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>
30 #include <linux/hmm.h>
32 #include <linux/nsproxy.h>
33 #include <linux/capability.h>
34 #include <linux/cpu.h>
35 #include <linux/cgroup.h>
36 #include <linux/security.h>
37 #include <linux/hugetlb.h>
38 #include <linux/seccomp.h>
39 #include <linux/swap.h>
40 #include <linux/syscalls.h>
41 #include <linux/jiffies.h>
42 #include <linux/futex.h>
43 #include <linux/compat.h>
44 #include <linux/kthread.h>
45 #include <linux/task_io_accounting_ops.h>
46 #include <linux/rcupdate.h>
47 #include <linux/ptrace.h>
48 #include <linux/mount.h>
49 #include <linux/audit.h>
50 #include <linux/memcontrol.h>
51 #include <linux/ftrace.h>
52 #include <linux/proc_fs.h>
53 #include <linux/profile.h>
54 #include <linux/rmap.h>
55 #include <linux/ksm.h>
56 #include <linux/acct.h>
57 #include <linux/tsacct_kern.h>
58 #include <linux/cn_proc.h>
59 #include <linux/freezer.h>
60 #include <linux/delayacct.h>
61 #include <linux/taskstats_kern.h>
62 #include <linux/random.h>
63 #include <linux/tty.h>
64 #include <linux/blkdev.h>
65 #include <linux/fs_struct.h>
66 #include <linux/magic.h>
67 #include <linux/perf_event.h>
68 #include <linux/posix-timers.h>
69 #include <linux/user-return-notifier.h>
70 #include <linux/oom.h>
71 #include <linux/khugepaged.h>
72 #include <linux/signalfd.h>
73 #include <linux/uprobes.h>
74 #include <linux/aio.h>
76 #include <asm/pgtable.h>
77 #include <asm/pgalloc.h>
78 #include <asm/uaccess.h>
79 #include <asm/mmu_context.h>
80 #include <asm/cacheflush.h>
81 #include <asm/tlbflush.h>
83 #include <trace/events/sched.h>
84 #include <trace/events/sys_calls.h>
86 #define CREATE_TRACE_POINTS
87 #include <trace/events/task.h>
90 * Protected counters by write_lock_irq(&tasklist_lock)
92 unsigned long total_forks; /* Handle normal Linux uptimes. */
93 int nr_threads; /* The idle threads do not count.. */
95 int max_threads; /* tunable limit on nr_threads */
97 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
99 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
101 #ifdef CONFIG_PROVE_RCU
102 int lockdep_tasklist_lock_is_held(void)
104 return lockdep_is_held(&tasklist_lock);
106 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
107 #endif /* #ifdef CONFIG_PROVE_RCU */
109 int nr_processes(void)
114 for_each_possible_cpu(cpu)
115 total += per_cpu(process_counts, cpu);
120 void __weak arch_release_task_struct(struct task_struct *tsk)
124 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
125 static struct kmem_cache *task_struct_cachep;
127 static inline struct task_struct *alloc_task_struct_node(int node)
129 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
132 static inline void free_task_struct(struct task_struct *tsk)
134 kmem_cache_free(task_struct_cachep, tsk);
138 void __weak arch_release_thread_info(struct thread_info *ti)
142 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
145 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
146 * kmemcache based allocator.
148 # if THREAD_SIZE >= PAGE_SIZE
149 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
152 struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
155 return page ? page_address(page) : NULL;
158 static inline void free_thread_info(struct thread_info *ti)
160 free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
163 static struct kmem_cache *thread_info_cache;
165 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
168 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
171 static void free_thread_info(struct thread_info *ti)
173 kmem_cache_free(thread_info_cache, ti);
176 void thread_info_cache_init(void)
178 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
179 THREAD_SIZE, 0, NULL);
180 BUG_ON(thread_info_cache == NULL);
185 /* SLAB cache for signal_struct structures (tsk->signal) */
186 static struct kmem_cache *signal_cachep;
188 /* SLAB cache for sighand_struct structures (tsk->sighand) */
189 struct kmem_cache *sighand_cachep;
191 /* SLAB cache for files_struct structures (tsk->files) */
192 struct kmem_cache *files_cachep;
194 /* SLAB cache for fs_struct structures (tsk->fs) */
195 struct kmem_cache *fs_cachep;
197 /* SLAB cache for vm_area_struct structures */
198 struct kmem_cache *vm_area_cachep;
200 /* SLAB cache for mm_struct structures (tsk->mm) */
201 static struct kmem_cache *mm_cachep;
203 /* Notifier list called when a task struct is freed */
204 static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
206 static void account_kernel_stack(struct thread_info *ti, int account)
208 struct zone *zone = page_zone(virt_to_page(ti));
210 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
213 void free_task(struct task_struct *tsk)
215 account_kernel_stack(tsk->stack, -1);
216 arch_release_thread_info(tsk->stack);
217 free_thread_info(tsk->stack);
218 rt_mutex_debug_task_free(tsk);
219 ftrace_graph_exit_task(tsk);
220 put_seccomp_filter(tsk);
221 arch_release_task_struct(tsk);
222 free_task_struct(tsk);
224 EXPORT_SYMBOL(free_task);
226 static inline void free_signal_struct(struct signal_struct *sig)
228 taskstats_tgid_free(sig);
229 sched_autogroup_exit(sig);
230 kmem_cache_free(signal_cachep, sig);
233 static inline void put_signal_struct(struct signal_struct *sig)
235 if (atomic_dec_and_test(&sig->sigcnt))
236 free_signal_struct(sig);
239 int task_free_register(struct notifier_block *n)
241 return atomic_notifier_chain_register(&task_free_notifier, n);
243 EXPORT_SYMBOL(task_free_register);
245 int task_free_unregister(struct notifier_block *n)
247 return atomic_notifier_chain_unregister(&task_free_notifier, n);
249 EXPORT_SYMBOL(task_free_unregister);
251 void __put_task_struct(struct task_struct *tsk)
253 WARN_ON(!tsk->exit_state);
254 WARN_ON(atomic_read(&tsk->usage));
255 WARN_ON(tsk == current);
257 security_task_free(tsk);
259 delayacct_tsk_free(tsk);
260 put_signal_struct(tsk->signal);
262 atomic_notifier_call_chain(&task_free_notifier, 0, tsk);
263 if (!profile_handoff_task(tsk))
266 EXPORT_SYMBOL_GPL(__put_task_struct);
268 void __init __weak arch_task_cache_init(void) { }
270 void __init fork_init(unsigned long mempages)
272 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
273 #ifndef ARCH_MIN_TASKALIGN
274 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
276 /* create a slab on which task_structs can be allocated */
278 kmem_cache_create("task_struct", sizeof(struct task_struct),
279 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
282 /* do the arch specific task caches init */
283 arch_task_cache_init();
286 * The default maximum number of threads is set to a safe
287 * value: the thread structures can take up at most half
290 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
293 * we need to allow at least 20 threads to boot a system
295 if (max_threads < 20)
298 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
299 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
300 init_task.signal->rlim[RLIMIT_SIGPENDING] =
301 init_task.signal->rlim[RLIMIT_NPROC];
304 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
305 struct task_struct *src)
311 static struct task_struct *dup_task_struct(struct task_struct *orig)
313 struct task_struct *tsk;
314 struct thread_info *ti;
315 unsigned long *stackend;
316 int node = tsk_fork_get_node(orig);
319 tsk = alloc_task_struct_node(node);
323 ti = alloc_thread_info_node(tsk, node);
327 err = arch_dup_task_struct(tsk, orig);
332 #ifdef CONFIG_SECCOMP
334 * We must handle setting up seccomp filters once we're under
335 * the sighand lock in case orig has changed between now and
336 * then. Until then, filter must be NULL to avoid messing up
337 * the usage counts on the error path calling free_task.
339 tsk->seccomp.filter = NULL;
342 setup_thread_stack(tsk, orig);
343 clear_user_return_notifier(tsk);
344 clear_tsk_need_resched(tsk);
345 stackend = end_of_stack(tsk);
346 *stackend = STACK_END_MAGIC; /* for overflow detection */
348 #ifdef CONFIG_CC_STACKPROTECTOR
349 tsk->stack_canary = get_random_int();
353 * One for us, one for whoever does the "release_task()" (usually
356 atomic_set(&tsk->usage, 2);
357 #ifdef CONFIG_BLK_DEV_IO_TRACE
360 tsk->splice_pipe = NULL;
361 tsk->task_frag.page = NULL;
363 account_kernel_stack(ti, 1);
368 free_thread_info(ti);
370 free_task_struct(tsk);
375 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
377 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
378 struct rb_node **rb_link, *rb_parent;
380 unsigned long charge;
381 struct mempolicy *pol;
383 uprobe_start_dup_mmap();
384 down_write(&oldmm->mmap_sem);
385 flush_cache_dup_mm(oldmm);
386 uprobe_dup_mmap(oldmm, mm);
388 * Not linked in yet - no deadlock potential:
390 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
394 mm->mmap_cache = NULL;
395 mm->free_area_cache = oldmm->mmap_base;
396 mm->cached_hole_size = ~0UL;
398 cpumask_clear(mm_cpumask(mm));
400 rb_link = &mm->mm_rb.rb_node;
403 retval = ksm_fork(mm, oldmm);
406 retval = khugepaged_fork(mm, oldmm);
411 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
414 if (mpnt->vm_flags & VM_DONTCOPY) {
415 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
420 if (mpnt->vm_flags & VM_ACCOUNT) {
421 unsigned long len = vma_pages(mpnt);
423 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
427 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
431 INIT_LIST_HEAD(&tmp->anon_vma_chain);
432 pol = mpol_dup(vma_policy(mpnt));
433 retval = PTR_ERR(pol);
435 goto fail_nomem_policy;
436 vma_set_policy(tmp, pol);
438 if (anon_vma_fork(tmp, mpnt))
439 goto fail_nomem_anon_vma_fork;
440 tmp->vm_flags &= ~VM_LOCKED;
441 tmp->vm_next = tmp->vm_prev = NULL;
444 struct inode *inode = file_inode(file);
445 struct address_space *mapping = file->f_mapping;
448 if (tmp->vm_flags & VM_DENYWRITE)
449 atomic_dec(&inode->i_writecount);
450 mutex_lock(&mapping->i_mmap_mutex);
451 if (tmp->vm_flags & VM_SHARED)
452 mapping->i_mmap_writable++;
453 flush_dcache_mmap_lock(mapping);
454 /* insert tmp into the share list, just after mpnt */
455 if (unlikely(tmp->vm_flags & VM_NONLINEAR))
456 vma_nonlinear_insert(tmp,
457 &mapping->i_mmap_nonlinear);
459 vma_interval_tree_insert_after(tmp, mpnt,
461 flush_dcache_mmap_unlock(mapping);
462 mutex_unlock(&mapping->i_mmap_mutex);
466 * Clear hugetlb-related page reserves for children. This only
467 * affects MAP_PRIVATE mappings. Faults generated by the child
468 * are not guaranteed to succeed, even if read-only
470 if (is_vm_hugetlb_page(tmp))
471 reset_vma_resv_huge_pages(tmp);
474 * Link in the new vma and copy the page table entries.
477 pprev = &tmp->vm_next;
481 __vma_link_rb(mm, tmp, rb_link, rb_parent);
482 rb_link = &tmp->vm_rb.rb_right;
483 rb_parent = &tmp->vm_rb;
486 retval = copy_page_range(mm, oldmm, mpnt);
488 if (tmp->vm_ops && tmp->vm_ops->open)
489 tmp->vm_ops->open(tmp);
494 /* a new mm has just been created */
495 arch_dup_mmap(oldmm, mm);
498 up_write(&mm->mmap_sem);
500 up_write(&oldmm->mmap_sem);
501 uprobe_end_dup_mmap();
503 fail_nomem_anon_vma_fork:
506 kmem_cache_free(vm_area_cachep, tmp);
509 vm_unacct_memory(charge);
513 static inline int mm_alloc_pgd(struct mm_struct *mm)
515 mm->pgd = pgd_alloc(mm);
516 if (unlikely(!mm->pgd))
521 static inline void mm_free_pgd(struct mm_struct *mm)
523 pgd_free(mm, mm->pgd);
526 #define dup_mmap(mm, oldmm) (0)
527 #define mm_alloc_pgd(mm) (0)
528 #define mm_free_pgd(mm)
529 #endif /* CONFIG_MMU */
531 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
533 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
534 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
536 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
538 static int __init coredump_filter_setup(char *s)
540 default_dump_filter =
541 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
542 MMF_DUMP_FILTER_MASK;
546 __setup("coredump_filter=", coredump_filter_setup);
548 #include <linux/init_task.h>
550 static void mm_init_aio(struct mm_struct *mm)
553 spin_lock_init(&mm->ioctx_lock);
554 INIT_HLIST_HEAD(&mm->ioctx_list);
558 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
560 atomic_set(&mm->mm_users, 1);
561 atomic_set(&mm->mm_count, 1);
562 init_rwsem(&mm->mmap_sem);
563 INIT_LIST_HEAD(&mm->mmlist);
564 mm->flags = (current->mm) ?
565 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
566 mm->core_state = NULL;
568 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
569 spin_lock_init(&mm->page_table_lock);
570 mm->free_area_cache = TASK_UNMAPPED_BASE;
571 mm->cached_hole_size = ~0UL;
573 mm_init_owner(mm, p);
575 clear_tlb_flush_pending(mm);
577 if (likely(!mm_alloc_pgd(mm))) {
579 mmu_notifier_mm_init(mm);
587 static void check_mm(struct mm_struct *mm)
591 for (i = 0; i < NR_MM_COUNTERS; i++) {
592 long x = atomic_long_read(&mm->rss_stat.count[i]);
595 printk(KERN_ALERT "BUG: Bad rss-counter state "
596 "mm:%p idx:%d val:%ld\n", mm, i, x);
599 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
600 VM_BUG_ON(mm->pmd_huge_pte);
605 * Allocate and initialize an mm_struct.
607 struct mm_struct *mm_alloc(void)
609 struct mm_struct *mm;
615 memset(mm, 0, sizeof(*mm));
617 return mm_init(mm, current);
621 * Called when the last reference to the mm
622 * is dropped: either by a lazy thread or by
623 * mmput. Free the page directory and the mm.
625 void __mmdrop(struct mm_struct *mm)
627 BUG_ON(mm == &init_mm);
630 mmu_notifier_mm_destroy(mm);
634 EXPORT_SYMBOL_GPL(__mmdrop);
637 * Decrement the use count and release all resources for an mm.
639 void mmput(struct mm_struct *mm)
643 if (atomic_dec_and_test(&mm->mm_users)) {
644 uprobe_clear_state(mm);
647 khugepaged_exit(mm); /* must run before exit_mmap */
649 set_mm_exe_file(mm, NULL);
650 if (!list_empty(&mm->mmlist)) {
651 spin_lock(&mmlist_lock);
652 list_del(&mm->mmlist);
653 spin_unlock(&mmlist_lock);
656 module_put(mm->binfmt->module);
660 EXPORT_SYMBOL_GPL(mmput);
662 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
665 get_file(new_exe_file);
668 mm->exe_file = new_exe_file;
671 struct file *get_mm_exe_file(struct mm_struct *mm)
673 struct file *exe_file;
675 /* We need mmap_sem to protect against races with removal of exe_file */
676 down_read(&mm->mmap_sem);
677 exe_file = mm->exe_file;
680 up_read(&mm->mmap_sem);
684 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
686 /* It's safe to write the exe_file pointer without exe_file_lock because
687 * this is called during fork when the task is not yet in /proc */
688 newmm->exe_file = get_mm_exe_file(oldmm);
692 * get_task_mm - acquire a reference to the task's mm
694 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
695 * this kernel workthread has transiently adopted a user mm with use_mm,
696 * to do its AIO) is not set and if so returns a reference to it, after
697 * bumping up the use count. User must release the mm via mmput()
698 * after use. Typically used by /proc and ptrace.
700 struct mm_struct *get_task_mm(struct task_struct *task)
702 struct mm_struct *mm;
707 if (task->flags & PF_KTHREAD)
710 atomic_inc(&mm->mm_users);
715 EXPORT_SYMBOL_GPL(get_task_mm);
717 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
719 struct mm_struct *mm;
722 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
726 mm = get_task_mm(task);
727 if (mm && mm != current->mm &&
728 !ptrace_may_access(task, mode) &&
729 !capable(CAP_SYS_RESOURCE)) {
731 mm = ERR_PTR(-EACCES);
733 mutex_unlock(&task->signal->cred_guard_mutex);
738 static void complete_vfork_done(struct task_struct *tsk)
740 struct completion *vfork;
743 vfork = tsk->vfork_done;
745 tsk->vfork_done = NULL;
751 static int wait_for_vfork_done(struct task_struct *child,
752 struct completion *vfork)
756 freezer_do_not_count();
757 killed = wait_for_completion_killable(vfork);
762 child->vfork_done = NULL;
766 put_task_struct(child);
770 /* Please note the differences between mmput and mm_release.
771 * mmput is called whenever we stop holding onto a mm_struct,
772 * error success whatever.
774 * mm_release is called after a mm_struct has been removed
775 * from the current process.
777 * This difference is important for error handling, when we
778 * only half set up a mm_struct for a new process and need to restore
779 * the old one. Because we mmput the new mm_struct before
780 * restoring the old one. . .
781 * Eric Biederman 10 January 1998
783 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
785 /* Get rid of any futexes when releasing the mm */
787 if (unlikely(tsk->robust_list)) {
788 exit_robust_list(tsk);
789 tsk->robust_list = NULL;
792 if (unlikely(tsk->compat_robust_list)) {
793 compat_exit_robust_list(tsk);
794 tsk->compat_robust_list = NULL;
797 if (unlikely(!list_empty(&tsk->pi_state_list)))
798 exit_pi_state_list(tsk);
801 uprobe_free_utask(tsk);
803 /* Get rid of any cached register state */
804 deactivate_mm(tsk, mm);
807 * If we're exiting normally, clear a user-space tid field if
808 * requested. We leave this alone when dying by signal, to leave
809 * the value intact in a core dump, and to save the unnecessary
810 * trouble, say, a killed vfork parent shouldn't touch this mm.
811 * Userland only wants this done for a sys_exit.
813 if (tsk->clear_child_tid) {
814 if (!(tsk->flags & PF_SIGNALED) &&
815 atomic_read(&mm->mm_users) > 1) {
817 * We don't check the error code - if userspace has
818 * not set up a proper pointer then tough luck.
820 put_user(0, tsk->clear_child_tid);
821 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
824 tsk->clear_child_tid = NULL;
828 * All done, finally we can wake up parent and return this mm to him.
829 * Also kthread_stop() uses this completion for synchronization.
832 complete_vfork_done(tsk);
836 * Allocate a new mm structure and copy contents from the
837 * mm structure of the passed in task structure.
839 struct mm_struct *dup_mm(struct task_struct *tsk)
841 struct mm_struct *mm, *oldmm = current->mm;
851 memcpy(mm, oldmm, sizeof(*mm));
854 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
855 mm->pmd_huge_pte = NULL;
857 #ifdef CONFIG_NUMA_BALANCING
858 mm->first_nid = NUMA_PTE_SCAN_INIT;
860 if (!mm_init(mm, tsk))
863 if (init_new_context(tsk, mm))
866 dup_mm_exe_file(oldmm, mm);
868 err = dup_mmap(mm, oldmm);
872 mm->hiwater_rss = get_mm_rss(mm);
873 mm->hiwater_vm = mm->total_vm;
875 if (mm->binfmt && !try_module_get(mm->binfmt->module))
881 /* don't put binfmt in mmput, we haven't got module yet */
890 * If init_new_context() failed, we cannot use mmput() to free the mm
891 * because it calls destroy_context()
898 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
900 struct mm_struct *mm, *oldmm;
903 tsk->min_flt = tsk->maj_flt = 0;
904 tsk->nvcsw = tsk->nivcsw = 0;
905 #ifdef CONFIG_DETECT_HUNG_TASK
906 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
910 tsk->active_mm = NULL;
913 * Are we cloning a kernel thread?
915 * We need to steal a active VM for that..
921 if (clone_flags & CLONE_VM) {
922 atomic_inc(&oldmm->mm_users);
941 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
943 struct fs_struct *fs = current->fs;
944 if (clone_flags & CLONE_FS) {
945 /* tsk->fs is already what we want */
946 spin_lock(&fs->lock);
948 spin_unlock(&fs->lock);
952 spin_unlock(&fs->lock);
955 tsk->fs = copy_fs_struct(fs);
961 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
963 struct files_struct *oldf, *newf;
967 * A background process may not have any files ...
969 oldf = current->files;
973 if (clone_flags & CLONE_FILES) {
974 atomic_inc(&oldf->count);
978 newf = dup_fd(oldf, &error);
988 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
991 struct io_context *ioc = current->io_context;
992 struct io_context *new_ioc;
997 * Share io context with parent, if CLONE_IO is set
999 if (clone_flags & CLONE_IO) {
1001 tsk->io_context = ioc;
1002 } else if (ioprio_valid(ioc->ioprio)) {
1003 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1004 if (unlikely(!new_ioc))
1007 new_ioc->ioprio = ioc->ioprio;
1008 put_io_context(new_ioc);
1014 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1016 struct sighand_struct *sig;
1018 if (clone_flags & CLONE_SIGHAND) {
1019 atomic_inc(¤t->sighand->count);
1022 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1023 rcu_assign_pointer(tsk->sighand, sig);
1026 atomic_set(&sig->count, 1);
1027 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1031 void __cleanup_sighand(struct sighand_struct *sighand)
1033 if (atomic_dec_and_test(&sighand->count)) {
1034 signalfd_cleanup(sighand);
1035 kmem_cache_free(sighand_cachep, sighand);
1041 * Initialize POSIX timer handling for a thread group.
1043 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1045 unsigned long cpu_limit;
1047 /* Thread group counters. */
1048 thread_group_cputime_init(sig);
1050 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1051 if (cpu_limit != RLIM_INFINITY) {
1052 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1053 sig->cputimer.running = 1;
1056 /* The timer lists. */
1057 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1058 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1059 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1062 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1064 struct signal_struct *sig;
1066 if (clone_flags & CLONE_THREAD)
1069 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1074 sig->nr_threads = 1;
1075 atomic_set(&sig->live, 1);
1076 atomic_set(&sig->sigcnt, 1);
1078 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1079 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1080 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1082 init_waitqueue_head(&sig->wait_chldexit);
1083 sig->curr_target = tsk;
1084 init_sigpending(&sig->shared_pending);
1085 INIT_LIST_HEAD(&sig->posix_timers);
1087 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1088 sig->real_timer.function = it_real_fn;
1090 task_lock(current->group_leader);
1091 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1092 task_unlock(current->group_leader);
1094 posix_cpu_timers_init_group(sig);
1096 tty_audit_fork(sig);
1097 sched_autogroup_fork(sig);
1099 #ifdef CONFIG_CGROUPS
1100 init_rwsem(&sig->group_rwsem);
1103 sig->oom_score_adj = current->signal->oom_score_adj;
1104 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1106 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1107 current->signal->is_child_subreaper;
1109 mutex_init(&sig->cred_guard_mutex);
1114 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1116 unsigned long new_flags = p->flags;
1118 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1119 new_flags |= PF_FORKNOEXEC;
1120 p->flags = new_flags;
1123 static void copy_seccomp(struct task_struct *p)
1125 #ifdef CONFIG_SECCOMP
1127 * Must be called with sighand->lock held, which is common to
1128 * all threads in the group. Holding cred_guard_mutex is not
1129 * needed because this new task is not yet running and cannot
1132 assert_spin_locked(¤t->sighand->siglock);
1134 /* Ref-count the new filter user, and assign it. */
1135 get_seccomp_filter(current);
1136 p->seccomp = current->seccomp;
1139 * Explicitly enable no_new_privs here in case it got set
1140 * between the task_struct being duplicated and holding the
1141 * sighand lock. The seccomp state and nnp must be in sync.
1143 if (task_no_new_privs(current))
1144 task_set_no_new_privs(p);
1147 * If the parent gained a seccomp mode after copying thread
1148 * flags and between before we held the sighand lock, we have
1149 * to manually enable the seccomp thread flag here.
1151 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1152 set_tsk_thread_flag(p, TIF_SECCOMP);
1156 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1158 current->clear_child_tid = tidptr;
1160 return task_pid_vnr(current);
1163 static void rt_mutex_init_task(struct task_struct *p)
1165 raw_spin_lock_init(&p->pi_lock);
1166 #ifdef CONFIG_RT_MUTEXES
1167 plist_head_init(&p->pi_waiters);
1168 p->pi_blocked_on = NULL;
1172 #ifdef CONFIG_MM_OWNER
1173 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1177 #endif /* CONFIG_MM_OWNER */
1180 * Initialize POSIX timer handling for a single task.
1182 static void posix_cpu_timers_init(struct task_struct *tsk)
1184 tsk->cputime_expires.prof_exp = 0;
1185 tsk->cputime_expires.virt_exp = 0;
1186 tsk->cputime_expires.sched_exp = 0;
1187 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1188 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1189 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1193 * This creates a new process as a copy of the old one,
1194 * but does not actually start it yet.
1196 * It copies the registers, and all the appropriate
1197 * parts of the process environment (as per the clone
1198 * flags). The actual kick-off is left to the caller.
1200 static struct task_struct *copy_process(unsigned long clone_flags,
1201 unsigned long stack_start,
1202 unsigned long stack_size,
1203 int __user *child_tidptr,
1208 struct task_struct *p;
1210 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1211 return ERR_PTR(-EINVAL);
1213 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1214 return ERR_PTR(-EINVAL);
1217 * Thread groups must share signals as well, and detached threads
1218 * can only be started up within the thread group.
1220 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1221 return ERR_PTR(-EINVAL);
1224 * Shared signal handlers imply shared VM. By way of the above,
1225 * thread groups also imply shared VM. Blocking this case allows
1226 * for various simplifications in other code.
1228 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1229 return ERR_PTR(-EINVAL);
1232 * Siblings of global init remain as zombies on exit since they are
1233 * not reaped by their parent (swapper). To solve this and to avoid
1234 * multi-rooted process trees, prevent global and container-inits
1235 * from creating siblings.
1237 if ((clone_flags & CLONE_PARENT) &&
1238 current->signal->flags & SIGNAL_UNKILLABLE)
1239 return ERR_PTR(-EINVAL);
1242 * If the new process will be in a different pid namespace don't
1243 * allow it to share a thread group or signal handlers with the
1246 if ((clone_flags & (CLONE_SIGHAND | CLONE_NEWPID)) &&
1247 (task_active_pid_ns(current) != current->nsproxy->pid_ns))
1248 return ERR_PTR(-EINVAL);
1250 retval = security_task_create(clone_flags);
1255 p = dup_task_struct(current);
1259 ftrace_graph_init_task(p);
1261 rt_mutex_init_task(p);
1263 #ifdef CONFIG_PROVE_LOCKING
1264 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1265 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1268 if (atomic_read(&p->real_cred->user->processes) >=
1269 task_rlimit(p, RLIMIT_NPROC)) {
1270 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1271 p->real_cred->user != INIT_USER)
1274 current->flags &= ~PF_NPROC_EXCEEDED;
1276 retval = copy_creds(p, clone_flags);
1281 * If multiple threads are within copy_process(), then this check
1282 * triggers too late. This doesn't hurt, the check is only there
1283 * to stop root fork bombs.
1286 if (nr_threads >= max_threads)
1287 goto bad_fork_cleanup_count;
1290 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1291 copy_flags(clone_flags, p);
1292 INIT_LIST_HEAD(&p->children);
1293 INIT_LIST_HEAD(&p->sibling);
1294 rcu_copy_process(p);
1295 p->vfork_done = NULL;
1296 spin_lock_init(&p->alloc_lock);
1298 init_sigpending(&p->pending);
1300 p->utime = p->stime = p->gtime = 0;
1301 p->utimescaled = p->stimescaled = 0;
1302 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1303 p->prev_cputime.utime = p->prev_cputime.stime = 0;
1305 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1306 seqlock_init(&p->vtime_seqlock);
1308 p->vtime_snap_whence = VTIME_SLEEPING;
1311 #if defined(SPLIT_RSS_COUNTING)
1312 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1315 p->default_timer_slack_ns = current->timer_slack_ns;
1317 task_io_accounting_init(&p->ioac);
1318 acct_clear_integrals(p);
1320 posix_cpu_timers_init(p);
1322 do_posix_clock_monotonic_gettime(&p->start_time);
1323 p->real_start_time = p->start_time;
1324 monotonic_to_bootbased(&p->real_start_time);
1325 p->io_context = NULL;
1326 p->audit_context = NULL;
1327 if (clone_flags & CLONE_THREAD)
1328 threadgroup_change_begin(current);
1331 p->mempolicy = mpol_dup(p->mempolicy);
1332 if (IS_ERR(p->mempolicy)) {
1333 retval = PTR_ERR(p->mempolicy);
1334 p->mempolicy = NULL;
1335 goto bad_fork_cleanup_cgroup;
1337 mpol_fix_fork_child_flag(p);
1339 #ifdef CONFIG_CPUSETS
1340 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1341 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1342 seqcount_init(&p->mems_allowed_seq);
1344 #ifdef CONFIG_TRACE_IRQFLAGS
1346 p->hardirqs_enabled = 0;
1347 p->hardirq_enable_ip = 0;
1348 p->hardirq_enable_event = 0;
1349 p->hardirq_disable_ip = _THIS_IP_;
1350 p->hardirq_disable_event = 0;
1351 p->softirqs_enabled = 1;
1352 p->softirq_enable_ip = _THIS_IP_;
1353 p->softirq_enable_event = 0;
1354 p->softirq_disable_ip = 0;
1355 p->softirq_disable_event = 0;
1356 p->hardirq_context = 0;
1357 p->softirq_context = 0;
1359 #ifdef CONFIG_LOCKDEP
1360 p->lockdep_depth = 0; /* no locks held yet */
1361 p->curr_chain_key = 0;
1362 p->lockdep_recursion = 0;
1365 #ifdef CONFIG_DEBUG_MUTEXES
1366 p->blocked_on = NULL; /* not blocked yet */
1369 p->memcg_batch.do_batch = 0;
1370 p->memcg_batch.memcg = NULL;
1372 #ifdef CONFIG_BCACHE
1373 p->sequential_io = 0;
1374 p->sequential_io_avg = 0;
1377 /* Perform scheduler related setup. Assign this task to a CPU. */
1380 retval = perf_event_init_task(p);
1382 goto bad_fork_cleanup_policy;
1383 retval = audit_alloc(p);
1385 goto bad_fork_cleanup_perf;
1386 /* copy all the process information */
1387 retval = copy_semundo(clone_flags, p);
1389 goto bad_fork_cleanup_audit;
1390 retval = copy_files(clone_flags, p);
1392 goto bad_fork_cleanup_semundo;
1393 retval = copy_fs(clone_flags, p);
1395 goto bad_fork_cleanup_files;
1396 retval = copy_sighand(clone_flags, p);
1398 goto bad_fork_cleanup_fs;
1399 retval = copy_signal(clone_flags, p);
1401 goto bad_fork_cleanup_sighand;
1402 retval = copy_mm(clone_flags, p);
1404 goto bad_fork_cleanup_signal;
1405 retval = copy_namespaces(clone_flags, p);
1407 goto bad_fork_cleanup_mm;
1408 retval = copy_io(clone_flags, p);
1410 goto bad_fork_cleanup_namespaces;
1411 retval = copy_thread(clone_flags, stack_start, stack_size, p);
1413 goto bad_fork_cleanup_io;
1415 if (pid != &init_struct_pid) {
1417 pid = alloc_pid(p->nsproxy->pid_ns);
1419 goto bad_fork_cleanup_io;
1422 p->pid = pid_nr(pid);
1424 if (clone_flags & CLONE_THREAD)
1425 p->tgid = current->tgid;
1427 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1429 * Clear TID on mm_release()?
1431 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1436 p->robust_list = NULL;
1437 #ifdef CONFIG_COMPAT
1438 p->compat_robust_list = NULL;
1440 INIT_LIST_HEAD(&p->pi_state_list);
1441 p->pi_state_cache = NULL;
1443 uprobe_copy_process(p);
1445 * sigaltstack should be cleared when sharing the same VM
1447 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1448 p->sas_ss_sp = p->sas_ss_size = 0;
1451 * Syscall tracing and stepping should be turned off in the
1452 * child regardless of CLONE_PTRACE.
1454 user_disable_single_step(p);
1455 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1456 #ifdef TIF_SYSCALL_EMU
1457 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1459 clear_all_latency_tracing(p);
1461 /* ok, now we should be set up.. */
1462 if (clone_flags & CLONE_THREAD)
1463 p->exit_signal = -1;
1464 else if (clone_flags & CLONE_PARENT)
1465 p->exit_signal = current->group_leader->exit_signal;
1467 p->exit_signal = (clone_flags & CSIGNAL);
1469 p->pdeath_signal = 0;
1473 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1474 p->dirty_paused_when = 0;
1477 * Ok, make it visible to the rest of the system.
1478 * We dont wake it up yet.
1480 p->group_leader = p;
1481 INIT_LIST_HEAD(&p->thread_group);
1482 p->task_works = NULL;
1484 /* Need tasklist lock for parent etc handling! */
1485 write_lock_irq(&tasklist_lock);
1487 /* CLONE_PARENT re-uses the old parent */
1488 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1489 p->real_parent = current->real_parent;
1490 p->parent_exec_id = current->parent_exec_id;
1492 p->real_parent = current;
1493 p->parent_exec_id = current->self_exec_id;
1496 spin_lock(¤t->sighand->siglock);
1499 * Copy seccomp details explicitly here, in case they were changed
1500 * before holding sighand lock.
1505 * Process group and session signals need to be delivered to just the
1506 * parent before the fork or both the parent and the child after the
1507 * fork. Restart if a signal comes in before we add the new process to
1508 * it's process group.
1509 * A fatal signal pending means that current will exit, so the new
1510 * thread can't slip out of an OOM kill (or normal SIGKILL).
1512 recalc_sigpending();
1513 if (signal_pending(current)) {
1514 spin_unlock(¤t->sighand->siglock);
1515 write_unlock_irq(&tasklist_lock);
1516 retval = -ERESTARTNOINTR;
1517 goto bad_fork_free_pid;
1520 if (likely(p->pid)) {
1521 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1523 if (thread_group_leader(p)) {
1524 if (is_child_reaper(pid)) {
1525 ns_of_pid(pid)->child_reaper = p;
1526 p->signal->flags |= SIGNAL_UNKILLABLE;
1529 p->signal->leader_pid = pid;
1530 p->signal->tty = tty_kref_get(current->signal->tty);
1531 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1532 attach_pid(p, PIDTYPE_SID, task_session(current));
1533 list_add_tail(&p->sibling, &p->real_parent->children);
1534 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1535 __this_cpu_inc(process_counts);
1537 current->signal->nr_threads++;
1538 atomic_inc(¤t->signal->live);
1539 atomic_inc(¤t->signal->sigcnt);
1540 p->group_leader = current->group_leader;
1541 list_add_tail_rcu(&p->thread_group,
1542 &p->group_leader->thread_group);
1543 list_add_tail_rcu(&p->thread_node,
1544 &p->signal->thread_head);
1546 attach_pid(p, PIDTYPE_PID, pid);
1551 spin_unlock(¤t->sighand->siglock);
1552 syscall_tracepoint_update(p);
1553 write_unlock_irq(&tasklist_lock);
1555 proc_fork_connector(p);
1556 cgroup_post_fork(p);
1557 if (clone_flags & CLONE_THREAD)
1558 threadgroup_change_end(current);
1561 trace_task_newtask(p, clone_flags);
1566 if (pid != &init_struct_pid)
1568 bad_fork_cleanup_io:
1571 bad_fork_cleanup_namespaces:
1572 exit_task_namespaces(p);
1573 bad_fork_cleanup_mm:
1576 bad_fork_cleanup_signal:
1577 if (!(clone_flags & CLONE_THREAD))
1578 free_signal_struct(p->signal);
1579 bad_fork_cleanup_sighand:
1580 __cleanup_sighand(p->sighand);
1581 bad_fork_cleanup_fs:
1582 exit_fs(p); /* blocking */
1583 bad_fork_cleanup_files:
1584 exit_files(p); /* blocking */
1585 bad_fork_cleanup_semundo:
1587 bad_fork_cleanup_audit:
1589 bad_fork_cleanup_perf:
1590 perf_event_free_task(p);
1591 bad_fork_cleanup_policy:
1593 mpol_put(p->mempolicy);
1594 bad_fork_cleanup_cgroup:
1596 if (clone_flags & CLONE_THREAD)
1597 threadgroup_change_end(current);
1599 delayacct_tsk_free(p);
1600 bad_fork_cleanup_count:
1601 atomic_dec(&p->cred->user->processes);
1606 return ERR_PTR(retval);
1609 static inline void init_idle_pids(struct pid_link *links)
1613 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1614 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1615 links[type].pid = &init_struct_pid;
1619 struct task_struct * __cpuinit fork_idle(int cpu)
1621 struct task_struct *task;
1622 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1623 if (!IS_ERR(task)) {
1624 init_idle_pids(task->pids);
1625 init_idle(task, cpu);
1632 * Ok, this is the main fork-routine.
1634 * It copies the process, and if successful kick-starts
1635 * it and waits for it to finish using the VM if required.
1637 long do_fork(unsigned long clone_flags,
1638 unsigned long stack_start,
1639 unsigned long stack_size,
1640 int __user *parent_tidptr,
1641 int __user *child_tidptr)
1643 struct task_struct *p;
1648 * Do some preliminary argument and permissions checking before we
1649 * actually start allocating stuff
1651 if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
1652 if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
1657 * Determine whether and which event to report to ptracer. When
1658 * called from kernel_thread or CLONE_UNTRACED is explicitly
1659 * requested, no event is reported; otherwise, report if the event
1660 * for the type of forking is enabled.
1662 if (!(clone_flags & CLONE_UNTRACED)) {
1663 if (clone_flags & CLONE_VFORK)
1664 trace = PTRACE_EVENT_VFORK;
1665 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1666 trace = PTRACE_EVENT_CLONE;
1668 trace = PTRACE_EVENT_FORK;
1670 if (likely(!ptrace_event_enabled(current, trace)))
1674 p = copy_process(clone_flags, stack_start, stack_size,
1675 child_tidptr, NULL, trace);
1677 * Do this prior waking up the new thread - the thread pointer
1678 * might get invalid after that point, if the thread exits quickly.
1681 struct completion vfork;
1684 trace_sched_process_fork(current, p);
1686 pid = get_task_pid(p, PIDTYPE_PID);
1689 if (clone_flags & CLONE_PARENT_SETTID)
1690 put_user(nr, parent_tidptr);
1692 if (clone_flags & CLONE_VFORK) {
1693 p->vfork_done = &vfork;
1694 init_completion(&vfork);
1698 wake_up_new_task(p);
1700 /* forking complete and child started to run, tell ptracer */
1701 if (unlikely(trace))
1702 ptrace_event_pid(trace, pid);
1704 if (clone_flags & CLONE_VFORK) {
1705 if (!wait_for_vfork_done(p, &vfork))
1706 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1717 * Create a kernel thread.
1719 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1721 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1722 (unsigned long)arg, NULL, NULL);
1725 #ifdef __ARCH_WANT_SYS_FORK
1726 SYSCALL_DEFINE0(fork)
1729 return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1731 /* can not support in nommu mode */
1737 #ifdef __ARCH_WANT_SYS_VFORK
1738 SYSCALL_DEFINE0(vfork)
1740 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1745 #ifdef __ARCH_WANT_SYS_CLONE
1746 #ifdef CONFIG_CLONE_BACKWARDS
1747 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1748 int __user *, parent_tidptr,
1750 int __user *, child_tidptr)
1751 #elif defined(CONFIG_CLONE_BACKWARDS2)
1752 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1753 int __user *, parent_tidptr,
1754 int __user *, child_tidptr,
1756 #elif defined(CONFIG_CLONE_BACKWARDS3)
1757 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1759 int __user *, parent_tidptr,
1760 int __user *, child_tidptr,
1763 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1764 int __user *, parent_tidptr,
1765 int __user *, child_tidptr,
1769 return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1773 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1774 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1777 static void sighand_ctor(void *data)
1779 struct sighand_struct *sighand = data;
1781 spin_lock_init(&sighand->siglock);
1782 init_waitqueue_head(&sighand->signalfd_wqh);
1785 void __init proc_caches_init(void)
1787 sighand_cachep = kmem_cache_create("sighand_cache",
1788 sizeof(struct sighand_struct), 0,
1789 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1790 SLAB_NOTRACK, sighand_ctor);
1791 signal_cachep = kmem_cache_create("signal_cache",
1792 sizeof(struct signal_struct), 0,
1793 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1794 files_cachep = kmem_cache_create("files_cache",
1795 sizeof(struct files_struct), 0,
1796 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1797 fs_cachep = kmem_cache_create("fs_cache",
1798 sizeof(struct fs_struct), 0,
1799 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1801 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1802 * whole struct cpumask for the OFFSTACK case. We could change
1803 * this to *only* allocate as much of it as required by the
1804 * maximum number of CPU's we can ever have. The cpumask_allocation
1805 * is at the end of the structure, exactly for that reason.
1807 mm_cachep = kmem_cache_create("mm_struct",
1808 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1809 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1810 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1812 nsproxy_cache_init();
1816 * Check constraints on flags passed to the unshare system call.
1818 static int check_unshare_flags(unsigned long unshare_flags)
1820 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1821 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1822 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1823 CLONE_NEWUSER|CLONE_NEWPID))
1826 * Not implemented, but pretend it works if there is nothing
1827 * to unshare. Note that unsharing the address space or the
1828 * signal handlers also need to unshare the signal queues (aka
1831 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1832 if (!thread_group_empty(current))
1835 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1836 if (atomic_read(¤t->sighand->count) > 1)
1839 if (unshare_flags & CLONE_VM) {
1840 if (!current_is_single_threaded())
1848 * Unshare the filesystem structure if it is being shared
1850 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1852 struct fs_struct *fs = current->fs;
1854 if (!(unshare_flags & CLONE_FS) || !fs)
1857 /* don't need lock here; in the worst case we'll do useless copy */
1861 *new_fsp = copy_fs_struct(fs);
1869 * Unshare file descriptor table if it is being shared
1871 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1873 struct files_struct *fd = current->files;
1876 if ((unshare_flags & CLONE_FILES) &&
1877 (fd && atomic_read(&fd->count) > 1)) {
1878 *new_fdp = dup_fd(fd, &error);
1887 * unshare allows a process to 'unshare' part of the process
1888 * context which was originally shared using clone. copy_*
1889 * functions used by do_fork() cannot be used here directly
1890 * because they modify an inactive task_struct that is being
1891 * constructed. Here we are modifying the current, active,
1894 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1896 struct fs_struct *fs, *new_fs = NULL;
1897 struct files_struct *fd, *new_fd = NULL;
1898 struct cred *new_cred = NULL;
1899 struct nsproxy *new_nsproxy = NULL;
1904 * If unsharing a user namespace must also unshare the thread.
1906 if (unshare_flags & CLONE_NEWUSER)
1907 unshare_flags |= CLONE_THREAD | CLONE_FS;
1909 * If unsharing a pid namespace must also unshare the thread.
1911 if (unshare_flags & CLONE_NEWPID)
1912 unshare_flags |= CLONE_THREAD;
1914 * If unsharing vm, must also unshare signal handlers.
1916 if (unshare_flags & CLONE_VM)
1917 unshare_flags |= CLONE_SIGHAND;
1919 * If unsharing a signal handlers, must also unshare the signal queues.
1921 if (unshare_flags & CLONE_SIGHAND)
1922 unshare_flags |= CLONE_THREAD;
1924 * If unsharing namespace, must also unshare filesystem information.
1926 if (unshare_flags & CLONE_NEWNS)
1927 unshare_flags |= CLONE_FS;
1929 err = check_unshare_flags(unshare_flags);
1931 goto bad_unshare_out;
1933 * CLONE_NEWIPC must also detach from the undolist: after switching
1934 * to a new ipc namespace, the semaphore arrays from the old
1935 * namespace are unreachable.
1937 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1939 err = unshare_fs(unshare_flags, &new_fs);
1941 goto bad_unshare_out;
1942 err = unshare_fd(unshare_flags, &new_fd);
1944 goto bad_unshare_cleanup_fs;
1945 err = unshare_userns(unshare_flags, &new_cred);
1947 goto bad_unshare_cleanup_fd;
1948 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1951 goto bad_unshare_cleanup_cred;
1953 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1956 * CLONE_SYSVSEM is equivalent to sys_exit().
1962 switch_task_namespaces(current, new_nsproxy);
1968 spin_lock(&fs->lock);
1969 current->fs = new_fs;
1974 spin_unlock(&fs->lock);
1978 fd = current->files;
1979 current->files = new_fd;
1983 task_unlock(current);
1986 /* Install the new user namespace */
1987 commit_creds(new_cred);
1992 bad_unshare_cleanup_cred:
1995 bad_unshare_cleanup_fd:
1997 put_files_struct(new_fd);
1999 bad_unshare_cleanup_fs:
2001 free_fs_struct(new_fs);
2008 * Helper to unshare the files of the current task.
2009 * We don't want to expose copy_files internals to
2010 * the exec layer of the kernel.
2013 int unshare_files(struct files_struct **displaced)
2015 struct task_struct *task = current;
2016 struct files_struct *copy = NULL;
2019 error = unshare_fd(CLONE_FILES, ©);
2020 if (error || !copy) {
2024 *displaced = task->files;