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/acct.h>
53 #include <linux/tsacct_kern.h>
54 #include <linux/cn_proc.h>
55 #include <linux/freezer.h>
56 #include <linux/delayacct.h>
57 #include <linux/taskstats_kern.h>
58 #include <linux/random.h>
59 #include <linux/tty.h>
60 #include <linux/proc_fs.h>
61 #include <linux/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_counter.h>
66 #include <asm/pgtable.h>
67 #include <asm/pgalloc.h>
68 #include <asm/uaccess.h>
69 #include <asm/mmu_context.h>
70 #include <asm/cacheflush.h>
71 #include <asm/tlbflush.h>
73 #include <trace/events/sched.h>
76 * Protected counters by write_lock_irq(&tasklist_lock)
78 unsigned long total_forks; /* Handle normal Linux uptimes. */
79 int nr_threads; /* The idle threads do not count.. */
81 int max_threads; /* tunable limit on nr_threads */
83 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
85 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
87 int nr_processes(void)
92 for_each_online_cpu(cpu)
93 total += per_cpu(process_counts, cpu);
98 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
99 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
100 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
101 static struct kmem_cache *task_struct_cachep;
104 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
105 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
107 #ifdef CONFIG_DEBUG_STACK_USAGE
108 gfp_t mask = GFP_KERNEL | __GFP_ZERO;
110 gfp_t mask = GFP_KERNEL;
112 return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
115 static inline void free_thread_info(struct thread_info *ti)
117 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
121 /* SLAB cache for signal_struct structures (tsk->signal) */
122 static struct kmem_cache *signal_cachep;
124 /* SLAB cache for sighand_struct structures (tsk->sighand) */
125 struct kmem_cache *sighand_cachep;
127 /* SLAB cache for files_struct structures (tsk->files) */
128 struct kmem_cache *files_cachep;
130 /* SLAB cache for fs_struct structures (tsk->fs) */
131 struct kmem_cache *fs_cachep;
133 /* SLAB cache for vm_area_struct structures */
134 struct kmem_cache *vm_area_cachep;
136 /* SLAB cache for mm_struct structures (tsk->mm) */
137 static struct kmem_cache *mm_cachep;
139 void free_task(struct task_struct *tsk)
141 prop_local_destroy_single(&tsk->dirties);
142 free_thread_info(tsk->stack);
143 rt_mutex_debug_task_free(tsk);
144 ftrace_graph_exit_task(tsk);
145 free_task_struct(tsk);
147 EXPORT_SYMBOL(free_task);
149 void __put_task_struct(struct task_struct *tsk)
151 WARN_ON(!tsk->exit_state);
152 WARN_ON(atomic_read(&tsk->usage));
153 WARN_ON(tsk == current);
156 delayacct_tsk_free(tsk);
158 if (!profile_handoff_task(tsk))
163 * macro override instead of weak attribute alias, to workaround
164 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
166 #ifndef arch_task_cache_init
167 #define arch_task_cache_init()
170 void __init fork_init(unsigned long mempages)
172 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
173 #ifndef ARCH_MIN_TASKALIGN
174 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
176 /* create a slab on which task_structs can be allocated */
178 kmem_cache_create("task_struct", sizeof(struct task_struct),
179 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
182 /* do the arch specific task caches init */
183 arch_task_cache_init();
186 * The default maximum number of threads is set to a safe
187 * value: the thread structures can take up at most half
190 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
193 * we need to allow at least 20 threads to boot a system
198 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
199 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
200 init_task.signal->rlim[RLIMIT_SIGPENDING] =
201 init_task.signal->rlim[RLIMIT_NPROC];
204 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
205 struct task_struct *src)
211 static struct task_struct *dup_task_struct(struct task_struct *orig)
213 struct task_struct *tsk;
214 struct thread_info *ti;
215 unsigned long *stackend;
219 prepare_to_copy(orig);
221 tsk = alloc_task_struct();
225 ti = alloc_thread_info(tsk);
227 free_task_struct(tsk);
231 err = arch_dup_task_struct(tsk, orig);
237 err = prop_local_init_single(&tsk->dirties);
241 setup_thread_stack(tsk, orig);
242 stackend = end_of_stack(tsk);
243 *stackend = STACK_END_MAGIC; /* for overflow detection */
245 #ifdef CONFIG_CC_STACKPROTECTOR
246 tsk->stack_canary = get_random_int();
249 /* One for us, one for whoever does the "release_task()" (usually parent) */
250 atomic_set(&tsk->usage,2);
251 atomic_set(&tsk->fs_excl, 0);
252 #ifdef CONFIG_BLK_DEV_IO_TRACE
255 tsk->splice_pipe = NULL;
259 free_thread_info(ti);
260 free_task_struct(tsk);
265 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
267 struct vm_area_struct *mpnt, *tmp, **pprev;
268 struct rb_node **rb_link, *rb_parent;
270 unsigned long charge;
271 struct mempolicy *pol;
273 down_write(&oldmm->mmap_sem);
274 flush_cache_dup_mm(oldmm);
276 * Not linked in yet - no deadlock potential:
278 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
282 mm->mmap_cache = NULL;
283 mm->free_area_cache = oldmm->mmap_base;
284 mm->cached_hole_size = ~0UL;
286 cpumask_clear(mm_cpumask(mm));
288 rb_link = &mm->mm_rb.rb_node;
292 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
295 if (mpnt->vm_flags & VM_DONTCOPY) {
296 long pages = vma_pages(mpnt);
297 mm->total_vm -= pages;
298 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
303 if (mpnt->vm_flags & VM_ACCOUNT) {
304 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
305 if (security_vm_enough_memory(len))
309 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
313 pol = mpol_dup(vma_policy(mpnt));
314 retval = PTR_ERR(pol);
316 goto fail_nomem_policy;
317 vma_set_policy(tmp, pol);
318 tmp->vm_flags &= ~VM_LOCKED;
324 struct inode *inode = file->f_path.dentry->d_inode;
325 struct address_space *mapping = file->f_mapping;
328 if (tmp->vm_flags & VM_DENYWRITE)
329 atomic_dec(&inode->i_writecount);
330 spin_lock(&mapping->i_mmap_lock);
331 if (tmp->vm_flags & VM_SHARED)
332 mapping->i_mmap_writable++;
333 tmp->vm_truncate_count = mpnt->vm_truncate_count;
334 flush_dcache_mmap_lock(mapping);
335 /* insert tmp into the share list, just after mpnt */
336 vma_prio_tree_add(tmp, mpnt);
337 flush_dcache_mmap_unlock(mapping);
338 spin_unlock(&mapping->i_mmap_lock);
342 * Clear hugetlb-related page reserves for children. This only
343 * affects MAP_PRIVATE mappings. Faults generated by the child
344 * are not guaranteed to succeed, even if read-only
346 if (is_vm_hugetlb_page(tmp))
347 reset_vma_resv_huge_pages(tmp);
350 * Link in the new vma and copy the page table entries.
353 pprev = &tmp->vm_next;
355 __vma_link_rb(mm, tmp, rb_link, rb_parent);
356 rb_link = &tmp->vm_rb.rb_right;
357 rb_parent = &tmp->vm_rb;
360 retval = copy_page_range(mm, oldmm, mpnt);
362 if (tmp->vm_ops && tmp->vm_ops->open)
363 tmp->vm_ops->open(tmp);
368 /* a new mm has just been created */
369 arch_dup_mmap(oldmm, mm);
372 up_write(&mm->mmap_sem);
374 up_write(&oldmm->mmap_sem);
377 kmem_cache_free(vm_area_cachep, tmp);
380 vm_unacct_memory(charge);
384 static inline int mm_alloc_pgd(struct mm_struct * mm)
386 mm->pgd = pgd_alloc(mm);
387 if (unlikely(!mm->pgd))
392 static inline void mm_free_pgd(struct mm_struct * mm)
394 pgd_free(mm, mm->pgd);
397 #define dup_mmap(mm, oldmm) (0)
398 #define mm_alloc_pgd(mm) (0)
399 #define mm_free_pgd(mm)
400 #endif /* CONFIG_MMU */
402 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
404 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
405 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
407 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
409 static int __init coredump_filter_setup(char *s)
411 default_dump_filter =
412 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
413 MMF_DUMP_FILTER_MASK;
417 __setup("coredump_filter=", coredump_filter_setup);
419 #include <linux/init_task.h>
421 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
423 atomic_set(&mm->mm_users, 1);
424 atomic_set(&mm->mm_count, 1);
425 init_rwsem(&mm->mmap_sem);
426 INIT_LIST_HEAD(&mm->mmlist);
427 mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
428 mm->core_state = NULL;
430 set_mm_counter(mm, file_rss, 0);
431 set_mm_counter(mm, anon_rss, 0);
432 spin_lock_init(&mm->page_table_lock);
433 spin_lock_init(&mm->ioctx_lock);
434 INIT_HLIST_HEAD(&mm->ioctx_list);
435 mm->free_area_cache = TASK_UNMAPPED_BASE;
436 mm->cached_hole_size = ~0UL;
437 mm_init_owner(mm, p);
439 if (likely(!mm_alloc_pgd(mm))) {
441 mmu_notifier_mm_init(mm);
450 * Allocate and initialize an mm_struct.
452 struct mm_struct * mm_alloc(void)
454 struct mm_struct * mm;
458 memset(mm, 0, sizeof(*mm));
459 mm = mm_init(mm, current);
465 * Called when the last reference to the mm
466 * is dropped: either by a lazy thread or by
467 * mmput. Free the page directory and the mm.
469 void __mmdrop(struct mm_struct *mm)
471 BUG_ON(mm == &init_mm);
474 mmu_notifier_mm_destroy(mm);
477 EXPORT_SYMBOL_GPL(__mmdrop);
480 * Decrement the use count and release all resources for an mm.
482 void mmput(struct mm_struct *mm)
486 if (atomic_dec_and_test(&mm->mm_users)) {
489 set_mm_exe_file(mm, NULL);
490 if (!list_empty(&mm->mmlist)) {
491 spin_lock(&mmlist_lock);
492 list_del(&mm->mmlist);
493 spin_unlock(&mmlist_lock);
499 EXPORT_SYMBOL_GPL(mmput);
502 * get_task_mm - acquire a reference to the task's mm
504 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
505 * this kernel workthread has transiently adopted a user mm with use_mm,
506 * to do its AIO) is not set and if so returns a reference to it, after
507 * bumping up the use count. User must release the mm via mmput()
508 * after use. Typically used by /proc and ptrace.
510 struct mm_struct *get_task_mm(struct task_struct *task)
512 struct mm_struct *mm;
517 if (task->flags & PF_KTHREAD)
520 atomic_inc(&mm->mm_users);
525 EXPORT_SYMBOL_GPL(get_task_mm);
527 /* Please note the differences between mmput and mm_release.
528 * mmput is called whenever we stop holding onto a mm_struct,
529 * error success whatever.
531 * mm_release is called after a mm_struct has been removed
532 * from the current process.
534 * This difference is important for error handling, when we
535 * only half set up a mm_struct for a new process and need to restore
536 * the old one. Because we mmput the new mm_struct before
537 * restoring the old one. . .
538 * Eric Biederman 10 January 1998
540 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
542 struct completion *vfork_done = tsk->vfork_done;
544 /* Get rid of any futexes when releasing the mm */
546 if (unlikely(tsk->robust_list)) {
547 exit_robust_list(tsk);
548 tsk->robust_list = NULL;
551 if (unlikely(tsk->compat_robust_list)) {
552 compat_exit_robust_list(tsk);
553 tsk->compat_robust_list = NULL;
558 /* Get rid of any cached register state */
559 deactivate_mm(tsk, mm);
561 /* notify parent sleeping on vfork() */
563 tsk->vfork_done = NULL;
564 complete(vfork_done);
568 * If we're exiting normally, clear a user-space tid field if
569 * requested. We leave this alone when dying by signal, to leave
570 * the value intact in a core dump, and to save the unnecessary
571 * trouble otherwise. Userland only wants this done for a sys_exit.
573 if (tsk->clear_child_tid) {
574 if (!(tsk->flags & PF_SIGNALED) &&
575 atomic_read(&mm->mm_users) > 1) {
577 * We don't check the error code - if userspace has
578 * not set up a proper pointer then tough luck.
580 put_user(0, tsk->clear_child_tid);
581 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
584 tsk->clear_child_tid = NULL;
589 * Allocate a new mm structure and copy contents from the
590 * mm structure of the passed in task structure.
592 struct mm_struct *dup_mm(struct task_struct *tsk)
594 struct mm_struct *mm, *oldmm = current->mm;
604 memcpy(mm, oldmm, sizeof(*mm));
606 /* Initializing for Swap token stuff */
607 mm->token_priority = 0;
608 mm->last_interval = 0;
610 if (!mm_init(mm, tsk))
613 if (init_new_context(tsk, mm))
616 dup_mm_exe_file(oldmm, mm);
618 err = dup_mmap(mm, oldmm);
622 mm->hiwater_rss = get_mm_rss(mm);
623 mm->hiwater_vm = mm->total_vm;
635 * If init_new_context() failed, we cannot use mmput() to free the mm
636 * because it calls destroy_context()
643 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
645 struct mm_struct * mm, *oldmm;
648 tsk->min_flt = tsk->maj_flt = 0;
649 tsk->nvcsw = tsk->nivcsw = 0;
650 #ifdef CONFIG_DETECT_HUNG_TASK
651 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
655 tsk->active_mm = NULL;
658 * Are we cloning a kernel thread?
660 * We need to steal a active VM for that..
666 if (clone_flags & CLONE_VM) {
667 atomic_inc(&oldmm->mm_users);
678 /* Initializing for Swap token stuff */
679 mm->token_priority = 0;
680 mm->last_interval = 0;
690 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
692 struct fs_struct *fs = current->fs;
693 if (clone_flags & CLONE_FS) {
694 /* tsk->fs is already what we want */
695 write_lock(&fs->lock);
697 write_unlock(&fs->lock);
701 write_unlock(&fs->lock);
704 tsk->fs = copy_fs_struct(fs);
710 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
712 struct files_struct *oldf, *newf;
716 * A background process may not have any files ...
718 oldf = current->files;
722 if (clone_flags & CLONE_FILES) {
723 atomic_inc(&oldf->count);
727 newf = dup_fd(oldf, &error);
737 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
740 struct io_context *ioc = current->io_context;
745 * Share io context with parent, if CLONE_IO is set
747 if (clone_flags & CLONE_IO) {
748 tsk->io_context = ioc_task_link(ioc);
749 if (unlikely(!tsk->io_context))
751 } else if (ioprio_valid(ioc->ioprio)) {
752 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
753 if (unlikely(!tsk->io_context))
756 tsk->io_context->ioprio = ioc->ioprio;
762 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
764 struct sighand_struct *sig;
766 if (clone_flags & CLONE_SIGHAND) {
767 atomic_inc(¤t->sighand->count);
770 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
771 rcu_assign_pointer(tsk->sighand, sig);
774 atomic_set(&sig->count, 1);
775 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
779 void __cleanup_sighand(struct sighand_struct *sighand)
781 if (atomic_dec_and_test(&sighand->count))
782 kmem_cache_free(sighand_cachep, sighand);
787 * Initialize POSIX timer handling for a thread group.
789 static void posix_cpu_timers_init_group(struct signal_struct *sig)
791 /* Thread group counters. */
792 thread_group_cputime_init(sig);
794 /* Expiration times and increments. */
795 sig->it_virt_expires = cputime_zero;
796 sig->it_virt_incr = cputime_zero;
797 sig->it_prof_expires = cputime_zero;
798 sig->it_prof_incr = cputime_zero;
800 /* Cached expiration times. */
801 sig->cputime_expires.prof_exp = cputime_zero;
802 sig->cputime_expires.virt_exp = cputime_zero;
803 sig->cputime_expires.sched_exp = 0;
805 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
806 sig->cputime_expires.prof_exp =
807 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
808 sig->cputimer.running = 1;
811 /* The timer lists. */
812 INIT_LIST_HEAD(&sig->cpu_timers[0]);
813 INIT_LIST_HEAD(&sig->cpu_timers[1]);
814 INIT_LIST_HEAD(&sig->cpu_timers[2]);
817 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
819 struct signal_struct *sig;
821 if (clone_flags & CLONE_THREAD)
824 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
829 atomic_set(&sig->count, 1);
830 atomic_set(&sig->live, 1);
831 init_waitqueue_head(&sig->wait_chldexit);
833 if (clone_flags & CLONE_NEWPID)
834 sig->flags |= SIGNAL_UNKILLABLE;
835 sig->group_exit_code = 0;
836 sig->group_exit_task = NULL;
837 sig->group_stop_count = 0;
838 sig->curr_target = tsk;
839 init_sigpending(&sig->shared_pending);
840 INIT_LIST_HEAD(&sig->posix_timers);
842 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
843 sig->it_real_incr.tv64 = 0;
844 sig->real_timer.function = it_real_fn;
846 sig->leader = 0; /* session leadership doesn't inherit */
847 sig->tty_old_pgrp = NULL;
850 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
851 sig->gtime = cputime_zero;
852 sig->cgtime = cputime_zero;
853 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
854 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
855 sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
856 task_io_accounting_init(&sig->ioac);
857 sig->sum_sched_runtime = 0;
858 taskstats_tgid_init(sig);
860 task_lock(current->group_leader);
861 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
862 task_unlock(current->group_leader);
864 posix_cpu_timers_init_group(sig);
866 acct_init_pacct(&sig->pacct);
873 void __cleanup_signal(struct signal_struct *sig)
875 thread_group_cputime_free(sig);
876 tty_kref_put(sig->tty);
877 kmem_cache_free(signal_cachep, sig);
880 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
882 unsigned long new_flags = p->flags;
884 new_flags &= ~PF_SUPERPRIV;
885 new_flags |= PF_FORKNOEXEC;
886 new_flags |= PF_STARTING;
887 p->flags = new_flags;
888 clear_freeze_flag(p);
891 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
893 current->clear_child_tid = tidptr;
895 return task_pid_vnr(current);
898 static void rt_mutex_init_task(struct task_struct *p)
900 spin_lock_init(&p->pi_lock);
901 #ifdef CONFIG_RT_MUTEXES
902 plist_head_init(&p->pi_waiters, &p->pi_lock);
903 p->pi_blocked_on = NULL;
907 #ifdef CONFIG_MM_OWNER
908 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
912 #endif /* CONFIG_MM_OWNER */
915 * Initialize POSIX timer handling for a single task.
917 static void posix_cpu_timers_init(struct task_struct *tsk)
919 tsk->cputime_expires.prof_exp = cputime_zero;
920 tsk->cputime_expires.virt_exp = cputime_zero;
921 tsk->cputime_expires.sched_exp = 0;
922 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
923 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
924 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
928 * This creates a new process as a copy of the old one,
929 * but does not actually start it yet.
931 * It copies the registers, and all the appropriate
932 * parts of the process environment (as per the clone
933 * flags). The actual kick-off is left to the caller.
935 static struct task_struct *copy_process(unsigned long clone_flags,
936 unsigned long stack_start,
937 struct pt_regs *regs,
938 unsigned long stack_size,
939 int __user *child_tidptr,
944 struct task_struct *p;
945 int cgroup_callbacks_done = 0;
947 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
948 return ERR_PTR(-EINVAL);
951 * Thread groups must share signals as well, and detached threads
952 * can only be started up within the thread group.
954 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
955 return ERR_PTR(-EINVAL);
958 * Shared signal handlers imply shared VM. By way of the above,
959 * thread groups also imply shared VM. Blocking this case allows
960 * for various simplifications in other code.
962 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
963 return ERR_PTR(-EINVAL);
965 retval = security_task_create(clone_flags);
970 p = dup_task_struct(current);
974 ftrace_graph_init_task(p);
976 rt_mutex_init_task(p);
978 #ifdef CONFIG_PROVE_LOCKING
979 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
980 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
983 if (atomic_read(&p->real_cred->user->processes) >=
984 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
985 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
986 p->real_cred->user != INIT_USER)
990 retval = copy_creds(p, clone_flags);
995 * If multiple threads are within copy_process(), then this check
996 * triggers too late. This doesn't hurt, the check is only there
997 * to stop root fork bombs.
1000 if (nr_threads >= max_threads)
1001 goto bad_fork_cleanup_count;
1003 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1004 goto bad_fork_cleanup_count;
1006 if (p->binfmt && !try_module_get(p->binfmt->module))
1007 goto bad_fork_cleanup_put_domain;
1010 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1011 copy_flags(clone_flags, p);
1012 INIT_LIST_HEAD(&p->children);
1013 INIT_LIST_HEAD(&p->sibling);
1014 rcu_copy_process(p);
1015 p->vfork_done = NULL;
1016 spin_lock_init(&p->alloc_lock);
1018 init_sigpending(&p->pending);
1020 p->utime = cputime_zero;
1021 p->stime = cputime_zero;
1022 p->gtime = cputime_zero;
1023 p->utimescaled = cputime_zero;
1024 p->stimescaled = cputime_zero;
1025 p->prev_utime = cputime_zero;
1026 p->prev_stime = cputime_zero;
1028 p->default_timer_slack_ns = current->timer_slack_ns;
1030 task_io_accounting_init(&p->ioac);
1031 acct_clear_integrals(p);
1033 posix_cpu_timers_init(p);
1035 p->lock_depth = -1; /* -1 = no lock */
1036 do_posix_clock_monotonic_gettime(&p->start_time);
1037 p->real_start_time = p->start_time;
1038 monotonic_to_bootbased(&p->real_start_time);
1039 p->io_context = NULL;
1040 p->audit_context = NULL;
1043 p->mempolicy = mpol_dup(p->mempolicy);
1044 if (IS_ERR(p->mempolicy)) {
1045 retval = PTR_ERR(p->mempolicy);
1046 p->mempolicy = NULL;
1047 goto bad_fork_cleanup_cgroup;
1049 mpol_fix_fork_child_flag(p);
1051 #ifdef CONFIG_TRACE_IRQFLAGS
1053 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1054 p->hardirqs_enabled = 1;
1056 p->hardirqs_enabled = 0;
1058 p->hardirq_enable_ip = 0;
1059 p->hardirq_enable_event = 0;
1060 p->hardirq_disable_ip = _THIS_IP_;
1061 p->hardirq_disable_event = 0;
1062 p->softirqs_enabled = 1;
1063 p->softirq_enable_ip = _THIS_IP_;
1064 p->softirq_enable_event = 0;
1065 p->softirq_disable_ip = 0;
1066 p->softirq_disable_event = 0;
1067 p->hardirq_context = 0;
1068 p->softirq_context = 0;
1070 #ifdef CONFIG_LOCKDEP
1071 p->lockdep_depth = 0; /* no locks held yet */
1072 p->curr_chain_key = 0;
1073 p->lockdep_recursion = 0;
1076 #ifdef CONFIG_DEBUG_MUTEXES
1077 p->blocked_on = NULL; /* not blocked yet */
1082 /* Perform scheduler related setup. Assign this task to a CPU. */
1083 sched_fork(p, clone_flags);
1085 retval = perf_counter_init_task(p);
1087 goto bad_fork_cleanup_policy;
1089 if ((retval = audit_alloc(p)))
1090 goto bad_fork_cleanup_policy;
1091 /* copy all the process information */
1092 if ((retval = copy_semundo(clone_flags, p)))
1093 goto bad_fork_cleanup_audit;
1094 if ((retval = copy_files(clone_flags, p)))
1095 goto bad_fork_cleanup_semundo;
1096 if ((retval = copy_fs(clone_flags, p)))
1097 goto bad_fork_cleanup_files;
1098 if ((retval = copy_sighand(clone_flags, p)))
1099 goto bad_fork_cleanup_fs;
1100 if ((retval = copy_signal(clone_flags, p)))
1101 goto bad_fork_cleanup_sighand;
1102 if ((retval = copy_mm(clone_flags, p)))
1103 goto bad_fork_cleanup_signal;
1104 if ((retval = copy_namespaces(clone_flags, p)))
1105 goto bad_fork_cleanup_mm;
1106 if ((retval = copy_io(clone_flags, p)))
1107 goto bad_fork_cleanup_namespaces;
1108 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1110 goto bad_fork_cleanup_io;
1112 if (pid != &init_struct_pid) {
1114 pid = alloc_pid(p->nsproxy->pid_ns);
1116 goto bad_fork_cleanup_io;
1118 if (clone_flags & CLONE_NEWPID) {
1119 retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1121 goto bad_fork_free_pid;
1125 p->pid = pid_nr(pid);
1127 if (clone_flags & CLONE_THREAD)
1128 p->tgid = current->tgid;
1130 if (current->nsproxy != p->nsproxy) {
1131 retval = ns_cgroup_clone(p, pid);
1133 goto bad_fork_free_pid;
1136 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1138 * Clear TID on mm_release()?
1140 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1142 p->robust_list = NULL;
1143 #ifdef CONFIG_COMPAT
1144 p->compat_robust_list = NULL;
1146 INIT_LIST_HEAD(&p->pi_state_list);
1147 p->pi_state_cache = NULL;
1150 * sigaltstack should be cleared when sharing the same VM
1152 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1153 p->sas_ss_sp = p->sas_ss_size = 0;
1156 * Syscall tracing should be turned off in the child regardless
1159 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1160 #ifdef TIF_SYSCALL_EMU
1161 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1163 clear_all_latency_tracing(p);
1165 /* ok, now we should be set up.. */
1166 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1167 p->pdeath_signal = 0;
1171 * Ok, make it visible to the rest of the system.
1172 * We dont wake it up yet.
1174 p->group_leader = p;
1175 INIT_LIST_HEAD(&p->thread_group);
1177 /* Now that the task is set up, run cgroup callbacks if
1178 * necessary. We need to run them before the task is visible
1179 * on the tasklist. */
1180 cgroup_fork_callbacks(p);
1181 cgroup_callbacks_done = 1;
1183 /* Need tasklist lock for parent etc handling! */
1184 write_lock_irq(&tasklist_lock);
1187 * The task hasn't been attached yet, so its cpus_allowed mask will
1188 * not be changed, nor will its assigned CPU.
1190 * The cpus_allowed mask of the parent may have changed after it was
1191 * copied first time - so re-copy it here, then check the child's CPU
1192 * to ensure it is on a valid CPU (and if not, just force it back to
1193 * parent's CPU). This avoids alot of nasty races.
1195 p->cpus_allowed = current->cpus_allowed;
1196 p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1197 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1198 !cpu_online(task_cpu(p))))
1199 set_task_cpu(p, smp_processor_id());
1201 /* CLONE_PARENT re-uses the old parent */
1202 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1203 p->real_parent = current->real_parent;
1204 p->parent_exec_id = current->parent_exec_id;
1206 p->real_parent = current;
1207 p->parent_exec_id = current->self_exec_id;
1210 spin_lock(¤t->sighand->siglock);
1213 * Process group and session signals need to be delivered to just the
1214 * parent before the fork or both the parent and the child after the
1215 * fork. Restart if a signal comes in before we add the new process to
1216 * it's process group.
1217 * A fatal signal pending means that current will exit, so the new
1218 * thread can't slip out of an OOM kill (or normal SIGKILL).
1220 recalc_sigpending();
1221 if (signal_pending(current)) {
1222 spin_unlock(¤t->sighand->siglock);
1223 write_unlock_irq(&tasklist_lock);
1224 retval = -ERESTARTNOINTR;
1225 goto bad_fork_free_pid;
1228 if (clone_flags & CLONE_THREAD) {
1229 atomic_inc(¤t->signal->count);
1230 atomic_inc(¤t->signal->live);
1231 p->group_leader = current->group_leader;
1232 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1235 if (likely(p->pid)) {
1236 list_add_tail(&p->sibling, &p->real_parent->children);
1237 tracehook_finish_clone(p, clone_flags, trace);
1239 if (thread_group_leader(p)) {
1240 if (clone_flags & CLONE_NEWPID)
1241 p->nsproxy->pid_ns->child_reaper = p;
1243 p->signal->leader_pid = pid;
1244 tty_kref_put(p->signal->tty);
1245 p->signal->tty = tty_kref_get(current->signal->tty);
1246 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1247 attach_pid(p, PIDTYPE_SID, task_session(current));
1248 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1249 __get_cpu_var(process_counts)++;
1251 attach_pid(p, PIDTYPE_PID, pid);
1256 spin_unlock(¤t->sighand->siglock);
1257 write_unlock_irq(&tasklist_lock);
1258 proc_fork_connector(p);
1259 cgroup_post_fork(p);
1260 perf_counter_fork(p);
1264 if (pid != &init_struct_pid)
1266 bad_fork_cleanup_io:
1267 put_io_context(p->io_context);
1268 bad_fork_cleanup_namespaces:
1269 exit_task_namespaces(p);
1270 bad_fork_cleanup_mm:
1273 bad_fork_cleanup_signal:
1274 if (!(clone_flags & CLONE_THREAD))
1275 __cleanup_signal(p->signal);
1276 bad_fork_cleanup_sighand:
1277 __cleanup_sighand(p->sighand);
1278 bad_fork_cleanup_fs:
1279 exit_fs(p); /* blocking */
1280 bad_fork_cleanup_files:
1281 exit_files(p); /* blocking */
1282 bad_fork_cleanup_semundo:
1284 bad_fork_cleanup_audit:
1286 bad_fork_cleanup_policy:
1287 perf_counter_free_task(p);
1289 mpol_put(p->mempolicy);
1290 bad_fork_cleanup_cgroup:
1292 cgroup_exit(p, cgroup_callbacks_done);
1293 delayacct_tsk_free(p);
1295 module_put(p->binfmt->module);
1296 bad_fork_cleanup_put_domain:
1297 module_put(task_thread_info(p)->exec_domain->module);
1298 bad_fork_cleanup_count:
1299 atomic_dec(&p->cred->user->processes);
1304 return ERR_PTR(retval);
1307 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1309 memset(regs, 0, sizeof(struct pt_regs));
1313 struct task_struct * __cpuinit fork_idle(int cpu)
1315 struct task_struct *task;
1316 struct pt_regs regs;
1318 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL,
1319 &init_struct_pid, 0);
1321 init_idle(task, cpu);
1327 * Ok, this is the main fork-routine.
1329 * It copies the process, and if successful kick-starts
1330 * it and waits for it to finish using the VM if required.
1332 long do_fork(unsigned long clone_flags,
1333 unsigned long stack_start,
1334 struct pt_regs *regs,
1335 unsigned long stack_size,
1336 int __user *parent_tidptr,
1337 int __user *child_tidptr)
1339 struct task_struct *p;
1344 * Do some preliminary argument and permissions checking before we
1345 * actually start allocating stuff
1347 if (clone_flags & CLONE_NEWUSER) {
1348 if (clone_flags & CLONE_THREAD)
1350 /* hopefully this check will go away when userns support is
1353 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1354 !capable(CAP_SETGID))
1359 * We hope to recycle these flags after 2.6.26
1361 if (unlikely(clone_flags & CLONE_STOPPED)) {
1362 static int __read_mostly count = 100;
1364 if (count > 0 && printk_ratelimit()) {
1365 char comm[TASK_COMM_LEN];
1368 printk(KERN_INFO "fork(): process `%s' used deprecated "
1369 "clone flags 0x%lx\n",
1370 get_task_comm(comm, current),
1371 clone_flags & CLONE_STOPPED);
1376 * When called from kernel_thread, don't do user tracing stuff.
1378 if (likely(user_mode(regs)))
1379 trace = tracehook_prepare_clone(clone_flags);
1381 p = copy_process(clone_flags, stack_start, regs, stack_size,
1382 child_tidptr, NULL, trace);
1384 * Do this prior waking up the new thread - the thread pointer
1385 * might get invalid after that point, if the thread exits quickly.
1388 struct completion vfork;
1390 trace_sched_process_fork(current, p);
1392 nr = task_pid_vnr(p);
1394 if (clone_flags & CLONE_PARENT_SETTID)
1395 put_user(nr, parent_tidptr);
1397 if (clone_flags & CLONE_VFORK) {
1398 p->vfork_done = &vfork;
1399 init_completion(&vfork);
1402 audit_finish_fork(p);
1403 tracehook_report_clone(regs, clone_flags, nr, p);
1406 * We set PF_STARTING at creation in case tracing wants to
1407 * use this to distinguish a fully live task from one that
1408 * hasn't gotten to tracehook_report_clone() yet. Now we
1409 * clear it and set the child going.
1411 p->flags &= ~PF_STARTING;
1413 if (unlikely(clone_flags & CLONE_STOPPED)) {
1415 * We'll start up with an immediate SIGSTOP.
1417 sigaddset(&p->pending.signal, SIGSTOP);
1418 set_tsk_thread_flag(p, TIF_SIGPENDING);
1419 __set_task_state(p, TASK_STOPPED);
1421 wake_up_new_task(p, clone_flags);
1424 tracehook_report_clone_complete(trace, regs,
1425 clone_flags, nr, p);
1427 if (clone_flags & CLONE_VFORK) {
1428 freezer_do_not_count();
1429 wait_for_completion(&vfork);
1431 tracehook_report_vfork_done(p, nr);
1439 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1440 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1443 static void sighand_ctor(void *data)
1445 struct sighand_struct *sighand = data;
1447 spin_lock_init(&sighand->siglock);
1448 init_waitqueue_head(&sighand->signalfd_wqh);
1451 void __init proc_caches_init(void)
1453 sighand_cachep = kmem_cache_create("sighand_cache",
1454 sizeof(struct sighand_struct), 0,
1455 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1456 SLAB_NOTRACK, sighand_ctor);
1457 signal_cachep = kmem_cache_create("signal_cache",
1458 sizeof(struct signal_struct), 0,
1459 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1460 files_cachep = kmem_cache_create("files_cache",
1461 sizeof(struct files_struct), 0,
1462 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1463 fs_cachep = kmem_cache_create("fs_cache",
1464 sizeof(struct fs_struct), 0,
1465 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1466 mm_cachep = kmem_cache_create("mm_struct",
1467 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1468 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1469 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1474 * Check constraints on flags passed to the unshare system call and
1475 * force unsharing of additional process context as appropriate.
1477 static void check_unshare_flags(unsigned long *flags_ptr)
1480 * If unsharing a thread from a thread group, must also
1483 if (*flags_ptr & CLONE_THREAD)
1484 *flags_ptr |= CLONE_VM;
1487 * If unsharing vm, must also unshare signal handlers.
1489 if (*flags_ptr & CLONE_VM)
1490 *flags_ptr |= CLONE_SIGHAND;
1493 * If unsharing signal handlers and the task was created
1494 * using CLONE_THREAD, then must unshare the thread
1496 if ((*flags_ptr & CLONE_SIGHAND) &&
1497 (atomic_read(¤t->signal->count) > 1))
1498 *flags_ptr |= CLONE_THREAD;
1501 * If unsharing namespace, must also unshare filesystem information.
1503 if (*flags_ptr & CLONE_NEWNS)
1504 *flags_ptr |= CLONE_FS;
1508 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1510 static int unshare_thread(unsigned long unshare_flags)
1512 if (unshare_flags & CLONE_THREAD)
1519 * Unshare the filesystem structure if it is being shared
1521 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1523 struct fs_struct *fs = current->fs;
1525 if (!(unshare_flags & CLONE_FS) || !fs)
1528 /* don't need lock here; in the worst case we'll do useless copy */
1532 *new_fsp = copy_fs_struct(fs);
1540 * Unsharing of sighand is not supported yet
1542 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1544 struct sighand_struct *sigh = current->sighand;
1546 if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1553 * Unshare vm if it is being shared
1555 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1557 struct mm_struct *mm = current->mm;
1559 if ((unshare_flags & CLONE_VM) &&
1560 (mm && atomic_read(&mm->mm_users) > 1)) {
1568 * Unshare file descriptor table if it is being shared
1570 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1572 struct files_struct *fd = current->files;
1575 if ((unshare_flags & CLONE_FILES) &&
1576 (fd && atomic_read(&fd->count) > 1)) {
1577 *new_fdp = dup_fd(fd, &error);
1586 * unshare allows a process to 'unshare' part of the process
1587 * context which was originally shared using clone. copy_*
1588 * functions used by do_fork() cannot be used here directly
1589 * because they modify an inactive task_struct that is being
1590 * constructed. Here we are modifying the current, active,
1593 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1596 struct fs_struct *fs, *new_fs = NULL;
1597 struct sighand_struct *new_sigh = NULL;
1598 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1599 struct files_struct *fd, *new_fd = NULL;
1600 struct nsproxy *new_nsproxy = NULL;
1603 check_unshare_flags(&unshare_flags);
1605 /* Return -EINVAL for all unsupported flags */
1607 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1608 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1609 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1610 goto bad_unshare_out;
1613 * CLONE_NEWIPC must also detach from the undolist: after switching
1614 * to a new ipc namespace, the semaphore arrays from the old
1615 * namespace are unreachable.
1617 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1619 if ((err = unshare_thread(unshare_flags)))
1620 goto bad_unshare_out;
1621 if ((err = unshare_fs(unshare_flags, &new_fs)))
1622 goto bad_unshare_cleanup_thread;
1623 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1624 goto bad_unshare_cleanup_fs;
1625 if ((err = unshare_vm(unshare_flags, &new_mm)))
1626 goto bad_unshare_cleanup_sigh;
1627 if ((err = unshare_fd(unshare_flags, &new_fd)))
1628 goto bad_unshare_cleanup_vm;
1629 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1631 goto bad_unshare_cleanup_fd;
1633 if (new_fs || new_mm || new_fd || do_sysvsem || new_nsproxy) {
1636 * CLONE_SYSVSEM is equivalent to sys_exit().
1642 switch_task_namespaces(current, new_nsproxy);
1650 write_lock(&fs->lock);
1651 current->fs = new_fs;
1656 write_unlock(&fs->lock);
1661 active_mm = current->active_mm;
1662 current->mm = new_mm;
1663 current->active_mm = new_mm;
1664 activate_mm(active_mm, new_mm);
1669 fd = current->files;
1670 current->files = new_fd;
1674 task_unlock(current);
1678 put_nsproxy(new_nsproxy);
1680 bad_unshare_cleanup_fd:
1682 put_files_struct(new_fd);
1684 bad_unshare_cleanup_vm:
1688 bad_unshare_cleanup_sigh:
1690 if (atomic_dec_and_test(&new_sigh->count))
1691 kmem_cache_free(sighand_cachep, new_sigh);
1693 bad_unshare_cleanup_fs:
1695 free_fs_struct(new_fs);
1697 bad_unshare_cleanup_thread:
1703 * Helper to unshare the files of the current task.
1704 * We don't want to expose copy_files internals to
1705 * the exec layer of the kernel.
1708 int unshare_files(struct files_struct **displaced)
1710 struct task_struct *task = current;
1711 struct files_struct *copy = NULL;
1714 error = unshare_fd(CLONE_FILES, ©);
1715 if (error || !copy) {
1719 *displaced = task->files;