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>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/seccomp.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/kthread.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/rmap.h>
54 #include <linux/ksm.h>
55 #include <linux/acct.h>
56 #include <linux/tsacct_kern.h>
57 #include <linux/cn_proc.h>
58 #include <linux/freezer.h>
59 #include <linux/delayacct.h>
60 #include <linux/taskstats_kern.h>
61 #include <linux/random.h>
62 #include <linux/tty.h>
63 #include <linux/blkdev.h>
64 #include <linux/fs_struct.h>
65 #include <linux/magic.h>
66 #include <linux/perf_event.h>
67 #include <linux/posix-timers.h>
68 #include <linux/user-return-notifier.h>
69 #include <linux/oom.h>
70 #include <linux/khugepaged.h>
71 #include <linux/signalfd.h>
72 #include <linux/uprobes.h>
73 #include <linux/aio.h>
75 #include <asm/pgtable.h>
76 #include <asm/pgalloc.h>
77 #include <asm/uaccess.h>
78 #include <asm/mmu_context.h>
79 #include <asm/cacheflush.h>
80 #include <asm/tlbflush.h>
82 #include <trace/events/sched.h>
84 #define CREATE_TRACE_POINTS
85 #include <trace/events/task.h>
88 * Protected counters by write_lock_irq(&tasklist_lock)
90 unsigned long total_forks; /* Handle normal Linux uptimes. */
91 int nr_threads; /* The idle threads do not count.. */
93 int max_threads; /* tunable limit on nr_threads */
95 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
97 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
99 #ifdef CONFIG_PROVE_RCU
100 int lockdep_tasklist_lock_is_held(void)
102 return lockdep_is_held(&tasklist_lock);
104 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
105 #endif /* #ifdef CONFIG_PROVE_RCU */
107 int nr_processes(void)
112 for_each_possible_cpu(cpu)
113 total += per_cpu(process_counts, cpu);
118 void __weak arch_release_task_struct(struct task_struct *tsk)
122 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
123 static struct kmem_cache *task_struct_cachep;
125 static inline struct task_struct *alloc_task_struct_node(int node)
127 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
130 static inline void free_task_struct(struct task_struct *tsk)
132 kmem_cache_free(task_struct_cachep, tsk);
136 void __weak arch_release_thread_info(struct thread_info *ti)
140 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
143 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
144 * kmemcache based allocator.
146 # if THREAD_SIZE >= PAGE_SIZE
147 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
150 struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
153 return page ? page_address(page) : NULL;
156 static inline void free_thread_info(struct thread_info *ti)
158 free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
161 static struct kmem_cache *thread_info_cache;
163 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
166 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
169 static void free_thread_info(struct thread_info *ti)
171 kmem_cache_free(thread_info_cache, ti);
174 void thread_info_cache_init(void)
176 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
177 THREAD_SIZE, 0, NULL);
178 BUG_ON(thread_info_cache == NULL);
183 /* SLAB cache for signal_struct structures (tsk->signal) */
184 static struct kmem_cache *signal_cachep;
186 /* SLAB cache for sighand_struct structures (tsk->sighand) */
187 struct kmem_cache *sighand_cachep;
189 /* SLAB cache for files_struct structures (tsk->files) */
190 struct kmem_cache *files_cachep;
192 /* SLAB cache for fs_struct structures (tsk->fs) */
193 struct kmem_cache *fs_cachep;
195 /* SLAB cache for vm_area_struct structures */
196 struct kmem_cache *vm_area_cachep;
198 /* SLAB cache for mm_struct structures (tsk->mm) */
199 static struct kmem_cache *mm_cachep;
201 /* Notifier list called when a task struct is freed */
202 static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
204 static void account_kernel_stack(struct thread_info *ti, int account)
206 struct zone *zone = page_zone(virt_to_page(ti));
208 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
211 void free_task(struct task_struct *tsk)
213 account_kernel_stack(tsk->stack, -1);
214 arch_release_thread_info(tsk->stack);
215 free_thread_info(tsk->stack);
216 rt_mutex_debug_task_free(tsk);
217 ftrace_graph_exit_task(tsk);
218 put_seccomp_filter(tsk);
219 arch_release_task_struct(tsk);
220 free_task_struct(tsk);
222 EXPORT_SYMBOL(free_task);
224 static inline void free_signal_struct(struct signal_struct *sig)
226 taskstats_tgid_free(sig);
227 sched_autogroup_exit(sig);
228 kmem_cache_free(signal_cachep, sig);
231 static inline void put_signal_struct(struct signal_struct *sig)
233 if (atomic_dec_and_test(&sig->sigcnt))
234 free_signal_struct(sig);
237 int task_free_register(struct notifier_block *n)
239 return atomic_notifier_chain_register(&task_free_notifier, n);
241 EXPORT_SYMBOL(task_free_register);
243 int task_free_unregister(struct notifier_block *n)
245 return atomic_notifier_chain_unregister(&task_free_notifier, n);
247 EXPORT_SYMBOL(task_free_unregister);
249 void __put_task_struct(struct task_struct *tsk)
251 WARN_ON(!tsk->exit_state);
252 WARN_ON(atomic_read(&tsk->usage));
253 WARN_ON(tsk == current);
255 security_task_free(tsk);
257 delayacct_tsk_free(tsk);
258 put_signal_struct(tsk->signal);
260 atomic_notifier_call_chain(&task_free_notifier, 0, tsk);
261 if (!profile_handoff_task(tsk))
264 EXPORT_SYMBOL_GPL(__put_task_struct);
266 void __init __weak arch_task_cache_init(void) { }
268 void __init fork_init(unsigned long mempages)
270 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
271 #ifndef ARCH_MIN_TASKALIGN
272 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
274 /* create a slab on which task_structs can be allocated */
276 kmem_cache_create("task_struct", sizeof(struct task_struct),
277 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
280 /* do the arch specific task caches init */
281 arch_task_cache_init();
284 * The default maximum number of threads is set to a safe
285 * value: the thread structures can take up at most half
288 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
291 * we need to allow at least 20 threads to boot a system
293 if (max_threads < 20)
296 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
297 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
298 init_task.signal->rlim[RLIMIT_SIGPENDING] =
299 init_task.signal->rlim[RLIMIT_NPROC];
302 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
303 struct task_struct *src)
309 static struct task_struct *dup_task_struct(struct task_struct *orig)
311 struct task_struct *tsk;
312 struct thread_info *ti;
313 unsigned long *stackend;
314 int node = tsk_fork_get_node(orig);
317 tsk = alloc_task_struct_node(node);
321 ti = alloc_thread_info_node(tsk, node);
325 err = arch_dup_task_struct(tsk, orig);
330 #ifdef CONFIG_SECCOMP
332 * We must handle setting up seccomp filters once we're under
333 * the sighand lock in case orig has changed between now and
334 * then. Until then, filter must be NULL to avoid messing up
335 * the usage counts on the error path calling free_task.
337 tsk->seccomp.filter = NULL;
340 setup_thread_stack(tsk, orig);
341 clear_user_return_notifier(tsk);
342 clear_tsk_need_resched(tsk);
343 stackend = end_of_stack(tsk);
344 *stackend = STACK_END_MAGIC; /* for overflow detection */
346 #ifdef CONFIG_CC_STACKPROTECTOR
347 tsk->stack_canary = get_random_int();
351 * One for us, one for whoever does the "release_task()" (usually
354 atomic_set(&tsk->usage, 2);
355 #ifdef CONFIG_BLK_DEV_IO_TRACE
358 tsk->splice_pipe = NULL;
359 tsk->task_frag.page = NULL;
361 account_kernel_stack(ti, 1);
366 free_thread_info(ti);
368 free_task_struct(tsk);
373 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
375 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
376 struct rb_node **rb_link, *rb_parent;
378 unsigned long charge;
379 struct mempolicy *pol;
381 uprobe_start_dup_mmap();
382 down_write(&oldmm->mmap_sem);
383 flush_cache_dup_mm(oldmm);
384 uprobe_dup_mmap(oldmm, mm);
386 * Not linked in yet - no deadlock potential:
388 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
392 mm->mmap_cache = NULL;
393 mm->free_area_cache = oldmm->mmap_base;
394 mm->cached_hole_size = ~0UL;
396 cpumask_clear(mm_cpumask(mm));
398 rb_link = &mm->mm_rb.rb_node;
401 retval = ksm_fork(mm, oldmm);
404 retval = khugepaged_fork(mm, oldmm);
409 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
412 if (mpnt->vm_flags & VM_DONTCOPY) {
413 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
418 if (mpnt->vm_flags & VM_ACCOUNT) {
419 unsigned long len = vma_pages(mpnt);
421 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
425 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
429 INIT_LIST_HEAD(&tmp->anon_vma_chain);
430 pol = mpol_dup(vma_policy(mpnt));
431 retval = PTR_ERR(pol);
433 goto fail_nomem_policy;
434 vma_set_policy(tmp, pol);
436 if (anon_vma_fork(tmp, mpnt))
437 goto fail_nomem_anon_vma_fork;
438 tmp->vm_flags &= ~VM_LOCKED;
439 tmp->vm_next = tmp->vm_prev = NULL;
442 struct inode *inode = file_inode(file);
443 struct address_space *mapping = file->f_mapping;
446 if (tmp->vm_flags & VM_DENYWRITE)
447 atomic_dec(&inode->i_writecount);
448 mutex_lock(&mapping->i_mmap_mutex);
449 if (tmp->vm_flags & VM_SHARED)
450 mapping->i_mmap_writable++;
451 flush_dcache_mmap_lock(mapping);
452 /* insert tmp into the share list, just after mpnt */
453 if (unlikely(tmp->vm_flags & VM_NONLINEAR))
454 vma_nonlinear_insert(tmp,
455 &mapping->i_mmap_nonlinear);
457 vma_interval_tree_insert_after(tmp, mpnt,
459 flush_dcache_mmap_unlock(mapping);
460 mutex_unlock(&mapping->i_mmap_mutex);
464 * Clear hugetlb-related page reserves for children. This only
465 * affects MAP_PRIVATE mappings. Faults generated by the child
466 * are not guaranteed to succeed, even if read-only
468 if (is_vm_hugetlb_page(tmp))
469 reset_vma_resv_huge_pages(tmp);
472 * Link in the new vma and copy the page table entries.
475 pprev = &tmp->vm_next;
479 __vma_link_rb(mm, tmp, rb_link, rb_parent);
480 rb_link = &tmp->vm_rb.rb_right;
481 rb_parent = &tmp->vm_rb;
484 retval = copy_page_range(mm, oldmm, mpnt);
486 if (tmp->vm_ops && tmp->vm_ops->open)
487 tmp->vm_ops->open(tmp);
492 /* a new mm has just been created */
493 arch_dup_mmap(oldmm, mm);
496 up_write(&mm->mmap_sem);
498 up_write(&oldmm->mmap_sem);
499 uprobe_end_dup_mmap();
501 fail_nomem_anon_vma_fork:
504 kmem_cache_free(vm_area_cachep, tmp);
507 vm_unacct_memory(charge);
511 static inline int mm_alloc_pgd(struct mm_struct *mm)
513 mm->pgd = pgd_alloc(mm);
514 if (unlikely(!mm->pgd))
519 static inline void mm_free_pgd(struct mm_struct *mm)
521 pgd_free(mm, mm->pgd);
524 #define dup_mmap(mm, oldmm) (0)
525 #define mm_alloc_pgd(mm) (0)
526 #define mm_free_pgd(mm)
527 #endif /* CONFIG_MMU */
529 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
531 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
532 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
534 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
536 static int __init coredump_filter_setup(char *s)
538 default_dump_filter =
539 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
540 MMF_DUMP_FILTER_MASK;
544 __setup("coredump_filter=", coredump_filter_setup);
546 #include <linux/init_task.h>
548 static void mm_init_aio(struct mm_struct *mm)
551 spin_lock_init(&mm->ioctx_lock);
552 INIT_HLIST_HEAD(&mm->ioctx_list);
556 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
558 atomic_set(&mm->mm_users, 1);
559 atomic_set(&mm->mm_count, 1);
560 init_rwsem(&mm->mmap_sem);
561 INIT_LIST_HEAD(&mm->mmlist);
562 mm->flags = (current->mm) ?
563 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
564 mm->core_state = NULL;
566 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
567 spin_lock_init(&mm->page_table_lock);
568 mm->free_area_cache = TASK_UNMAPPED_BASE;
569 mm->cached_hole_size = ~0UL;
571 mm_init_owner(mm, p);
572 clear_tlb_flush_pending(mm);
574 if (likely(!mm_alloc_pgd(mm))) {
576 mmu_notifier_mm_init(mm);
584 static void check_mm(struct mm_struct *mm)
588 for (i = 0; i < NR_MM_COUNTERS; i++) {
589 long x = atomic_long_read(&mm->rss_stat.count[i]);
592 printk(KERN_ALERT "BUG: Bad rss-counter state "
593 "mm:%p idx:%d val:%ld\n", mm, i, x);
596 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
597 VM_BUG_ON(mm->pmd_huge_pte);
602 * Allocate and initialize an mm_struct.
604 struct mm_struct *mm_alloc(void)
606 struct mm_struct *mm;
612 memset(mm, 0, sizeof(*mm));
614 return mm_init(mm, current);
618 * Called when the last reference to the mm
619 * is dropped: either by a lazy thread or by
620 * mmput. Free the page directory and the mm.
622 void __mmdrop(struct mm_struct *mm)
624 BUG_ON(mm == &init_mm);
627 mmu_notifier_mm_destroy(mm);
631 EXPORT_SYMBOL_GPL(__mmdrop);
634 * Decrement the use count and release all resources for an mm.
636 void mmput(struct mm_struct *mm)
640 if (atomic_dec_and_test(&mm->mm_users)) {
641 uprobe_clear_state(mm);
644 khugepaged_exit(mm); /* must run before exit_mmap */
646 set_mm_exe_file(mm, NULL);
647 if (!list_empty(&mm->mmlist)) {
648 spin_lock(&mmlist_lock);
649 list_del(&mm->mmlist);
650 spin_unlock(&mmlist_lock);
653 module_put(mm->binfmt->module);
657 EXPORT_SYMBOL_GPL(mmput);
659 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
662 get_file(new_exe_file);
665 mm->exe_file = new_exe_file;
668 struct file *get_mm_exe_file(struct mm_struct *mm)
670 struct file *exe_file;
672 /* We need mmap_sem to protect against races with removal of exe_file */
673 down_read(&mm->mmap_sem);
674 exe_file = mm->exe_file;
677 up_read(&mm->mmap_sem);
681 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
683 /* It's safe to write the exe_file pointer without exe_file_lock because
684 * this is called during fork when the task is not yet in /proc */
685 newmm->exe_file = get_mm_exe_file(oldmm);
689 * get_task_mm - acquire a reference to the task's mm
691 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
692 * this kernel workthread has transiently adopted a user mm with use_mm,
693 * to do its AIO) is not set and if so returns a reference to it, after
694 * bumping up the use count. User must release the mm via mmput()
695 * after use. Typically used by /proc and ptrace.
697 struct mm_struct *get_task_mm(struct task_struct *task)
699 struct mm_struct *mm;
704 if (task->flags & PF_KTHREAD)
707 atomic_inc(&mm->mm_users);
712 EXPORT_SYMBOL_GPL(get_task_mm);
714 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
716 struct mm_struct *mm;
719 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
723 mm = get_task_mm(task);
724 if (mm && mm != current->mm &&
725 !ptrace_may_access(task, mode) &&
726 !capable(CAP_SYS_RESOURCE)) {
728 mm = ERR_PTR(-EACCES);
730 mutex_unlock(&task->signal->cred_guard_mutex);
735 static void complete_vfork_done(struct task_struct *tsk)
737 struct completion *vfork;
740 vfork = tsk->vfork_done;
742 tsk->vfork_done = NULL;
748 static int wait_for_vfork_done(struct task_struct *child,
749 struct completion *vfork)
753 freezer_do_not_count();
754 killed = wait_for_completion_killable(vfork);
759 child->vfork_done = NULL;
763 put_task_struct(child);
767 /* Please note the differences between mmput and mm_release.
768 * mmput is called whenever we stop holding onto a mm_struct,
769 * error success whatever.
771 * mm_release is called after a mm_struct has been removed
772 * from the current process.
774 * This difference is important for error handling, when we
775 * only half set up a mm_struct for a new process and need to restore
776 * the old one. Because we mmput the new mm_struct before
777 * restoring the old one. . .
778 * Eric Biederman 10 January 1998
780 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
782 /* Get rid of any futexes when releasing the mm */
784 if (unlikely(tsk->robust_list)) {
785 exit_robust_list(tsk);
786 tsk->robust_list = NULL;
789 if (unlikely(tsk->compat_robust_list)) {
790 compat_exit_robust_list(tsk);
791 tsk->compat_robust_list = NULL;
794 if (unlikely(!list_empty(&tsk->pi_state_list)))
795 exit_pi_state_list(tsk);
798 uprobe_free_utask(tsk);
800 /* Get rid of any cached register state */
801 deactivate_mm(tsk, mm);
804 * If we're exiting normally, clear a user-space tid field if
805 * requested. We leave this alone when dying by signal, to leave
806 * the value intact in a core dump, and to save the unnecessary
807 * trouble, say, a killed vfork parent shouldn't touch this mm.
808 * Userland only wants this done for a sys_exit.
810 if (tsk->clear_child_tid) {
811 if (!(tsk->flags & PF_SIGNALED) &&
812 atomic_read(&mm->mm_users) > 1) {
814 * We don't check the error code - if userspace has
815 * not set up a proper pointer then tough luck.
817 put_user(0, tsk->clear_child_tid);
818 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
821 tsk->clear_child_tid = NULL;
825 * All done, finally we can wake up parent and return this mm to him.
826 * Also kthread_stop() uses this completion for synchronization.
829 complete_vfork_done(tsk);
833 * Allocate a new mm structure and copy contents from the
834 * mm structure of the passed in task structure.
836 struct mm_struct *dup_mm(struct task_struct *tsk)
838 struct mm_struct *mm, *oldmm = current->mm;
848 memcpy(mm, oldmm, sizeof(*mm));
851 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
852 mm->pmd_huge_pte = NULL;
854 #ifdef CONFIG_NUMA_BALANCING
855 mm->first_nid = NUMA_PTE_SCAN_INIT;
857 if (!mm_init(mm, tsk))
860 if (init_new_context(tsk, mm))
863 dup_mm_exe_file(oldmm, mm);
865 err = dup_mmap(mm, oldmm);
869 mm->hiwater_rss = get_mm_rss(mm);
870 mm->hiwater_vm = mm->total_vm;
872 if (mm->binfmt && !try_module_get(mm->binfmt->module))
878 /* don't put binfmt in mmput, we haven't got module yet */
887 * If init_new_context() failed, we cannot use mmput() to free the mm
888 * because it calls destroy_context()
895 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
897 struct mm_struct *mm, *oldmm;
900 tsk->min_flt = tsk->maj_flt = 0;
901 tsk->nvcsw = tsk->nivcsw = 0;
902 #ifdef CONFIG_DETECT_HUNG_TASK
903 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
907 tsk->active_mm = NULL;
910 * Are we cloning a kernel thread?
912 * We need to steal a active VM for that..
918 if (clone_flags & CLONE_VM) {
919 atomic_inc(&oldmm->mm_users);
938 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
940 struct fs_struct *fs = current->fs;
941 if (clone_flags & CLONE_FS) {
942 /* tsk->fs is already what we want */
943 spin_lock(&fs->lock);
945 spin_unlock(&fs->lock);
949 spin_unlock(&fs->lock);
952 tsk->fs = copy_fs_struct(fs);
958 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
960 struct files_struct *oldf, *newf;
964 * A background process may not have any files ...
966 oldf = current->files;
970 if (clone_flags & CLONE_FILES) {
971 atomic_inc(&oldf->count);
975 newf = dup_fd(oldf, &error);
985 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
988 struct io_context *ioc = current->io_context;
989 struct io_context *new_ioc;
994 * Share io context with parent, if CLONE_IO is set
996 if (clone_flags & CLONE_IO) {
998 tsk->io_context = ioc;
999 } else if (ioprio_valid(ioc->ioprio)) {
1000 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1001 if (unlikely(!new_ioc))
1004 new_ioc->ioprio = ioc->ioprio;
1005 put_io_context(new_ioc);
1011 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1013 struct sighand_struct *sig;
1015 if (clone_flags & CLONE_SIGHAND) {
1016 atomic_inc(¤t->sighand->count);
1019 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1020 rcu_assign_pointer(tsk->sighand, sig);
1023 atomic_set(&sig->count, 1);
1024 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1028 void __cleanup_sighand(struct sighand_struct *sighand)
1030 if (atomic_dec_and_test(&sighand->count)) {
1031 signalfd_cleanup(sighand);
1032 kmem_cache_free(sighand_cachep, sighand);
1038 * Initialize POSIX timer handling for a thread group.
1040 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1042 unsigned long cpu_limit;
1044 /* Thread group counters. */
1045 thread_group_cputime_init(sig);
1047 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1048 if (cpu_limit != RLIM_INFINITY) {
1049 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1050 sig->cputimer.running = 1;
1053 /* The timer lists. */
1054 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1055 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1056 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1059 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1061 struct signal_struct *sig;
1063 if (clone_flags & CLONE_THREAD)
1066 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1071 sig->nr_threads = 1;
1072 atomic_set(&sig->live, 1);
1073 atomic_set(&sig->sigcnt, 1);
1074 init_waitqueue_head(&sig->wait_chldexit);
1075 sig->curr_target = tsk;
1076 init_sigpending(&sig->shared_pending);
1077 INIT_LIST_HEAD(&sig->posix_timers);
1079 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1080 sig->real_timer.function = it_real_fn;
1082 task_lock(current->group_leader);
1083 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1084 task_unlock(current->group_leader);
1086 posix_cpu_timers_init_group(sig);
1088 tty_audit_fork(sig);
1089 sched_autogroup_fork(sig);
1091 #ifdef CONFIG_CGROUPS
1092 init_rwsem(&sig->group_rwsem);
1095 sig->oom_score_adj = current->signal->oom_score_adj;
1096 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1098 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1099 current->signal->is_child_subreaper;
1101 mutex_init(&sig->cred_guard_mutex);
1106 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1108 unsigned long new_flags = p->flags;
1110 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1111 new_flags |= PF_FORKNOEXEC;
1112 p->flags = new_flags;
1115 static void copy_seccomp(struct task_struct *p)
1117 #ifdef CONFIG_SECCOMP
1119 * Must be called with sighand->lock held, which is common to
1120 * all threads in the group. Holding cred_guard_mutex is not
1121 * needed because this new task is not yet running and cannot
1124 BUG_ON(!spin_is_locked(¤t->sighand->siglock));
1126 /* Ref-count the new filter user, and assign it. */
1127 get_seccomp_filter(current);
1128 p->seccomp = current->seccomp;
1131 * Explicitly enable no_new_privs here in case it got set
1132 * between the task_struct being duplicated and holding the
1133 * sighand lock. The seccomp state and nnp must be in sync.
1135 if (task_no_new_privs(current))
1136 task_set_no_new_privs(p);
1139 * If the parent gained a seccomp mode after copying thread
1140 * flags and between before we held the sighand lock, we have
1141 * to manually enable the seccomp thread flag here.
1143 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1144 set_tsk_thread_flag(p, TIF_SECCOMP);
1148 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1150 current->clear_child_tid = tidptr;
1152 return task_pid_vnr(current);
1155 static void rt_mutex_init_task(struct task_struct *p)
1157 raw_spin_lock_init(&p->pi_lock);
1158 #ifdef CONFIG_RT_MUTEXES
1159 plist_head_init(&p->pi_waiters);
1160 p->pi_blocked_on = NULL;
1164 #ifdef CONFIG_MM_OWNER
1165 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1169 #endif /* CONFIG_MM_OWNER */
1172 * Initialize POSIX timer handling for a single task.
1174 static void posix_cpu_timers_init(struct task_struct *tsk)
1176 tsk->cputime_expires.prof_exp = 0;
1177 tsk->cputime_expires.virt_exp = 0;
1178 tsk->cputime_expires.sched_exp = 0;
1179 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1180 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1181 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1185 * This creates a new process as a copy of the old one,
1186 * but does not actually start it yet.
1188 * It copies the registers, and all the appropriate
1189 * parts of the process environment (as per the clone
1190 * flags). The actual kick-off is left to the caller.
1192 static struct task_struct *copy_process(unsigned long clone_flags,
1193 unsigned long stack_start,
1194 unsigned long stack_size,
1195 int __user *child_tidptr,
1200 struct task_struct *p;
1202 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1203 return ERR_PTR(-EINVAL);
1205 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1206 return ERR_PTR(-EINVAL);
1209 * Thread groups must share signals as well, and detached threads
1210 * can only be started up within the thread group.
1212 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1213 return ERR_PTR(-EINVAL);
1216 * Shared signal handlers imply shared VM. By way of the above,
1217 * thread groups also imply shared VM. Blocking this case allows
1218 * for various simplifications in other code.
1220 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1221 return ERR_PTR(-EINVAL);
1224 * Siblings of global init remain as zombies on exit since they are
1225 * not reaped by their parent (swapper). To solve this and to avoid
1226 * multi-rooted process trees, prevent global and container-inits
1227 * from creating siblings.
1229 if ((clone_flags & CLONE_PARENT) &&
1230 current->signal->flags & SIGNAL_UNKILLABLE)
1231 return ERR_PTR(-EINVAL);
1234 * If the new process will be in a different pid namespace don't
1235 * allow it to share a thread group or signal handlers with the
1238 if ((clone_flags & (CLONE_SIGHAND | CLONE_NEWPID)) &&
1239 (task_active_pid_ns(current) != current->nsproxy->pid_ns))
1240 return ERR_PTR(-EINVAL);
1242 retval = security_task_create(clone_flags);
1247 p = dup_task_struct(current);
1251 ftrace_graph_init_task(p);
1253 rt_mutex_init_task(p);
1255 #ifdef CONFIG_PROVE_LOCKING
1256 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1257 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1260 if (atomic_read(&p->real_cred->user->processes) >=
1261 task_rlimit(p, RLIMIT_NPROC)) {
1262 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1263 p->real_cred->user != INIT_USER)
1266 current->flags &= ~PF_NPROC_EXCEEDED;
1268 retval = copy_creds(p, clone_flags);
1273 * If multiple threads are within copy_process(), then this check
1274 * triggers too late. This doesn't hurt, the check is only there
1275 * to stop root fork bombs.
1278 if (nr_threads >= max_threads)
1279 goto bad_fork_cleanup_count;
1281 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1282 goto bad_fork_cleanup_count;
1285 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1286 copy_flags(clone_flags, p);
1287 INIT_LIST_HEAD(&p->children);
1288 INIT_LIST_HEAD(&p->sibling);
1289 rcu_copy_process(p);
1290 p->vfork_done = NULL;
1291 spin_lock_init(&p->alloc_lock);
1293 init_sigpending(&p->pending);
1295 p->utime = p->stime = p->gtime = 0;
1296 p->utimescaled = p->stimescaled = 0;
1297 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1298 p->prev_cputime.utime = p->prev_cputime.stime = 0;
1300 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1301 seqlock_init(&p->vtime_seqlock);
1303 p->vtime_snap_whence = VTIME_SLEEPING;
1306 #if defined(SPLIT_RSS_COUNTING)
1307 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1310 p->default_timer_slack_ns = current->timer_slack_ns;
1312 task_io_accounting_init(&p->ioac);
1313 acct_clear_integrals(p);
1315 posix_cpu_timers_init(p);
1317 do_posix_clock_monotonic_gettime(&p->start_time);
1318 p->real_start_time = p->start_time;
1319 monotonic_to_bootbased(&p->real_start_time);
1320 p->io_context = NULL;
1321 p->audit_context = NULL;
1322 if (clone_flags & CLONE_THREAD)
1323 threadgroup_change_begin(current);
1326 p->mempolicy = mpol_dup(p->mempolicy);
1327 if (IS_ERR(p->mempolicy)) {
1328 retval = PTR_ERR(p->mempolicy);
1329 p->mempolicy = NULL;
1330 goto bad_fork_cleanup_cgroup;
1332 mpol_fix_fork_child_flag(p);
1334 #ifdef CONFIG_CPUSETS
1335 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1336 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1337 seqcount_init(&p->mems_allowed_seq);
1339 #ifdef CONFIG_TRACE_IRQFLAGS
1341 p->hardirqs_enabled = 0;
1342 p->hardirq_enable_ip = 0;
1343 p->hardirq_enable_event = 0;
1344 p->hardirq_disable_ip = _THIS_IP_;
1345 p->hardirq_disable_event = 0;
1346 p->softirqs_enabled = 1;
1347 p->softirq_enable_ip = _THIS_IP_;
1348 p->softirq_enable_event = 0;
1349 p->softirq_disable_ip = 0;
1350 p->softirq_disable_event = 0;
1351 p->hardirq_context = 0;
1352 p->softirq_context = 0;
1354 #ifdef CONFIG_LOCKDEP
1355 p->lockdep_depth = 0; /* no locks held yet */
1356 p->curr_chain_key = 0;
1357 p->lockdep_recursion = 0;
1360 #ifdef CONFIG_DEBUG_MUTEXES
1361 p->blocked_on = NULL; /* not blocked yet */
1364 p->memcg_batch.do_batch = 0;
1365 p->memcg_batch.memcg = NULL;
1367 #ifdef CONFIG_BCACHE
1368 p->sequential_io = 0;
1369 p->sequential_io_avg = 0;
1372 /* Perform scheduler related setup. Assign this task to a CPU. */
1375 retval = perf_event_init_task(p);
1377 goto bad_fork_cleanup_policy;
1378 retval = audit_alloc(p);
1380 goto bad_fork_cleanup_policy;
1381 /* copy all the process information */
1382 retval = copy_semundo(clone_flags, p);
1384 goto bad_fork_cleanup_audit;
1385 retval = copy_files(clone_flags, p);
1387 goto bad_fork_cleanup_semundo;
1388 retval = copy_fs(clone_flags, p);
1390 goto bad_fork_cleanup_files;
1391 retval = copy_sighand(clone_flags, p);
1393 goto bad_fork_cleanup_fs;
1394 retval = copy_signal(clone_flags, p);
1396 goto bad_fork_cleanup_sighand;
1397 retval = copy_mm(clone_flags, p);
1399 goto bad_fork_cleanup_signal;
1400 retval = copy_namespaces(clone_flags, p);
1402 goto bad_fork_cleanup_mm;
1403 retval = copy_io(clone_flags, p);
1405 goto bad_fork_cleanup_namespaces;
1406 retval = copy_thread(clone_flags, stack_start, stack_size, p);
1408 goto bad_fork_cleanup_io;
1410 if (pid != &init_struct_pid) {
1412 pid = alloc_pid(p->nsproxy->pid_ns);
1414 goto bad_fork_cleanup_io;
1417 p->pid = pid_nr(pid);
1419 if (clone_flags & CLONE_THREAD)
1420 p->tgid = current->tgid;
1422 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1424 * Clear TID on mm_release()?
1426 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1431 p->robust_list = NULL;
1432 #ifdef CONFIG_COMPAT
1433 p->compat_robust_list = NULL;
1435 INIT_LIST_HEAD(&p->pi_state_list);
1436 p->pi_state_cache = NULL;
1438 uprobe_copy_process(p);
1440 * sigaltstack should be cleared when sharing the same VM
1442 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1443 p->sas_ss_sp = p->sas_ss_size = 0;
1446 * Syscall tracing and stepping should be turned off in the
1447 * child regardless of CLONE_PTRACE.
1449 user_disable_single_step(p);
1450 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1451 #ifdef TIF_SYSCALL_EMU
1452 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1454 clear_all_latency_tracing(p);
1456 /* ok, now we should be set up.. */
1457 if (clone_flags & CLONE_THREAD)
1458 p->exit_signal = -1;
1459 else if (clone_flags & CLONE_PARENT)
1460 p->exit_signal = current->group_leader->exit_signal;
1462 p->exit_signal = (clone_flags & CSIGNAL);
1464 p->pdeath_signal = 0;
1468 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1469 p->dirty_paused_when = 0;
1472 * Ok, make it visible to the rest of the system.
1473 * We dont wake it up yet.
1475 p->group_leader = p;
1476 INIT_LIST_HEAD(&p->thread_group);
1477 p->task_works = NULL;
1479 /* Need tasklist lock for parent etc handling! */
1480 write_lock_irq(&tasklist_lock);
1482 /* CLONE_PARENT re-uses the old parent */
1483 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1484 p->real_parent = current->real_parent;
1485 p->parent_exec_id = current->parent_exec_id;
1487 p->real_parent = current;
1488 p->parent_exec_id = current->self_exec_id;
1491 spin_lock(¤t->sighand->siglock);
1494 * Copy seccomp details explicitly here, in case they were changed
1495 * before holding sighand lock.
1500 * Process group and session signals need to be delivered to just the
1501 * parent before the fork or both the parent and the child after the
1502 * fork. Restart if a signal comes in before we add the new process to
1503 * it's process group.
1504 * A fatal signal pending means that current will exit, so the new
1505 * thread can't slip out of an OOM kill (or normal SIGKILL).
1507 recalc_sigpending();
1508 if (signal_pending(current)) {
1509 spin_unlock(¤t->sighand->siglock);
1510 write_unlock_irq(&tasklist_lock);
1511 retval = -ERESTARTNOINTR;
1512 goto bad_fork_free_pid;
1515 if (clone_flags & CLONE_THREAD) {
1516 current->signal->nr_threads++;
1517 atomic_inc(¤t->signal->live);
1518 atomic_inc(¤t->signal->sigcnt);
1519 p->group_leader = current->group_leader;
1520 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1523 if (likely(p->pid)) {
1524 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1526 if (thread_group_leader(p)) {
1527 if (is_child_reaper(pid)) {
1528 ns_of_pid(pid)->child_reaper = p;
1529 p->signal->flags |= SIGNAL_UNKILLABLE;
1532 p->signal->leader_pid = pid;
1533 p->signal->tty = tty_kref_get(current->signal->tty);
1534 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1535 attach_pid(p, PIDTYPE_SID, task_session(current));
1536 list_add_tail(&p->sibling, &p->real_parent->children);
1537 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1538 __this_cpu_inc(process_counts);
1540 attach_pid(p, PIDTYPE_PID, pid);
1545 spin_unlock(¤t->sighand->siglock);
1546 syscall_tracepoint_update(p);
1547 write_unlock_irq(&tasklist_lock);
1549 proc_fork_connector(p);
1550 cgroup_post_fork(p);
1551 if (clone_flags & CLONE_THREAD)
1552 threadgroup_change_end(current);
1555 trace_task_newtask(p, clone_flags);
1560 if (pid != &init_struct_pid)
1562 bad_fork_cleanup_io:
1565 bad_fork_cleanup_namespaces:
1566 exit_task_namespaces(p);
1567 bad_fork_cleanup_mm:
1570 bad_fork_cleanup_signal:
1571 if (!(clone_flags & CLONE_THREAD))
1572 free_signal_struct(p->signal);
1573 bad_fork_cleanup_sighand:
1574 __cleanup_sighand(p->sighand);
1575 bad_fork_cleanup_fs:
1576 exit_fs(p); /* blocking */
1577 bad_fork_cleanup_files:
1578 exit_files(p); /* blocking */
1579 bad_fork_cleanup_semundo:
1581 bad_fork_cleanup_audit:
1583 bad_fork_cleanup_policy:
1584 perf_event_free_task(p);
1586 mpol_put(p->mempolicy);
1587 bad_fork_cleanup_cgroup:
1589 if (clone_flags & CLONE_THREAD)
1590 threadgroup_change_end(current);
1592 delayacct_tsk_free(p);
1593 module_put(task_thread_info(p)->exec_domain->module);
1594 bad_fork_cleanup_count:
1595 atomic_dec(&p->cred->user->processes);
1600 return ERR_PTR(retval);
1603 static inline void init_idle_pids(struct pid_link *links)
1607 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1608 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1609 links[type].pid = &init_struct_pid;
1613 struct task_struct * __cpuinit fork_idle(int cpu)
1615 struct task_struct *task;
1616 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1617 if (!IS_ERR(task)) {
1618 init_idle_pids(task->pids);
1619 init_idle(task, cpu);
1626 * Ok, this is the main fork-routine.
1628 * It copies the process, and if successful kick-starts
1629 * it and waits for it to finish using the VM if required.
1631 long do_fork(unsigned long clone_flags,
1632 unsigned long stack_start,
1633 unsigned long stack_size,
1634 int __user *parent_tidptr,
1635 int __user *child_tidptr)
1637 struct task_struct *p;
1642 * Do some preliminary argument and permissions checking before we
1643 * actually start allocating stuff
1645 if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
1646 if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
1651 * Determine whether and which event to report to ptracer. When
1652 * called from kernel_thread or CLONE_UNTRACED is explicitly
1653 * requested, no event is reported; otherwise, report if the event
1654 * for the type of forking is enabled.
1656 if (!(clone_flags & CLONE_UNTRACED)) {
1657 if (clone_flags & CLONE_VFORK)
1658 trace = PTRACE_EVENT_VFORK;
1659 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1660 trace = PTRACE_EVENT_CLONE;
1662 trace = PTRACE_EVENT_FORK;
1664 if (likely(!ptrace_event_enabled(current, trace)))
1668 p = copy_process(clone_flags, stack_start, stack_size,
1669 child_tidptr, NULL, trace);
1671 * Do this prior waking up the new thread - the thread pointer
1672 * might get invalid after that point, if the thread exits quickly.
1675 struct completion vfork;
1678 trace_sched_process_fork(current, p);
1680 pid = get_task_pid(p, PIDTYPE_PID);
1683 if (clone_flags & CLONE_PARENT_SETTID)
1684 put_user(nr, parent_tidptr);
1686 if (clone_flags & CLONE_VFORK) {
1687 p->vfork_done = &vfork;
1688 init_completion(&vfork);
1692 wake_up_new_task(p);
1694 /* forking complete and child started to run, tell ptracer */
1695 if (unlikely(trace))
1696 ptrace_event_pid(trace, pid);
1698 if (clone_flags & CLONE_VFORK) {
1699 if (!wait_for_vfork_done(p, &vfork))
1700 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1711 * Create a kernel thread.
1713 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1715 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1716 (unsigned long)arg, NULL, NULL);
1719 #ifdef __ARCH_WANT_SYS_FORK
1720 SYSCALL_DEFINE0(fork)
1723 return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1725 /* can not support in nommu mode */
1731 #ifdef __ARCH_WANT_SYS_VFORK
1732 SYSCALL_DEFINE0(vfork)
1734 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1739 #ifdef __ARCH_WANT_SYS_CLONE
1740 #ifdef CONFIG_CLONE_BACKWARDS
1741 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1742 int __user *, parent_tidptr,
1744 int __user *, child_tidptr)
1745 #elif defined(CONFIG_CLONE_BACKWARDS2)
1746 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1747 int __user *, parent_tidptr,
1748 int __user *, child_tidptr,
1750 #elif defined(CONFIG_CLONE_BACKWARDS3)
1751 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1753 int __user *, parent_tidptr,
1754 int __user *, child_tidptr,
1757 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1758 int __user *, parent_tidptr,
1759 int __user *, child_tidptr,
1763 return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1767 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1768 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1771 static void sighand_ctor(void *data)
1773 struct sighand_struct *sighand = data;
1775 spin_lock_init(&sighand->siglock);
1776 init_waitqueue_head(&sighand->signalfd_wqh);
1779 void __init proc_caches_init(void)
1781 sighand_cachep = kmem_cache_create("sighand_cache",
1782 sizeof(struct sighand_struct), 0,
1783 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1784 SLAB_NOTRACK, sighand_ctor);
1785 signal_cachep = kmem_cache_create("signal_cache",
1786 sizeof(struct signal_struct), 0,
1787 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1788 files_cachep = kmem_cache_create("files_cache",
1789 sizeof(struct files_struct), 0,
1790 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1791 fs_cachep = kmem_cache_create("fs_cache",
1792 sizeof(struct fs_struct), 0,
1793 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1795 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1796 * whole struct cpumask for the OFFSTACK case. We could change
1797 * this to *only* allocate as much of it as required by the
1798 * maximum number of CPU's we can ever have. The cpumask_allocation
1799 * is at the end of the structure, exactly for that reason.
1801 mm_cachep = kmem_cache_create("mm_struct",
1802 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1803 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1804 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1806 nsproxy_cache_init();
1810 * Check constraints on flags passed to the unshare system call.
1812 static int check_unshare_flags(unsigned long unshare_flags)
1814 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1815 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1816 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1817 CLONE_NEWUSER|CLONE_NEWPID))
1820 * Not implemented, but pretend it works if there is nothing to
1821 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1822 * needs to unshare vm.
1824 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1825 /* FIXME: get_task_mm() increments ->mm_users */
1826 if (atomic_read(¤t->mm->mm_users) > 1)
1834 * Unshare the filesystem structure if it is being shared
1836 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1838 struct fs_struct *fs = current->fs;
1840 if (!(unshare_flags & CLONE_FS) || !fs)
1843 /* don't need lock here; in the worst case we'll do useless copy */
1847 *new_fsp = copy_fs_struct(fs);
1855 * Unshare file descriptor table if it is being shared
1857 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1859 struct files_struct *fd = current->files;
1862 if ((unshare_flags & CLONE_FILES) &&
1863 (fd && atomic_read(&fd->count) > 1)) {
1864 *new_fdp = dup_fd(fd, &error);
1873 * unshare allows a process to 'unshare' part of the process
1874 * context which was originally shared using clone. copy_*
1875 * functions used by do_fork() cannot be used here directly
1876 * because they modify an inactive task_struct that is being
1877 * constructed. Here we are modifying the current, active,
1880 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1882 struct fs_struct *fs, *new_fs = NULL;
1883 struct files_struct *fd, *new_fd = NULL;
1884 struct cred *new_cred = NULL;
1885 struct nsproxy *new_nsproxy = NULL;
1890 * If unsharing a user namespace must also unshare the thread.
1892 if (unshare_flags & CLONE_NEWUSER)
1893 unshare_flags |= CLONE_THREAD | CLONE_FS;
1895 * If unsharing a pid namespace must also unshare the thread.
1897 if (unshare_flags & CLONE_NEWPID)
1898 unshare_flags |= CLONE_THREAD;
1900 * If unsharing a thread from a thread group, must also unshare vm.
1902 if (unshare_flags & CLONE_THREAD)
1903 unshare_flags |= CLONE_VM;
1905 * If unsharing vm, must also unshare signal handlers.
1907 if (unshare_flags & CLONE_VM)
1908 unshare_flags |= CLONE_SIGHAND;
1910 * If unsharing namespace, must also unshare filesystem information.
1912 if (unshare_flags & CLONE_NEWNS)
1913 unshare_flags |= CLONE_FS;
1915 err = check_unshare_flags(unshare_flags);
1917 goto bad_unshare_out;
1919 * CLONE_NEWIPC must also detach from the undolist: after switching
1920 * to a new ipc namespace, the semaphore arrays from the old
1921 * namespace are unreachable.
1923 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1925 err = unshare_fs(unshare_flags, &new_fs);
1927 goto bad_unshare_out;
1928 err = unshare_fd(unshare_flags, &new_fd);
1930 goto bad_unshare_cleanup_fs;
1931 err = unshare_userns(unshare_flags, &new_cred);
1933 goto bad_unshare_cleanup_fd;
1934 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1937 goto bad_unshare_cleanup_cred;
1939 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1942 * CLONE_SYSVSEM is equivalent to sys_exit().
1948 switch_task_namespaces(current, new_nsproxy);
1954 spin_lock(&fs->lock);
1955 current->fs = new_fs;
1960 spin_unlock(&fs->lock);
1964 fd = current->files;
1965 current->files = new_fd;
1969 task_unlock(current);
1972 /* Install the new user namespace */
1973 commit_creds(new_cred);
1978 bad_unshare_cleanup_cred:
1981 bad_unshare_cleanup_fd:
1983 put_files_struct(new_fd);
1985 bad_unshare_cleanup_fs:
1987 free_fs_struct(new_fs);
1994 * Helper to unshare the files of the current task.
1995 * We don't want to expose copy_files internals to
1996 * the exec layer of the kernel.
1999 int unshare_files(struct files_struct **displaced)
2001 struct task_struct *task = current;
2002 struct files_struct *copy = NULL;
2005 error = unshare_fd(CLONE_FILES, ©);
2006 if (error || !copy) {
2010 *displaced = task->files;