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/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/tracehook.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/task_io_accounting_ops.h>
44 #include <linux/rcupdate.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/audit.h>
48 #include <linux/memcontrol.h>
49 #include <linux/ftrace.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/ksm.h>
53 #include <linux/acct.h>
54 #include <linux/tsacct_kern.h>
55 #include <linux/cn_proc.h>
56 #include <linux/freezer.h>
57 #include <linux/delayacct.h>
58 #include <linux/taskstats_kern.h>
59 #include <linux/random.h>
60 #include <linux/tty.h>
61 #include <linux/proc_fs.h>
62 #include <linux/blkdev.h>
63 #include <linux/fs_struct.h>
64 #include <linux/magic.h>
65 #include <linux/perf_event.h>
66 #include <linux/posix-timers.h>
67 #include <linux/user-return-notifier.h>
69 #include <asm/pgtable.h>
70 #include <asm/pgalloc.h>
71 #include <asm/uaccess.h>
72 #include <asm/mmu_context.h>
73 #include <asm/cacheflush.h>
74 #include <asm/tlbflush.h>
76 #include <trace/events/sched.h>
79 * Protected counters by write_lock_irq(&tasklist_lock)
81 unsigned long total_forks; /* Handle normal Linux uptimes. */
82 int nr_threads; /* The idle threads do not count.. */
84 int max_threads; /* tunable limit on nr_threads */
86 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
88 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
89 EXPORT_SYMBOL_GPL(tasklist_lock);
91 int nr_processes(void)
96 for_each_possible_cpu(cpu)
97 total += per_cpu(process_counts, cpu);
102 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
103 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
104 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
105 static struct kmem_cache *task_struct_cachep;
108 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
109 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
111 #ifdef CONFIG_DEBUG_STACK_USAGE
112 gfp_t mask = GFP_KERNEL | __GFP_ZERO;
114 gfp_t mask = GFP_KERNEL;
116 return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
119 static inline void free_thread_info(struct thread_info *ti)
121 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
125 /* SLAB cache for signal_struct structures (tsk->signal) */
126 static struct kmem_cache *signal_cachep;
128 /* SLAB cache for sighand_struct structures (tsk->sighand) */
129 struct kmem_cache *sighand_cachep;
131 /* SLAB cache for files_struct structures (tsk->files) */
132 struct kmem_cache *files_cachep;
134 /* SLAB cache for fs_struct structures (tsk->fs) */
135 struct kmem_cache *fs_cachep;
137 /* SLAB cache for vm_area_struct structures */
138 struct kmem_cache *vm_area_cachep;
140 /* SLAB cache for mm_struct structures (tsk->mm) */
141 static struct kmem_cache *mm_cachep;
143 static void account_kernel_stack(struct thread_info *ti, int account)
145 struct zone *zone = page_zone(virt_to_page(ti));
147 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
150 void free_task(struct task_struct *tsk)
152 prop_local_destroy_single(&tsk->dirties);
153 account_kernel_stack(tsk->stack, -1);
154 free_thread_info(tsk->stack);
155 rt_mutex_debug_task_free(tsk);
156 ftrace_graph_exit_task(tsk);
157 free_task_struct(tsk);
159 EXPORT_SYMBOL(free_task);
161 void __put_task_struct(struct task_struct *tsk)
163 WARN_ON(!tsk->exit_state);
164 WARN_ON(atomic_read(&tsk->usage));
165 WARN_ON(tsk == current);
168 delayacct_tsk_free(tsk);
170 if (!profile_handoff_task(tsk))
175 * macro override instead of weak attribute alias, to workaround
176 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
178 #ifndef arch_task_cache_init
179 #define arch_task_cache_init()
182 void __init fork_init(unsigned long mempages)
184 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
185 #ifndef ARCH_MIN_TASKALIGN
186 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
188 /* create a slab on which task_structs can be allocated */
190 kmem_cache_create("task_struct", sizeof(struct task_struct),
191 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
194 /* do the arch specific task caches init */
195 arch_task_cache_init();
198 * The default maximum number of threads is set to a safe
199 * value: the thread structures can take up at most half
202 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
205 * we need to allow at least 20 threads to boot a system
210 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
211 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
212 init_task.signal->rlim[RLIMIT_SIGPENDING] =
213 init_task.signal->rlim[RLIMIT_NPROC];
216 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
217 struct task_struct *src)
223 static struct task_struct *dup_task_struct(struct task_struct *orig)
225 struct task_struct *tsk;
226 struct thread_info *ti;
227 unsigned long *stackend;
231 prepare_to_copy(orig);
233 tsk = alloc_task_struct();
237 ti = alloc_thread_info(tsk);
239 free_task_struct(tsk);
243 err = arch_dup_task_struct(tsk, orig);
249 err = prop_local_init_single(&tsk->dirties);
253 setup_thread_stack(tsk, orig);
254 clear_user_return_notifier(tsk);
255 stackend = end_of_stack(tsk);
256 *stackend = STACK_END_MAGIC; /* for overflow detection */
258 #ifdef CONFIG_CC_STACKPROTECTOR
259 tsk->stack_canary = get_random_int();
262 /* One for us, one for whoever does the "release_task()" (usually parent) */
263 atomic_set(&tsk->usage,2);
264 atomic_set(&tsk->fs_excl, 0);
265 #ifdef CONFIG_BLK_DEV_IO_TRACE
268 tsk->splice_pipe = NULL;
270 account_kernel_stack(ti, 1);
275 free_thread_info(ti);
276 free_task_struct(tsk);
281 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
283 struct vm_area_struct *mpnt, *tmp, **pprev;
284 struct rb_node **rb_link, *rb_parent;
286 unsigned long charge;
287 struct mempolicy *pol;
289 down_write(&oldmm->mmap_sem);
290 flush_cache_dup_mm(oldmm);
292 * Not linked in yet - no deadlock potential:
294 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
298 mm->mmap_cache = NULL;
299 mm->free_area_cache = oldmm->mmap_base;
300 mm->cached_hole_size = ~0UL;
302 cpumask_clear(mm_cpumask(mm));
304 rb_link = &mm->mm_rb.rb_node;
307 retval = ksm_fork(mm, oldmm);
311 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
314 if (mpnt->vm_flags & VM_DONTCOPY) {
315 long pages = vma_pages(mpnt);
316 mm->total_vm -= pages;
317 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
322 if (mpnt->vm_flags & VM_ACCOUNT) {
323 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
324 if (security_vm_enough_memory(len))
328 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
332 INIT_LIST_HEAD(&tmp->anon_vma_chain);
333 pol = mpol_dup(vma_policy(mpnt));
334 retval = PTR_ERR(pol);
336 goto fail_nomem_policy;
337 vma_set_policy(tmp, pol);
338 if (anon_vma_fork(tmp, mpnt))
339 goto fail_nomem_anon_vma_fork;
340 tmp->vm_flags &= ~VM_LOCKED;
345 struct inode *inode = file->f_path.dentry->d_inode;
346 struct address_space *mapping = file->f_mapping;
349 if (tmp->vm_flags & VM_DENYWRITE)
350 atomic_dec(&inode->i_writecount);
351 spin_lock(&mapping->i_mmap_lock);
352 if (tmp->vm_flags & VM_SHARED)
353 mapping->i_mmap_writable++;
354 tmp->vm_truncate_count = mpnt->vm_truncate_count;
355 flush_dcache_mmap_lock(mapping);
356 /* insert tmp into the share list, just after mpnt */
357 vma_prio_tree_add(tmp, mpnt);
358 flush_dcache_mmap_unlock(mapping);
359 spin_unlock(&mapping->i_mmap_lock);
363 * Clear hugetlb-related page reserves for children. This only
364 * affects MAP_PRIVATE mappings. Faults generated by the child
365 * are not guaranteed to succeed, even if read-only
367 if (is_vm_hugetlb_page(tmp))
368 reset_vma_resv_huge_pages(tmp);
371 * Link in the new vma and copy the page table entries.
374 pprev = &tmp->vm_next;
376 __vma_link_rb(mm, tmp, rb_link, rb_parent);
377 rb_link = &tmp->vm_rb.rb_right;
378 rb_parent = &tmp->vm_rb;
381 retval = copy_page_range(mm, oldmm, mpnt);
383 if (tmp->vm_ops && tmp->vm_ops->open)
384 tmp->vm_ops->open(tmp);
389 /* a new mm has just been created */
390 arch_dup_mmap(oldmm, mm);
393 up_write(&mm->mmap_sem);
395 up_write(&oldmm->mmap_sem);
397 fail_nomem_anon_vma_fork:
400 kmem_cache_free(vm_area_cachep, tmp);
403 vm_unacct_memory(charge);
407 static inline int mm_alloc_pgd(struct mm_struct * mm)
409 mm->pgd = pgd_alloc(mm);
410 if (unlikely(!mm->pgd))
415 static inline void mm_free_pgd(struct mm_struct * mm)
417 pgd_free(mm, mm->pgd);
420 #define dup_mmap(mm, oldmm) (0)
421 #define mm_alloc_pgd(mm) (0)
422 #define mm_free_pgd(mm)
423 #endif /* CONFIG_MMU */
425 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
427 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
428 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
430 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
432 static int __init coredump_filter_setup(char *s)
434 default_dump_filter =
435 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
436 MMF_DUMP_FILTER_MASK;
440 __setup("coredump_filter=", coredump_filter_setup);
442 #include <linux/init_task.h>
444 static void mm_init_aio(struct mm_struct *mm)
447 spin_lock_init(&mm->ioctx_lock);
448 INIT_HLIST_HEAD(&mm->ioctx_list);
452 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
454 atomic_set(&mm->mm_users, 1);
455 atomic_set(&mm->mm_count, 1);
456 init_rwsem(&mm->mmap_sem);
457 INIT_LIST_HEAD(&mm->mmlist);
458 mm->flags = (current->mm) ?
459 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
460 mm->core_state = NULL;
462 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
463 spin_lock_init(&mm->page_table_lock);
464 mm->free_area_cache = TASK_UNMAPPED_BASE;
465 mm->cached_hole_size = ~0UL;
467 mm_init_owner(mm, p);
469 if (likely(!mm_alloc_pgd(mm))) {
471 mmu_notifier_mm_init(mm);
480 * Allocate and initialize an mm_struct.
482 struct mm_struct * mm_alloc(void)
484 struct mm_struct * mm;
488 memset(mm, 0, sizeof(*mm));
489 mm = mm_init(mm, current);
495 * Called when the last reference to the mm
496 * is dropped: either by a lazy thread or by
497 * mmput. Free the page directory and the mm.
499 void __mmdrop(struct mm_struct *mm)
501 BUG_ON(mm == &init_mm);
504 mmu_notifier_mm_destroy(mm);
507 EXPORT_SYMBOL_GPL(__mmdrop);
510 * Decrement the use count and release all resources for an mm.
512 void mmput(struct mm_struct *mm)
516 if (atomic_dec_and_test(&mm->mm_users)) {
520 set_mm_exe_file(mm, NULL);
521 if (!list_empty(&mm->mmlist)) {
522 spin_lock(&mmlist_lock);
523 list_del(&mm->mmlist);
524 spin_unlock(&mmlist_lock);
528 module_put(mm->binfmt->module);
532 EXPORT_SYMBOL_GPL(mmput);
535 * get_task_mm - acquire a reference to the task's mm
537 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
538 * this kernel workthread has transiently adopted a user mm with use_mm,
539 * to do its AIO) is not set and if so returns a reference to it, after
540 * bumping up the use count. User must release the mm via mmput()
541 * after use. Typically used by /proc and ptrace.
543 struct mm_struct *get_task_mm(struct task_struct *task)
545 struct mm_struct *mm;
550 if (task->flags & PF_KTHREAD)
553 atomic_inc(&mm->mm_users);
558 EXPORT_SYMBOL_GPL(get_task_mm);
560 /* Please note the differences between mmput and mm_release.
561 * mmput is called whenever we stop holding onto a mm_struct,
562 * error success whatever.
564 * mm_release is called after a mm_struct has been removed
565 * from the current process.
567 * This difference is important for error handling, when we
568 * only half set up a mm_struct for a new process and need to restore
569 * the old one. Because we mmput the new mm_struct before
570 * restoring the old one. . .
571 * Eric Biederman 10 January 1998
573 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
575 struct completion *vfork_done = tsk->vfork_done;
577 /* Get rid of any futexes when releasing the mm */
579 if (unlikely(tsk->robust_list)) {
580 exit_robust_list(tsk);
581 tsk->robust_list = NULL;
584 if (unlikely(tsk->compat_robust_list)) {
585 compat_exit_robust_list(tsk);
586 tsk->compat_robust_list = NULL;
589 if (unlikely(!list_empty(&tsk->pi_state_list)))
590 exit_pi_state_list(tsk);
593 /* Get rid of any cached register state */
594 deactivate_mm(tsk, mm);
596 /* notify parent sleeping on vfork() */
598 tsk->vfork_done = NULL;
599 complete(vfork_done);
603 * If we're exiting normally, clear a user-space tid field if
604 * requested. We leave this alone when dying by signal, to leave
605 * the value intact in a core dump, and to save the unnecessary
606 * trouble otherwise. Userland only wants this done for a sys_exit.
608 if (tsk->clear_child_tid) {
609 if (!(tsk->flags & PF_SIGNALED) &&
610 atomic_read(&mm->mm_users) > 1) {
612 * We don't check the error code - if userspace has
613 * not set up a proper pointer then tough luck.
615 put_user(0, tsk->clear_child_tid);
616 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
619 tsk->clear_child_tid = NULL;
624 * Allocate a new mm structure and copy contents from the
625 * mm structure of the passed in task structure.
627 struct mm_struct *dup_mm(struct task_struct *tsk)
629 struct mm_struct *mm, *oldmm = current->mm;
639 memcpy(mm, oldmm, sizeof(*mm));
641 /* Initializing for Swap token stuff */
642 mm->token_priority = 0;
643 mm->last_interval = 0;
645 if (!mm_init(mm, tsk))
648 if (init_new_context(tsk, mm))
651 dup_mm_exe_file(oldmm, mm);
653 err = dup_mmap(mm, oldmm);
657 mm->hiwater_rss = get_mm_rss(mm);
658 mm->hiwater_vm = mm->total_vm;
660 if (mm->binfmt && !try_module_get(mm->binfmt->module))
666 /* don't put binfmt in mmput, we haven't got module yet */
675 * If init_new_context() failed, we cannot use mmput() to free the mm
676 * because it calls destroy_context()
683 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
685 struct mm_struct * mm, *oldmm;
688 tsk->min_flt = tsk->maj_flt = 0;
689 tsk->nvcsw = tsk->nivcsw = 0;
690 #ifdef CONFIG_DETECT_HUNG_TASK
691 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
695 tsk->active_mm = NULL;
698 * Are we cloning a kernel thread?
700 * We need to steal a active VM for that..
706 if (clone_flags & CLONE_VM) {
707 atomic_inc(&oldmm->mm_users);
718 /* Initializing for Swap token stuff */
719 mm->token_priority = 0;
720 mm->last_interval = 0;
730 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
732 struct fs_struct *fs = current->fs;
733 if (clone_flags & CLONE_FS) {
734 /* tsk->fs is already what we want */
735 write_lock(&fs->lock);
737 write_unlock(&fs->lock);
741 write_unlock(&fs->lock);
744 tsk->fs = copy_fs_struct(fs);
750 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
752 struct files_struct *oldf, *newf;
756 * A background process may not have any files ...
758 oldf = current->files;
762 if (clone_flags & CLONE_FILES) {
763 atomic_inc(&oldf->count);
767 newf = dup_fd(oldf, &error);
777 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
780 struct io_context *ioc = current->io_context;
785 * Share io context with parent, if CLONE_IO is set
787 if (clone_flags & CLONE_IO) {
788 tsk->io_context = ioc_task_link(ioc);
789 if (unlikely(!tsk->io_context))
791 } else if (ioprio_valid(ioc->ioprio)) {
792 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
793 if (unlikely(!tsk->io_context))
796 tsk->io_context->ioprio = ioc->ioprio;
802 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
804 struct sighand_struct *sig;
806 if (clone_flags & CLONE_SIGHAND) {
807 atomic_inc(¤t->sighand->count);
810 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
811 rcu_assign_pointer(tsk->sighand, sig);
814 atomic_set(&sig->count, 1);
815 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
819 void __cleanup_sighand(struct sighand_struct *sighand)
821 if (atomic_dec_and_test(&sighand->count))
822 kmem_cache_free(sighand_cachep, sighand);
827 * Initialize POSIX timer handling for a thread group.
829 static void posix_cpu_timers_init_group(struct signal_struct *sig)
831 unsigned long cpu_limit;
833 /* Thread group counters. */
834 thread_group_cputime_init(sig);
836 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
837 if (cpu_limit != RLIM_INFINITY) {
838 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
839 sig->cputimer.running = 1;
842 /* The timer lists. */
843 INIT_LIST_HEAD(&sig->cpu_timers[0]);
844 INIT_LIST_HEAD(&sig->cpu_timers[1]);
845 INIT_LIST_HEAD(&sig->cpu_timers[2]);
848 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
850 struct signal_struct *sig;
852 if (clone_flags & CLONE_THREAD)
855 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
860 atomic_set(&sig->count, 1);
861 atomic_set(&sig->live, 1);
862 init_waitqueue_head(&sig->wait_chldexit);
863 if (clone_flags & CLONE_NEWPID)
864 sig->flags |= SIGNAL_UNKILLABLE;
865 sig->curr_target = tsk;
866 init_sigpending(&sig->shared_pending);
867 INIT_LIST_HEAD(&sig->posix_timers);
869 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
870 sig->real_timer.function = it_real_fn;
872 task_lock(current->group_leader);
873 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
874 task_unlock(current->group_leader);
876 posix_cpu_timers_init_group(sig);
880 sig->oom_adj = current->signal->oom_adj;
885 void __cleanup_signal(struct signal_struct *sig)
887 thread_group_cputime_free(sig);
888 tty_kref_put(sig->tty);
889 kmem_cache_free(signal_cachep, sig);
892 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
894 unsigned long new_flags = p->flags;
896 new_flags &= ~PF_SUPERPRIV;
897 new_flags |= PF_FORKNOEXEC;
898 new_flags |= PF_STARTING;
899 p->flags = new_flags;
900 clear_freeze_flag(p);
903 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
905 current->clear_child_tid = tidptr;
907 return task_pid_vnr(current);
910 static void rt_mutex_init_task(struct task_struct *p)
912 raw_spin_lock_init(&p->pi_lock);
913 #ifdef CONFIG_RT_MUTEXES
914 plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
915 p->pi_blocked_on = NULL;
919 #ifdef CONFIG_MM_OWNER
920 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
924 #endif /* CONFIG_MM_OWNER */
927 * Initialize POSIX timer handling for a single task.
929 static void posix_cpu_timers_init(struct task_struct *tsk)
931 tsk->cputime_expires.prof_exp = cputime_zero;
932 tsk->cputime_expires.virt_exp = cputime_zero;
933 tsk->cputime_expires.sched_exp = 0;
934 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
935 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
936 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
940 * This creates a new process as a copy of the old one,
941 * but does not actually start it yet.
943 * It copies the registers, and all the appropriate
944 * parts of the process environment (as per the clone
945 * flags). The actual kick-off is left to the caller.
947 static struct task_struct *copy_process(unsigned long clone_flags,
948 unsigned long stack_start,
949 struct pt_regs *regs,
950 unsigned long stack_size,
951 int __user *child_tidptr,
956 struct task_struct *p;
957 int cgroup_callbacks_done = 0;
959 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
960 return ERR_PTR(-EINVAL);
963 * Thread groups must share signals as well, and detached threads
964 * can only be started up within the thread group.
966 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
967 return ERR_PTR(-EINVAL);
970 * Shared signal handlers imply shared VM. By way of the above,
971 * thread groups also imply shared VM. Blocking this case allows
972 * for various simplifications in other code.
974 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
975 return ERR_PTR(-EINVAL);
978 * Siblings of global init remain as zombies on exit since they are
979 * not reaped by their parent (swapper). To solve this and to avoid
980 * multi-rooted process trees, prevent global and container-inits
981 * from creating siblings.
983 if ((clone_flags & CLONE_PARENT) &&
984 current->signal->flags & SIGNAL_UNKILLABLE)
985 return ERR_PTR(-EINVAL);
987 retval = security_task_create(clone_flags);
992 p = dup_task_struct(current);
996 ftrace_graph_init_task(p);
998 rt_mutex_init_task(p);
1000 #ifdef CONFIG_PROVE_LOCKING
1001 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1002 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1005 if (atomic_read(&p->real_cred->user->processes) >=
1006 task_rlimit(p, RLIMIT_NPROC)) {
1007 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1008 p->real_cred->user != INIT_USER)
1012 retval = copy_creds(p, clone_flags);
1017 * If multiple threads are within copy_process(), then this check
1018 * triggers too late. This doesn't hurt, the check is only there
1019 * to stop root fork bombs.
1022 if (nr_threads >= max_threads)
1023 goto bad_fork_cleanup_count;
1025 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1026 goto bad_fork_cleanup_count;
1029 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1030 copy_flags(clone_flags, p);
1031 INIT_LIST_HEAD(&p->children);
1032 INIT_LIST_HEAD(&p->sibling);
1033 rcu_copy_process(p);
1034 p->vfork_done = NULL;
1035 spin_lock_init(&p->alloc_lock);
1037 init_sigpending(&p->pending);
1039 p->utime = cputime_zero;
1040 p->stime = cputime_zero;
1041 p->gtime = cputime_zero;
1042 p->utimescaled = cputime_zero;
1043 p->stimescaled = cputime_zero;
1044 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1045 p->prev_utime = cputime_zero;
1046 p->prev_stime = cputime_zero;
1049 p->default_timer_slack_ns = current->timer_slack_ns;
1051 task_io_accounting_init(&p->ioac);
1052 acct_clear_integrals(p);
1054 posix_cpu_timers_init(p);
1056 p->lock_depth = -1; /* -1 = no lock */
1057 do_posix_clock_monotonic_gettime(&p->start_time);
1058 p->real_start_time = p->start_time;
1059 monotonic_to_bootbased(&p->real_start_time);
1060 p->io_context = NULL;
1061 p->audit_context = NULL;
1064 p->mempolicy = mpol_dup(p->mempolicy);
1065 if (IS_ERR(p->mempolicy)) {
1066 retval = PTR_ERR(p->mempolicy);
1067 p->mempolicy = NULL;
1068 goto bad_fork_cleanup_cgroup;
1070 mpol_fix_fork_child_flag(p);
1072 #ifdef CONFIG_TRACE_IRQFLAGS
1074 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1075 p->hardirqs_enabled = 1;
1077 p->hardirqs_enabled = 0;
1079 p->hardirq_enable_ip = 0;
1080 p->hardirq_enable_event = 0;
1081 p->hardirq_disable_ip = _THIS_IP_;
1082 p->hardirq_disable_event = 0;
1083 p->softirqs_enabled = 1;
1084 p->softirq_enable_ip = _THIS_IP_;
1085 p->softirq_enable_event = 0;
1086 p->softirq_disable_ip = 0;
1087 p->softirq_disable_event = 0;
1088 p->hardirq_context = 0;
1089 p->softirq_context = 0;
1091 #ifdef CONFIG_LOCKDEP
1092 p->lockdep_depth = 0; /* no locks held yet */
1093 p->curr_chain_key = 0;
1094 p->lockdep_recursion = 0;
1097 #ifdef CONFIG_DEBUG_MUTEXES
1098 p->blocked_on = NULL; /* not blocked yet */
1100 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1101 p->memcg_batch.do_batch = 0;
1102 p->memcg_batch.memcg = NULL;
1107 p->stack_start = stack_start;
1109 /* Perform scheduler related setup. Assign this task to a CPU. */
1110 sched_fork(p, clone_flags);
1112 retval = perf_event_init_task(p);
1114 goto bad_fork_cleanup_policy;
1116 if ((retval = audit_alloc(p)))
1117 goto bad_fork_cleanup_policy;
1118 /* copy all the process information */
1119 if ((retval = copy_semundo(clone_flags, p)))
1120 goto bad_fork_cleanup_audit;
1121 if ((retval = copy_files(clone_flags, p)))
1122 goto bad_fork_cleanup_semundo;
1123 if ((retval = copy_fs(clone_flags, p)))
1124 goto bad_fork_cleanup_files;
1125 if ((retval = copy_sighand(clone_flags, p)))
1126 goto bad_fork_cleanup_fs;
1127 if ((retval = copy_signal(clone_flags, p)))
1128 goto bad_fork_cleanup_sighand;
1129 if ((retval = copy_mm(clone_flags, p)))
1130 goto bad_fork_cleanup_signal;
1131 if ((retval = copy_namespaces(clone_flags, p)))
1132 goto bad_fork_cleanup_mm;
1133 if ((retval = copy_io(clone_flags, p)))
1134 goto bad_fork_cleanup_namespaces;
1135 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1137 goto bad_fork_cleanup_io;
1139 if (pid != &init_struct_pid) {
1141 pid = alloc_pid(p->nsproxy->pid_ns);
1143 goto bad_fork_cleanup_io;
1145 if (clone_flags & CLONE_NEWPID) {
1146 retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1148 goto bad_fork_free_pid;
1152 p->pid = pid_nr(pid);
1154 if (clone_flags & CLONE_THREAD)
1155 p->tgid = current->tgid;
1157 if (current->nsproxy != p->nsproxy) {
1158 retval = ns_cgroup_clone(p, pid);
1160 goto bad_fork_free_pid;
1163 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1165 * Clear TID on mm_release()?
1167 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1169 p->robust_list = NULL;
1170 #ifdef CONFIG_COMPAT
1171 p->compat_robust_list = NULL;
1173 INIT_LIST_HEAD(&p->pi_state_list);
1174 p->pi_state_cache = NULL;
1177 * sigaltstack should be cleared when sharing the same VM
1179 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1180 p->sas_ss_sp = p->sas_ss_size = 0;
1183 * Syscall tracing and stepping should be turned off in the
1184 * child regardless of CLONE_PTRACE.
1186 user_disable_single_step(p);
1187 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1188 #ifdef TIF_SYSCALL_EMU
1189 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1191 clear_all_latency_tracing(p);
1193 /* ok, now we should be set up.. */
1194 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1195 p->pdeath_signal = 0;
1199 * Ok, make it visible to the rest of the system.
1200 * We dont wake it up yet.
1202 p->group_leader = p;
1203 INIT_LIST_HEAD(&p->thread_group);
1205 /* Now that the task is set up, run cgroup callbacks if
1206 * necessary. We need to run them before the task is visible
1207 * on the tasklist. */
1208 cgroup_fork_callbacks(p);
1209 cgroup_callbacks_done = 1;
1211 /* Need tasklist lock for parent etc handling! */
1212 write_lock_irq(&tasklist_lock);
1214 /* CLONE_PARENT re-uses the old parent */
1215 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1216 p->real_parent = current->real_parent;
1217 p->parent_exec_id = current->parent_exec_id;
1219 p->real_parent = current;
1220 p->parent_exec_id = current->self_exec_id;
1223 spin_lock(¤t->sighand->siglock);
1226 * Process group and session signals need to be delivered to just the
1227 * parent before the fork or both the parent and the child after the
1228 * fork. Restart if a signal comes in before we add the new process to
1229 * it's process group.
1230 * A fatal signal pending means that current will exit, so the new
1231 * thread can't slip out of an OOM kill (or normal SIGKILL).
1233 recalc_sigpending();
1234 if (signal_pending(current)) {
1235 spin_unlock(¤t->sighand->siglock);
1236 write_unlock_irq(&tasklist_lock);
1237 retval = -ERESTARTNOINTR;
1238 goto bad_fork_free_pid;
1241 if (clone_flags & CLONE_THREAD) {
1242 atomic_inc(¤t->signal->count);
1243 atomic_inc(¤t->signal->live);
1244 p->group_leader = current->group_leader;
1245 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1248 if (likely(p->pid)) {
1249 tracehook_finish_clone(p, clone_flags, trace);
1251 if (thread_group_leader(p)) {
1252 if (clone_flags & CLONE_NEWPID)
1253 p->nsproxy->pid_ns->child_reaper = p;
1255 p->signal->leader_pid = pid;
1256 tty_kref_put(p->signal->tty);
1257 p->signal->tty = tty_kref_get(current->signal->tty);
1258 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1259 attach_pid(p, PIDTYPE_SID, task_session(current));
1260 list_add_tail(&p->sibling, &p->real_parent->children);
1261 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1262 __get_cpu_var(process_counts)++;
1264 attach_pid(p, PIDTYPE_PID, pid);
1269 spin_unlock(¤t->sighand->siglock);
1270 write_unlock_irq(&tasklist_lock);
1271 proc_fork_connector(p);
1272 cgroup_post_fork(p);
1277 if (pid != &init_struct_pid)
1279 bad_fork_cleanup_io:
1282 bad_fork_cleanup_namespaces:
1283 exit_task_namespaces(p);
1284 bad_fork_cleanup_mm:
1287 bad_fork_cleanup_signal:
1288 if (!(clone_flags & CLONE_THREAD))
1289 __cleanup_signal(p->signal);
1290 bad_fork_cleanup_sighand:
1291 __cleanup_sighand(p->sighand);
1292 bad_fork_cleanup_fs:
1293 exit_fs(p); /* blocking */
1294 bad_fork_cleanup_files:
1295 exit_files(p); /* blocking */
1296 bad_fork_cleanup_semundo:
1298 bad_fork_cleanup_audit:
1300 bad_fork_cleanup_policy:
1301 perf_event_free_task(p);
1303 mpol_put(p->mempolicy);
1304 bad_fork_cleanup_cgroup:
1306 cgroup_exit(p, cgroup_callbacks_done);
1307 delayacct_tsk_free(p);
1308 module_put(task_thread_info(p)->exec_domain->module);
1309 bad_fork_cleanup_count:
1310 atomic_dec(&p->cred->user->processes);
1315 return ERR_PTR(retval);
1318 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1320 memset(regs, 0, sizeof(struct pt_regs));
1324 struct task_struct * __cpuinit fork_idle(int cpu)
1326 struct task_struct *task;
1327 struct pt_regs regs;
1329 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL,
1330 &init_struct_pid, 0);
1332 init_idle(task, cpu);
1338 * Ok, this is the main fork-routine.
1340 * It copies the process, and if successful kick-starts
1341 * it and waits for it to finish using the VM if required.
1343 long do_fork(unsigned long clone_flags,
1344 unsigned long stack_start,
1345 struct pt_regs *regs,
1346 unsigned long stack_size,
1347 int __user *parent_tidptr,
1348 int __user *child_tidptr)
1350 struct task_struct *p;
1355 * Do some preliminary argument and permissions checking before we
1356 * actually start allocating stuff
1358 if (clone_flags & CLONE_NEWUSER) {
1359 if (clone_flags & CLONE_THREAD)
1361 /* hopefully this check will go away when userns support is
1364 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1365 !capable(CAP_SETGID))
1370 * We hope to recycle these flags after 2.6.26
1372 if (unlikely(clone_flags & CLONE_STOPPED)) {
1373 static int __read_mostly count = 100;
1375 if (count > 0 && printk_ratelimit()) {
1376 char comm[TASK_COMM_LEN];
1379 printk(KERN_INFO "fork(): process `%s' used deprecated "
1380 "clone flags 0x%lx\n",
1381 get_task_comm(comm, current),
1382 clone_flags & CLONE_STOPPED);
1387 * When called from kernel_thread, don't do user tracing stuff.
1389 if (likely(user_mode(regs)))
1390 trace = tracehook_prepare_clone(clone_flags);
1392 p = copy_process(clone_flags, stack_start, regs, stack_size,
1393 child_tidptr, NULL, trace);
1395 * Do this prior waking up the new thread - the thread pointer
1396 * might get invalid after that point, if the thread exits quickly.
1399 struct completion vfork;
1401 trace_sched_process_fork(current, p);
1403 nr = task_pid_vnr(p);
1405 if (clone_flags & CLONE_PARENT_SETTID)
1406 put_user(nr, parent_tidptr);
1408 if (clone_flags & CLONE_VFORK) {
1409 p->vfork_done = &vfork;
1410 init_completion(&vfork);
1413 audit_finish_fork(p);
1414 tracehook_report_clone(regs, clone_flags, nr, p);
1417 * We set PF_STARTING at creation in case tracing wants to
1418 * use this to distinguish a fully live task from one that
1419 * hasn't gotten to tracehook_report_clone() yet. Now we
1420 * clear it and set the child going.
1422 p->flags &= ~PF_STARTING;
1424 if (unlikely(clone_flags & CLONE_STOPPED)) {
1426 * We'll start up with an immediate SIGSTOP.
1428 sigaddset(&p->pending.signal, SIGSTOP);
1429 set_tsk_thread_flag(p, TIF_SIGPENDING);
1430 __set_task_state(p, TASK_STOPPED);
1432 wake_up_new_task(p, clone_flags);
1435 tracehook_report_clone_complete(trace, regs,
1436 clone_flags, nr, p);
1438 if (clone_flags & CLONE_VFORK) {
1439 freezer_do_not_count();
1440 wait_for_completion(&vfork);
1442 tracehook_report_vfork_done(p, nr);
1450 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1451 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1454 static void sighand_ctor(void *data)
1456 struct sighand_struct *sighand = data;
1458 spin_lock_init(&sighand->siglock);
1459 init_waitqueue_head(&sighand->signalfd_wqh);
1462 void __init proc_caches_init(void)
1464 sighand_cachep = kmem_cache_create("sighand_cache",
1465 sizeof(struct sighand_struct), 0,
1466 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1467 SLAB_NOTRACK, sighand_ctor);
1468 signal_cachep = kmem_cache_create("signal_cache",
1469 sizeof(struct signal_struct), 0,
1470 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1471 files_cachep = kmem_cache_create("files_cache",
1472 sizeof(struct files_struct), 0,
1473 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1474 fs_cachep = kmem_cache_create("fs_cache",
1475 sizeof(struct fs_struct), 0,
1476 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1477 mm_cachep = kmem_cache_create("mm_struct",
1478 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1479 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1480 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1485 * Check constraints on flags passed to the unshare system call and
1486 * force unsharing of additional process context as appropriate.
1488 static void check_unshare_flags(unsigned long *flags_ptr)
1491 * If unsharing a thread from a thread group, must also
1494 if (*flags_ptr & CLONE_THREAD)
1495 *flags_ptr |= CLONE_VM;
1498 * If unsharing vm, must also unshare signal handlers.
1500 if (*flags_ptr & CLONE_VM)
1501 *flags_ptr |= CLONE_SIGHAND;
1504 * If unsharing signal handlers and the task was created
1505 * using CLONE_THREAD, then must unshare the thread
1507 if ((*flags_ptr & CLONE_SIGHAND) &&
1508 (atomic_read(¤t->signal->count) > 1))
1509 *flags_ptr |= CLONE_THREAD;
1512 * If unsharing namespace, must also unshare filesystem information.
1514 if (*flags_ptr & CLONE_NEWNS)
1515 *flags_ptr |= CLONE_FS;
1519 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1521 static int unshare_thread(unsigned long unshare_flags)
1523 if (unshare_flags & CLONE_THREAD)
1530 * Unshare the filesystem structure if it is being shared
1532 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1534 struct fs_struct *fs = current->fs;
1536 if (!(unshare_flags & CLONE_FS) || !fs)
1539 /* don't need lock here; in the worst case we'll do useless copy */
1543 *new_fsp = copy_fs_struct(fs);
1551 * Unsharing of sighand is not supported yet
1553 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1555 struct sighand_struct *sigh = current->sighand;
1557 if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1564 * Unshare vm if it is being shared
1566 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1568 struct mm_struct *mm = current->mm;
1570 if ((unshare_flags & CLONE_VM) &&
1571 (mm && atomic_read(&mm->mm_users) > 1)) {
1579 * Unshare file descriptor table if it is being shared
1581 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1583 struct files_struct *fd = current->files;
1586 if ((unshare_flags & CLONE_FILES) &&
1587 (fd && atomic_read(&fd->count) > 1)) {
1588 *new_fdp = dup_fd(fd, &error);
1597 * unshare allows a process to 'unshare' part of the process
1598 * context which was originally shared using clone. copy_*
1599 * functions used by do_fork() cannot be used here directly
1600 * because they modify an inactive task_struct that is being
1601 * constructed. Here we are modifying the current, active,
1604 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1607 struct fs_struct *fs, *new_fs = NULL;
1608 struct sighand_struct *new_sigh = NULL;
1609 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1610 struct files_struct *fd, *new_fd = NULL;
1611 struct nsproxy *new_nsproxy = NULL;
1614 check_unshare_flags(&unshare_flags);
1616 /* Return -EINVAL for all unsupported flags */
1618 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1619 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1620 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1621 goto bad_unshare_out;
1624 * CLONE_NEWIPC must also detach from the undolist: after switching
1625 * to a new ipc namespace, the semaphore arrays from the old
1626 * namespace are unreachable.
1628 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1630 if ((err = unshare_thread(unshare_flags)))
1631 goto bad_unshare_out;
1632 if ((err = unshare_fs(unshare_flags, &new_fs)))
1633 goto bad_unshare_cleanup_thread;
1634 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1635 goto bad_unshare_cleanup_fs;
1636 if ((err = unshare_vm(unshare_flags, &new_mm)))
1637 goto bad_unshare_cleanup_sigh;
1638 if ((err = unshare_fd(unshare_flags, &new_fd)))
1639 goto bad_unshare_cleanup_vm;
1640 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1642 goto bad_unshare_cleanup_fd;
1644 if (new_fs || new_mm || new_fd || do_sysvsem || new_nsproxy) {
1647 * CLONE_SYSVSEM is equivalent to sys_exit().
1653 switch_task_namespaces(current, new_nsproxy);
1661 write_lock(&fs->lock);
1662 current->fs = new_fs;
1667 write_unlock(&fs->lock);
1672 active_mm = current->active_mm;
1673 current->mm = new_mm;
1674 current->active_mm = new_mm;
1675 activate_mm(active_mm, new_mm);
1680 fd = current->files;
1681 current->files = new_fd;
1685 task_unlock(current);
1689 put_nsproxy(new_nsproxy);
1691 bad_unshare_cleanup_fd:
1693 put_files_struct(new_fd);
1695 bad_unshare_cleanup_vm:
1699 bad_unshare_cleanup_sigh:
1701 if (atomic_dec_and_test(&new_sigh->count))
1702 kmem_cache_free(sighand_cachep, new_sigh);
1704 bad_unshare_cleanup_fs:
1706 free_fs_struct(new_fs);
1708 bad_unshare_cleanup_thread:
1714 * Helper to unshare the files of the current task.
1715 * We don't want to expose copy_files internals to
1716 * the exec layer of the kernel.
1719 int unshare_files(struct files_struct **displaced)
1721 struct task_struct *task = current;
1722 struct files_struct *copy = NULL;
1725 error = unshare_fd(CLONE_FILES, ©);
1726 if (error || !copy) {
1730 *displaced = task->files;