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 void __weak arch_release_task_struct(struct task_struct *tsk)
121 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
122 static struct kmem_cache *task_struct_cachep;
124 static inline struct task_struct *alloc_task_struct_node(int node)
126 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
129 static inline void free_task_struct(struct task_struct *tsk)
131 kmem_cache_free(task_struct_cachep, tsk);
135 void __weak arch_release_thread_info(struct thread_info *ti)
139 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
142 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
143 * kmemcache based allocator.
145 # if THREAD_SIZE >= PAGE_SIZE
146 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
149 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
152 return page ? page_address(page) : NULL;
155 static inline void free_thread_info(struct thread_info *ti)
157 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
160 static struct kmem_cache *thread_info_cache;
162 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
165 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
168 static void free_thread_info(struct thread_info *ti)
170 kmem_cache_free(thread_info_cache, ti);
173 void thread_info_cache_init(void)
175 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
176 THREAD_SIZE, 0, NULL);
177 BUG_ON(thread_info_cache == NULL);
182 /* SLAB cache for signal_struct structures (tsk->signal) */
183 static struct kmem_cache *signal_cachep;
185 /* SLAB cache for sighand_struct structures (tsk->sighand) */
186 struct kmem_cache *sighand_cachep;
188 /* SLAB cache for files_struct structures (tsk->files) */
189 struct kmem_cache *files_cachep;
191 /* SLAB cache for fs_struct structures (tsk->fs) */
192 struct kmem_cache *fs_cachep;
194 /* SLAB cache for vm_area_struct structures */
195 struct kmem_cache *vm_area_cachep;
197 /* SLAB cache for mm_struct structures (tsk->mm) */
198 static struct kmem_cache *mm_cachep;
200 static void account_kernel_stack(struct thread_info *ti, int account)
202 struct zone *zone = page_zone(virt_to_page(ti));
204 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
207 void free_task(struct task_struct *tsk)
209 account_kernel_stack(tsk->stack, -1);
210 arch_release_thread_info(tsk->stack);
211 free_thread_info(tsk->stack);
212 rt_mutex_debug_task_free(tsk);
213 ftrace_graph_exit_task(tsk);
214 put_seccomp_filter(tsk);
215 arch_release_task_struct(tsk);
216 free_task_struct(tsk);
218 EXPORT_SYMBOL(free_task);
220 static inline void free_signal_struct(struct signal_struct *sig)
222 taskstats_tgid_free(sig);
223 sched_autogroup_exit(sig);
224 kmem_cache_free(signal_cachep, sig);
227 static inline void put_signal_struct(struct signal_struct *sig)
229 if (atomic_dec_and_test(&sig->sigcnt))
230 free_signal_struct(sig);
233 void __put_task_struct(struct task_struct *tsk)
235 WARN_ON(!tsk->exit_state);
236 WARN_ON(atomic_read(&tsk->usage));
237 WARN_ON(tsk == current);
239 security_task_free(tsk);
241 delayacct_tsk_free(tsk);
242 put_signal_struct(tsk->signal);
244 if (!profile_handoff_task(tsk))
247 EXPORT_SYMBOL_GPL(__put_task_struct);
249 void __init __weak arch_task_cache_init(void) { }
251 void __init fork_init(unsigned long mempages)
253 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
254 #ifndef ARCH_MIN_TASKALIGN
255 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
257 /* create a slab on which task_structs can be allocated */
259 kmem_cache_create("task_struct", sizeof(struct task_struct),
260 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
263 /* do the arch specific task caches init */
264 arch_task_cache_init();
267 * The default maximum number of threads is set to a safe
268 * value: the thread structures can take up at most half
271 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
274 * we need to allow at least 20 threads to boot a system
276 if (max_threads < 20)
279 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
280 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
281 init_task.signal->rlim[RLIMIT_SIGPENDING] =
282 init_task.signal->rlim[RLIMIT_NPROC];
285 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
286 struct task_struct *src)
292 static struct task_struct *dup_task_struct(struct task_struct *orig)
294 struct task_struct *tsk;
295 struct thread_info *ti;
296 unsigned long *stackend;
297 int node = tsk_fork_get_node(orig);
300 tsk = alloc_task_struct_node(node);
304 ti = alloc_thread_info_node(tsk, node);
308 err = arch_dup_task_struct(tsk, orig);
314 setup_thread_stack(tsk, orig);
315 clear_user_return_notifier(tsk);
316 clear_tsk_need_resched(tsk);
317 stackend = end_of_stack(tsk);
318 *stackend = STACK_END_MAGIC; /* for overflow detection */
320 #ifdef CONFIG_CC_STACKPROTECTOR
321 tsk->stack_canary = get_random_int();
325 * One for us, one for whoever does the "release_task()" (usually
328 atomic_set(&tsk->usage, 2);
329 #ifdef CONFIG_BLK_DEV_IO_TRACE
332 tsk->splice_pipe = NULL;
333 tsk->task_frag.page = NULL;
335 account_kernel_stack(ti, 1);
340 free_thread_info(ti);
342 free_task_struct(tsk);
347 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
349 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
350 struct rb_node **rb_link, *rb_parent;
352 unsigned long charge;
353 struct mempolicy *pol;
355 down_write(&oldmm->mmap_sem);
356 flush_cache_dup_mm(oldmm);
358 * Not linked in yet - no deadlock potential:
360 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
364 mm->mmap_cache = NULL;
365 mm->free_area_cache = oldmm->mmap_base;
366 mm->cached_hole_size = ~0UL;
368 cpumask_clear(mm_cpumask(mm));
370 rb_link = &mm->mm_rb.rb_node;
373 retval = ksm_fork(mm, oldmm);
376 retval = khugepaged_fork(mm, oldmm);
381 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
384 if (mpnt->vm_flags & VM_DONTCOPY) {
385 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
390 if (mpnt->vm_flags & VM_ACCOUNT) {
391 unsigned long len = vma_pages(mpnt);
393 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
397 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
401 INIT_LIST_HEAD(&tmp->anon_vma_chain);
402 pol = mpol_dup(vma_policy(mpnt));
403 retval = PTR_ERR(pol);
405 goto fail_nomem_policy;
406 vma_set_policy(tmp, pol);
408 if (anon_vma_fork(tmp, mpnt))
409 goto fail_nomem_anon_vma_fork;
410 tmp->vm_flags &= ~VM_LOCKED;
411 tmp->vm_next = tmp->vm_prev = NULL;
414 struct inode *inode = file->f_path.dentry->d_inode;
415 struct address_space *mapping = file->f_mapping;
418 if (tmp->vm_flags & VM_DENYWRITE)
419 atomic_dec(&inode->i_writecount);
420 mutex_lock(&mapping->i_mmap_mutex);
421 if (tmp->vm_flags & VM_SHARED)
422 mapping->i_mmap_writable++;
423 flush_dcache_mmap_lock(mapping);
424 /* insert tmp into the share list, just after mpnt */
425 vma_prio_tree_add(tmp, mpnt);
426 flush_dcache_mmap_unlock(mapping);
427 mutex_unlock(&mapping->i_mmap_mutex);
431 * Clear hugetlb-related page reserves for children. This only
432 * affects MAP_PRIVATE mappings. Faults generated by the child
433 * are not guaranteed to succeed, even if read-only
435 if (is_vm_hugetlb_page(tmp))
436 reset_vma_resv_huge_pages(tmp);
439 * Link in the new vma and copy the page table entries.
442 pprev = &tmp->vm_next;
446 __vma_link_rb(mm, tmp, rb_link, rb_parent);
447 rb_link = &tmp->vm_rb.rb_right;
448 rb_parent = &tmp->vm_rb;
451 retval = copy_page_range(mm, oldmm, mpnt);
453 if (tmp->vm_ops && tmp->vm_ops->open)
454 tmp->vm_ops->open(tmp);
462 /* a new mm has just been created */
463 arch_dup_mmap(oldmm, mm);
466 up_write(&mm->mmap_sem);
468 up_write(&oldmm->mmap_sem);
470 fail_nomem_anon_vma_fork:
473 kmem_cache_free(vm_area_cachep, tmp);
476 vm_unacct_memory(charge);
480 static inline int mm_alloc_pgd(struct mm_struct *mm)
482 mm->pgd = pgd_alloc(mm);
483 if (unlikely(!mm->pgd))
488 static inline void mm_free_pgd(struct mm_struct *mm)
490 pgd_free(mm, mm->pgd);
493 #define dup_mmap(mm, oldmm) (0)
494 #define mm_alloc_pgd(mm) (0)
495 #define mm_free_pgd(mm)
496 #endif /* CONFIG_MMU */
498 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
500 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
501 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
503 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
505 static int __init coredump_filter_setup(char *s)
507 default_dump_filter =
508 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
509 MMF_DUMP_FILTER_MASK;
513 __setup("coredump_filter=", coredump_filter_setup);
515 #include <linux/init_task.h>
517 static void mm_init_aio(struct mm_struct *mm)
520 spin_lock_init(&mm->ioctx_lock);
521 INIT_HLIST_HEAD(&mm->ioctx_list);
525 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
527 atomic_set(&mm->mm_users, 1);
528 atomic_set(&mm->mm_count, 1);
529 init_rwsem(&mm->mmap_sem);
530 INIT_LIST_HEAD(&mm->mmlist);
531 mm->flags = (current->mm) ?
532 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
533 mm->core_state = NULL;
535 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
536 spin_lock_init(&mm->page_table_lock);
537 mm->free_area_cache = TASK_UNMAPPED_BASE;
538 mm->cached_hole_size = ~0UL;
540 mm_init_owner(mm, p);
542 if (likely(!mm_alloc_pgd(mm))) {
544 mmu_notifier_mm_init(mm);
552 static void check_mm(struct mm_struct *mm)
556 for (i = 0; i < NR_MM_COUNTERS; i++) {
557 long x = atomic_long_read(&mm->rss_stat.count[i]);
560 printk(KERN_ALERT "BUG: Bad rss-counter state "
561 "mm:%p idx:%d val:%ld\n", mm, i, x);
564 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
565 VM_BUG_ON(mm->pmd_huge_pte);
570 * Allocate and initialize an mm_struct.
572 struct mm_struct *mm_alloc(void)
574 struct mm_struct *mm;
580 memset(mm, 0, sizeof(*mm));
582 return mm_init(mm, current);
586 * Called when the last reference to the mm
587 * is dropped: either by a lazy thread or by
588 * mmput. Free the page directory and the mm.
590 void __mmdrop(struct mm_struct *mm)
592 BUG_ON(mm == &init_mm);
595 mmu_notifier_mm_destroy(mm);
599 EXPORT_SYMBOL_GPL(__mmdrop);
602 * Decrement the use count and release all resources for an mm.
604 void mmput(struct mm_struct *mm)
608 if (atomic_dec_and_test(&mm->mm_users)) {
609 uprobe_clear_state(mm);
612 khugepaged_exit(mm); /* must run before exit_mmap */
614 set_mm_exe_file(mm, NULL);
615 if (!list_empty(&mm->mmlist)) {
616 spin_lock(&mmlist_lock);
617 list_del(&mm->mmlist);
618 spin_unlock(&mmlist_lock);
621 module_put(mm->binfmt->module);
625 EXPORT_SYMBOL_GPL(mmput);
628 * We added or removed a vma mapping the executable. The vmas are only mapped
629 * during exec and are not mapped with the mmap system call.
630 * Callers must hold down_write() on the mm's mmap_sem for these
632 void added_exe_file_vma(struct mm_struct *mm)
634 mm->num_exe_file_vmas++;
637 void removed_exe_file_vma(struct mm_struct *mm)
639 mm->num_exe_file_vmas--;
640 if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
647 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
650 get_file(new_exe_file);
653 mm->exe_file = new_exe_file;
654 mm->num_exe_file_vmas = 0;
657 struct file *get_mm_exe_file(struct mm_struct *mm)
659 struct file *exe_file;
661 /* We need mmap_sem to protect against races with removal of
662 * VM_EXECUTABLE vmas */
663 down_read(&mm->mmap_sem);
664 exe_file = mm->exe_file;
667 up_read(&mm->mmap_sem);
671 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
673 /* It's safe to write the exe_file pointer without exe_file_lock because
674 * this is called during fork when the task is not yet in /proc */
675 newmm->exe_file = get_mm_exe_file(oldmm);
679 * get_task_mm - acquire a reference to the task's mm
681 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
682 * this kernel workthread has transiently adopted a user mm with use_mm,
683 * to do its AIO) is not set and if so returns a reference to it, after
684 * bumping up the use count. User must release the mm via mmput()
685 * after use. Typically used by /proc and ptrace.
687 struct mm_struct *get_task_mm(struct task_struct *task)
689 struct mm_struct *mm;
694 if (task->flags & PF_KTHREAD)
697 atomic_inc(&mm->mm_users);
702 EXPORT_SYMBOL_GPL(get_task_mm);
704 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
706 struct mm_struct *mm;
709 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
713 mm = get_task_mm(task);
714 if (mm && mm != current->mm &&
715 !ptrace_may_access(task, mode)) {
717 mm = ERR_PTR(-EACCES);
719 mutex_unlock(&task->signal->cred_guard_mutex);
724 static void complete_vfork_done(struct task_struct *tsk)
726 struct completion *vfork;
729 vfork = tsk->vfork_done;
731 tsk->vfork_done = NULL;
737 static int wait_for_vfork_done(struct task_struct *child,
738 struct completion *vfork)
742 freezer_do_not_count();
743 killed = wait_for_completion_killable(vfork);
748 child->vfork_done = NULL;
752 put_task_struct(child);
756 /* Please note the differences between mmput and mm_release.
757 * mmput is called whenever we stop holding onto a mm_struct,
758 * error success whatever.
760 * mm_release is called after a mm_struct has been removed
761 * from the current process.
763 * This difference is important for error handling, when we
764 * only half set up a mm_struct for a new process and need to restore
765 * the old one. Because we mmput the new mm_struct before
766 * restoring the old one. . .
767 * Eric Biederman 10 January 1998
769 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
771 /* Get rid of any futexes when releasing the mm */
773 if (unlikely(tsk->robust_list)) {
774 exit_robust_list(tsk);
775 tsk->robust_list = NULL;
778 if (unlikely(tsk->compat_robust_list)) {
779 compat_exit_robust_list(tsk);
780 tsk->compat_robust_list = NULL;
783 if (unlikely(!list_empty(&tsk->pi_state_list)))
784 exit_pi_state_list(tsk);
787 uprobe_free_utask(tsk);
789 /* Get rid of any cached register state */
790 deactivate_mm(tsk, mm);
793 * If we're exiting normally, clear a user-space tid field if
794 * requested. We leave this alone when dying by signal, to leave
795 * the value intact in a core dump, and to save the unnecessary
796 * trouble, say, a killed vfork parent shouldn't touch this mm.
797 * Userland only wants this done for a sys_exit.
799 if (tsk->clear_child_tid) {
800 if (!(tsk->flags & PF_SIGNALED) &&
801 atomic_read(&mm->mm_users) > 1) {
803 * We don't check the error code - if userspace has
804 * not set up a proper pointer then tough luck.
806 put_user(0, tsk->clear_child_tid);
807 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
810 tsk->clear_child_tid = NULL;
814 * All done, finally we can wake up parent and return this mm to him.
815 * Also kthread_stop() uses this completion for synchronization.
818 complete_vfork_done(tsk);
822 * Allocate a new mm structure and copy contents from the
823 * mm structure of the passed in task structure.
825 struct mm_struct *dup_mm(struct task_struct *tsk)
827 struct mm_struct *mm, *oldmm = current->mm;
837 memcpy(mm, oldmm, sizeof(*mm));
840 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
841 mm->pmd_huge_pte = NULL;
843 uprobe_reset_state(mm);
845 if (!mm_init(mm, tsk))
848 if (init_new_context(tsk, mm))
851 dup_mm_exe_file(oldmm, mm);
853 err = dup_mmap(mm, oldmm);
857 mm->hiwater_rss = get_mm_rss(mm);
858 mm->hiwater_vm = mm->total_vm;
860 if (mm->binfmt && !try_module_get(mm->binfmt->module))
866 /* don't put binfmt in mmput, we haven't got module yet */
875 * If init_new_context() failed, we cannot use mmput() to free the mm
876 * because it calls destroy_context()
883 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
885 struct mm_struct *mm, *oldmm;
888 tsk->min_flt = tsk->maj_flt = 0;
889 tsk->nvcsw = tsk->nivcsw = 0;
890 #ifdef CONFIG_DETECT_HUNG_TASK
891 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
895 tsk->active_mm = NULL;
898 * Are we cloning a kernel thread?
900 * We need to steal a active VM for that..
906 if (clone_flags & CLONE_VM) {
907 atomic_inc(&oldmm->mm_users);
926 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
928 struct fs_struct *fs = current->fs;
929 if (clone_flags & CLONE_FS) {
930 /* tsk->fs is already what we want */
931 spin_lock(&fs->lock);
933 spin_unlock(&fs->lock);
937 spin_unlock(&fs->lock);
940 tsk->fs = copy_fs_struct(fs);
946 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
948 struct files_struct *oldf, *newf;
952 * A background process may not have any files ...
954 oldf = current->files;
958 if (clone_flags & CLONE_FILES) {
959 atomic_inc(&oldf->count);
963 newf = dup_fd(oldf, &error);
973 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
976 struct io_context *ioc = current->io_context;
977 struct io_context *new_ioc;
982 * Share io context with parent, if CLONE_IO is set
984 if (clone_flags & CLONE_IO) {
986 tsk->io_context = ioc;
987 } else if (ioprio_valid(ioc->ioprio)) {
988 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
989 if (unlikely(!new_ioc))
992 new_ioc->ioprio = ioc->ioprio;
993 put_io_context(new_ioc);
999 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1001 struct sighand_struct *sig;
1003 if (clone_flags & CLONE_SIGHAND) {
1004 atomic_inc(¤t->sighand->count);
1007 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1008 rcu_assign_pointer(tsk->sighand, sig);
1011 atomic_set(&sig->count, 1);
1012 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1016 void __cleanup_sighand(struct sighand_struct *sighand)
1018 if (atomic_dec_and_test(&sighand->count)) {
1019 signalfd_cleanup(sighand);
1020 kmem_cache_free(sighand_cachep, sighand);
1026 * Initialize POSIX timer handling for a thread group.
1028 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1030 unsigned long cpu_limit;
1032 /* Thread group counters. */
1033 thread_group_cputime_init(sig);
1035 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1036 if (cpu_limit != RLIM_INFINITY) {
1037 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1038 sig->cputimer.running = 1;
1041 /* The timer lists. */
1042 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1043 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1044 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1047 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1049 struct signal_struct *sig;
1051 if (clone_flags & CLONE_THREAD)
1054 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1059 sig->nr_threads = 1;
1060 atomic_set(&sig->live, 1);
1061 atomic_set(&sig->sigcnt, 1);
1062 init_waitqueue_head(&sig->wait_chldexit);
1063 if (clone_flags & CLONE_NEWPID)
1064 sig->flags |= SIGNAL_UNKILLABLE;
1065 sig->curr_target = tsk;
1066 init_sigpending(&sig->shared_pending);
1067 INIT_LIST_HEAD(&sig->posix_timers);
1069 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1070 sig->real_timer.function = it_real_fn;
1072 task_lock(current->group_leader);
1073 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1074 task_unlock(current->group_leader);
1076 posix_cpu_timers_init_group(sig);
1078 tty_audit_fork(sig);
1079 sched_autogroup_fork(sig);
1081 #ifdef CONFIG_CGROUPS
1082 init_rwsem(&sig->group_rwsem);
1085 sig->oom_adj = current->signal->oom_adj;
1086 sig->oom_score_adj = current->signal->oom_score_adj;
1087 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1089 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1090 current->signal->is_child_subreaper;
1092 mutex_init(&sig->cred_guard_mutex);
1097 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1099 unsigned long new_flags = p->flags;
1101 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1102 new_flags |= PF_FORKNOEXEC;
1103 p->flags = new_flags;
1106 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1108 current->clear_child_tid = tidptr;
1110 return task_pid_vnr(current);
1113 static void rt_mutex_init_task(struct task_struct *p)
1115 raw_spin_lock_init(&p->pi_lock);
1116 #ifdef CONFIG_RT_MUTEXES
1117 plist_head_init(&p->pi_waiters);
1118 p->pi_blocked_on = NULL;
1122 #ifdef CONFIG_MM_OWNER
1123 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1127 #endif /* CONFIG_MM_OWNER */
1130 * Initialize POSIX timer handling for a single task.
1132 static void posix_cpu_timers_init(struct task_struct *tsk)
1134 tsk->cputime_expires.prof_exp = 0;
1135 tsk->cputime_expires.virt_exp = 0;
1136 tsk->cputime_expires.sched_exp = 0;
1137 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1138 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1139 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1143 * This creates a new process as a copy of the old one,
1144 * but does not actually start it yet.
1146 * It copies the registers, and all the appropriate
1147 * parts of the process environment (as per the clone
1148 * flags). The actual kick-off is left to the caller.
1150 static struct task_struct *copy_process(unsigned long clone_flags,
1151 unsigned long stack_start,
1152 struct pt_regs *regs,
1153 unsigned long stack_size,
1154 int __user *child_tidptr,
1159 struct task_struct *p;
1160 int cgroup_callbacks_done = 0;
1162 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1163 return ERR_PTR(-EINVAL);
1166 * Thread groups must share signals as well, and detached threads
1167 * can only be started up within the thread group.
1169 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1170 return ERR_PTR(-EINVAL);
1173 * Shared signal handlers imply shared VM. By way of the above,
1174 * thread groups also imply shared VM. Blocking this case allows
1175 * for various simplifications in other code.
1177 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1178 return ERR_PTR(-EINVAL);
1181 * Siblings of global init remain as zombies on exit since they are
1182 * not reaped by their parent (swapper). To solve this and to avoid
1183 * multi-rooted process trees, prevent global and container-inits
1184 * from creating siblings.
1186 if ((clone_flags & CLONE_PARENT) &&
1187 current->signal->flags & SIGNAL_UNKILLABLE)
1188 return ERR_PTR(-EINVAL);
1190 retval = security_task_create(clone_flags);
1195 p = dup_task_struct(current);
1199 ftrace_graph_init_task(p);
1200 get_seccomp_filter(p);
1202 rt_mutex_init_task(p);
1204 #ifdef CONFIG_PROVE_LOCKING
1205 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1206 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1209 if (atomic_read(&p->real_cred->user->processes) >=
1210 task_rlimit(p, RLIMIT_NPROC)) {
1211 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1212 p->real_cred->user != INIT_USER)
1215 current->flags &= ~PF_NPROC_EXCEEDED;
1217 retval = copy_creds(p, clone_flags);
1222 * If multiple threads are within copy_process(), then this check
1223 * triggers too late. This doesn't hurt, the check is only there
1224 * to stop root fork bombs.
1227 if (nr_threads >= max_threads)
1228 goto bad_fork_cleanup_count;
1230 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1231 goto bad_fork_cleanup_count;
1234 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1235 copy_flags(clone_flags, p);
1236 INIT_LIST_HEAD(&p->children);
1237 INIT_LIST_HEAD(&p->sibling);
1238 rcu_copy_process(p);
1239 p->vfork_done = NULL;
1240 spin_lock_init(&p->alloc_lock);
1242 init_sigpending(&p->pending);
1244 p->utime = p->stime = p->gtime = 0;
1245 p->utimescaled = p->stimescaled = 0;
1246 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1247 p->prev_utime = p->prev_stime = 0;
1249 #if defined(SPLIT_RSS_COUNTING)
1250 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1253 p->default_timer_slack_ns = current->timer_slack_ns;
1255 task_io_accounting_init(&p->ioac);
1256 acct_clear_integrals(p);
1258 posix_cpu_timers_init(p);
1260 do_posix_clock_monotonic_gettime(&p->start_time);
1261 p->real_start_time = p->start_time;
1262 monotonic_to_bootbased(&p->real_start_time);
1263 p->io_context = NULL;
1264 p->audit_context = NULL;
1265 if (clone_flags & CLONE_THREAD)
1266 threadgroup_change_begin(current);
1269 p->mempolicy = mpol_dup(p->mempolicy);
1270 if (IS_ERR(p->mempolicy)) {
1271 retval = PTR_ERR(p->mempolicy);
1272 p->mempolicy = NULL;
1273 goto bad_fork_cleanup_cgroup;
1275 mpol_fix_fork_child_flag(p);
1277 #ifdef CONFIG_CPUSETS
1278 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1279 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1280 seqcount_init(&p->mems_allowed_seq);
1282 #ifdef CONFIG_TRACE_IRQFLAGS
1284 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1285 p->hardirqs_enabled = 1;
1287 p->hardirqs_enabled = 0;
1289 p->hardirq_enable_ip = 0;
1290 p->hardirq_enable_event = 0;
1291 p->hardirq_disable_ip = _THIS_IP_;
1292 p->hardirq_disable_event = 0;
1293 p->softirqs_enabled = 1;
1294 p->softirq_enable_ip = _THIS_IP_;
1295 p->softirq_enable_event = 0;
1296 p->softirq_disable_ip = 0;
1297 p->softirq_disable_event = 0;
1298 p->hardirq_context = 0;
1299 p->softirq_context = 0;
1301 #ifdef CONFIG_LOCKDEP
1302 p->lockdep_depth = 0; /* no locks held yet */
1303 p->curr_chain_key = 0;
1304 p->lockdep_recursion = 0;
1307 #ifdef CONFIG_DEBUG_MUTEXES
1308 p->blocked_on = NULL; /* not blocked yet */
1311 p->memcg_batch.do_batch = 0;
1312 p->memcg_batch.memcg = NULL;
1315 /* Perform scheduler related setup. Assign this task to a CPU. */
1318 retval = perf_event_init_task(p);
1320 goto bad_fork_cleanup_policy;
1321 retval = audit_alloc(p);
1323 goto bad_fork_cleanup_policy;
1324 /* copy all the process information */
1325 retval = copy_semundo(clone_flags, p);
1327 goto bad_fork_cleanup_audit;
1328 retval = copy_files(clone_flags, p);
1330 goto bad_fork_cleanup_semundo;
1331 retval = copy_fs(clone_flags, p);
1333 goto bad_fork_cleanup_files;
1334 retval = copy_sighand(clone_flags, p);
1336 goto bad_fork_cleanup_fs;
1337 retval = copy_signal(clone_flags, p);
1339 goto bad_fork_cleanup_sighand;
1340 retval = copy_mm(clone_flags, p);
1342 goto bad_fork_cleanup_signal;
1343 retval = copy_namespaces(clone_flags, p);
1345 goto bad_fork_cleanup_mm;
1346 retval = copy_io(clone_flags, p);
1348 goto bad_fork_cleanup_namespaces;
1349 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1351 goto bad_fork_cleanup_io;
1353 if (pid != &init_struct_pid) {
1355 pid = alloc_pid(p->nsproxy->pid_ns);
1357 goto bad_fork_cleanup_io;
1360 p->pid = pid_nr(pid);
1362 if (clone_flags & CLONE_THREAD)
1363 p->tgid = current->tgid;
1365 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1367 * Clear TID on mm_release()?
1369 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1374 p->robust_list = NULL;
1375 #ifdef CONFIG_COMPAT
1376 p->compat_robust_list = NULL;
1378 INIT_LIST_HEAD(&p->pi_state_list);
1379 p->pi_state_cache = NULL;
1381 uprobe_copy_process(p);
1383 * sigaltstack should be cleared when sharing the same VM
1385 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1386 p->sas_ss_sp = p->sas_ss_size = 0;
1389 * Syscall tracing and stepping should be turned off in the
1390 * child regardless of CLONE_PTRACE.
1392 user_disable_single_step(p);
1393 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1394 #ifdef TIF_SYSCALL_EMU
1395 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1397 clear_all_latency_tracing(p);
1399 /* ok, now we should be set up.. */
1400 if (clone_flags & CLONE_THREAD)
1401 p->exit_signal = -1;
1402 else if (clone_flags & CLONE_PARENT)
1403 p->exit_signal = current->group_leader->exit_signal;
1405 p->exit_signal = (clone_flags & CSIGNAL);
1407 p->pdeath_signal = 0;
1411 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1412 p->dirty_paused_when = 0;
1415 * Ok, make it visible to the rest of the system.
1416 * We dont wake it up yet.
1418 p->group_leader = p;
1419 INIT_LIST_HEAD(&p->thread_group);
1420 p->task_works = NULL;
1422 /* Now that the task is set up, run cgroup callbacks if
1423 * necessary. We need to run them before the task is visible
1424 * on the tasklist. */
1425 cgroup_fork_callbacks(p);
1426 cgroup_callbacks_done = 1;
1428 /* Need tasklist lock for parent etc handling! */
1429 write_lock_irq(&tasklist_lock);
1431 /* CLONE_PARENT re-uses the old parent */
1432 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1433 p->real_parent = current->real_parent;
1434 p->parent_exec_id = current->parent_exec_id;
1436 p->real_parent = current;
1437 p->parent_exec_id = current->self_exec_id;
1440 spin_lock(¤t->sighand->siglock);
1443 * Process group and session signals need to be delivered to just the
1444 * parent before the fork or both the parent and the child after the
1445 * fork. Restart if a signal comes in before we add the new process to
1446 * it's process group.
1447 * A fatal signal pending means that current will exit, so the new
1448 * thread can't slip out of an OOM kill (or normal SIGKILL).
1450 recalc_sigpending();
1451 if (signal_pending(current)) {
1452 spin_unlock(¤t->sighand->siglock);
1453 write_unlock_irq(&tasklist_lock);
1454 retval = -ERESTARTNOINTR;
1455 goto bad_fork_free_pid;
1458 if (clone_flags & CLONE_THREAD) {
1459 current->signal->nr_threads++;
1460 atomic_inc(¤t->signal->live);
1461 atomic_inc(¤t->signal->sigcnt);
1462 p->group_leader = current->group_leader;
1463 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1466 if (likely(p->pid)) {
1467 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1469 if (thread_group_leader(p)) {
1470 if (is_child_reaper(pid))
1471 p->nsproxy->pid_ns->child_reaper = p;
1473 p->signal->leader_pid = pid;
1474 p->signal->tty = tty_kref_get(current->signal->tty);
1475 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1476 attach_pid(p, PIDTYPE_SID, task_session(current));
1477 list_add_tail(&p->sibling, &p->real_parent->children);
1478 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1479 __this_cpu_inc(process_counts);
1481 attach_pid(p, PIDTYPE_PID, pid);
1486 spin_unlock(¤t->sighand->siglock);
1487 write_unlock_irq(&tasklist_lock);
1488 proc_fork_connector(p);
1489 cgroup_post_fork(p);
1490 if (clone_flags & CLONE_THREAD)
1491 threadgroup_change_end(current);
1494 trace_task_newtask(p, clone_flags);
1499 if (pid != &init_struct_pid)
1501 bad_fork_cleanup_io:
1504 bad_fork_cleanup_namespaces:
1505 if (unlikely(clone_flags & CLONE_NEWPID))
1506 pid_ns_release_proc(p->nsproxy->pid_ns);
1507 exit_task_namespaces(p);
1508 bad_fork_cleanup_mm:
1511 bad_fork_cleanup_signal:
1512 if (!(clone_flags & CLONE_THREAD))
1513 free_signal_struct(p->signal);
1514 bad_fork_cleanup_sighand:
1515 __cleanup_sighand(p->sighand);
1516 bad_fork_cleanup_fs:
1517 exit_fs(p); /* blocking */
1518 bad_fork_cleanup_files:
1519 exit_files(p); /* blocking */
1520 bad_fork_cleanup_semundo:
1522 bad_fork_cleanup_audit:
1524 bad_fork_cleanup_policy:
1525 perf_event_free_task(p);
1527 mpol_put(p->mempolicy);
1528 bad_fork_cleanup_cgroup:
1530 if (clone_flags & CLONE_THREAD)
1531 threadgroup_change_end(current);
1532 cgroup_exit(p, cgroup_callbacks_done);
1533 delayacct_tsk_free(p);
1534 module_put(task_thread_info(p)->exec_domain->module);
1535 bad_fork_cleanup_count:
1536 atomic_dec(&p->cred->user->processes);
1541 return ERR_PTR(retval);
1544 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1546 memset(regs, 0, sizeof(struct pt_regs));
1550 static inline void init_idle_pids(struct pid_link *links)
1554 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1555 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1556 links[type].pid = &init_struct_pid;
1560 struct task_struct * __cpuinit fork_idle(int cpu)
1562 struct task_struct *task;
1563 struct pt_regs regs;
1565 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL,
1566 &init_struct_pid, 0);
1567 if (!IS_ERR(task)) {
1568 init_idle_pids(task->pids);
1569 init_idle(task, cpu);
1576 * Ok, this is the main fork-routine.
1578 * It copies the process, and if successful kick-starts
1579 * it and waits for it to finish using the VM if required.
1581 long do_fork(unsigned long clone_flags,
1582 unsigned long stack_start,
1583 struct pt_regs *regs,
1584 unsigned long stack_size,
1585 int __user *parent_tidptr,
1586 int __user *child_tidptr)
1588 struct task_struct *p;
1593 * Do some preliminary argument and permissions checking before we
1594 * actually start allocating stuff
1596 if (clone_flags & CLONE_NEWUSER) {
1597 if (clone_flags & CLONE_THREAD)
1599 /* hopefully this check will go away when userns support is
1602 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1603 !capable(CAP_SETGID))
1608 * Determine whether and which event to report to ptracer. When
1609 * called from kernel_thread or CLONE_UNTRACED is explicitly
1610 * requested, no event is reported; otherwise, report if the event
1611 * for the type of forking is enabled.
1613 if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1614 if (clone_flags & CLONE_VFORK)
1615 trace = PTRACE_EVENT_VFORK;
1616 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1617 trace = PTRACE_EVENT_CLONE;
1619 trace = PTRACE_EVENT_FORK;
1621 if (likely(!ptrace_event_enabled(current, trace)))
1625 p = copy_process(clone_flags, stack_start, regs, stack_size,
1626 child_tidptr, NULL, trace);
1628 * Do this prior waking up the new thread - the thread pointer
1629 * might get invalid after that point, if the thread exits quickly.
1632 struct completion vfork;
1634 trace_sched_process_fork(current, p);
1636 nr = task_pid_vnr(p);
1638 if (clone_flags & CLONE_PARENT_SETTID)
1639 put_user(nr, parent_tidptr);
1641 if (clone_flags & CLONE_VFORK) {
1642 p->vfork_done = &vfork;
1643 init_completion(&vfork);
1647 wake_up_new_task(p);
1649 /* forking complete and child started to run, tell ptracer */
1650 if (unlikely(trace))
1651 ptrace_event(trace, nr);
1653 if (clone_flags & CLONE_VFORK) {
1654 if (!wait_for_vfork_done(p, &vfork))
1655 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1663 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1664 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1667 static void sighand_ctor(void *data)
1669 struct sighand_struct *sighand = data;
1671 spin_lock_init(&sighand->siglock);
1672 init_waitqueue_head(&sighand->signalfd_wqh);
1675 void __init proc_caches_init(void)
1677 sighand_cachep = kmem_cache_create("sighand_cache",
1678 sizeof(struct sighand_struct), 0,
1679 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1680 SLAB_NOTRACK, sighand_ctor);
1681 signal_cachep = kmem_cache_create("signal_cache",
1682 sizeof(struct signal_struct), 0,
1683 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1684 files_cachep = kmem_cache_create("files_cache",
1685 sizeof(struct files_struct), 0,
1686 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1687 fs_cachep = kmem_cache_create("fs_cache",
1688 sizeof(struct fs_struct), 0,
1689 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1691 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1692 * whole struct cpumask for the OFFSTACK case. We could change
1693 * this to *only* allocate as much of it as required by the
1694 * maximum number of CPU's we can ever have. The cpumask_allocation
1695 * is at the end of the structure, exactly for that reason.
1697 mm_cachep = kmem_cache_create("mm_struct",
1698 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1699 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1700 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1702 nsproxy_cache_init();
1706 * Check constraints on flags passed to the unshare system call.
1708 static int check_unshare_flags(unsigned long unshare_flags)
1710 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1711 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1712 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1715 * Not implemented, but pretend it works if there is nothing to
1716 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1717 * needs to unshare vm.
1719 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1720 /* FIXME: get_task_mm() increments ->mm_users */
1721 if (atomic_read(¤t->mm->mm_users) > 1)
1729 * Unshare the filesystem structure if it is being shared
1731 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1733 struct fs_struct *fs = current->fs;
1735 if (!(unshare_flags & CLONE_FS) || !fs)
1738 /* don't need lock here; in the worst case we'll do useless copy */
1742 *new_fsp = copy_fs_struct(fs);
1750 * Unshare file descriptor table if it is being shared
1752 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1754 struct files_struct *fd = current->files;
1757 if ((unshare_flags & CLONE_FILES) &&
1758 (fd && atomic_read(&fd->count) > 1)) {
1759 *new_fdp = dup_fd(fd, &error);
1768 * unshare allows a process to 'unshare' part of the process
1769 * context which was originally shared using clone. copy_*
1770 * functions used by do_fork() cannot be used here directly
1771 * because they modify an inactive task_struct that is being
1772 * constructed. Here we are modifying the current, active,
1775 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1777 struct fs_struct *fs, *new_fs = NULL;
1778 struct files_struct *fd, *new_fd = NULL;
1779 struct nsproxy *new_nsproxy = NULL;
1783 err = check_unshare_flags(unshare_flags);
1785 goto bad_unshare_out;
1788 * If unsharing namespace, must also unshare filesystem information.
1790 if (unshare_flags & CLONE_NEWNS)
1791 unshare_flags |= CLONE_FS;
1793 * CLONE_NEWIPC must also detach from the undolist: after switching
1794 * to a new ipc namespace, the semaphore arrays from the old
1795 * namespace are unreachable.
1797 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1799 err = unshare_fs(unshare_flags, &new_fs);
1801 goto bad_unshare_out;
1802 err = unshare_fd(unshare_flags, &new_fd);
1804 goto bad_unshare_cleanup_fs;
1805 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1807 goto bad_unshare_cleanup_fd;
1809 if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1812 * CLONE_SYSVSEM is equivalent to sys_exit().
1818 switch_task_namespaces(current, new_nsproxy);
1826 spin_lock(&fs->lock);
1827 current->fs = new_fs;
1832 spin_unlock(&fs->lock);
1836 fd = current->files;
1837 current->files = new_fd;
1841 task_unlock(current);
1845 put_nsproxy(new_nsproxy);
1847 bad_unshare_cleanup_fd:
1849 put_files_struct(new_fd);
1851 bad_unshare_cleanup_fs:
1853 free_fs_struct(new_fs);
1860 * Helper to unshare the files of the current task.
1861 * We don't want to expose copy_files internals to
1862 * the exec layer of the kernel.
1865 int unshare_files(struct files_struct **displaced)
1867 struct task_struct *task = current;
1868 struct files_struct *copy = NULL;
1871 error = unshare_fd(CLONE_FILES, ©);
1872 if (error || !copy) {
1876 *displaced = task->files;