (init_new_context(p, mm))
644 static void check_mm(struct mm_struct *mm)
648 for (i = 0; i < NR_MM_COUNTERS; i++) {
649 long x = atomic_long_read(&mm->rss_stat.count[i]);
652 printk(KERN_ALERT "BUG: Bad rss-counter state "
653 "mm:%p idx:%d val:%ld\n", mm, i, x);
656 if (atomic_long_read(&mm->nr_ptes))
657 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
658 atomic_long_read(&mm->nr_ptes));
660 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
663 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
664 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
669 * Allocate and initialize an mm_struct.
671 struct mm_struct *mm_alloc(void)
673 struct mm_struct *mm;
679 memset(mm, 0, sizeof(*mm));
680 return mm_init(mm, current);
684 * Called when the last reference to the mm
685 * is dropped: either by a lazy thread or by
686 * mmput. Free the page directory and the mm.
688 void __mmdrop(struct mm_struct *mm)
690 BUG_ON(mm == &init_mm);
693 mmu_notifier_mm_destroy(mm);
697 EXPORT_SYMBOL_GPL(__mmdrop);
700 * Decrement the use count and release all resources for an mm.
702 void mmput(struct mm_struct *mm)
706 if (atomic_dec_and_test(&mm->mm_users)) {
707 uprobe_clear_state(mm);
710 khugepaged_exit(mm); /* must run before exit_mmap */
712 set_mm_exe_file(mm, NULL);
713 if (!list_empty(&mm->mmlist)) {
714 spin_lock(&mmlist_lock);
715 list_del(&mm->mmlist);
716 spin_unlock(&mmlist_lock);
719 module_put(mm->binfmt->module);
723 EXPORT_SYMBOL_GPL(mmput);
726 * set_mm_exe_file - change a reference to the mm's executable file
728 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
730 * Main users are mmput() and sys_execve(). Callers prevent concurrent
731 * invocations: in mmput() nobody alive left, in execve task is single
732 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
733 * mm->exe_file, but does so without using set_mm_exe_file() in order
734 * to do avoid the need for any locks.
736 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
738 struct file *old_exe_file;
741 * It is safe to dereference the exe_file without RCU as
742 * this function is only called if nobody else can access
743 * this mm -- see comment above for justification.
745 old_exe_file = rcu_dereference_raw(mm->exe_file);
748 get_file(new_exe_file);
749 rcu_assign_pointer(mm->exe_file, new_exe_file);
755 * get_mm_exe_file - acquire a reference to the mm's executable file
757 * Returns %NULL if mm has no associated executable file.
758 * User must release file via fput().
760 struct file *get_mm_exe_file(struct mm_struct *mm)
762 struct file *exe_file;
765 exe_file = rcu_dereference(mm->exe_file);
766 if (exe_file && !get_file_rcu(exe_file))
771 EXPORT_SYMBOL(get_mm_exe_file);
774 * get_task_mm - acquire a reference to the task's mm
776 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
777 * this kernel workthread has transiently adopted a user mm with use_mm,
778 * to do its AIO) is not set and if so returns a reference to it, after
779 * bumping up the use count. User must release the mm via mmput()
780 * after use. Typically used by /proc and ptrace.
782 struct mm_struct *get_task_mm(struct task_struct *task)
784 struct mm_struct *mm;
789 if (task->flags & PF_KTHREAD)
792 atomic_inc(&mm->mm_users);
797 EXPORT_SYMBOL_GPL(get_task_mm);
799 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
801 struct mm_struct *mm;
804 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
808 mm = get_task_mm(task);
809 if (mm && mm != current->mm &&
810 !ptrace_may_access(task, mode)) {
812 mm = ERR_PTR(-EACCES);
814 mutex_unlock(&task->signal->cred_guard_mutex);
819 static void complete_vfork_done(struct task_struct *tsk)
821 struct completion *vfork;
824 vfork = tsk->vfork_done;
826 tsk->vfork_done = NULL;
832 static int wait_for_vfork_done(struct task_struct *child,
833 struct completion *vfork)
837 freezer_do_not_count();
838 killed = wait_for_completion_killable(vfork);
843 child->vfork_done = NULL;
847 put_task_struct(child);
851 /* Please note the differences between mmput and mm_release.
852 * mmput is called whenever we stop holding onto a mm_struct,
853 * error success whatever.
855 * mm_release is called after a mm_struct has been removed
856 * from the current process.
858 * This difference is important for error handling, when we
859 * only half set up a mm_struct for a new process and need to restore
860 * the old one. Because we mmput the new mm_struct before
861 * restoring the old one. . .
862 * Eric Biederman 10 January 1998
864 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
866 /* Get rid of any futexes when releasing the mm */
868 if (unlikely(tsk->robust_list)) {
869 exit_robust_list(tsk);
870 tsk->robust_list = NULL;
873 if (unlikely(tsk->compat_robust_list)) {
874 compat_exit_robust_list(tsk);
875 tsk->compat_robust_list = NULL;
878 if (unlikely(!list_empty(&tsk->pi_state_list)))
879 exit_pi_state_list(tsk);
882 uprobe_free_utask(tsk);
884 /* Get rid of any cached register state */
885 deactivate_mm(tsk, mm);
888 * If we're exiting normally, clear a user-space tid field if
889 * requested. We leave this alone when dying by signal, to leave
890 * the value intact in a core dump, and to save the unnecessary
891 * trouble, say, a killed vfork parent shouldn't touch this mm.
892 * Userland only wants this done for a sys_exit.
894 if (tsk->clear_child_tid) {
895 if (!(tsk->flags & PF_SIGNALED) &&
896 atomic_read(&mm->mm_users) > 1) {
898 * We don't check the error code - if userspace has
899 * not set up a proper pointer then tough luck.
901 put_user(0, tsk->clear_child_tid);
902 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
905 tsk->clear_child_tid = NULL;
909 * All done, finally we can wake up parent and return this mm to him.
910 * Also kthread_stop() uses this completion for synchronization.
913 complete_vfork_done(tsk);
917 * Allocate a new mm structure and copy contents from the
918 * mm structure of the passed in task structure.
920 static struct mm_struct *dup_mm(struct task_struct *tsk)
922 struct mm_struct *mm, *oldmm = current->mm;
929 memcpy(mm, oldmm, sizeof(*mm));
931 if (!mm_init(mm, tsk))
934 err = dup_mmap(mm, oldmm);
938 mm->hiwater_rss = get_mm_rss(mm);
939 mm->hiwater_vm = mm->total_vm;
941 if (mm->binfmt && !try_module_get(mm->binfmt->module))
947 /* don't put binfmt in mmput, we haven't got module yet */
955 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
957 struct mm_struct *mm, *oldmm;
960 tsk->min_flt = tsk->maj_flt = 0;
961 tsk->nvcsw = tsk->nivcsw = 0;
962 #ifdef CONFIG_DETECT_HUNG_TASK
963 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
967 tsk->active_mm = NULL;
970 * Are we cloning a kernel thread?
972 * We need to steal a active VM for that..
978 /* initialize the new vmacache entries */
981 if (clone_flags & CLONE_VM) {
982 atomic_inc(&oldmm->mm_users);
1001 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1003 struct fs_struct *fs = current->fs;
1004 if (clone_flags & CLONE_FS) {
1005 /* tsk->fs is already what we want */
1006 spin_lock(&fs->lock);
1008 spin_unlock(&fs->lock);
1012 spin_unlock(&fs->lock);
1015 tsk->fs = copy_fs_struct(fs);
1021 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1023 struct files_struct *oldf, *newf;
1027 * A background process may not have any files ...
1029 oldf = current->files;
1033 if (clone_flags & CLONE_FILES) {
1034 atomic_inc(&oldf->count);
1038 newf = dup_fd(oldf, &error);
1048 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1051 struct io_context *ioc = current->io_context;
1052 struct io_context *new_ioc;
1057 * Share io context with parent, if CLONE_IO is set
1059 if (clone_flags & CLONE_IO) {
1061 tsk->io_context = ioc;
1062 } else if (ioprio_valid(ioc->ioprio)) {
1063 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1064 if (unlikely(!new_ioc))
1067 new_ioc->ioprio = ioc->ioprio;
1068 put_io_context(new_ioc);
1074 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1076 struct sighand_struct *sig;
1078 if (clone_flags & CLONE_SIGHAND) {
1079 atomic_inc(¤t->sighand->count);
1082 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1083 rcu_assign_pointer(tsk->sighand, sig);
1087 atomic_set(&sig->count, 1);
1088 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1092 void __cleanup_sighand(struct sighand_struct *sighand)
1094 if (atomic_dec_and_test(&sighand->count)) {
1095 signalfd_cleanup(sighand);
1097 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1098 * without an RCU grace period, see __lock_task_sighand().
1100 kmem_cache_free(sighand_cachep, sighand);
1105 * Initialize POSIX timer handling for a thread group.
1107 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1109 unsigned long cpu_limit;
1111 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1112 if (cpu_limit != RLIM_INFINITY) {
1113 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1114 sig->cputimer.running = true;
1117 /* The timer lists. */
1118 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1119 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1120 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1123 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1125 struct signal_struct *sig;
1127 if (clone_flags & CLONE_THREAD)
1130 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1135 sig->nr_threads = 1;
1136 atomic_set(&sig->live, 1);
1137 atomic_set(&sig->sigcnt, 1);
1139 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1140 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1141 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1143 init_waitqueue_head(&sig->wait_chldexit);
1144 sig->curr_target = tsk;
1145 init_sigpending(&sig->shared_pending);
1146 INIT_LIST_HEAD(&sig->posix_timers);
1147 seqlock_init(&sig->stats_lock);
1148 prev_cputime_init(&sig->prev_cputime);
1150 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1151 sig->real_timer.function = it_real_fn;
1153 task_lock(current->group_leader);
1154 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1155 task_unlock(current->group_leader);
1157 posix_cpu_timers_init_group(sig);
1159 tty_audit_fork(sig);
1160 sched_autogroup_fork(sig);
1162 sig->oom_score_adj = current->signal->oom_score_adj;
1163 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1165 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1166 current->signal->is_child_subreaper;
1168 mutex_init(&sig->cred_guard_mutex);
1173 static void copy_seccomp(struct task_struct *p)
1175 #ifdef CONFIG_SECCOMP
1177 * Must be called with sighand->lock held, which is common to
1178 * all threads in the group. Holding cred_guard_mutex is not
1179 * needed because this new task is not yet running and cannot
1182 assert_spin_locked(¤t->sighand->siglock);
1184 /* Ref-count the new filter user, and assign it. */
1185 get_seccomp_filter(current);
1186 p->seccomp = current->seccomp;
1189 * Explicitly enable no_new_privs here in case it got set
1190 * between the task_struct being duplicated and holding the
1191 * sighand lock. The seccomp state and nnp must be in sync.
1193 if (task_no_new_privs(current))
1194 task_set_no_new_privs(p);
1197 * If the parent gained a seccomp mode after copying thread
1198 * flags and between before we held the sighand lock, we have
1199 * to manually enable the seccomp thread flag here.
1201 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1202 set_tsk_thread_flag(p, TIF_SECCOMP);
1206 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1208 current->clear_child_tid = tidptr;
1210 return task_pid_vnr(current);
1213 static void rt_mutex_init_task(struct task_struct *p)
1215 raw_spin_lock_init(&p->pi_lock);
1216 #ifdef CONFIG_RT_MUTEXES
1217 p->pi_waiters = RB_ROOT;
1218 p->pi_waiters_leftmost = NULL;
1219 p->pi_blocked_on = NULL;
1224 * Initialize POSIX timer handling for a single task.
1226 static void posix_cpu_timers_init(struct task_struct *tsk)
1228 tsk->cputime_expires.prof_exp = 0;
1229 tsk->cputime_expires.virt_exp = 0;
1230 tsk->cputime_expires.sched_exp = 0;
1231 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1232 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1233 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1237 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1239 task->pids[type].pid = pid;
1243 * This creates a new process as a copy of the old one,
1244 * but does not actually start it yet.
1246 * It copies the registers, and all the appropriate
1247 * parts of the process environment (as per the clone
1248 * flags). The actual kick-off is left to the caller.
1250 static struct task_struct *copy_process(unsigned long clone_flags,
1251 unsigned long stack_start,
1252 unsigned long stack_size,
1253 int __user *child_tidptr,
1259 struct task_struct *p;
1261 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1262 return ERR_PTR(-EINVAL);
1264 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1265 return ERR_PTR(-EINVAL);
1268 * Thread groups must share signals as well, and detached threads
1269 * can only be started up within the thread group.
1271 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1272 return ERR_PTR(-EINVAL);
1275 * Shared signal handlers imply shared VM. By way of the above,
1276 * thread groups also imply shared VM. Blocking this case allows
1277 * for various simplifications in other code.
1279 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1280 return ERR_PTR(-EINVAL);
1283 * Siblings of global init remain as zombies on exit since they are
1284 * not reaped by their parent (swapper). To solve this and to avoid
1285 * multi-rooted process trees, prevent global and container-inits
1286 * from creating siblings.
1288 if ((clone_flags & CLONE_PARENT) &&
1289 current->signal->flags & SIGNAL_UNKILLABLE)
1290 return ERR_PTR(-EINVAL);
1293 * If the new process will be in a different pid or user namespace
1294 * do not allow it to share a thread group with the forking task.
1296 if (clone_flags & CLONE_THREAD) {
1297 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1298 (task_active_pid_ns(current) !=
1299 current->nsproxy->pid_ns_for_children))
1300 return ERR_PTR(-EINVAL);
1303 retval = security_task_create(clone_flags);
1308 p = dup_task_struct(current);
1312 ftrace_graph_init_task(p);
1314 rt_mutex_init_task(p);
1316 #ifdef CONFIG_PROVE_LOCKING
1317 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1318 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1321 if (atomic_read(&p->real_cred->user->processes) >=
1322 task_rlimit(p, RLIMIT_NPROC)) {
1323 if (p->real_cred->user != INIT_USER &&
1324 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1327 current->flags &= ~PF_NPROC_EXCEEDED;
1329 retval = copy_creds(p, clone_flags);
1334 * If multiple threads are within copy_process(), then this check
1335 * triggers too late. This doesn't hurt, the check is only there
1336 * to stop root fork bombs.
1339 if (nr_threads >= max_threads)
1340 goto bad_fork_cleanup_count;
1342 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1343 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1344 p->flags |= PF_FORKNOEXEC;
1345 INIT_LIST_HEAD(&p->children);
1346 INIT_LIST_HEAD(&p->sibling);
1347 rcu_copy_process(p);
1348 p->vfork_done = NULL;
1349 spin_lock_init(&p->alloc_lock);
1351 init_sigpending(&p->pending);
1353 p->utime = p->stime = p->gtime = 0;
1354 p->utimescaled = p->stimescaled = 0;
1355 prev_cputime_init(&p->prev_cputime);
1357 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1358 seqcount_init(&p->vtime_seqcount);
1360 p->vtime_snap_whence = VTIME_INACTIVE;
1363 #if defined(SPLIT_RSS_COUNTING)
1364 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1367 p->default_timer_slack_ns = current->timer_slack_ns;
1369 task_io_accounting_init(&p->ioac);
1370 acct_clear_integrals(p);
1372 posix_cpu_timers_init(p);
1374 p->start_time = ktime_get_ns();
1375 p->real_start_time = ktime_get_boot_ns();
1376 p->io_context = NULL;
1377 p->audit_context = NULL;
1378 threadgroup_change_begin(current);
1381 p->mempolicy = mpol_dup(p->mempolicy);
1382 if (IS_ERR(p->mempolicy)) {
1383 retval = PTR_ERR(p->mempolicy);
1384 p->mempolicy = NULL;
1385 goto bad_fork_cleanup_threadgroup_lock;
1388 #ifdef CONFIG_CPUSETS
1389 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1390 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1391 seqcount_init(&p->mems_allowed_seq);
1393 #ifdef CONFIG_TRACE_IRQFLAGS
1395 p->hardirqs_enabled = 0;
1396 p->hardirq_enable_ip = 0;
1397 p->hardirq_enable_event = 0;
1398 p->hardirq_disable_ip = _THIS_IP_;
1399 p->hardirq_disable_event = 0;
1400 p->softirqs_enabled = 1;
1401 p->softirq_enable_ip = _THIS_IP_;
1402 p->softirq_enable_event = 0;
1403 p->softirq_disable_ip = 0;
1404 p->softirq_disable_event = 0;
1405 p->hardirq_context = 0;
1406 p->softirq_context = 0;
1409 p->pagefault_disabled = 0;
1411 #ifdef CONFIG_LOCKDEP
1412 p->lockdep_depth = 0; /* no locks held yet */
1413 p->curr_chain_key = 0;
1414 p->lockdep_recursion = 0;
1417 #ifdef CONFIG_DEBUG_MUTEXES
1418 p->blocked_on = NULL; /* not blocked yet */
1420 #ifdef CONFIG_BCACHE
1421 p->sequential_io = 0;
1422 p->sequential_io_avg = 0;
1425 /* Perform scheduler related setup. Assign this task to a CPU. */
1426 retval = sched_fork(clone_flags, p);
1428 goto bad_fork_cleanup_policy;
1430 retval = perf_event_init_task(p);
1432 goto bad_fork_cleanup_policy;
1433 retval = audit_alloc(p);
1435 goto bad_fork_cleanup_perf;
1436 /* copy all the process information */
1438 retval = copy_semundo(clone_flags, p);
1440 goto bad_fork_cleanup_audit;
1441 retval = copy_files(clone_flags, p);
1443 goto bad_fork_cleanup_semundo;
1444 retval = copy_fs(clone_flags, p);
1446 goto bad_fork_cleanup_files;
1447 retval = copy_sighand(clone_flags, p);
1449 goto bad_fork_cleanup_fs;
1450 retval = copy_signal(clone_flags, p);
1452 goto bad_fork_cleanup_sighand;
1453 retval = copy_mm(clone_flags, p);
1455 goto bad_fork_cleanup_signal;
1456 retval = copy_namespaces(clone_flags, p);
1458 goto bad_fork_cleanup_mm;
1459 retval = copy_io(clone_flags, p);
1461 goto bad_fork_cleanup_namespaces;
1462 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1464 goto bad_fork_cleanup_io;
1466 if (pid != &init_struct_pid) {
1467 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1469 retval = PTR_ERR(pid);
1470 goto bad_fork_cleanup_io;
1474 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1476 * Clear TID on mm_release()?
1478 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1483 p->robust_list = NULL;
1484 #ifdef CONFIG_COMPAT
1485 p->compat_robust_list = NULL;
1487 INIT_LIST_HEAD(&p->pi_state_list);
1488 p->pi_state_cache = NULL;
1491 * sigaltstack should be cleared when sharing the same VM
1493 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1494 p->sas_ss_sp = p->sas_ss_size = 0;
1497 * Syscall tracing and stepping should be turned off in the
1498 * child regardless of CLONE_PTRACE.
1500 user_disable_single_step(p);
1501 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1502 #ifdef TIF_SYSCALL_EMU
1503 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1505 clear_all_latency_tracing(p);
1507 /* ok, now we should be set up.. */
1508 p->pid = pid_nr(pid);
1509 if (clone_flags & CLONE_THREAD) {
1510 p->exit_signal = -1;
1511 p->group_leader = current->group_leader;
1512 p->tgid = current->tgid;
1514 if (clone_flags & CLONE_PARENT)
1515 p->exit_signal = current->group_leader->exit_signal;
1517 p->exit_signal = (clone_flags & CSIGNAL);
1518 p->group_leader = p;
1523 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1524 p->dirty_paused_when = 0;
1526 p->pdeath_signal = 0;
1527 INIT_LIST_HEAD(&p->thread_group);
1528 p->task_works = NULL;
1531 * Ensure that the cgroup subsystem policies allow the new process to be
1532 * forked. It should be noted the the new process's css_set can be changed
1533 * between here and cgroup_post_fork() if an organisation operation is in
1536 retval = cgroup_can_fork(p);
1538 goto bad_fork_free_pid;
1541 * Make it visible to the rest of the system, but dont wake it up yet.
1542 * Need tasklist lock for parent etc handling!
1544 write_lock_irq(&tasklist_lock);
1546 /* CLONE_PARENT re-uses the old parent */
1547 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1548 p->real_parent = current->real_parent;
1549 p->parent_exec_id = current->parent_exec_id;
1551 p->real_parent = current;
1552 p->parent_exec_id = current->self_exec_id;
1555 spin_lock(¤t->sighand->siglock);
1558 * Copy seccomp details explicitly here, in case they were changed
1559 * before holding sighand lock.
1564 * Process group and session signals need to be delivered to just the
1565 * parent before the fork or both the parent and the child after the
1566 * fork. Restart if a signal comes in before we add the new process to
1567 * it's process group.
1568 * A fatal signal pending means that current will exit, so the new
1569 * thread can't slip out of an OOM kill (or normal SIGKILL).
1571 recalc_sigpending();
1572 if (signal_pending(current)) {
1573 spin_unlock(¤t->sighand->siglock);
1574 write_unlock_irq(&tasklist_lock);
1575 retval = -ERESTARTNOINTR;
1576 goto bad_fork_cancel_cgroup;
1579 if (likely(p->pid)) {
1580 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1582 init_task_pid(p, PIDTYPE_PID, pid);
1583 if (thread_group_leader(p)) {
1584 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1585 init_task_pid(p, PIDTYPE_SID, task_session(current));
1587 if (is_child_reaper(pid)) {
1588 ns_of_pid(pid)->child_reaper = p;
1589 p->signal->flags |= SIGNAL_UNKILLABLE;
1592 p->signal->leader_pid = pid;
1593 p->signal->tty = tty_kref_get(current->signal->tty);
1594 list_add_tail(&p->sibling, &p->real_parent->children);
1595 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1596 attach_pid(p, PIDTYPE_PGID);
1597 attach_pid(p, PIDTYPE_SID);
1598 __this_cpu_inc(process_counts);
1600 current->signal->nr_threads++;
1601 atomic_inc(¤t->signal->live);
1602 atomic_inc(¤t->signal->sigcnt);
1603 list_add_tail_rcu(&p->thread_group,
1604 &p->group_leader->thread_group);
1605 list_add_tail_rcu(&p->thread_node,
1606 &p->signal->thread_head);
1608 attach_pid(p, PIDTYPE_PID);
1613 spin_unlock(¤t->sighand->siglock);
1614 syscall_tracepoint_update(p);
1615 write_unlock_irq(&tasklist_lock);
1617 proc_fork_connector(p);
1618 cgroup_post_fork(p);
1619 threadgroup_change_end(current);
1622 trace_task_newtask(p, clone_flags);
1623 uprobe_copy_process(p, clone_flags);
1627 bad_fork_cancel_cgroup:
1628 cgroup_cancel_fork(p);
1630 if (pid != &init_struct_pid)
1632 bad_fork_cleanup_io:
1635 bad_fork_cleanup_namespaces:
1636 exit_task_namespaces(p);
1637 bad_fork_cleanup_mm:
1640 bad_fork_cleanup_signal:
1641 if (!(clone_flags & CLONE_THREAD))
1642 free_signal_struct(p->signal);
1643 bad_fork_cleanup_sighand:
1644 __cleanup_sighand(p->sighand);
1645 bad_fork_cleanup_fs:
1646 exit_fs(p); /* blocking */
1647 bad_fork_cleanup_files:
1648 exit_files(p); /* blocking */
1649 bad_fork_cleanup_semundo:
1651 bad_fork_cleanup_audit:
1653 bad_fork_cleanup_perf:
1654 perf_event_free_task(p);
1655 bad_fork_cleanup_policy:
1657 mpol_put(p->mempolicy);
1658 bad_fork_cleanup_threadgroup_lock:
1660 threadgroup_change_end(current);
1661 delayacct_tsk_free(p);
1662 bad_fork_cleanup_count:
1663 atomic_dec(&p->cred->user->processes);
1668 return ERR_PTR(retval);
1671 static inline void init_idle_pids(struct pid_link *links)
1675 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1676 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1677 links[type].pid = &init_struct_pid;
1681 struct task_struct *fork_idle(int cpu)
1683 struct task_struct *task;
1684 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1685 if (!IS_ERR(task)) {
1686 init_idle_pids(task->pids);
1687 init_idle(task, cpu);
1694 * Ok, this is the main fork-routine.
1696 * It copies the process, and if successful kick-starts
1697 * it and waits for it to finish using the VM if required.
1699 long _do_fork(unsigned long clone_flags,
1700 unsigned long stack_start,
1701 unsigned long stack_size,
1702 int __user *parent_tidptr,
1703 int __user *child_tidptr,
1706 struct task_struct *p;
1711 * Determine whether and which event to report to ptracer. When
1712 * called from kernel_thread or CLONE_UNTRACED is explicitly
1713 * requested, no event is reported; otherwise, report if the event
1714 * for the type of forking is enabled.
1716 if (!(clone_flags & CLONE_UNTRACED)) {
1717 if (clone_flags & CLONE_VFORK)
1718 trace = PTRACE_EVENT_VFORK;
1719 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1720 trace = PTRACE_EVENT_CLONE;
1722 trace = PTRACE_EVENT_FORK;
1724 if (likely(!ptrace_event_enabled(current, trace)))
1728 p = copy_process(clone_flags, stack_start, stack_size,
1729 child_tidptr, NULL, trace, tls);
1731 * Do this prior waking up the new thread - the thread pointer
1732 * might get invalid after that point, if the thread exits quickly.
1735 struct completion vfork;
1738 trace_sched_process_fork(current, p);
1740 pid = get_task_pid(p, PIDTYPE_PID);
1743 if (clone_flags & CLONE_PARENT_SETTID)
1744 put_user(nr, parent_tidptr);
1746 if (clone_flags & CLONE_VFORK) {
1747 p->vfork_done = &vfork;
1748 init_completion(&vfork);
1752 wake_up_new_task(p);
1754 /* forking complete and child started to run, tell ptracer */
1755 if (unlikely(trace))
1756 ptrace_event_pid(trace, pid);
1758 if (clone_flags & CLONE_VFORK) {
1759 if (!wait_for_vfork_done(p, &vfork))
1760 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1770 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1771 /* For compatibility with architectures that call do_fork directly rather than
1772 * using the syscall entry points below. */
1773 long do_fork(unsigned long clone_flags,
1774 unsigned long stack_start,
1775 unsigned long stack_size,
1776 int __user *parent_tidptr,
1777 int __user *child_tidptr)
1779 return _do_fork(clone_flags, stack_start, stack_size,
1780 parent_tidptr, child_tidptr, 0);
1785 * Create a kernel thread.
1787 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1789 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1790 (unsigned long)arg, NULL, NULL, 0);
1793 #ifdef __ARCH_WANT_SYS_FORK
1794 SYSCALL_DEFINE0(fork)
1797 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1799 /* can not support in nommu mode */
1805 #ifdef __ARCH_WANT_SYS_VFORK
1806 SYSCALL_DEFINE0(vfork)
1808 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1813 #ifdef __ARCH_WANT_SYS_CLONE
1814 #ifdef CONFIG_CLONE_BACKWARDS
1815 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1816 int __user *, parent_tidptr,
1818 int __user *, child_tidptr)
1819 #elif defined(CONFIG_CLONE_BACKWARDS2)
1820 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1821 int __user *, parent_tidptr,
1822 int __user *, child_tidptr,
1824 #elif defined(CONFIG_CLONE_BACKWARDS3)
1825 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1827 int __user *, parent_tidptr,
1828 int __user *, child_tidptr,
1831 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1832 int __user *, parent_tidptr,
1833 int __user *, child_tidptr,
1837 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1841 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1842 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1845 static void sighand_ctor(void *data)
1847 struct sighand_struct *sighand = data;
1849 spin_lock_init(&sighand->siglock);
1850 init_waitqueue_head(&sighand->signalfd_wqh);
1853 void __init proc_caches_init(void)
1855 sighand_cachep = kmem_cache_create("sighand_cache",
1856 sizeof(struct sighand_struct), 0,
1857 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1858 SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
1859 signal_cachep = kmem_cache_create("signal_cache",
1860 sizeof(struct signal_struct), 0,
1861 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1863 files_cachep = kmem_cache_create("files_cache",
1864 sizeof(struct files_struct), 0,
1865 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1867 fs_cachep = kmem_cache_create("fs_cache",
1868 sizeof(struct fs_struct), 0,
1869 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1872 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1873 * whole struct cpumask for the OFFSTACK case. We could change
1874 * this to *only* allocate as much of it as required by the
1875 * maximum number of CPU's we can ever have. The cpumask_allocation
1876 * is at the end of the structure, exactly for that reason.
1878 mm_cachep = kmem_cache_create("mm_struct",
1879 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1880 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1882 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
1884 nsproxy_cache_init();
1888 * Check constraints on flags passed to the unshare system call.
1890 static int check_unshare_flags(unsigned long unshare_flags)
1892 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1893 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1894 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1895 CLONE_NEWUSER|CLONE_NEWPID))
1898 * Not implemented, but pretend it works if there is nothing
1899 * to unshare. Note that unsharing the address space or the
1900 * signal handlers also need to unshare the signal queues (aka
1903 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1904 if (!thread_group_empty(current))
1907 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1908 if (atomic_read(¤t->sighand->count) > 1)
1911 if (unshare_flags & CLONE_VM) {
1912 if (!current_is_single_threaded())
1920 * Unshare the filesystem structure if it is being shared
1922 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1924 struct fs_struct *fs = current->fs;
1926 if (!(unshare_flags & CLONE_FS) || !fs)
1929 /* don't need lock here; in the worst case we'll do useless copy */
1933 *new_fsp = copy_fs_struct(fs);
1941 * Unshare file descriptor table if it is being shared
1943 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1945 struct files_struct *fd = current->files;
1948 if ((unshare_flags & CLONE_FILES) &&
1949 (fd && atomic_read(&fd->count) > 1)) {
1950 *new_fdp = dup_fd(fd, &error);
1959 * unshare allows a process to 'unshare' part of the process
1960 * context which was originally shared using clone. copy_*
1961 * functions used by do_fork() cannot be used here directly
1962 * because they modify an inactive task_struct that is being
1963 * constructed. Here we are modifying the current, active,
1966 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1968 struct fs_struct *fs, *new_fs = NULL;
1969 struct files_struct *fd, *new_fd = NULL;
1970 struct cred *new_cred = NULL;
1971 struct nsproxy *new_nsproxy = NULL;
1976 * If unsharing a user namespace must also unshare the thread group
1977 * and unshare the filesystem root and working directories.
1979 if (unshare_flags & CLONE_NEWUSER)
1980 unshare_flags |= CLONE_THREAD | CLONE_FS;
1982 * If unsharing vm, must also unshare signal handlers.
1984 if (unshare_flags & CLONE_VM)
1985 unshare_flags |= CLONE_SIGHAND;
1987 * If unsharing a signal handlers, must also unshare the signal queues.
1989 if (unshare_flags & CLONE_SIGHAND)
1990 unshare_flags |= CLONE_THREAD;
1992 * If unsharing namespace, must also unshare filesystem information.
1994 if (unshare_flags & CLONE_NEWNS)
1995 unshare_flags |= CLONE_FS;
1997 err = check_unshare_flags(unshare_flags);
1999 goto bad_unshare_out;
2001 * CLONE_NEWIPC must also detach from the undolist: after switching
2002 * to a new ipc namespace, the semaphore arrays from the old
2003 * namespace are unreachable.
2005 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2007 err = unshare_fs(unshare_flags, &new_fs);
2009 goto bad_unshare_out;
2010 err = unshare_fd(unshare_flags, &new_fd);
2012 goto bad_unshare_cleanup_fs;
2013 err = unshare_userns(unshare_flags, &new_cred);
2015 goto bad_unshare_cleanup_fd;
2016 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2019 goto bad_unshare_cleanup_cred;
2021 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2024 * CLONE_SYSVSEM is equivalent to sys_exit().
2028 if (unshare_flags & CLONE_NEWIPC) {
2029 /* Orphan segments in old ns (see sem above). */
2031 shm_init_task(current);
2035 switch_task_namespaces(current, new_nsproxy);
2041 spin_lock(&fs->lock);
2042 current->fs = new_fs;
2047 spin_unlock(&fs->lock);
2051 fd = current->files;
2052 current->files = new_fd;
2056 task_unlock(current);
2059 /* Install the new user namespace */
2060 commit_creds(new_cred);
2065 bad_unshare_cleanup_cred:
2068 bad_unshare_cleanup_fd:
2070 put_files_struct(new_fd);
2072 bad_unshare_cleanup_fs:
2074 free_fs_struct(new_fs);
2081 * Helper to unshare the files of the current task.
2082 * We don't want to expose copy_files internals to
2083 * the exec layer of the kernel.
2086 int unshare_files(struct files_struct **displaced)
2088 struct task_struct *task = current;
2089 struct files_struct *copy = NULL;
2092 error = unshare_fd(CLONE_FILES, ©);
2093 if (error || !copy) {
2097 *displaced = task->files;
2104 int sysctl_max_threads(struct ctl_table *table, int write,
2105 void __user *buffer, size_t *lenp, loff_t *ppos)
2109 int threads = max_threads;
2110 int min = MIN_THREADS;
2111 int max = MAX_THREADS;
2118 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2122 set_max_threads(threads);