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>
74 #include <asm/pgtable.h>
75 #include <asm/pgalloc.h>
76 #include <asm/uaccess.h>
77 #include <asm/mmu_context.h>
78 #include <asm/cacheflush.h>
79 #include <asm/tlbflush.h>
81 #include <trace/events/sched.h>
83 #define CREATE_TRACE_POINTS
84 #include <trace/events/task.h>
87 * Protected counters by write_lock_irq(&tasklist_lock)
89 unsigned long total_forks; /* Handle normal Linux uptimes. */
90 int nr_threads; /* The idle threads do not count.. */
92 int max_threads; /* tunable limit on nr_threads */
94 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
96 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
98 #ifdef CONFIG_PROVE_RCU
99 int lockdep_tasklist_lock_is_held(void)
101 return lockdep_is_held(&tasklist_lock);
103 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
104 #endif /* #ifdef CONFIG_PROVE_RCU */
106 int nr_processes(void)
111 for_each_possible_cpu(cpu)
112 total += per_cpu(process_counts, cpu);
117 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
118 static struct kmem_cache *task_struct_cachep;
120 static inline struct task_struct *alloc_task_struct_node(int node)
122 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
125 void __weak arch_release_task_struct(struct task_struct *tsk) { }
127 static inline void free_task_struct(struct task_struct *tsk)
129 arch_release_task_struct(tsk);
130 kmem_cache_free(task_struct_cachep, tsk);
134 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
135 void __weak arch_release_thread_info(struct thread_info *ti) { }
138 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
139 * kmemcache based allocator.
141 # if THREAD_SIZE >= PAGE_SIZE
142 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
145 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
148 return page ? page_address(page) : NULL;
151 static inline void free_thread_info(struct thread_info *ti)
153 arch_release_thread_info(ti);
154 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
157 static struct kmem_cache *thread_info_cache;
159 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
162 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
165 static void free_thread_info(struct thread_info *ti)
167 arch_release_thread_info(ti);
168 kmem_cache_free(thread_info_cache, ti);
171 void thread_info_cache_init(void)
173 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
174 THREAD_SIZE, 0, NULL);
175 BUG_ON(thread_info_cache == NULL);
180 /* SLAB cache for signal_struct structures (tsk->signal) */
181 static struct kmem_cache *signal_cachep;
183 /* SLAB cache for sighand_struct structures (tsk->sighand) */
184 struct kmem_cache *sighand_cachep;
186 /* SLAB cache for files_struct structures (tsk->files) */
187 struct kmem_cache *files_cachep;
189 /* SLAB cache for fs_struct structures (tsk->fs) */
190 struct kmem_cache *fs_cachep;
192 /* SLAB cache for vm_area_struct structures */
193 struct kmem_cache *vm_area_cachep;
195 /* SLAB cache for mm_struct structures (tsk->mm) */
196 static struct kmem_cache *mm_cachep;
198 static void account_kernel_stack(struct thread_info *ti, int account)
200 struct zone *zone = page_zone(virt_to_page(ti));
202 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
205 void free_task(struct task_struct *tsk)
207 account_kernel_stack(tsk->stack, -1);
208 free_thread_info(tsk->stack);
209 rt_mutex_debug_task_free(tsk);
210 ftrace_graph_exit_task(tsk);
211 put_seccomp_filter(tsk);
212 free_task_struct(tsk);
214 EXPORT_SYMBOL(free_task);
216 static inline void free_signal_struct(struct signal_struct *sig)
218 taskstats_tgid_free(sig);
219 sched_autogroup_exit(sig);
220 kmem_cache_free(signal_cachep, sig);
223 static inline void put_signal_struct(struct signal_struct *sig)
225 if (atomic_dec_and_test(&sig->sigcnt))
226 free_signal_struct(sig);
229 void __put_task_struct(struct task_struct *tsk)
231 WARN_ON(!tsk->exit_state);
232 WARN_ON(atomic_read(&tsk->usage));
233 WARN_ON(tsk == current);
235 security_task_free(tsk);
237 delayacct_tsk_free(tsk);
238 put_signal_struct(tsk->signal);
240 if (!profile_handoff_task(tsk))
243 EXPORT_SYMBOL_GPL(__put_task_struct);
245 void __init __weak arch_task_cache_init(void) { }
247 void __init fork_init(unsigned long mempages)
249 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
250 #ifndef ARCH_MIN_TASKALIGN
251 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
253 /* create a slab on which task_structs can be allocated */
255 kmem_cache_create("task_struct", sizeof(struct task_struct),
256 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
259 /* do the arch specific task caches init */
260 arch_task_cache_init();
263 * The default maximum number of threads is set to a safe
264 * value: the thread structures can take up at most half
267 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
270 * we need to allow at least 20 threads to boot a system
272 if (max_threads < 20)
275 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
276 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
277 init_task.signal->rlim[RLIMIT_SIGPENDING] =
278 init_task.signal->rlim[RLIMIT_NPROC];
281 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
282 struct task_struct *src)
288 static struct task_struct *dup_task_struct(struct task_struct *orig)
290 struct task_struct *tsk;
291 struct thread_info *ti;
292 unsigned long *stackend;
293 int node = tsk_fork_get_node(orig);
296 tsk = alloc_task_struct_node(node);
300 ti = alloc_thread_info_node(tsk, node);
302 free_task_struct(tsk);
306 err = arch_dup_task_struct(tsk, orig);
312 setup_thread_stack(tsk, orig);
313 clear_user_return_notifier(tsk);
314 clear_tsk_need_resched(tsk);
315 stackend = end_of_stack(tsk);
316 *stackend = STACK_END_MAGIC; /* for overflow detection */
318 #ifdef CONFIG_CC_STACKPROTECTOR
319 tsk->stack_canary = get_random_int();
323 * One for us, one for whoever does the "release_task()" (usually
326 atomic_set(&tsk->usage, 2);
327 #ifdef CONFIG_BLK_DEV_IO_TRACE
330 tsk->splice_pipe = NULL;
332 account_kernel_stack(ti, 1);
337 free_thread_info(ti);
338 free_task_struct(tsk);
343 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
345 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
346 struct rb_node **rb_link, *rb_parent;
348 unsigned long charge;
349 struct mempolicy *pol;
351 down_write(&oldmm->mmap_sem);
352 flush_cache_dup_mm(oldmm);
354 * Not linked in yet - no deadlock potential:
356 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
360 mm->mmap_cache = NULL;
361 mm->free_area_cache = oldmm->mmap_base;
362 mm->cached_hole_size = ~0UL;
364 cpumask_clear(mm_cpumask(mm));
366 rb_link = &mm->mm_rb.rb_node;
369 retval = ksm_fork(mm, oldmm);
372 retval = khugepaged_fork(mm, oldmm);
377 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
380 if (mpnt->vm_flags & VM_DONTCOPY) {
381 long pages = vma_pages(mpnt);
382 mm->total_vm -= pages;
383 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
388 if (mpnt->vm_flags & VM_ACCOUNT) {
389 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
390 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
394 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
398 INIT_LIST_HEAD(&tmp->anon_vma_chain);
399 pol = mpol_dup(vma_policy(mpnt));
400 retval = PTR_ERR(pol);
402 goto fail_nomem_policy;
403 vma_set_policy(tmp, pol);
405 if (anon_vma_fork(tmp, mpnt))
406 goto fail_nomem_anon_vma_fork;
407 tmp->vm_flags &= ~VM_LOCKED;
408 tmp->vm_next = tmp->vm_prev = NULL;
411 struct inode *inode = file->f_path.dentry->d_inode;
412 struct address_space *mapping = file->f_mapping;
415 if (tmp->vm_flags & VM_DENYWRITE)
416 atomic_dec(&inode->i_writecount);
417 mutex_lock(&mapping->i_mmap_mutex);
418 if (tmp->vm_flags & VM_SHARED)
419 mapping->i_mmap_writable++;
420 flush_dcache_mmap_lock(mapping);
421 /* insert tmp into the share list, just after mpnt */
422 vma_prio_tree_add(tmp, mpnt);
423 flush_dcache_mmap_unlock(mapping);
424 mutex_unlock(&mapping->i_mmap_mutex);
428 * Clear hugetlb-related page reserves for children. This only
429 * affects MAP_PRIVATE mappings. Faults generated by the child
430 * are not guaranteed to succeed, even if read-only
432 if (is_vm_hugetlb_page(tmp))
433 reset_vma_resv_huge_pages(tmp);
436 * Link in the new vma and copy the page table entries.
439 pprev = &tmp->vm_next;
443 __vma_link_rb(mm, tmp, rb_link, rb_parent);
444 rb_link = &tmp->vm_rb.rb_right;
445 rb_parent = &tmp->vm_rb;
448 retval = copy_page_range(mm, oldmm, mpnt);
450 if (tmp->vm_ops && tmp->vm_ops->open)
451 tmp->vm_ops->open(tmp);
456 if (file && uprobe_mmap(tmp))
459 /* a new mm has just been created */
460 arch_dup_mmap(oldmm, mm);
463 up_write(&mm->mmap_sem);
465 up_write(&oldmm->mmap_sem);
467 fail_nomem_anon_vma_fork:
470 kmem_cache_free(vm_area_cachep, tmp);
473 vm_unacct_memory(charge);
477 static inline int mm_alloc_pgd(struct mm_struct *mm)
479 mm->pgd = pgd_alloc(mm);
480 if (unlikely(!mm->pgd))
485 static inline void mm_free_pgd(struct mm_struct *mm)
487 pgd_free(mm, mm->pgd);
490 #define dup_mmap(mm, oldmm) (0)
491 #define mm_alloc_pgd(mm) (0)
492 #define mm_free_pgd(mm)
493 #endif /* CONFIG_MMU */
495 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
497 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
498 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
500 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
502 static int __init coredump_filter_setup(char *s)
504 default_dump_filter =
505 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
506 MMF_DUMP_FILTER_MASK;
510 __setup("coredump_filter=", coredump_filter_setup);
512 #include <linux/init_task.h>
514 static void mm_init_aio(struct mm_struct *mm)
517 spin_lock_init(&mm->ioctx_lock);
518 INIT_HLIST_HEAD(&mm->ioctx_list);
522 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
524 atomic_set(&mm->mm_users, 1);
525 atomic_set(&mm->mm_count, 1);
526 init_rwsem(&mm->mmap_sem);
527 INIT_LIST_HEAD(&mm->mmlist);
528 mm->flags = (current->mm) ?
529 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
530 mm->core_state = NULL;
532 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
533 spin_lock_init(&mm->page_table_lock);
534 mm->free_area_cache = TASK_UNMAPPED_BASE;
535 mm->cached_hole_size = ~0UL;
537 mm_init_owner(mm, p);
539 if (likely(!mm_alloc_pgd(mm))) {
541 mmu_notifier_mm_init(mm);
549 static void check_mm(struct mm_struct *mm)
553 for (i = 0; i < NR_MM_COUNTERS; i++) {
554 long x = atomic_long_read(&mm->rss_stat.count[i]);
557 printk(KERN_ALERT "BUG: Bad rss-counter state "
558 "mm:%p idx:%d val:%ld\n", mm, i, x);
561 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
562 VM_BUG_ON(mm->pmd_huge_pte);
567 * Allocate and initialize an mm_struct.
569 struct mm_struct *mm_alloc(void)
571 struct mm_struct *mm;
577 memset(mm, 0, sizeof(*mm));
579 return mm_init(mm, current);
583 * Called when the last reference to the mm
584 * is dropped: either by a lazy thread or by
585 * mmput. Free the page directory and the mm.
587 void __mmdrop(struct mm_struct *mm)
589 BUG_ON(mm == &init_mm);
592 mmu_notifier_mm_destroy(mm);
596 EXPORT_SYMBOL_GPL(__mmdrop);
599 * Decrement the use count and release all resources for an mm.
601 void mmput(struct mm_struct *mm)
605 if (atomic_dec_and_test(&mm->mm_users)) {
606 uprobe_clear_state(mm);
609 khugepaged_exit(mm); /* must run before exit_mmap */
611 set_mm_exe_file(mm, NULL);
612 if (!list_empty(&mm->mmlist)) {
613 spin_lock(&mmlist_lock);
614 list_del(&mm->mmlist);
615 spin_unlock(&mmlist_lock);
619 module_put(mm->binfmt->module);
623 EXPORT_SYMBOL_GPL(mmput);
626 * We added or removed a vma mapping the executable. The vmas are only mapped
627 * during exec and are not mapped with the mmap system call.
628 * Callers must hold down_write() on the mm's mmap_sem for these
630 void added_exe_file_vma(struct mm_struct *mm)
632 mm->num_exe_file_vmas++;
635 void removed_exe_file_vma(struct mm_struct *mm)
637 mm->num_exe_file_vmas--;
638 if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
645 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
648 get_file(new_exe_file);
651 mm->exe_file = new_exe_file;
652 mm->num_exe_file_vmas = 0;
655 struct file *get_mm_exe_file(struct mm_struct *mm)
657 struct file *exe_file;
659 /* We need mmap_sem to protect against races with removal of
660 * VM_EXECUTABLE vmas */
661 down_read(&mm->mmap_sem);
662 exe_file = mm->exe_file;
665 up_read(&mm->mmap_sem);
669 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
671 /* It's safe to write the exe_file pointer without exe_file_lock because
672 * this is called during fork when the task is not yet in /proc */
673 newmm->exe_file = get_mm_exe_file(oldmm);
677 * get_task_mm - acquire a reference to the task's mm
679 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
680 * this kernel workthread has transiently adopted a user mm with use_mm,
681 * to do its AIO) is not set and if so returns a reference to it, after
682 * bumping up the use count. User must release the mm via mmput()
683 * after use. Typically used by /proc and ptrace.
685 struct mm_struct *get_task_mm(struct task_struct *task)
687 struct mm_struct *mm;
692 if (task->flags & PF_KTHREAD)
695 atomic_inc(&mm->mm_users);
700 EXPORT_SYMBOL_GPL(get_task_mm);
702 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
704 struct mm_struct *mm;
707 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
711 mm = get_task_mm(task);
712 if (mm && mm != current->mm &&
713 !ptrace_may_access(task, mode)) {
715 mm = ERR_PTR(-EACCES);
717 mutex_unlock(&task->signal->cred_guard_mutex);
722 static void complete_vfork_done(struct task_struct *tsk)
724 struct completion *vfork;
727 vfork = tsk->vfork_done;
729 tsk->vfork_done = NULL;
735 static int wait_for_vfork_done(struct task_struct *child,
736 struct completion *vfork)
740 freezer_do_not_count();
741 killed = wait_for_completion_killable(vfork);
746 child->vfork_done = NULL;
750 put_task_struct(child);
754 /* Please note the differences between mmput and mm_release.
755 * mmput is called whenever we stop holding onto a mm_struct,
756 * error success whatever.
758 * mm_release is called after a mm_struct has been removed
759 * from the current process.
761 * This difference is important for error handling, when we
762 * only half set up a mm_struct for a new process and need to restore
763 * the old one. Because we mmput the new mm_struct before
764 * restoring the old one. . .
765 * Eric Biederman 10 January 1998
767 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
769 /* Get rid of any futexes when releasing the mm */
771 if (unlikely(tsk->robust_list)) {
772 exit_robust_list(tsk);
773 tsk->robust_list = NULL;
776 if (unlikely(tsk->compat_robust_list)) {
777 compat_exit_robust_list(tsk);
778 tsk->compat_robust_list = NULL;
781 if (unlikely(!list_empty(&tsk->pi_state_list)))
782 exit_pi_state_list(tsk);
785 uprobe_free_utask(tsk);
787 /* Get rid of any cached register state */
788 deactivate_mm(tsk, mm);
791 complete_vfork_done(tsk);
794 * If we're exiting normally, clear a user-space tid field if
795 * requested. We leave this alone when dying by signal, to leave
796 * the value intact in a core dump, and to save the unnecessary
797 * trouble, say, a killed vfork parent shouldn't touch this mm.
798 * Userland only wants this done for a sys_exit.
800 if (tsk->clear_child_tid) {
801 if (!(tsk->flags & PF_SIGNALED) &&
802 atomic_read(&mm->mm_users) > 1) {
804 * We don't check the error code - if userspace has
805 * not set up a proper pointer then tough luck.
807 put_user(0, tsk->clear_child_tid);
808 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
811 tsk->clear_child_tid = NULL;
816 * Allocate a new mm structure and copy contents from the
817 * mm structure of the passed in task structure.
819 struct mm_struct *dup_mm(struct task_struct *tsk)
821 struct mm_struct *mm, *oldmm = current->mm;
831 memcpy(mm, oldmm, sizeof(*mm));
834 /* Initializing for Swap token stuff */
835 mm->token_priority = 0;
836 mm->last_interval = 0;
838 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
839 mm->pmd_huge_pte = NULL;
841 uprobe_reset_state(mm);
843 if (!mm_init(mm, tsk))
846 if (init_new_context(tsk, mm))
849 dup_mm_exe_file(oldmm, mm);
851 err = dup_mmap(mm, oldmm);
855 mm->hiwater_rss = get_mm_rss(mm);
856 mm->hiwater_vm = mm->total_vm;
858 if (mm->binfmt && !try_module_get(mm->binfmt->module))
864 /* don't put binfmt in mmput, we haven't got module yet */
873 * If init_new_context() failed, we cannot use mmput() to free the mm
874 * because it calls destroy_context()
881 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
883 struct mm_struct *mm, *oldmm;
886 tsk->min_flt = tsk->maj_flt = 0;
887 tsk->nvcsw = tsk->nivcsw = 0;
888 #ifdef CONFIG_DETECT_HUNG_TASK
889 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
893 tsk->active_mm = NULL;
896 * Are we cloning a kernel thread?
898 * We need to steal a active VM for that..
904 if (clone_flags & CLONE_VM) {
905 atomic_inc(&oldmm->mm_users);
916 /* Initializing for Swap token stuff */
917 mm->token_priority = 0;
918 mm->last_interval = 0;
928 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
930 struct fs_struct *fs = current->fs;
931 if (clone_flags & CLONE_FS) {
932 /* tsk->fs is already what we want */
933 spin_lock(&fs->lock);
935 spin_unlock(&fs->lock);
939 spin_unlock(&fs->lock);
942 tsk->fs = copy_fs_struct(fs);
948 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
950 struct files_struct *oldf, *newf;
954 * A background process may not have any files ...
956 oldf = current->files;
960 if (clone_flags & CLONE_FILES) {
961 atomic_inc(&oldf->count);
965 newf = dup_fd(oldf, &error);
975 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
978 struct io_context *ioc = current->io_context;
979 struct io_context *new_ioc;
984 * Share io context with parent, if CLONE_IO is set
986 if (clone_flags & CLONE_IO) {
987 tsk->io_context = ioc_task_link(ioc);
988 if (unlikely(!tsk->io_context))
990 } else if (ioprio_valid(ioc->ioprio)) {
991 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
992 if (unlikely(!new_ioc))
995 new_ioc->ioprio = ioc->ioprio;
996 put_io_context(new_ioc);
1002 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1004 struct sighand_struct *sig;
1006 if (clone_flags & CLONE_SIGHAND) {
1007 atomic_inc(¤t->sighand->count);
1010 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1011 rcu_assign_pointer(tsk->sighand, sig);
1014 atomic_set(&sig->count, 1);
1015 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1019 void __cleanup_sighand(struct sighand_struct *sighand)
1021 if (atomic_dec_and_test(&sighand->count)) {
1022 signalfd_cleanup(sighand);
1023 kmem_cache_free(sighand_cachep, sighand);
1029 * Initialize POSIX timer handling for a thread group.
1031 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1033 unsigned long cpu_limit;
1035 /* Thread group counters. */
1036 thread_group_cputime_init(sig);
1038 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1039 if (cpu_limit != RLIM_INFINITY) {
1040 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1041 sig->cputimer.running = 1;
1044 /* The timer lists. */
1045 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1046 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1047 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1050 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1052 struct signal_struct *sig;
1054 if (clone_flags & CLONE_THREAD)
1057 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1062 sig->nr_threads = 1;
1063 atomic_set(&sig->live, 1);
1064 atomic_set(&sig->sigcnt, 1);
1065 init_waitqueue_head(&sig->wait_chldexit);
1066 if (clone_flags & CLONE_NEWPID)
1067 sig->flags |= SIGNAL_UNKILLABLE;
1068 sig->curr_target = tsk;
1069 init_sigpending(&sig->shared_pending);
1070 INIT_LIST_HEAD(&sig->posix_timers);
1072 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1073 sig->real_timer.function = it_real_fn;
1075 task_lock(current->group_leader);
1076 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1077 task_unlock(current->group_leader);
1079 posix_cpu_timers_init_group(sig);
1081 tty_audit_fork(sig);
1082 sched_autogroup_fork(sig);
1084 #ifdef CONFIG_CGROUPS
1085 init_rwsem(&sig->group_rwsem);
1088 sig->oom_adj = current->signal->oom_adj;
1089 sig->oom_score_adj = current->signal->oom_score_adj;
1090 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1092 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1093 current->signal->is_child_subreaper;
1095 mutex_init(&sig->cred_guard_mutex);
1100 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1102 unsigned long new_flags = p->flags;
1104 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1105 new_flags |= PF_FORKNOEXEC;
1106 p->flags = new_flags;
1109 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1111 current->clear_child_tid = tidptr;
1113 return task_pid_vnr(current);
1116 static void rt_mutex_init_task(struct task_struct *p)
1118 raw_spin_lock_init(&p->pi_lock);
1119 #ifdef CONFIG_RT_MUTEXES
1120 plist_head_init(&p->pi_waiters);
1121 p->pi_blocked_on = NULL;
1125 #ifdef CONFIG_MM_OWNER
1126 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1130 #endif /* CONFIG_MM_OWNER */
1133 * Initialize POSIX timer handling for a single task.
1135 static void posix_cpu_timers_init(struct task_struct *tsk)
1137 tsk->cputime_expires.prof_exp = 0;
1138 tsk->cputime_expires.virt_exp = 0;
1139 tsk->cputime_expires.sched_exp = 0;
1140 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1141 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1142 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1146 * This creates a new process as a copy of the old one,
1147 * but does not actually start it yet.
1149 * It copies the registers, and all the appropriate
1150 * parts of the process environment (as per the clone
1151 * flags). The actual kick-off is left to the caller.
1153 static struct task_struct *copy_process(unsigned long clone_flags,
1154 unsigned long stack_start,
1155 struct pt_regs *regs,
1156 unsigned long stack_size,
1157 int __user *child_tidptr,
1162 struct task_struct *p;
1163 int cgroup_callbacks_done = 0;
1165 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1166 return ERR_PTR(-EINVAL);
1169 * Thread groups must share signals as well, and detached threads
1170 * can only be started up within the thread group.
1172 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1173 return ERR_PTR(-EINVAL);
1176 * Shared signal handlers imply shared VM. By way of the above,
1177 * thread groups also imply shared VM. Blocking this case allows
1178 * for various simplifications in other code.
1180 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1181 return ERR_PTR(-EINVAL);
1184 * Siblings of global init remain as zombies on exit since they are
1185 * not reaped by their parent (swapper). To solve this and to avoid
1186 * multi-rooted process trees, prevent global and container-inits
1187 * from creating siblings.
1189 if ((clone_flags & CLONE_PARENT) &&
1190 current->signal->flags & SIGNAL_UNKILLABLE)
1191 return ERR_PTR(-EINVAL);
1193 retval = security_task_create(clone_flags);
1198 p = dup_task_struct(current);
1202 ftrace_graph_init_task(p);
1203 get_seccomp_filter(p);
1205 rt_mutex_init_task(p);
1207 #ifdef CONFIG_PROVE_LOCKING
1208 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1209 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1212 if (atomic_read(&p->real_cred->user->processes) >=
1213 task_rlimit(p, RLIMIT_NPROC)) {
1214 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1215 p->real_cred->user != INIT_USER)
1218 current->flags &= ~PF_NPROC_EXCEEDED;
1220 retval = copy_creds(p, clone_flags);
1225 * If multiple threads are within copy_process(), then this check
1226 * triggers too late. This doesn't hurt, the check is only there
1227 * to stop root fork bombs.
1230 if (nr_threads >= max_threads)
1231 goto bad_fork_cleanup_count;
1233 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1234 goto bad_fork_cleanup_count;
1237 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1238 copy_flags(clone_flags, p);
1239 INIT_LIST_HEAD(&p->children);
1240 INIT_LIST_HEAD(&p->sibling);
1241 rcu_copy_process(p);
1242 p->vfork_done = NULL;
1243 spin_lock_init(&p->alloc_lock);
1245 init_sigpending(&p->pending);
1247 p->utime = p->stime = p->gtime = 0;
1248 p->utimescaled = p->stimescaled = 0;
1249 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1250 p->prev_utime = p->prev_stime = 0;
1252 #if defined(SPLIT_RSS_COUNTING)
1253 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1256 p->default_timer_slack_ns = current->timer_slack_ns;
1258 task_io_accounting_init(&p->ioac);
1259 acct_clear_integrals(p);
1261 posix_cpu_timers_init(p);
1263 do_posix_clock_monotonic_gettime(&p->start_time);
1264 p->real_start_time = p->start_time;
1265 monotonic_to_bootbased(&p->real_start_time);
1266 p->io_context = NULL;
1267 p->audit_context = NULL;
1268 if (clone_flags & CLONE_THREAD)
1269 threadgroup_change_begin(current);
1272 p->mempolicy = mpol_dup(p->mempolicy);
1273 if (IS_ERR(p->mempolicy)) {
1274 retval = PTR_ERR(p->mempolicy);
1275 p->mempolicy = NULL;
1276 goto bad_fork_cleanup_cgroup;
1278 mpol_fix_fork_child_flag(p);
1280 #ifdef CONFIG_CPUSETS
1281 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1282 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1283 seqcount_init(&p->mems_allowed_seq);
1285 #ifdef CONFIG_TRACE_IRQFLAGS
1287 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1288 p->hardirqs_enabled = 1;
1290 p->hardirqs_enabled = 0;
1292 p->hardirq_enable_ip = 0;
1293 p->hardirq_enable_event = 0;
1294 p->hardirq_disable_ip = _THIS_IP_;
1295 p->hardirq_disable_event = 0;
1296 p->softirqs_enabled = 1;
1297 p->softirq_enable_ip = _THIS_IP_;
1298 p->softirq_enable_event = 0;
1299 p->softirq_disable_ip = 0;
1300 p->softirq_disable_event = 0;
1301 p->hardirq_context = 0;
1302 p->softirq_context = 0;
1304 #ifdef CONFIG_LOCKDEP
1305 p->lockdep_depth = 0; /* no locks held yet */
1306 p->curr_chain_key = 0;
1307 p->lockdep_recursion = 0;
1310 #ifdef CONFIG_DEBUG_MUTEXES
1311 p->blocked_on = NULL; /* not blocked yet */
1313 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1314 p->memcg_batch.do_batch = 0;
1315 p->memcg_batch.memcg = NULL;
1318 /* Perform scheduler related setup. Assign this task to a CPU. */
1321 retval = perf_event_init_task(p);
1323 goto bad_fork_cleanup_policy;
1324 retval = audit_alloc(p);
1326 goto bad_fork_cleanup_policy;
1327 /* copy all the process information */
1328 retval = copy_semundo(clone_flags, p);
1330 goto bad_fork_cleanup_audit;
1331 retval = copy_files(clone_flags, p);
1333 goto bad_fork_cleanup_semundo;
1334 retval = copy_fs(clone_flags, p);
1336 goto bad_fork_cleanup_files;
1337 retval = copy_sighand(clone_flags, p);
1339 goto bad_fork_cleanup_fs;
1340 retval = copy_signal(clone_flags, p);
1342 goto bad_fork_cleanup_sighand;
1343 retval = copy_mm(clone_flags, p);
1345 goto bad_fork_cleanup_signal;
1346 retval = copy_namespaces(clone_flags, p);
1348 goto bad_fork_cleanup_mm;
1349 retval = copy_io(clone_flags, p);
1351 goto bad_fork_cleanup_namespaces;
1352 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1354 goto bad_fork_cleanup_io;
1356 if (pid != &init_struct_pid) {
1358 pid = alloc_pid(p->nsproxy->pid_ns);
1360 goto bad_fork_cleanup_io;
1363 p->pid = pid_nr(pid);
1365 if (clone_flags & CLONE_THREAD)
1366 p->tgid = current->tgid;
1368 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1370 * Clear TID on mm_release()?
1372 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1377 p->robust_list = NULL;
1378 #ifdef CONFIG_COMPAT
1379 p->compat_robust_list = NULL;
1381 INIT_LIST_HEAD(&p->pi_state_list);
1382 p->pi_state_cache = NULL;
1384 uprobe_copy_process(p);
1386 * sigaltstack should be cleared when sharing the same VM
1388 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1389 p->sas_ss_sp = p->sas_ss_size = 0;
1392 * Syscall tracing and stepping should be turned off in the
1393 * child regardless of CLONE_PTRACE.
1395 user_disable_single_step(p);
1396 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1397 #ifdef TIF_SYSCALL_EMU
1398 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1400 clear_all_latency_tracing(p);
1402 /* ok, now we should be set up.. */
1403 if (clone_flags & CLONE_THREAD)
1404 p->exit_signal = -1;
1405 else if (clone_flags & CLONE_PARENT)
1406 p->exit_signal = current->group_leader->exit_signal;
1408 p->exit_signal = (clone_flags & CSIGNAL);
1410 p->pdeath_signal = 0;
1414 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1415 p->dirty_paused_when = 0;
1418 * Ok, make it visible to the rest of the system.
1419 * We dont wake it up yet.
1421 p->group_leader = p;
1422 INIT_LIST_HEAD(&p->thread_group);
1424 /* Now that the task is set up, run cgroup callbacks if
1425 * necessary. We need to run them before the task is visible
1426 * on the tasklist. */
1427 cgroup_fork_callbacks(p);
1428 cgroup_callbacks_done = 1;
1430 /* Need tasklist lock for parent etc handling! */
1431 write_lock_irq(&tasklist_lock);
1433 /* CLONE_PARENT re-uses the old parent */
1434 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1435 p->real_parent = current->real_parent;
1436 p->parent_exec_id = current->parent_exec_id;
1438 p->real_parent = current;
1439 p->parent_exec_id = current->self_exec_id;
1442 spin_lock(¤t->sighand->siglock);
1445 * Process group and session signals need to be delivered to just the
1446 * parent before the fork or both the parent and the child after the
1447 * fork. Restart if a signal comes in before we add the new process to
1448 * it's process group.
1449 * A fatal signal pending means that current will exit, so the new
1450 * thread can't slip out of an OOM kill (or normal SIGKILL).
1452 recalc_sigpending();
1453 if (signal_pending(current)) {
1454 spin_unlock(¤t->sighand->siglock);
1455 write_unlock_irq(&tasklist_lock);
1456 retval = -ERESTARTNOINTR;
1457 goto bad_fork_free_pid;
1460 if (clone_flags & CLONE_THREAD) {
1461 current->signal->nr_threads++;
1462 atomic_inc(¤t->signal->live);
1463 atomic_inc(¤t->signal->sigcnt);
1464 p->group_leader = current->group_leader;
1465 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1468 if (likely(p->pid)) {
1469 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1471 if (thread_group_leader(p)) {
1472 if (is_child_reaper(pid))
1473 p->nsproxy->pid_ns->child_reaper = p;
1475 p->signal->leader_pid = pid;
1476 p->signal->tty = tty_kref_get(current->signal->tty);
1477 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1478 attach_pid(p, PIDTYPE_SID, task_session(current));
1479 list_add_tail(&p->sibling, &p->real_parent->children);
1480 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1481 __this_cpu_inc(process_counts);
1483 attach_pid(p, PIDTYPE_PID, pid);
1488 spin_unlock(¤t->sighand->siglock);
1489 write_unlock_irq(&tasklist_lock);
1490 proc_fork_connector(p);
1491 cgroup_post_fork(p);
1492 if (clone_flags & CLONE_THREAD)
1493 threadgroup_change_end(current);
1496 trace_task_newtask(p, clone_flags);
1501 if (pid != &init_struct_pid)
1503 bad_fork_cleanup_io:
1506 bad_fork_cleanup_namespaces:
1507 if (unlikely(clone_flags & CLONE_NEWPID))
1508 pid_ns_release_proc(p->nsproxy->pid_ns);
1509 exit_task_namespaces(p);
1510 bad_fork_cleanup_mm:
1513 bad_fork_cleanup_signal:
1514 if (!(clone_flags & CLONE_THREAD))
1515 free_signal_struct(p->signal);
1516 bad_fork_cleanup_sighand:
1517 __cleanup_sighand(p->sighand);
1518 bad_fork_cleanup_fs:
1519 exit_fs(p); /* blocking */
1520 bad_fork_cleanup_files:
1521 exit_files(p); /* blocking */
1522 bad_fork_cleanup_semundo:
1524 bad_fork_cleanup_audit:
1526 bad_fork_cleanup_policy:
1527 perf_event_free_task(p);
1529 mpol_put(p->mempolicy);
1530 bad_fork_cleanup_cgroup:
1532 if (clone_flags & CLONE_THREAD)
1533 threadgroup_change_end(current);
1534 cgroup_exit(p, cgroup_callbacks_done);
1535 delayacct_tsk_free(p);
1536 module_put(task_thread_info(p)->exec_domain->module);
1537 bad_fork_cleanup_count:
1538 atomic_dec(&p->cred->user->processes);
1543 return ERR_PTR(retval);
1546 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1548 memset(regs, 0, sizeof(struct pt_regs));
1552 static inline void init_idle_pids(struct pid_link *links)
1556 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1557 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1558 links[type].pid = &init_struct_pid;
1562 struct task_struct * __cpuinit fork_idle(int cpu)
1564 struct task_struct *task;
1565 struct pt_regs regs;
1567 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL,
1568 &init_struct_pid, 0);
1569 if (!IS_ERR(task)) {
1570 init_idle_pids(task->pids);
1571 init_idle(task, cpu);
1578 * Ok, this is the main fork-routine.
1580 * It copies the process, and if successful kick-starts
1581 * it and waits for it to finish using the VM if required.
1583 long do_fork(unsigned long clone_flags,
1584 unsigned long stack_start,
1585 struct pt_regs *regs,
1586 unsigned long stack_size,
1587 int __user *parent_tidptr,
1588 int __user *child_tidptr)
1590 struct task_struct *p;
1595 * Do some preliminary argument and permissions checking before we
1596 * actually start allocating stuff
1598 if (clone_flags & CLONE_NEWUSER) {
1599 if (clone_flags & CLONE_THREAD)
1601 /* hopefully this check will go away when userns support is
1604 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1605 !capable(CAP_SETGID))
1610 * Determine whether and which event to report to ptracer. When
1611 * called from kernel_thread or CLONE_UNTRACED is explicitly
1612 * requested, no event is reported; otherwise, report if the event
1613 * for the type of forking is enabled.
1615 if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1616 if (clone_flags & CLONE_VFORK)
1617 trace = PTRACE_EVENT_VFORK;
1618 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1619 trace = PTRACE_EVENT_CLONE;
1621 trace = PTRACE_EVENT_FORK;
1623 if (likely(!ptrace_event_enabled(current, trace)))
1627 p = copy_process(clone_flags, stack_start, regs, stack_size,
1628 child_tidptr, NULL, trace);
1630 * Do this prior waking up the new thread - the thread pointer
1631 * might get invalid after that point, if the thread exits quickly.
1634 struct completion vfork;
1636 trace_sched_process_fork(current, p);
1638 nr = task_pid_vnr(p);
1640 if (clone_flags & CLONE_PARENT_SETTID)
1641 put_user(nr, parent_tidptr);
1643 if (clone_flags & CLONE_VFORK) {
1644 p->vfork_done = &vfork;
1645 init_completion(&vfork);
1649 wake_up_new_task(p);
1651 /* forking complete and child started to run, tell ptracer */
1652 if (unlikely(trace))
1653 ptrace_event(trace, nr);
1655 if (clone_flags & CLONE_VFORK) {
1656 if (!wait_for_vfork_done(p, &vfork))
1657 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1665 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1666 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1669 static void sighand_ctor(void *data)
1671 struct sighand_struct *sighand = data;
1673 spin_lock_init(&sighand->siglock);
1674 init_waitqueue_head(&sighand->signalfd_wqh);
1677 void __init proc_caches_init(void)
1679 sighand_cachep = kmem_cache_create("sighand_cache",
1680 sizeof(struct sighand_struct), 0,
1681 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1682 SLAB_NOTRACK, sighand_ctor);
1683 signal_cachep = kmem_cache_create("signal_cache",
1684 sizeof(struct signal_struct), 0,
1685 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1686 files_cachep = kmem_cache_create("files_cache",
1687 sizeof(struct files_struct), 0,
1688 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1689 fs_cachep = kmem_cache_create("fs_cache",
1690 sizeof(struct fs_struct), 0,
1691 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1693 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1694 * whole struct cpumask for the OFFSTACK case. We could change
1695 * this to *only* allocate as much of it as required by the
1696 * maximum number of CPU's we can ever have. The cpumask_allocation
1697 * is at the end of the structure, exactly for that reason.
1699 mm_cachep = kmem_cache_create("mm_struct",
1700 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1701 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1702 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1704 nsproxy_cache_init();
1708 * Check constraints on flags passed to the unshare system call.
1710 static int check_unshare_flags(unsigned long unshare_flags)
1712 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1713 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1714 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1717 * Not implemented, but pretend it works if there is nothing to
1718 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1719 * needs to unshare vm.
1721 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1722 /* FIXME: get_task_mm() increments ->mm_users */
1723 if (atomic_read(¤t->mm->mm_users) > 1)
1731 * Unshare the filesystem structure if it is being shared
1733 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1735 struct fs_struct *fs = current->fs;
1737 if (!(unshare_flags & CLONE_FS) || !fs)
1740 /* don't need lock here; in the worst case we'll do useless copy */
1744 *new_fsp = copy_fs_struct(fs);
1752 * Unshare file descriptor table if it is being shared
1754 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1756 struct files_struct *fd = current->files;
1759 if ((unshare_flags & CLONE_FILES) &&
1760 (fd && atomic_read(&fd->count) > 1)) {
1761 *new_fdp = dup_fd(fd, &error);
1770 * unshare allows a process to 'unshare' part of the process
1771 * context which was originally shared using clone. copy_*
1772 * functions used by do_fork() cannot be used here directly
1773 * because they modify an inactive task_struct that is being
1774 * constructed. Here we are modifying the current, active,
1777 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1779 struct fs_struct *fs, *new_fs = NULL;
1780 struct files_struct *fd, *new_fd = NULL;
1781 struct nsproxy *new_nsproxy = NULL;
1785 err = check_unshare_flags(unshare_flags);
1787 goto bad_unshare_out;
1790 * If unsharing namespace, must also unshare filesystem information.
1792 if (unshare_flags & CLONE_NEWNS)
1793 unshare_flags |= CLONE_FS;
1795 * CLONE_NEWIPC must also detach from the undolist: after switching
1796 * to a new ipc namespace, the semaphore arrays from the old
1797 * namespace are unreachable.
1799 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1801 err = unshare_fs(unshare_flags, &new_fs);
1803 goto bad_unshare_out;
1804 err = unshare_fd(unshare_flags, &new_fd);
1806 goto bad_unshare_cleanup_fs;
1807 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1809 goto bad_unshare_cleanup_fd;
1811 if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1814 * CLONE_SYSVSEM is equivalent to sys_exit().
1820 switch_task_namespaces(current, new_nsproxy);
1828 spin_lock(&fs->lock);
1829 current->fs = new_fs;
1834 spin_unlock(&fs->lock);
1838 fd = current->files;
1839 current->files = new_fd;
1843 task_unlock(current);
1847 put_nsproxy(new_nsproxy);
1849 bad_unshare_cleanup_fd:
1851 put_files_struct(new_fd);
1853 bad_unshare_cleanup_fs:
1855 free_fs_struct(new_fs);
1862 * Helper to unshare the files of the current task.
1863 * We don't want to expose copy_files internals to
1864 * the exec layer of the kernel.
1867 int unshare_files(struct files_struct **displaced)
1869 struct task_struct *task = current;
1870 struct files_struct *copy = NULL;
1873 error = unshare_fd(CLONE_FILES, ©);
1874 if (error || !copy) {
1878 *displaced = task->files;