4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 'fork.c' contains the help-routines for the 'fork' system call
9 * (see also entry.S and others).
10 * Fork is rather simple, once you get the hang of it, but the memory
11 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/seccomp.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/kthread.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/rmap.h>
54 #include <linux/ksm.h>
55 #include <linux/acct.h>
56 #include <linux/tsacct_kern.h>
57 #include <linux/cn_proc.h>
58 #include <linux/freezer.h>
59 #include <linux/delayacct.h>
60 #include <linux/taskstats_kern.h>
61 #include <linux/random.h>
62 #include <linux/tty.h>
63 #include <linux/blkdev.h>
64 #include <linux/fs_struct.h>
65 #include <linux/magic.h>
66 #include <linux/perf_event.h>
67 #include <linux/posix-timers.h>
68 #include <linux/user-return-notifier.h>
69 #include <linux/oom.h>
70 #include <linux/khugepaged.h>
71 #include <linux/signalfd.h>
72 #include <linux/uprobes.h>
73 #include <linux/aio.h>
75 #include <asm/pgtable.h>
76 #include <asm/pgalloc.h>
77 #include <asm/uaccess.h>
78 #include <asm/mmu_context.h>
79 #include <asm/cacheflush.h>
80 #include <asm/tlbflush.h>
82 #include <trace/events/sched.h>
84 #define CREATE_TRACE_POINTS
85 #include <trace/events/task.h>
88 * Protected counters by write_lock_irq(&tasklist_lock)
90 unsigned long total_forks; /* Handle normal Linux uptimes. */
91 int nr_threads; /* The idle threads do not count.. */
93 int max_threads; /* tunable limit on nr_threads */
95 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
97 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
99 #ifdef CONFIG_PROVE_RCU
100 int lockdep_tasklist_lock_is_held(void)
102 return lockdep_is_held(&tasklist_lock);
104 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
105 #endif /* #ifdef CONFIG_PROVE_RCU */
107 int nr_processes(void)
112 for_each_possible_cpu(cpu)
113 total += per_cpu(process_counts, cpu);
118 void __weak arch_release_task_struct(struct task_struct *tsk)
122 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
123 static struct kmem_cache *task_struct_cachep;
125 static inline struct task_struct *alloc_task_struct_node(int node)
127 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
130 static inline void free_task_struct(struct task_struct *tsk)
132 kmem_cache_free(task_struct_cachep, tsk);
136 void __weak arch_release_thread_info(struct thread_info *ti)
140 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
143 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
144 * kmemcache based allocator.
146 # if THREAD_SIZE >= PAGE_SIZE
147 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
150 struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
153 return page ? page_address(page) : NULL;
156 static inline void free_thread_info(struct thread_info *ti)
158 free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
161 static struct kmem_cache *thread_info_cache;
163 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
166 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
169 static void free_thread_info(struct thread_info *ti)
171 kmem_cache_free(thread_info_cache, ti);
174 void thread_info_cache_init(void)
176 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
177 THREAD_SIZE, 0, NULL);
178 BUG_ON(thread_info_cache == NULL);
183 /* SLAB cache for signal_struct structures (tsk->signal) */
184 static struct kmem_cache *signal_cachep;
186 /* SLAB cache for sighand_struct structures (tsk->sighand) */
187 struct kmem_cache *sighand_cachep;
189 /* SLAB cache for files_struct structures (tsk->files) */
190 struct kmem_cache *files_cachep;
192 /* SLAB cache for fs_struct structures (tsk->fs) */
193 struct kmem_cache *fs_cachep;
195 /* SLAB cache for vm_area_struct structures */
196 struct kmem_cache *vm_area_cachep;
198 /* SLAB cache for mm_struct structures (tsk->mm) */
199 static struct kmem_cache *mm_cachep;
201 static void account_kernel_stack(struct thread_info *ti, int account)
203 struct zone *zone = page_zone(virt_to_page(ti));
205 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
208 void free_task(struct task_struct *tsk)
210 account_kernel_stack(tsk->stack, -1);
211 arch_release_thread_info(tsk->stack);
212 free_thread_info(tsk->stack);
213 rt_mutex_debug_task_free(tsk);
214 ftrace_graph_exit_task(tsk);
215 put_seccomp_filter(tsk);
216 arch_release_task_struct(tsk);
217 free_task_struct(tsk);
219 EXPORT_SYMBOL(free_task);
221 static inline void free_signal_struct(struct signal_struct *sig)
223 taskstats_tgid_free(sig);
224 sched_autogroup_exit(sig);
225 kmem_cache_free(signal_cachep, sig);
228 static inline void put_signal_struct(struct signal_struct *sig)
230 if (atomic_dec_and_test(&sig->sigcnt))
231 free_signal_struct(sig);
234 void __put_task_struct(struct task_struct *tsk)
236 WARN_ON(!tsk->exit_state);
237 WARN_ON(atomic_read(&tsk->usage));
238 WARN_ON(tsk == current);
241 security_task_free(tsk);
243 delayacct_tsk_free(tsk);
244 put_signal_struct(tsk->signal);
246 if (!profile_handoff_task(tsk))
249 EXPORT_SYMBOL_GPL(__put_task_struct);
251 void __init __weak arch_task_cache_init(void) { }
253 void __init fork_init(unsigned long mempages)
255 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
256 #ifndef ARCH_MIN_TASKALIGN
257 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
259 /* create a slab on which task_structs can be allocated */
261 kmem_cache_create("task_struct", sizeof(struct task_struct),
262 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
265 /* do the arch specific task caches init */
266 arch_task_cache_init();
269 * The default maximum number of threads is set to a safe
270 * value: the thread structures can take up at most half
273 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
276 * we need to allow at least 20 threads to boot a system
278 if (max_threads < 20)
281 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
282 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
283 init_task.signal->rlim[RLIMIT_SIGPENDING] =
284 init_task.signal->rlim[RLIMIT_NPROC];
287 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
288 struct task_struct *src)
294 static struct task_struct *dup_task_struct(struct task_struct *orig)
296 struct task_struct *tsk;
297 struct thread_info *ti;
298 unsigned long *stackend;
299 int node = tsk_fork_get_node(orig);
302 tsk = alloc_task_struct_node(node);
306 ti = alloc_thread_info_node(tsk, node);
310 err = arch_dup_task_struct(tsk, orig);
316 setup_thread_stack(tsk, orig);
317 clear_user_return_notifier(tsk);
318 clear_tsk_need_resched(tsk);
319 stackend = end_of_stack(tsk);
320 *stackend = STACK_END_MAGIC; /* for overflow detection */
322 #ifdef CONFIG_CC_STACKPROTECTOR
323 tsk->stack_canary = get_random_int();
327 * One for us, one for whoever does the "release_task()" (usually
330 atomic_set(&tsk->usage, 2);
331 #ifdef CONFIG_BLK_DEV_IO_TRACE
334 tsk->splice_pipe = NULL;
335 tsk->task_frag.page = NULL;
337 account_kernel_stack(ti, 1);
342 free_thread_info(ti);
344 free_task_struct(tsk);
349 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
351 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
352 struct rb_node **rb_link, *rb_parent;
354 unsigned long charge;
356 uprobe_start_dup_mmap();
357 down_write(&oldmm->mmap_sem);
358 flush_cache_dup_mm(oldmm);
359 uprobe_dup_mmap(oldmm, mm);
361 * Not linked in yet - no deadlock potential:
363 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
367 mm->mmap_cache = NULL;
369 cpumask_clear(mm_cpumask(mm));
371 rb_link = &mm->mm_rb.rb_node;
374 retval = ksm_fork(mm, oldmm);
377 retval = khugepaged_fork(mm, oldmm);
382 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
385 if (mpnt->vm_flags & VM_DONTCOPY) {
386 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
391 if (mpnt->vm_flags & VM_ACCOUNT) {
392 unsigned long len = vma_pages(mpnt);
394 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
398 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
402 INIT_LIST_HEAD(&tmp->anon_vma_chain);
403 retval = vma_dup_policy(mpnt, tmp);
405 goto fail_nomem_policy;
407 if (anon_vma_fork(tmp, mpnt))
408 goto fail_nomem_anon_vma_fork;
409 tmp->vm_flags &= ~VM_LOCKED;
410 tmp->vm_next = tmp->vm_prev = NULL;
413 struct inode *inode = file_inode(file);
414 struct address_space *mapping = file->f_mapping;
417 if (tmp->vm_flags & VM_DENYWRITE)
418 atomic_dec(&inode->i_writecount);
419 mutex_lock(&mapping->i_mmap_mutex);
420 if (tmp->vm_flags & VM_SHARED)
421 mapping->i_mmap_writable++;
422 flush_dcache_mmap_lock(mapping);
423 /* insert tmp into the share list, just after mpnt */
424 if (unlikely(tmp->vm_flags & VM_NONLINEAR))
425 vma_nonlinear_insert(tmp,
426 &mapping->i_mmap_nonlinear);
428 vma_interval_tree_insert_after(tmp, mpnt,
430 flush_dcache_mmap_unlock(mapping);
431 mutex_unlock(&mapping->i_mmap_mutex);
435 * Clear hugetlb-related page reserves for children. This only
436 * affects MAP_PRIVATE mappings. Faults generated by the child
437 * are not guaranteed to succeed, even if read-only
439 if (is_vm_hugetlb_page(tmp))
440 reset_vma_resv_huge_pages(tmp);
443 * Link in the new vma and copy the page table entries.
446 pprev = &tmp->vm_next;
450 __vma_link_rb(mm, tmp, rb_link, rb_parent);
451 rb_link = &tmp->vm_rb.rb_right;
452 rb_parent = &tmp->vm_rb;
455 retval = copy_page_range(mm, oldmm, mpnt);
457 if (tmp->vm_ops && tmp->vm_ops->open)
458 tmp->vm_ops->open(tmp);
463 /* a new mm has just been created */
464 arch_dup_mmap(oldmm, mm);
467 up_write(&mm->mmap_sem);
469 up_write(&oldmm->mmap_sem);
470 uprobe_end_dup_mmap();
472 fail_nomem_anon_vma_fork:
473 mpol_put(vma_policy(tmp));
475 kmem_cache_free(vm_area_cachep, tmp);
478 vm_unacct_memory(charge);
482 static inline int mm_alloc_pgd(struct mm_struct *mm)
484 mm->pgd = pgd_alloc(mm);
485 if (unlikely(!mm->pgd))
490 static inline void mm_free_pgd(struct mm_struct *mm)
492 pgd_free(mm, mm->pgd);
495 #define dup_mmap(mm, oldmm) (0)
496 #define mm_alloc_pgd(mm) (0)
497 #define mm_free_pgd(mm)
498 #endif /* CONFIG_MMU */
500 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
502 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
503 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
505 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
507 static int __init coredump_filter_setup(char *s)
509 default_dump_filter =
510 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
511 MMF_DUMP_FILTER_MASK;
515 __setup("coredump_filter=", coredump_filter_setup);
517 #include <linux/init_task.h>
519 static void mm_init_aio(struct mm_struct *mm)
522 spin_lock_init(&mm->ioctx_lock);
523 mm->ioctx_table = NULL;
527 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
529 atomic_set(&mm->mm_users, 1);
530 atomic_set(&mm->mm_count, 1);
531 init_rwsem(&mm->mmap_sem);
532 INIT_LIST_HEAD(&mm->mmlist);
533 mm->core_state = NULL;
534 atomic_long_set(&mm->nr_ptes, 0);
535 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
536 spin_lock_init(&mm->page_table_lock);
538 mm_init_owner(mm, p);
539 clear_tlb_flush_pending(mm);
542 mm->flags = current->mm->flags & MMF_INIT_MASK;
543 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
545 mm->flags = default_dump_filter;
549 if (likely(!mm_alloc_pgd(mm))) {
550 mmu_notifier_mm_init(mm);
558 static void check_mm(struct mm_struct *mm)
562 for (i = 0; i < NR_MM_COUNTERS; i++) {
563 long x = atomic_long_read(&mm->rss_stat.count[i]);
566 printk(KERN_ALERT "BUG: Bad rss-counter state "
567 "mm:%p idx:%d val:%ld\n", mm, i, x);
570 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
571 VM_BUG_ON(mm->pmd_huge_pte);
576 * Allocate and initialize an mm_struct.
578 struct mm_struct *mm_alloc(void)
580 struct mm_struct *mm;
586 memset(mm, 0, sizeof(*mm));
588 return mm_init(mm, current);
592 * Called when the last reference to the mm
593 * is dropped: either by a lazy thread or by
594 * mmput. Free the page directory and the mm.
596 void __mmdrop(struct mm_struct *mm)
598 BUG_ON(mm == &init_mm);
601 mmu_notifier_mm_destroy(mm);
605 EXPORT_SYMBOL_GPL(__mmdrop);
608 * Decrement the use count and release all resources for an mm.
610 void mmput(struct mm_struct *mm)
614 if (atomic_dec_and_test(&mm->mm_users)) {
615 uprobe_clear_state(mm);
618 khugepaged_exit(mm); /* must run before exit_mmap */
620 set_mm_exe_file(mm, NULL);
621 if (!list_empty(&mm->mmlist)) {
622 spin_lock(&mmlist_lock);
623 list_del(&mm->mmlist);
624 spin_unlock(&mmlist_lock);
627 module_put(mm->binfmt->module);
631 EXPORT_SYMBOL_GPL(mmput);
633 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
636 get_file(new_exe_file);
639 mm->exe_file = new_exe_file;
642 struct file *get_mm_exe_file(struct mm_struct *mm)
644 struct file *exe_file;
646 /* We need mmap_sem to protect against races with removal of exe_file */
647 down_read(&mm->mmap_sem);
648 exe_file = mm->exe_file;
651 up_read(&mm->mmap_sem);
655 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
657 /* It's safe to write the exe_file pointer without exe_file_lock because
658 * this is called during fork when the task is not yet in /proc */
659 newmm->exe_file = get_mm_exe_file(oldmm);
663 * get_task_mm - acquire a reference to the task's mm
665 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
666 * this kernel workthread has transiently adopted a user mm with use_mm,
667 * to do its AIO) is not set and if so returns a reference to it, after
668 * bumping up the use count. User must release the mm via mmput()
669 * after use. Typically used by /proc and ptrace.
671 struct mm_struct *get_task_mm(struct task_struct *task)
673 struct mm_struct *mm;
678 if (task->flags & PF_KTHREAD)
681 atomic_inc(&mm->mm_users);
686 EXPORT_SYMBOL_GPL(get_task_mm);
688 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
690 struct mm_struct *mm;
693 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
697 mm = get_task_mm(task);
698 if (mm && mm != current->mm &&
699 !ptrace_may_access(task, mode)) {
701 mm = ERR_PTR(-EACCES);
703 mutex_unlock(&task->signal->cred_guard_mutex);
708 static void complete_vfork_done(struct task_struct *tsk)
710 struct completion *vfork;
713 vfork = tsk->vfork_done;
715 tsk->vfork_done = NULL;
721 static int wait_for_vfork_done(struct task_struct *child,
722 struct completion *vfork)
726 freezer_do_not_count();
727 killed = wait_for_completion_killable(vfork);
732 child->vfork_done = NULL;
736 put_task_struct(child);
740 /* Please note the differences between mmput and mm_release.
741 * mmput is called whenever we stop holding onto a mm_struct,
742 * error success whatever.
744 * mm_release is called after a mm_struct has been removed
745 * from the current process.
747 * This difference is important for error handling, when we
748 * only half set up a mm_struct for a new process and need to restore
749 * the old one. Because we mmput the new mm_struct before
750 * restoring the old one. . .
751 * Eric Biederman 10 January 1998
753 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
755 /* Get rid of any futexes when releasing the mm */
757 if (unlikely(tsk->robust_list)) {
758 exit_robust_list(tsk);
759 tsk->robust_list = NULL;
762 if (unlikely(tsk->compat_robust_list)) {
763 compat_exit_robust_list(tsk);
764 tsk->compat_robust_list = NULL;
767 if (unlikely(!list_empty(&tsk->pi_state_list)))
768 exit_pi_state_list(tsk);
771 uprobe_free_utask(tsk);
773 /* Get rid of any cached register state */
774 deactivate_mm(tsk, mm);
777 * If we're exiting normally, clear a user-space tid field if
778 * requested. We leave this alone when dying by signal, to leave
779 * the value intact in a core dump, and to save the unnecessary
780 * trouble, say, a killed vfork parent shouldn't touch this mm.
781 * Userland only wants this done for a sys_exit.
783 if (tsk->clear_child_tid) {
784 if (!(tsk->flags & PF_SIGNALED) &&
785 atomic_read(&mm->mm_users) > 1) {
787 * We don't check the error code - if userspace has
788 * not set up a proper pointer then tough luck.
790 put_user(0, tsk->clear_child_tid);
791 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
794 tsk->clear_child_tid = NULL;
798 * All done, finally we can wake up parent and return this mm to him.
799 * Also kthread_stop() uses this completion for synchronization.
802 complete_vfork_done(tsk);
806 * Allocate a new mm structure and copy contents from the
807 * mm structure of the passed in task structure.
809 static struct mm_struct *dup_mm(struct task_struct *tsk)
811 struct mm_struct *mm, *oldmm = current->mm;
818 memcpy(mm, oldmm, sizeof(*mm));
821 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
822 mm->pmd_huge_pte = NULL;
824 if (!mm_init(mm, tsk))
827 if (init_new_context(tsk, mm))
830 dup_mm_exe_file(oldmm, mm);
832 err = dup_mmap(mm, oldmm);
836 mm->hiwater_rss = get_mm_rss(mm);
837 mm->hiwater_vm = mm->total_vm;
839 if (mm->binfmt && !try_module_get(mm->binfmt->module))
845 /* don't put binfmt in mmput, we haven't got module yet */
854 * If init_new_context() failed, we cannot use mmput() to free the mm
855 * because it calls destroy_context()
862 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
864 struct mm_struct *mm, *oldmm;
867 tsk->min_flt = tsk->maj_flt = 0;
868 tsk->nvcsw = tsk->nivcsw = 0;
869 #ifdef CONFIG_DETECT_HUNG_TASK
870 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
874 tsk->active_mm = NULL;
877 * Are we cloning a kernel thread?
879 * We need to steal a active VM for that..
885 if (clone_flags & CLONE_VM) {
886 atomic_inc(&oldmm->mm_users);
905 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
907 struct fs_struct *fs = current->fs;
908 if (clone_flags & CLONE_FS) {
909 /* tsk->fs is already what we want */
910 spin_lock(&fs->lock);
912 spin_unlock(&fs->lock);
916 spin_unlock(&fs->lock);
919 tsk->fs = copy_fs_struct(fs);
925 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
927 struct files_struct *oldf, *newf;
931 * A background process may not have any files ...
933 oldf = current->files;
937 if (clone_flags & CLONE_FILES) {
938 atomic_inc(&oldf->count);
942 newf = dup_fd(oldf, &error);
952 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
955 struct io_context *ioc = current->io_context;
956 struct io_context *new_ioc;
961 * Share io context with parent, if CLONE_IO is set
963 if (clone_flags & CLONE_IO) {
965 tsk->io_context = ioc;
966 } else if (ioprio_valid(ioc->ioprio)) {
967 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
968 if (unlikely(!new_ioc))
971 new_ioc->ioprio = ioc->ioprio;
972 put_io_context(new_ioc);
978 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
980 struct sighand_struct *sig;
982 if (clone_flags & CLONE_SIGHAND) {
983 atomic_inc(¤t->sighand->count);
986 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
987 rcu_assign_pointer(tsk->sighand, sig);
990 atomic_set(&sig->count, 1);
991 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
995 void __cleanup_sighand(struct sighand_struct *sighand)
997 if (atomic_dec_and_test(&sighand->count)) {
998 signalfd_cleanup(sighand);
999 kmem_cache_free(sighand_cachep, sighand);
1005 * Initialize POSIX timer handling for a thread group.
1007 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1009 unsigned long cpu_limit;
1011 /* Thread group counters. */
1012 thread_group_cputime_init(sig);
1014 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1015 if (cpu_limit != RLIM_INFINITY) {
1016 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1017 sig->cputimer.running = 1;
1020 /* The timer lists. */
1021 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1022 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1023 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1026 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1028 struct signal_struct *sig;
1030 if (clone_flags & CLONE_THREAD)
1033 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1038 sig->nr_threads = 1;
1039 atomic_set(&sig->live, 1);
1040 atomic_set(&sig->sigcnt, 1);
1042 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1043 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1044 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1046 init_waitqueue_head(&sig->wait_chldexit);
1047 sig->curr_target = tsk;
1048 init_sigpending(&sig->shared_pending);
1049 INIT_LIST_HEAD(&sig->posix_timers);
1051 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1052 sig->real_timer.function = it_real_fn;
1054 task_lock(current->group_leader);
1055 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1056 task_unlock(current->group_leader);
1058 posix_cpu_timers_init_group(sig);
1060 tty_audit_fork(sig);
1061 sched_autogroup_fork(sig);
1063 #ifdef CONFIG_CGROUPS
1064 init_rwsem(&sig->group_rwsem);
1067 sig->oom_score_adj = current->signal->oom_score_adj;
1068 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1070 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1071 current->signal->is_child_subreaper;
1073 mutex_init(&sig->cred_guard_mutex);
1078 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1080 unsigned long new_flags = p->flags;
1082 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1083 new_flags |= PF_FORKNOEXEC;
1084 p->flags = new_flags;
1087 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1089 current->clear_child_tid = tidptr;
1091 return task_pid_vnr(current);
1094 static void rt_mutex_init_task(struct task_struct *p)
1096 raw_spin_lock_init(&p->pi_lock);
1097 #ifdef CONFIG_RT_MUTEXES
1098 p->pi_waiters = RB_ROOT;
1099 p->pi_waiters_leftmost = NULL;
1100 p->pi_blocked_on = NULL;
1101 p->pi_top_task = NULL;
1105 #ifdef CONFIG_MM_OWNER
1106 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1110 #endif /* CONFIG_MM_OWNER */
1113 * Initialize POSIX timer handling for a single task.
1115 static void posix_cpu_timers_init(struct task_struct *tsk)
1117 tsk->cputime_expires.prof_exp = 0;
1118 tsk->cputime_expires.virt_exp = 0;
1119 tsk->cputime_expires.sched_exp = 0;
1120 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1121 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1122 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1126 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1128 task->pids[type].pid = pid;
1132 * This creates a new process as a copy of the old one,
1133 * but does not actually start it yet.
1135 * It copies the registers, and all the appropriate
1136 * parts of the process environment (as per the clone
1137 * flags). The actual kick-off is left to the caller.
1139 static struct task_struct *copy_process(unsigned long clone_flags,
1140 unsigned long stack_start,
1141 unsigned long stack_size,
1142 int __user *child_tidptr,
1147 struct task_struct *p;
1149 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1150 return ERR_PTR(-EINVAL);
1152 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1153 return ERR_PTR(-EINVAL);
1156 * Thread groups must share signals as well, and detached threads
1157 * can only be started up within the thread group.
1159 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1160 return ERR_PTR(-EINVAL);
1163 * Shared signal handlers imply shared VM. By way of the above,
1164 * thread groups also imply shared VM. Blocking this case allows
1165 * for various simplifications in other code.
1167 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1168 return ERR_PTR(-EINVAL);
1171 * Siblings of global init remain as zombies on exit since they are
1172 * not reaped by their parent (swapper). To solve this and to avoid
1173 * multi-rooted process trees, prevent global and container-inits
1174 * from creating siblings.
1176 if ((clone_flags & CLONE_PARENT) &&
1177 current->signal->flags & SIGNAL_UNKILLABLE)
1178 return ERR_PTR(-EINVAL);
1181 * If the new process will be in a different pid or user namespace
1182 * do not allow it to share a thread group or signal handlers or
1183 * parent with the forking task.
1185 if (clone_flags & CLONE_SIGHAND) {
1186 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1187 (task_active_pid_ns(current) !=
1188 current->nsproxy->pid_ns_for_children))
1189 return ERR_PTR(-EINVAL);
1192 retval = security_task_create(clone_flags);
1197 p = dup_task_struct(current);
1201 ftrace_graph_init_task(p);
1202 get_seccomp_filter(p);
1204 rt_mutex_init_task(p);
1206 #ifdef CONFIG_PROVE_LOCKING
1207 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1208 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1211 if (atomic_read(&p->real_cred->user->processes) >=
1212 task_rlimit(p, RLIMIT_NPROC)) {
1213 if (p->real_cred->user != INIT_USER &&
1214 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1217 current->flags &= ~PF_NPROC_EXCEEDED;
1219 retval = copy_creds(p, clone_flags);
1224 * If multiple threads are within copy_process(), then this check
1225 * triggers too late. This doesn't hurt, the check is only there
1226 * to stop root fork bombs.
1229 if (nr_threads >= max_threads)
1230 goto bad_fork_cleanup_count;
1232 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1233 goto bad_fork_cleanup_count;
1235 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1236 copy_flags(clone_flags, p);
1237 INIT_LIST_HEAD(&p->children);
1238 INIT_LIST_HEAD(&p->sibling);
1239 rcu_copy_process(p);
1240 p->vfork_done = NULL;
1241 spin_lock_init(&p->alloc_lock);
1243 init_sigpending(&p->pending);
1245 p->utime = p->stime = p->gtime = 0;
1246 p->utimescaled = p->stimescaled = 0;
1247 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1248 p->prev_cputime.utime = p->prev_cputime.stime = 0;
1250 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1251 seqlock_init(&p->vtime_seqlock);
1253 p->vtime_snap_whence = VTIME_SLEEPING;
1256 #if defined(SPLIT_RSS_COUNTING)
1257 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1260 p->default_timer_slack_ns = current->timer_slack_ns;
1262 task_io_accounting_init(&p->ioac);
1263 acct_clear_integrals(p);
1265 posix_cpu_timers_init(p);
1267 do_posix_clock_monotonic_gettime(&p->start_time);
1268 p->real_start_time = p->start_time;
1269 monotonic_to_bootbased(&p->real_start_time);
1270 p->io_context = NULL;
1271 p->audit_context = NULL;
1272 if (clone_flags & CLONE_THREAD)
1273 threadgroup_change_begin(current);
1276 p->mempolicy = mpol_dup(p->mempolicy);
1277 if (IS_ERR(p->mempolicy)) {
1278 retval = PTR_ERR(p->mempolicy);
1279 p->mempolicy = NULL;
1280 goto bad_fork_cleanup_threadgroup_lock;
1282 mpol_fix_fork_child_flag(p);
1284 #ifdef CONFIG_CPUSETS
1285 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1286 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1287 seqcount_init(&p->mems_allowed_seq);
1289 #ifdef CONFIG_TRACE_IRQFLAGS
1291 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 */
1314 p->memcg_batch.do_batch = 0;
1315 p->memcg_batch.memcg = NULL;
1317 #ifdef CONFIG_BCACHE
1318 p->sequential_io = 0;
1319 p->sequential_io_avg = 0;
1322 /* Perform scheduler related setup. Assign this task to a CPU. */
1323 retval = sched_fork(clone_flags, p);
1325 goto bad_fork_cleanup_policy;
1327 retval = perf_event_init_task(p);
1329 goto bad_fork_cleanup_policy;
1330 retval = audit_alloc(p);
1332 goto bad_fork_cleanup_policy;
1333 /* copy all the process information */
1334 retval = copy_semundo(clone_flags, p);
1336 goto bad_fork_cleanup_audit;
1337 retval = copy_files(clone_flags, p);
1339 goto bad_fork_cleanup_semundo;
1340 retval = copy_fs(clone_flags, p);
1342 goto bad_fork_cleanup_files;
1343 retval = copy_sighand(clone_flags, p);
1345 goto bad_fork_cleanup_fs;
1346 retval = copy_signal(clone_flags, p);
1348 goto bad_fork_cleanup_sighand;
1349 retval = copy_mm(clone_flags, p);
1351 goto bad_fork_cleanup_signal;
1352 retval = copy_namespaces(clone_flags, p);
1354 goto bad_fork_cleanup_mm;
1355 retval = copy_io(clone_flags, p);
1357 goto bad_fork_cleanup_namespaces;
1358 retval = copy_thread(clone_flags, stack_start, stack_size, p);
1360 goto bad_fork_cleanup_io;
1362 if (pid != &init_struct_pid) {
1364 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1366 goto bad_fork_cleanup_io;
1369 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1371 * Clear TID on mm_release()?
1373 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1378 p->robust_list = NULL;
1379 #ifdef CONFIG_COMPAT
1380 p->compat_robust_list = NULL;
1382 INIT_LIST_HEAD(&p->pi_state_list);
1383 p->pi_state_cache = NULL;
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 p->pid = pid_nr(pid);
1404 if (clone_flags & CLONE_THREAD) {
1405 p->exit_signal = -1;
1406 p->group_leader = current->group_leader;
1407 p->tgid = current->tgid;
1409 if (clone_flags & CLONE_PARENT)
1410 p->exit_signal = current->group_leader->exit_signal;
1412 p->exit_signal = (clone_flags & CSIGNAL);
1413 p->group_leader = p;
1418 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1419 p->dirty_paused_when = 0;
1421 p->pdeath_signal = 0;
1422 INIT_LIST_HEAD(&p->thread_group);
1423 p->task_works = NULL;
1426 * Make it visible to the rest of the system, but dont wake it up yet.
1427 * 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 (likely(p->pid)) {
1459 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1461 init_task_pid(p, PIDTYPE_PID, pid);
1462 if (thread_group_leader(p)) {
1463 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1464 init_task_pid(p, PIDTYPE_SID, task_session(current));
1466 if (is_child_reaper(pid)) {
1467 ns_of_pid(pid)->child_reaper = p;
1468 p->signal->flags |= SIGNAL_UNKILLABLE;
1471 p->signal->leader_pid = pid;
1472 p->signal->tty = tty_kref_get(current->signal->tty);
1473 list_add_tail(&p->sibling, &p->real_parent->children);
1474 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1475 attach_pid(p, PIDTYPE_PGID);
1476 attach_pid(p, PIDTYPE_SID);
1477 __this_cpu_inc(process_counts);
1479 current->signal->nr_threads++;
1480 atomic_inc(¤t->signal->live);
1481 atomic_inc(¤t->signal->sigcnt);
1482 list_add_tail_rcu(&p->thread_group,
1483 &p->group_leader->thread_group);
1484 list_add_tail_rcu(&p->thread_node,
1485 &p->signal->thread_head);
1487 attach_pid(p, PIDTYPE_PID);
1492 spin_unlock(¤t->sighand->siglock);
1493 write_unlock_irq(&tasklist_lock);
1494 proc_fork_connector(p);
1495 cgroup_post_fork(p);
1496 if (clone_flags & CLONE_THREAD)
1497 threadgroup_change_end(current);
1500 trace_task_newtask(p, clone_flags);
1501 uprobe_copy_process(p, clone_flags);
1506 if (pid != &init_struct_pid)
1508 bad_fork_cleanup_io:
1511 bad_fork_cleanup_namespaces:
1512 exit_task_namespaces(p);
1513 bad_fork_cleanup_mm:
1516 bad_fork_cleanup_signal:
1517 if (!(clone_flags & CLONE_THREAD))
1518 free_signal_struct(p->signal);
1519 bad_fork_cleanup_sighand:
1520 __cleanup_sighand(p->sighand);
1521 bad_fork_cleanup_fs:
1522 exit_fs(p); /* blocking */
1523 bad_fork_cleanup_files:
1524 exit_files(p); /* blocking */
1525 bad_fork_cleanup_semundo:
1527 bad_fork_cleanup_audit:
1529 bad_fork_cleanup_policy:
1530 perf_event_free_task(p);
1532 mpol_put(p->mempolicy);
1533 bad_fork_cleanup_threadgroup_lock:
1535 if (clone_flags & CLONE_THREAD)
1536 threadgroup_change_end(current);
1537 delayacct_tsk_free(p);
1538 module_put(task_thread_info(p)->exec_domain->module);
1539 bad_fork_cleanup_count:
1540 atomic_dec(&p->cred->user->processes);
1545 return ERR_PTR(retval);
1548 static inline void init_idle_pids(struct pid_link *links)
1552 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1553 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1554 links[type].pid = &init_struct_pid;
1558 struct task_struct *fork_idle(int cpu)
1560 struct task_struct *task;
1561 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1562 if (!IS_ERR(task)) {
1563 init_idle_pids(task->pids);
1564 init_idle(task, cpu);
1571 * Ok, this is the main fork-routine.
1573 * It copies the process, and if successful kick-starts
1574 * it and waits for it to finish using the VM if required.
1576 long do_fork(unsigned long clone_flags,
1577 unsigned long stack_start,
1578 unsigned long stack_size,
1579 int __user *parent_tidptr,
1580 int __user *child_tidptr)
1582 struct task_struct *p;
1587 * Determine whether and which event to report to ptracer. When
1588 * called from kernel_thread or CLONE_UNTRACED is explicitly
1589 * requested, no event is reported; otherwise, report if the event
1590 * for the type of forking is enabled.
1592 if (!(clone_flags & CLONE_UNTRACED)) {
1593 if (clone_flags & CLONE_VFORK)
1594 trace = PTRACE_EVENT_VFORK;
1595 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1596 trace = PTRACE_EVENT_CLONE;
1598 trace = PTRACE_EVENT_FORK;
1600 if (likely(!ptrace_event_enabled(current, trace)))
1604 p = copy_process(clone_flags, stack_start, stack_size,
1605 child_tidptr, NULL, trace);
1607 * Do this prior waking up the new thread - the thread pointer
1608 * might get invalid after that point, if the thread exits quickly.
1611 struct completion vfork;
1613 trace_sched_process_fork(current, p);
1615 nr = task_pid_vnr(p);
1617 if (clone_flags & CLONE_PARENT_SETTID)
1618 put_user(nr, parent_tidptr);
1620 if (clone_flags & CLONE_VFORK) {
1621 p->vfork_done = &vfork;
1622 init_completion(&vfork);
1626 wake_up_new_task(p);
1628 /* forking complete and child started to run, tell ptracer */
1629 if (unlikely(trace))
1630 ptrace_event(trace, nr);
1632 if (clone_flags & CLONE_VFORK) {
1633 if (!wait_for_vfork_done(p, &vfork))
1634 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1643 * Create a kernel thread.
1645 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1647 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1648 (unsigned long)arg, NULL, NULL);
1651 #ifdef __ARCH_WANT_SYS_FORK
1652 SYSCALL_DEFINE0(fork)
1655 return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1657 /* can not support in nommu mode */
1663 #ifdef __ARCH_WANT_SYS_VFORK
1664 SYSCALL_DEFINE0(vfork)
1666 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1671 #ifdef __ARCH_WANT_SYS_CLONE
1672 #ifdef CONFIG_CLONE_BACKWARDS
1673 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1674 int __user *, parent_tidptr,
1676 int __user *, child_tidptr)
1677 #elif defined(CONFIG_CLONE_BACKWARDS2)
1678 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1679 int __user *, parent_tidptr,
1680 int __user *, child_tidptr,
1682 #elif defined(CONFIG_CLONE_BACKWARDS3)
1683 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1685 int __user *, parent_tidptr,
1686 int __user *, child_tidptr,
1689 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1690 int __user *, parent_tidptr,
1691 int __user *, child_tidptr,
1695 return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1699 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1700 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1703 static void sighand_ctor(void *data)
1705 struct sighand_struct *sighand = data;
1707 spin_lock_init(&sighand->siglock);
1708 init_waitqueue_head(&sighand->signalfd_wqh);
1711 void __init proc_caches_init(void)
1713 sighand_cachep = kmem_cache_create("sighand_cache",
1714 sizeof(struct sighand_struct), 0,
1715 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1716 SLAB_NOTRACK, sighand_ctor);
1717 signal_cachep = kmem_cache_create("signal_cache",
1718 sizeof(struct signal_struct), 0,
1719 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1720 files_cachep = kmem_cache_create("files_cache",
1721 sizeof(struct files_struct), 0,
1722 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1723 fs_cachep = kmem_cache_create("fs_cache",
1724 sizeof(struct fs_struct), 0,
1725 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1727 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1728 * whole struct cpumask for the OFFSTACK case. We could change
1729 * this to *only* allocate as much of it as required by the
1730 * maximum number of CPU's we can ever have. The cpumask_allocation
1731 * is at the end of the structure, exactly for that reason.
1733 mm_cachep = kmem_cache_create("mm_struct",
1734 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1735 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1736 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1738 nsproxy_cache_init();
1742 * Check constraints on flags passed to the unshare system call.
1744 static int check_unshare_flags(unsigned long unshare_flags)
1746 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1747 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1748 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1749 CLONE_NEWUSER|CLONE_NEWPID))
1752 * Not implemented, but pretend it works if there is nothing to
1753 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1754 * needs to unshare vm.
1756 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1757 /* FIXME: get_task_mm() increments ->mm_users */
1758 if (atomic_read(¤t->mm->mm_users) > 1)
1766 * Unshare the filesystem structure if it is being shared
1768 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1770 struct fs_struct *fs = current->fs;
1772 if (!(unshare_flags & CLONE_FS) || !fs)
1775 /* don't need lock here; in the worst case we'll do useless copy */
1779 *new_fsp = copy_fs_struct(fs);
1787 * Unshare file descriptor table if it is being shared
1789 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1791 struct files_struct *fd = current->files;
1794 if ((unshare_flags & CLONE_FILES) &&
1795 (fd && atomic_read(&fd->count) > 1)) {
1796 *new_fdp = dup_fd(fd, &error);
1805 * unshare allows a process to 'unshare' part of the process
1806 * context which was originally shared using clone. copy_*
1807 * functions used by do_fork() cannot be used here directly
1808 * because they modify an inactive task_struct that is being
1809 * constructed. Here we are modifying the current, active,
1812 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1814 struct fs_struct *fs, *new_fs = NULL;
1815 struct files_struct *fd, *new_fd = NULL;
1816 struct cred *new_cred = NULL;
1817 struct nsproxy *new_nsproxy = NULL;
1822 * If unsharing a user namespace must also unshare the thread.
1824 if (unshare_flags & CLONE_NEWUSER)
1825 unshare_flags |= CLONE_THREAD | CLONE_FS;
1827 * If unsharing a thread from a thread group, must also unshare vm.
1829 if (unshare_flags & CLONE_THREAD)
1830 unshare_flags |= CLONE_VM;
1832 * If unsharing vm, must also unshare signal handlers.
1834 if (unshare_flags & CLONE_VM)
1835 unshare_flags |= CLONE_SIGHAND;
1837 * If unsharing namespace, must also unshare filesystem information.
1839 if (unshare_flags & CLONE_NEWNS)
1840 unshare_flags |= CLONE_FS;
1842 err = check_unshare_flags(unshare_flags);
1844 goto bad_unshare_out;
1846 * CLONE_NEWIPC must also detach from the undolist: after switching
1847 * to a new ipc namespace, the semaphore arrays from the old
1848 * namespace are unreachable.
1850 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1852 err = unshare_fs(unshare_flags, &new_fs);
1854 goto bad_unshare_out;
1855 err = unshare_fd(unshare_flags, &new_fd);
1857 goto bad_unshare_cleanup_fs;
1858 err = unshare_userns(unshare_flags, &new_cred);
1860 goto bad_unshare_cleanup_fd;
1861 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1864 goto bad_unshare_cleanup_cred;
1866 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1869 * CLONE_SYSVSEM is equivalent to sys_exit().
1875 switch_task_namespaces(current, new_nsproxy);
1881 spin_lock(&fs->lock);
1882 current->fs = new_fs;
1887 spin_unlock(&fs->lock);
1891 fd = current->files;
1892 current->files = new_fd;
1896 task_unlock(current);
1899 /* Install the new user namespace */
1900 commit_creds(new_cred);
1905 bad_unshare_cleanup_cred:
1908 bad_unshare_cleanup_fd:
1910 put_files_struct(new_fd);
1912 bad_unshare_cleanup_fs:
1914 free_fs_struct(new_fs);
1921 * Helper to unshare the files of the current task.
1922 * We don't want to expose copy_files internals to
1923 * the exec layer of the kernel.
1926 int unshare_files(struct files_struct **displaced)
1928 struct task_struct *task = current;
1929 struct files_struct *copy = NULL;
1932 error = unshare_fd(CLONE_FILES, ©);
1933 if (error || !copy) {
1937 *displaced = task->files;