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/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/profile.h>
52 #include <linux/rmap.h>
53 #include <linux/ksm.h>
54 #include <linux/acct.h>
55 #include <linux/tsacct_kern.h>
56 #include <linux/cn_proc.h>
57 #include <linux/freezer.h>
58 #include <linux/delayacct.h>
59 #include <linux/taskstats_kern.h>
60 #include <linux/random.h>
61 #include <linux/tty.h>
62 #include <linux/proc_fs.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>
72 #include <asm/pgtable.h>
73 #include <asm/pgalloc.h>
74 #include <asm/uaccess.h>
75 #include <asm/mmu_context.h>
76 #include <asm/cacheflush.h>
77 #include <asm/tlbflush.h>
79 #include <trace/events/sched.h>
82 * Protected counters by write_lock_irq(&tasklist_lock)
84 unsigned long total_forks; /* Handle normal Linux uptimes. */
85 int nr_threads; /* The idle threads do not count.. */
87 int max_threads; /* tunable limit on nr_threads */
89 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
91 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
93 #ifdef CONFIG_PROVE_RCU
94 int lockdep_tasklist_lock_is_held(void)
96 return lockdep_is_held(&tasklist_lock);
98 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
99 #endif /* #ifdef CONFIG_PROVE_RCU */
101 int nr_processes(void)
106 for_each_possible_cpu(cpu)
107 total += per_cpu(process_counts, cpu);
112 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
113 # define alloc_task_struct_node(node) \
114 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
115 # define free_task_struct(tsk) \
116 kmem_cache_free(task_struct_cachep, (tsk))
117 static struct kmem_cache *task_struct_cachep;
120 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
121 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
124 #ifdef CONFIG_DEBUG_STACK_USAGE
125 gfp_t mask = GFP_KERNEL | __GFP_ZERO;
127 gfp_t mask = GFP_KERNEL;
129 struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
131 return page ? page_address(page) : NULL;
134 static inline void free_thread_info(struct thread_info *ti)
136 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
140 /* SLAB cache for signal_struct structures (tsk->signal) */
141 static struct kmem_cache *signal_cachep;
143 /* SLAB cache for sighand_struct structures (tsk->sighand) */
144 struct kmem_cache *sighand_cachep;
146 /* SLAB cache for files_struct structures (tsk->files) */
147 struct kmem_cache *files_cachep;
149 /* SLAB cache for fs_struct structures (tsk->fs) */
150 struct kmem_cache *fs_cachep;
152 /* SLAB cache for vm_area_struct structures */
153 struct kmem_cache *vm_area_cachep;
155 /* SLAB cache for mm_struct structures (tsk->mm) */
156 static struct kmem_cache *mm_cachep;
158 static void account_kernel_stack(struct thread_info *ti, int account)
160 struct zone *zone = page_zone(virt_to_page(ti));
162 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
165 void free_task(struct task_struct *tsk)
167 prop_local_destroy_single(&tsk->dirties);
168 account_kernel_stack(tsk->stack, -1);
169 free_thread_info(tsk->stack);
170 rt_mutex_debug_task_free(tsk);
171 ftrace_graph_exit_task(tsk);
172 free_task_struct(tsk);
174 EXPORT_SYMBOL(free_task);
176 static inline void free_signal_struct(struct signal_struct *sig)
178 taskstats_tgid_free(sig);
179 sched_autogroup_exit(sig);
180 kmem_cache_free(signal_cachep, sig);
183 static inline void put_signal_struct(struct signal_struct *sig)
185 if (atomic_dec_and_test(&sig->sigcnt))
186 free_signal_struct(sig);
189 void __put_task_struct(struct task_struct *tsk)
191 WARN_ON(!tsk->exit_state);
192 WARN_ON(atomic_read(&tsk->usage));
193 WARN_ON(tsk == current);
196 delayacct_tsk_free(tsk);
197 put_signal_struct(tsk->signal);
199 if (!profile_handoff_task(tsk))
202 EXPORT_SYMBOL_GPL(__put_task_struct);
205 * macro override instead of weak attribute alias, to workaround
206 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
208 #ifndef arch_task_cache_init
209 #define arch_task_cache_init()
212 void __init fork_init(unsigned long mempages)
214 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
215 #ifndef ARCH_MIN_TASKALIGN
216 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
218 /* create a slab on which task_structs can be allocated */
220 kmem_cache_create("task_struct", sizeof(struct task_struct),
221 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
224 /* do the arch specific task caches init */
225 arch_task_cache_init();
228 * The default maximum number of threads is set to a safe
229 * value: the thread structures can take up at most half
232 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
235 * we need to allow at least 20 threads to boot a system
240 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
241 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
242 init_task.signal->rlim[RLIMIT_SIGPENDING] =
243 init_task.signal->rlim[RLIMIT_NPROC];
246 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
247 struct task_struct *src)
253 static struct task_struct *dup_task_struct(struct task_struct *orig)
255 struct task_struct *tsk;
256 struct thread_info *ti;
257 unsigned long *stackend;
258 int node = tsk_fork_get_node(orig);
261 prepare_to_copy(orig);
263 tsk = alloc_task_struct_node(node);
267 ti = alloc_thread_info_node(tsk, node);
269 free_task_struct(tsk);
273 err = arch_dup_task_struct(tsk, orig);
279 err = prop_local_init_single(&tsk->dirties);
283 setup_thread_stack(tsk, orig);
284 clear_user_return_notifier(tsk);
285 clear_tsk_need_resched(tsk);
286 stackend = end_of_stack(tsk);
287 *stackend = STACK_END_MAGIC; /* for overflow detection */
289 #ifdef CONFIG_CC_STACKPROTECTOR
290 tsk->stack_canary = get_random_int();
293 /* One for us, one for whoever does the "release_task()" (usually parent) */
294 atomic_set(&tsk->usage,2);
295 atomic_set(&tsk->fs_excl, 0);
296 #ifdef CONFIG_BLK_DEV_IO_TRACE
299 tsk->splice_pipe = NULL;
301 account_kernel_stack(ti, 1);
306 free_thread_info(ti);
307 free_task_struct(tsk);
312 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
314 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
315 struct rb_node **rb_link, *rb_parent;
317 unsigned long charge;
318 struct mempolicy *pol;
320 down_write(&oldmm->mmap_sem);
321 flush_cache_dup_mm(oldmm);
323 * Not linked in yet - no deadlock potential:
325 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
329 mm->mmap_cache = NULL;
330 mm->free_area_cache = oldmm->mmap_base;
331 mm->cached_hole_size = ~0UL;
333 cpumask_clear(mm_cpumask(mm));
335 rb_link = &mm->mm_rb.rb_node;
338 retval = ksm_fork(mm, oldmm);
341 retval = khugepaged_fork(mm, oldmm);
346 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
349 if (mpnt->vm_flags & VM_DONTCOPY) {
350 long pages = vma_pages(mpnt);
351 mm->total_vm -= pages;
352 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
357 if (mpnt->vm_flags & VM_ACCOUNT) {
358 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
359 if (security_vm_enough_memory(len))
363 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
367 INIT_LIST_HEAD(&tmp->anon_vma_chain);
368 pol = mpol_dup(vma_policy(mpnt));
369 retval = PTR_ERR(pol);
371 goto fail_nomem_policy;
372 vma_set_policy(tmp, pol);
374 if (anon_vma_fork(tmp, mpnt))
375 goto fail_nomem_anon_vma_fork;
376 tmp->vm_flags &= ~VM_LOCKED;
377 tmp->vm_next = tmp->vm_prev = NULL;
380 struct inode *inode = file->f_path.dentry->d_inode;
381 struct address_space *mapping = file->f_mapping;
384 if (tmp->vm_flags & VM_DENYWRITE)
385 atomic_dec(&inode->i_writecount);
386 spin_lock(&mapping->i_mmap_lock);
387 if (tmp->vm_flags & VM_SHARED)
388 mapping->i_mmap_writable++;
389 flush_dcache_mmap_lock(mapping);
390 /* insert tmp into the share list, just after mpnt */
391 vma_prio_tree_add(tmp, mpnt);
392 flush_dcache_mmap_unlock(mapping);
393 spin_unlock(&mapping->i_mmap_lock);
397 * Clear hugetlb-related page reserves for children. This only
398 * affects MAP_PRIVATE mappings. Faults generated by the child
399 * are not guaranteed to succeed, even if read-only
401 if (is_vm_hugetlb_page(tmp))
402 reset_vma_resv_huge_pages(tmp);
405 * Link in the new vma and copy the page table entries.
408 pprev = &tmp->vm_next;
412 __vma_link_rb(mm, tmp, rb_link, rb_parent);
413 rb_link = &tmp->vm_rb.rb_right;
414 rb_parent = &tmp->vm_rb;
417 retval = copy_page_range(mm, oldmm, mpnt);
419 if (tmp->vm_ops && tmp->vm_ops->open)
420 tmp->vm_ops->open(tmp);
425 /* a new mm has just been created */
426 arch_dup_mmap(oldmm, mm);
429 up_write(&mm->mmap_sem);
431 up_write(&oldmm->mmap_sem);
433 fail_nomem_anon_vma_fork:
436 kmem_cache_free(vm_area_cachep, tmp);
439 vm_unacct_memory(charge);
443 static inline int mm_alloc_pgd(struct mm_struct * mm)
445 mm->pgd = pgd_alloc(mm);
446 if (unlikely(!mm->pgd))
451 static inline void mm_free_pgd(struct mm_struct * mm)
453 pgd_free(mm, mm->pgd);
456 #define dup_mmap(mm, oldmm) (0)
457 #define mm_alloc_pgd(mm) (0)
458 #define mm_free_pgd(mm)
459 #endif /* CONFIG_MMU */
461 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
463 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
464 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
466 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
468 static int __init coredump_filter_setup(char *s)
470 default_dump_filter =
471 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
472 MMF_DUMP_FILTER_MASK;
476 __setup("coredump_filter=", coredump_filter_setup);
478 #include <linux/init_task.h>
480 static void mm_init_aio(struct mm_struct *mm)
483 spin_lock_init(&mm->ioctx_lock);
484 INIT_HLIST_HEAD(&mm->ioctx_list);
488 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
490 atomic_set(&mm->mm_users, 1);
491 atomic_set(&mm->mm_count, 1);
492 init_rwsem(&mm->mmap_sem);
493 INIT_LIST_HEAD(&mm->mmlist);
494 mm->flags = (current->mm) ?
495 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
496 mm->core_state = NULL;
498 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
499 spin_lock_init(&mm->page_table_lock);
500 mm->free_area_cache = TASK_UNMAPPED_BASE;
501 mm->cached_hole_size = ~0UL;
503 mm_init_owner(mm, p);
504 atomic_set(&mm->oom_disable_count, 0);
506 if (likely(!mm_alloc_pgd(mm))) {
508 mmu_notifier_mm_init(mm);
517 * Allocate and initialize an mm_struct.
519 struct mm_struct * mm_alloc(void)
521 struct mm_struct * mm;
525 memset(mm, 0, sizeof(*mm));
526 mm = mm_init(mm, current);
532 * Called when the last reference to the mm
533 * is dropped: either by a lazy thread or by
534 * mmput. Free the page directory and the mm.
536 void __mmdrop(struct mm_struct *mm)
538 BUG_ON(mm == &init_mm);
541 mmu_notifier_mm_destroy(mm);
542 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
543 VM_BUG_ON(mm->pmd_huge_pte);
547 EXPORT_SYMBOL_GPL(__mmdrop);
550 * Decrement the use count and release all resources for an mm.
552 void mmput(struct mm_struct *mm)
556 if (atomic_dec_and_test(&mm->mm_users)) {
559 khugepaged_exit(mm); /* must run before exit_mmap */
561 set_mm_exe_file(mm, NULL);
562 if (!list_empty(&mm->mmlist)) {
563 spin_lock(&mmlist_lock);
564 list_del(&mm->mmlist);
565 spin_unlock(&mmlist_lock);
569 module_put(mm->binfmt->module);
573 EXPORT_SYMBOL_GPL(mmput);
576 * get_task_mm - acquire a reference to the task's mm
578 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
579 * this kernel workthread has transiently adopted a user mm with use_mm,
580 * to do its AIO) is not set and if so returns a reference to it, after
581 * bumping up the use count. User must release the mm via mmput()
582 * after use. Typically used by /proc and ptrace.
584 struct mm_struct *get_task_mm(struct task_struct *task)
586 struct mm_struct *mm;
591 if (task->flags & PF_KTHREAD)
594 atomic_inc(&mm->mm_users);
599 EXPORT_SYMBOL_GPL(get_task_mm);
601 /* Please note the differences between mmput and mm_release.
602 * mmput is called whenever we stop holding onto a mm_struct,
603 * error success whatever.
605 * mm_release is called after a mm_struct has been removed
606 * from the current process.
608 * This difference is important for error handling, when we
609 * only half set up a mm_struct for a new process and need to restore
610 * the old one. Because we mmput the new mm_struct before
611 * restoring the old one. . .
612 * Eric Biederman 10 January 1998
614 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
616 struct completion *vfork_done = tsk->vfork_done;
618 /* Get rid of any futexes when releasing the mm */
620 if (unlikely(tsk->robust_list)) {
621 exit_robust_list(tsk);
622 tsk->robust_list = NULL;
625 if (unlikely(tsk->compat_robust_list)) {
626 compat_exit_robust_list(tsk);
627 tsk->compat_robust_list = NULL;
630 if (unlikely(!list_empty(&tsk->pi_state_list)))
631 exit_pi_state_list(tsk);
634 /* Get rid of any cached register state */
635 deactivate_mm(tsk, mm);
637 /* notify parent sleeping on vfork() */
639 tsk->vfork_done = NULL;
640 complete(vfork_done);
644 * If we're exiting normally, clear a user-space tid field if
645 * requested. We leave this alone when dying by signal, to leave
646 * the value intact in a core dump, and to save the unnecessary
647 * trouble otherwise. Userland only wants this done for a sys_exit.
649 if (tsk->clear_child_tid) {
650 if (!(tsk->flags & PF_SIGNALED) &&
651 atomic_read(&mm->mm_users) > 1) {
653 * We don't check the error code - if userspace has
654 * not set up a proper pointer then tough luck.
656 put_user(0, tsk->clear_child_tid);
657 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
660 tsk->clear_child_tid = NULL;
665 * Allocate a new mm structure and copy contents from the
666 * mm structure of the passed in task structure.
668 struct mm_struct *dup_mm(struct task_struct *tsk)
670 struct mm_struct *mm, *oldmm = current->mm;
680 memcpy(mm, oldmm, sizeof(*mm));
682 /* Initializing for Swap token stuff */
683 mm->token_priority = 0;
684 mm->last_interval = 0;
686 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
687 mm->pmd_huge_pte = NULL;
690 if (!mm_init(mm, tsk))
693 if (init_new_context(tsk, mm))
696 dup_mm_exe_file(oldmm, mm);
698 err = dup_mmap(mm, oldmm);
702 mm->hiwater_rss = get_mm_rss(mm);
703 mm->hiwater_vm = mm->total_vm;
705 if (mm->binfmt && !try_module_get(mm->binfmt->module))
711 /* don't put binfmt in mmput, we haven't got module yet */
720 * If init_new_context() failed, we cannot use mmput() to free the mm
721 * because it calls destroy_context()
728 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
730 struct mm_struct * mm, *oldmm;
733 tsk->min_flt = tsk->maj_flt = 0;
734 tsk->nvcsw = tsk->nivcsw = 0;
735 #ifdef CONFIG_DETECT_HUNG_TASK
736 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
740 tsk->active_mm = NULL;
743 * Are we cloning a kernel thread?
745 * We need to steal a active VM for that..
751 if (clone_flags & CLONE_VM) {
752 atomic_inc(&oldmm->mm_users);
763 /* Initializing for Swap token stuff */
764 mm->token_priority = 0;
765 mm->last_interval = 0;
766 if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
767 atomic_inc(&mm->oom_disable_count);
777 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
779 struct fs_struct *fs = current->fs;
780 if (clone_flags & CLONE_FS) {
781 /* tsk->fs is already what we want */
782 spin_lock(&fs->lock);
784 spin_unlock(&fs->lock);
788 spin_unlock(&fs->lock);
791 tsk->fs = copy_fs_struct(fs);
797 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
799 struct files_struct *oldf, *newf;
803 * A background process may not have any files ...
805 oldf = current->files;
809 if (clone_flags & CLONE_FILES) {
810 atomic_inc(&oldf->count);
814 newf = dup_fd(oldf, &error);
824 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
827 struct io_context *ioc = current->io_context;
832 * Share io context with parent, if CLONE_IO is set
834 if (clone_flags & CLONE_IO) {
835 tsk->io_context = ioc_task_link(ioc);
836 if (unlikely(!tsk->io_context))
838 } else if (ioprio_valid(ioc->ioprio)) {
839 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
840 if (unlikely(!tsk->io_context))
843 tsk->io_context->ioprio = ioc->ioprio;
849 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
851 struct sighand_struct *sig;
853 if (clone_flags & CLONE_SIGHAND) {
854 atomic_inc(¤t->sighand->count);
857 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
858 rcu_assign_pointer(tsk->sighand, sig);
861 atomic_set(&sig->count, 1);
862 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
866 void __cleanup_sighand(struct sighand_struct *sighand)
868 if (atomic_dec_and_test(&sighand->count))
869 kmem_cache_free(sighand_cachep, sighand);
874 * Initialize POSIX timer handling for a thread group.
876 static void posix_cpu_timers_init_group(struct signal_struct *sig)
878 unsigned long cpu_limit;
880 /* Thread group counters. */
881 thread_group_cputime_init(sig);
883 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
884 if (cpu_limit != RLIM_INFINITY) {
885 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
886 sig->cputimer.running = 1;
889 /* The timer lists. */
890 INIT_LIST_HEAD(&sig->cpu_timers[0]);
891 INIT_LIST_HEAD(&sig->cpu_timers[1]);
892 INIT_LIST_HEAD(&sig->cpu_timers[2]);
895 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
897 struct signal_struct *sig;
899 if (clone_flags & CLONE_THREAD)
902 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
908 atomic_set(&sig->live, 1);
909 atomic_set(&sig->sigcnt, 1);
910 init_waitqueue_head(&sig->wait_chldexit);
911 if (clone_flags & CLONE_NEWPID)
912 sig->flags |= SIGNAL_UNKILLABLE;
913 sig->curr_target = tsk;
914 init_sigpending(&sig->shared_pending);
915 INIT_LIST_HEAD(&sig->posix_timers);
917 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
918 sig->real_timer.function = it_real_fn;
920 task_lock(current->group_leader);
921 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
922 task_unlock(current->group_leader);
924 posix_cpu_timers_init_group(sig);
927 sched_autogroup_fork(sig);
929 sig->oom_adj = current->signal->oom_adj;
930 sig->oom_score_adj = current->signal->oom_score_adj;
931 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
933 mutex_init(&sig->cred_guard_mutex);
938 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
940 unsigned long new_flags = p->flags;
942 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
943 new_flags |= PF_FORKNOEXEC;
944 new_flags |= PF_STARTING;
945 p->flags = new_flags;
946 clear_freeze_flag(p);
949 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
951 current->clear_child_tid = tidptr;
953 return task_pid_vnr(current);
956 static void rt_mutex_init_task(struct task_struct *p)
958 raw_spin_lock_init(&p->pi_lock);
959 #ifdef CONFIG_RT_MUTEXES
960 plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
961 p->pi_blocked_on = NULL;
965 #ifdef CONFIG_MM_OWNER
966 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
970 #endif /* CONFIG_MM_OWNER */
973 * Initialize POSIX timer handling for a single task.
975 static void posix_cpu_timers_init(struct task_struct *tsk)
977 tsk->cputime_expires.prof_exp = cputime_zero;
978 tsk->cputime_expires.virt_exp = cputime_zero;
979 tsk->cputime_expires.sched_exp = 0;
980 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
981 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
982 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
986 * This creates a new process as a copy of the old one,
987 * but does not actually start it yet.
989 * It copies the registers, and all the appropriate
990 * parts of the process environment (as per the clone
991 * flags). The actual kick-off is left to the caller.
993 static struct task_struct *copy_process(unsigned long clone_flags,
994 unsigned long stack_start,
995 struct pt_regs *regs,
996 unsigned long stack_size,
997 int __user *child_tidptr,
1002 struct task_struct *p;
1003 int cgroup_callbacks_done = 0;
1005 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1006 return ERR_PTR(-EINVAL);
1009 * Thread groups must share signals as well, and detached threads
1010 * can only be started up within the thread group.
1012 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1013 return ERR_PTR(-EINVAL);
1016 * Shared signal handlers imply shared VM. By way of the above,
1017 * thread groups also imply shared VM. Blocking this case allows
1018 * for various simplifications in other code.
1020 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1021 return ERR_PTR(-EINVAL);
1024 * Siblings of global init remain as zombies on exit since they are
1025 * not reaped by their parent (swapper). To solve this and to avoid
1026 * multi-rooted process trees, prevent global and container-inits
1027 * from creating siblings.
1029 if ((clone_flags & CLONE_PARENT) &&
1030 current->signal->flags & SIGNAL_UNKILLABLE)
1031 return ERR_PTR(-EINVAL);
1033 retval = security_task_create(clone_flags);
1038 p = dup_task_struct(current);
1042 ftrace_graph_init_task(p);
1044 rt_mutex_init_task(p);
1046 #ifdef CONFIG_PROVE_LOCKING
1047 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1048 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1051 if (atomic_read(&p->real_cred->user->processes) >=
1052 task_rlimit(p, RLIMIT_NPROC)) {
1053 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1054 p->real_cred->user != INIT_USER)
1058 retval = copy_creds(p, clone_flags);
1063 * If multiple threads are within copy_process(), then this check
1064 * triggers too late. This doesn't hurt, the check is only there
1065 * to stop root fork bombs.
1068 if (nr_threads >= max_threads)
1069 goto bad_fork_cleanup_count;
1071 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1072 goto bad_fork_cleanup_count;
1075 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1076 copy_flags(clone_flags, p);
1077 INIT_LIST_HEAD(&p->children);
1078 INIT_LIST_HEAD(&p->sibling);
1079 rcu_copy_process(p);
1080 p->vfork_done = NULL;
1081 spin_lock_init(&p->alloc_lock);
1083 init_sigpending(&p->pending);
1085 p->utime = cputime_zero;
1086 p->stime = cputime_zero;
1087 p->gtime = cputime_zero;
1088 p->utimescaled = cputime_zero;
1089 p->stimescaled = cputime_zero;
1090 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1091 p->prev_utime = cputime_zero;
1092 p->prev_stime = cputime_zero;
1094 #if defined(SPLIT_RSS_COUNTING)
1095 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1098 p->default_timer_slack_ns = current->timer_slack_ns;
1100 task_io_accounting_init(&p->ioac);
1101 acct_clear_integrals(p);
1103 posix_cpu_timers_init(p);
1105 do_posix_clock_monotonic_gettime(&p->start_time);
1106 p->real_start_time = p->start_time;
1107 monotonic_to_bootbased(&p->real_start_time);
1108 p->io_context = NULL;
1109 p->audit_context = NULL;
1112 p->mempolicy = mpol_dup(p->mempolicy);
1113 if (IS_ERR(p->mempolicy)) {
1114 retval = PTR_ERR(p->mempolicy);
1115 p->mempolicy = NULL;
1116 goto bad_fork_cleanup_cgroup;
1118 mpol_fix_fork_child_flag(p);
1120 #ifdef CONFIG_TRACE_IRQFLAGS
1122 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1123 p->hardirqs_enabled = 1;
1125 p->hardirqs_enabled = 0;
1127 p->hardirq_enable_ip = 0;
1128 p->hardirq_enable_event = 0;
1129 p->hardirq_disable_ip = _THIS_IP_;
1130 p->hardirq_disable_event = 0;
1131 p->softirqs_enabled = 1;
1132 p->softirq_enable_ip = _THIS_IP_;
1133 p->softirq_enable_event = 0;
1134 p->softirq_disable_ip = 0;
1135 p->softirq_disable_event = 0;
1136 p->hardirq_context = 0;
1137 p->softirq_context = 0;
1139 #ifdef CONFIG_LOCKDEP
1140 p->lockdep_depth = 0; /* no locks held yet */
1141 p->curr_chain_key = 0;
1142 p->lockdep_recursion = 0;
1145 #ifdef CONFIG_DEBUG_MUTEXES
1146 p->blocked_on = NULL; /* not blocked yet */
1148 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1149 p->memcg_batch.do_batch = 0;
1150 p->memcg_batch.memcg = NULL;
1153 /* Perform scheduler related setup. Assign this task to a CPU. */
1156 retval = perf_event_init_task(p);
1158 goto bad_fork_cleanup_policy;
1160 if ((retval = audit_alloc(p)))
1161 goto bad_fork_cleanup_policy;
1162 /* copy all the process information */
1163 if ((retval = copy_semundo(clone_flags, p)))
1164 goto bad_fork_cleanup_audit;
1165 if ((retval = copy_files(clone_flags, p)))
1166 goto bad_fork_cleanup_semundo;
1167 if ((retval = copy_fs(clone_flags, p)))
1168 goto bad_fork_cleanup_files;
1169 if ((retval = copy_sighand(clone_flags, p)))
1170 goto bad_fork_cleanup_fs;
1171 if ((retval = copy_signal(clone_flags, p)))
1172 goto bad_fork_cleanup_sighand;
1173 if ((retval = copy_mm(clone_flags, p)))
1174 goto bad_fork_cleanup_signal;
1175 if ((retval = copy_namespaces(clone_flags, p)))
1176 goto bad_fork_cleanup_mm;
1177 if ((retval = copy_io(clone_flags, p)))
1178 goto bad_fork_cleanup_namespaces;
1179 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1181 goto bad_fork_cleanup_io;
1183 if (pid != &init_struct_pid) {
1185 pid = alloc_pid(p->nsproxy->pid_ns);
1187 goto bad_fork_cleanup_io;
1190 p->pid = pid_nr(pid);
1192 if (clone_flags & CLONE_THREAD)
1193 p->tgid = current->tgid;
1195 if (current->nsproxy != p->nsproxy) {
1196 retval = ns_cgroup_clone(p, pid);
1198 goto bad_fork_free_pid;
1201 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1203 * Clear TID on mm_release()?
1205 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1210 p->robust_list = NULL;
1211 #ifdef CONFIG_COMPAT
1212 p->compat_robust_list = NULL;
1214 INIT_LIST_HEAD(&p->pi_state_list);
1215 p->pi_state_cache = NULL;
1218 * sigaltstack should be cleared when sharing the same VM
1220 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1221 p->sas_ss_sp = p->sas_ss_size = 0;
1224 * Syscall tracing and stepping should be turned off in the
1225 * child regardless of CLONE_PTRACE.
1227 user_disable_single_step(p);
1228 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1229 #ifdef TIF_SYSCALL_EMU
1230 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1232 clear_all_latency_tracing(p);
1234 /* ok, now we should be set up.. */
1235 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1236 p->pdeath_signal = 0;
1240 * Ok, make it visible to the rest of the system.
1241 * We dont wake it up yet.
1243 p->group_leader = p;
1244 INIT_LIST_HEAD(&p->thread_group);
1246 /* Now that the task is set up, run cgroup callbacks if
1247 * necessary. We need to run them before the task is visible
1248 * on the tasklist. */
1249 cgroup_fork_callbacks(p);
1250 cgroup_callbacks_done = 1;
1252 /* Need tasklist lock for parent etc handling! */
1253 write_lock_irq(&tasklist_lock);
1255 /* CLONE_PARENT re-uses the old parent */
1256 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1257 p->real_parent = current->real_parent;
1258 p->parent_exec_id = current->parent_exec_id;
1260 p->real_parent = current;
1261 p->parent_exec_id = current->self_exec_id;
1264 spin_lock(¤t->sighand->siglock);
1267 * Process group and session signals need to be delivered to just the
1268 * parent before the fork or both the parent and the child after the
1269 * fork. Restart if a signal comes in before we add the new process to
1270 * it's process group.
1271 * A fatal signal pending means that current will exit, so the new
1272 * thread can't slip out of an OOM kill (or normal SIGKILL).
1274 recalc_sigpending();
1275 if (signal_pending(current)) {
1276 spin_unlock(¤t->sighand->siglock);
1277 write_unlock_irq(&tasklist_lock);
1278 retval = -ERESTARTNOINTR;
1279 goto bad_fork_free_pid;
1282 if (clone_flags & CLONE_THREAD) {
1283 current->signal->nr_threads++;
1284 atomic_inc(¤t->signal->live);
1285 atomic_inc(¤t->signal->sigcnt);
1286 p->group_leader = current->group_leader;
1287 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1290 if (likely(p->pid)) {
1291 tracehook_finish_clone(p, clone_flags, trace);
1293 if (thread_group_leader(p)) {
1294 if (is_child_reaper(pid))
1295 p->nsproxy->pid_ns->child_reaper = p;
1297 p->signal->leader_pid = pid;
1298 p->signal->tty = tty_kref_get(current->signal->tty);
1299 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1300 attach_pid(p, PIDTYPE_SID, task_session(current));
1301 list_add_tail(&p->sibling, &p->real_parent->children);
1302 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1303 __this_cpu_inc(process_counts);
1305 attach_pid(p, PIDTYPE_PID, pid);
1310 spin_unlock(¤t->sighand->siglock);
1311 write_unlock_irq(&tasklist_lock);
1312 proc_fork_connector(p);
1313 cgroup_post_fork(p);
1318 if (pid != &init_struct_pid)
1320 bad_fork_cleanup_io:
1323 bad_fork_cleanup_namespaces:
1324 exit_task_namespaces(p);
1325 bad_fork_cleanup_mm:
1328 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1329 atomic_dec(&p->mm->oom_disable_count);
1333 bad_fork_cleanup_signal:
1334 if (!(clone_flags & CLONE_THREAD))
1335 free_signal_struct(p->signal);
1336 bad_fork_cleanup_sighand:
1337 __cleanup_sighand(p->sighand);
1338 bad_fork_cleanup_fs:
1339 exit_fs(p); /* blocking */
1340 bad_fork_cleanup_files:
1341 exit_files(p); /* blocking */
1342 bad_fork_cleanup_semundo:
1344 bad_fork_cleanup_audit:
1346 bad_fork_cleanup_policy:
1347 perf_event_free_task(p);
1349 mpol_put(p->mempolicy);
1350 bad_fork_cleanup_cgroup:
1352 cgroup_exit(p, cgroup_callbacks_done);
1353 delayacct_tsk_free(p);
1354 module_put(task_thread_info(p)->exec_domain->module);
1355 bad_fork_cleanup_count:
1356 atomic_dec(&p->cred->user->processes);
1361 return ERR_PTR(retval);
1364 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1366 memset(regs, 0, sizeof(struct pt_regs));
1370 static inline void init_idle_pids(struct pid_link *links)
1374 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1375 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1376 links[type].pid = &init_struct_pid;
1380 struct task_struct * __cpuinit fork_idle(int cpu)
1382 struct task_struct *task;
1383 struct pt_regs regs;
1385 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL,
1386 &init_struct_pid, 0);
1387 if (!IS_ERR(task)) {
1388 init_idle_pids(task->pids);
1389 init_idle(task, cpu);
1396 * Ok, this is the main fork-routine.
1398 * It copies the process, and if successful kick-starts
1399 * it and waits for it to finish using the VM if required.
1401 long do_fork(unsigned long clone_flags,
1402 unsigned long stack_start,
1403 struct pt_regs *regs,
1404 unsigned long stack_size,
1405 int __user *parent_tidptr,
1406 int __user *child_tidptr)
1408 struct task_struct *p;
1413 * Do some preliminary argument and permissions checking before we
1414 * actually start allocating stuff
1416 if (clone_flags & CLONE_NEWUSER) {
1417 if (clone_flags & CLONE_THREAD)
1419 /* hopefully this check will go away when userns support is
1422 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1423 !capable(CAP_SETGID))
1428 * When called from kernel_thread, don't do user tracing stuff.
1430 if (likely(user_mode(regs)))
1431 trace = tracehook_prepare_clone(clone_flags);
1433 p = copy_process(clone_flags, stack_start, regs, stack_size,
1434 child_tidptr, NULL, trace);
1436 * Do this prior waking up the new thread - the thread pointer
1437 * might get invalid after that point, if the thread exits quickly.
1440 struct completion vfork;
1442 trace_sched_process_fork(current, p);
1444 nr = task_pid_vnr(p);
1446 if (clone_flags & CLONE_PARENT_SETTID)
1447 put_user(nr, parent_tidptr);
1449 if (clone_flags & CLONE_VFORK) {
1450 p->vfork_done = &vfork;
1451 init_completion(&vfork);
1454 audit_finish_fork(p);
1455 tracehook_report_clone(regs, clone_flags, nr, p);
1458 * We set PF_STARTING at creation in case tracing wants to
1459 * use this to distinguish a fully live task from one that
1460 * hasn't gotten to tracehook_report_clone() yet. Now we
1461 * clear it and set the child going.
1463 p->flags &= ~PF_STARTING;
1465 wake_up_new_task(p);
1467 tracehook_report_clone_complete(trace, regs,
1468 clone_flags, nr, p);
1470 if (clone_flags & CLONE_VFORK) {
1471 freezer_do_not_count();
1472 wait_for_completion(&vfork);
1474 tracehook_report_vfork_done(p, nr);
1482 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1483 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1486 static void sighand_ctor(void *data)
1488 struct sighand_struct *sighand = data;
1490 spin_lock_init(&sighand->siglock);
1491 init_waitqueue_head(&sighand->signalfd_wqh);
1494 void __init proc_caches_init(void)
1496 sighand_cachep = kmem_cache_create("sighand_cache",
1497 sizeof(struct sighand_struct), 0,
1498 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1499 SLAB_NOTRACK, sighand_ctor);
1500 signal_cachep = kmem_cache_create("signal_cache",
1501 sizeof(struct signal_struct), 0,
1502 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1503 files_cachep = kmem_cache_create("files_cache",
1504 sizeof(struct files_struct), 0,
1505 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1506 fs_cachep = kmem_cache_create("fs_cache",
1507 sizeof(struct fs_struct), 0,
1508 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1509 mm_cachep = kmem_cache_create("mm_struct",
1510 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1511 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1512 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1517 * Check constraints on flags passed to the unshare system call.
1519 static int check_unshare_flags(unsigned long unshare_flags)
1521 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1522 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1523 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1526 * Not implemented, but pretend it works if there is nothing to
1527 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1528 * needs to unshare vm.
1530 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1531 /* FIXME: get_task_mm() increments ->mm_users */
1532 if (atomic_read(¤t->mm->mm_users) > 1)
1540 * Unshare the filesystem structure if it is being shared
1542 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1544 struct fs_struct *fs = current->fs;
1546 if (!(unshare_flags & CLONE_FS) || !fs)
1549 /* don't need lock here; in the worst case we'll do useless copy */
1553 *new_fsp = copy_fs_struct(fs);
1561 * Unshare file descriptor table if it is being shared
1563 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1565 struct files_struct *fd = current->files;
1568 if ((unshare_flags & CLONE_FILES) &&
1569 (fd && atomic_read(&fd->count) > 1)) {
1570 *new_fdp = dup_fd(fd, &error);
1579 * unshare allows a process to 'unshare' part of the process
1580 * context which was originally shared using clone. copy_*
1581 * functions used by do_fork() cannot be used here directly
1582 * because they modify an inactive task_struct that is being
1583 * constructed. Here we are modifying the current, active,
1586 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1588 struct fs_struct *fs, *new_fs = NULL;
1589 struct files_struct *fd, *new_fd = NULL;
1590 struct nsproxy *new_nsproxy = NULL;
1594 err = check_unshare_flags(unshare_flags);
1596 goto bad_unshare_out;
1599 * If unsharing namespace, must also unshare filesystem information.
1601 if (unshare_flags & CLONE_NEWNS)
1602 unshare_flags |= CLONE_FS;
1604 * CLONE_NEWIPC must also detach from the undolist: after switching
1605 * to a new ipc namespace, the semaphore arrays from the old
1606 * namespace are unreachable.
1608 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1610 if ((err = unshare_fs(unshare_flags, &new_fs)))
1611 goto bad_unshare_out;
1612 if ((err = unshare_fd(unshare_flags, &new_fd)))
1613 goto bad_unshare_cleanup_fs;
1614 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1616 goto bad_unshare_cleanup_fd;
1618 if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1621 * CLONE_SYSVSEM is equivalent to sys_exit().
1627 switch_task_namespaces(current, new_nsproxy);
1635 spin_lock(&fs->lock);
1636 current->fs = new_fs;
1641 spin_unlock(&fs->lock);
1645 fd = current->files;
1646 current->files = new_fd;
1650 task_unlock(current);
1654 put_nsproxy(new_nsproxy);
1656 bad_unshare_cleanup_fd:
1658 put_files_struct(new_fd);
1660 bad_unshare_cleanup_fs:
1662 free_fs_struct(new_fs);
1669 * Helper to unshare the files of the current task.
1670 * We don't want to expose copy_files internals to
1671 * the exec layer of the kernel.
1674 int unshare_files(struct files_struct **displaced)
1676 struct task_struct *task = current;
1677 struct files_struct *copy = NULL;
1680 error = unshare_fd(CLONE_FILES, ©);
1681 if (error || !copy) {
1685 *displaced = task->files;