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/mnt_namespace.h>
21 #include <linux/personality.h>
22 #include <linux/mempolicy.h>
23 #include <linux/sem.h>
24 #include <linux/file.h>
25 #include <linux/fdtable.h>
26 #include <linux/iocontext.h>
27 #include <linux/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.h>
30 #include <linux/mmu_notifier.h>
32 #include <linux/nsproxy.h>
33 #include <linux/capability.h>
34 #include <linux/cpu.h>
35 #include <linux/cgroup.h>
36 #include <linux/security.h>
37 #include <linux/hugetlb.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/tracehook.h>
42 #include <linux/futex.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/profile.h>
50 #include <linux/rmap.h>
51 #include <linux/acct.h>
52 #include <linux/tsacct_kern.h>
53 #include <linux/cn_proc.h>
54 #include <linux/freezer.h>
55 #include <linux/delayacct.h>
56 #include <linux/taskstats_kern.h>
57 #include <linux/random.h>
58 #include <linux/tty.h>
59 #include <linux/proc_fs.h>
60 #include <linux/blkdev.h>
61 #include <trace/sched.h>
63 #include <asm/pgtable.h>
64 #include <asm/pgalloc.h>
65 #include <asm/uaccess.h>
66 #include <asm/mmu_context.h>
67 #include <asm/cacheflush.h>
68 #include <asm/tlbflush.h>
71 * Protected counters by write_lock_irq(&tasklist_lock)
73 unsigned long total_forks; /* Handle normal Linux uptimes. */
74 int nr_threads; /* The idle threads do not count.. */
76 int max_threads; /* tunable limit on nr_threads */
78 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
80 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
82 int nr_processes(void)
87 for_each_online_cpu(cpu)
88 total += per_cpu(process_counts, cpu);
93 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
94 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
95 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
96 static struct kmem_cache *task_struct_cachep;
99 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
100 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
102 #ifdef CONFIG_DEBUG_STACK_USAGE
103 gfp_t mask = GFP_KERNEL | __GFP_ZERO;
105 gfp_t mask = GFP_KERNEL;
107 return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
110 static inline void free_thread_info(struct thread_info *ti)
112 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
116 /* SLAB cache for signal_struct structures (tsk->signal) */
117 static struct kmem_cache *signal_cachep;
119 /* SLAB cache for sighand_struct structures (tsk->sighand) */
120 struct kmem_cache *sighand_cachep;
122 /* SLAB cache for files_struct structures (tsk->files) */
123 struct kmem_cache *files_cachep;
125 /* SLAB cache for fs_struct structures (tsk->fs) */
126 struct kmem_cache *fs_cachep;
128 /* SLAB cache for vm_area_struct structures */
129 struct kmem_cache *vm_area_cachep;
131 /* SLAB cache for mm_struct structures (tsk->mm) */
132 static struct kmem_cache *mm_cachep;
134 void free_task(struct task_struct *tsk)
136 prop_local_destroy_single(&tsk->dirties);
137 free_thread_info(tsk->stack);
138 rt_mutex_debug_task_free(tsk);
139 free_task_struct(tsk);
141 EXPORT_SYMBOL(free_task);
143 void __put_task_struct(struct task_struct *tsk)
145 WARN_ON(!tsk->exit_state);
146 WARN_ON(atomic_read(&tsk->usage));
147 WARN_ON(tsk == current);
150 delayacct_tsk_free(tsk);
152 if (!profile_handoff_task(tsk))
157 * macro override instead of weak attribute alias, to workaround
158 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
160 #ifndef arch_task_cache_init
161 #define arch_task_cache_init()
164 void __init fork_init(unsigned long mempages)
166 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
167 #ifndef ARCH_MIN_TASKALIGN
168 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
170 /* create a slab on which task_structs can be allocated */
172 kmem_cache_create("task_struct", sizeof(struct task_struct),
173 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
176 /* do the arch specific task caches init */
177 arch_task_cache_init();
180 * The default maximum number of threads is set to a safe
181 * value: the thread structures can take up at most half
184 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
187 * we need to allow at least 20 threads to boot a system
192 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
193 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
194 init_task.signal->rlim[RLIMIT_SIGPENDING] =
195 init_task.signal->rlim[RLIMIT_NPROC];
198 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
199 struct task_struct *src)
205 static struct task_struct *dup_task_struct(struct task_struct *orig)
207 struct task_struct *tsk;
208 struct thread_info *ti;
211 prepare_to_copy(orig);
213 tsk = alloc_task_struct();
217 ti = alloc_thread_info(tsk);
219 free_task_struct(tsk);
223 err = arch_dup_task_struct(tsk, orig);
229 err = prop_local_init_single(&tsk->dirties);
233 setup_thread_stack(tsk, orig);
235 #ifdef CONFIG_CC_STACKPROTECTOR
236 tsk->stack_canary = get_random_int();
239 /* One for us, one for whoever does the "release_task()" (usually parent) */
240 atomic_set(&tsk->usage,2);
241 atomic_set(&tsk->fs_excl, 0);
242 #ifdef CONFIG_BLK_DEV_IO_TRACE
245 tsk->splice_pipe = NULL;
249 free_thread_info(ti);
250 free_task_struct(tsk);
255 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
257 struct vm_area_struct *mpnt, *tmp, **pprev;
258 struct rb_node **rb_link, *rb_parent;
260 unsigned long charge;
261 struct mempolicy *pol;
263 down_write(&oldmm->mmap_sem);
264 flush_cache_dup_mm(oldmm);
266 * Not linked in yet - no deadlock potential:
268 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
272 mm->mmap_cache = NULL;
273 mm->free_area_cache = oldmm->mmap_base;
274 mm->cached_hole_size = ~0UL;
276 cpus_clear(mm->cpu_vm_mask);
278 rb_link = &mm->mm_rb.rb_node;
282 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
285 if (mpnt->vm_flags & VM_DONTCOPY) {
286 long pages = vma_pages(mpnt);
287 mm->total_vm -= pages;
288 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
293 if (mpnt->vm_flags & VM_ACCOUNT) {
294 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
295 if (security_vm_enough_memory(len))
299 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
303 pol = mpol_dup(vma_policy(mpnt));
304 retval = PTR_ERR(pol);
306 goto fail_nomem_policy;
307 vma_set_policy(tmp, pol);
308 tmp->vm_flags &= ~VM_LOCKED;
314 struct inode *inode = file->f_path.dentry->d_inode;
316 if (tmp->vm_flags & VM_DENYWRITE)
317 atomic_dec(&inode->i_writecount);
319 /* insert tmp into the share list, just after mpnt */
320 spin_lock(&file->f_mapping->i_mmap_lock);
321 tmp->vm_truncate_count = mpnt->vm_truncate_count;
322 flush_dcache_mmap_lock(file->f_mapping);
323 vma_prio_tree_add(tmp, mpnt);
324 flush_dcache_mmap_unlock(file->f_mapping);
325 spin_unlock(&file->f_mapping->i_mmap_lock);
329 * Clear hugetlb-related page reserves for children. This only
330 * affects MAP_PRIVATE mappings. Faults generated by the child
331 * are not guaranteed to succeed, even if read-only
333 if (is_vm_hugetlb_page(tmp))
334 reset_vma_resv_huge_pages(tmp);
337 * Link in the new vma and copy the page table entries.
340 pprev = &tmp->vm_next;
342 __vma_link_rb(mm, tmp, rb_link, rb_parent);
343 rb_link = &tmp->vm_rb.rb_right;
344 rb_parent = &tmp->vm_rb;
347 retval = copy_page_range(mm, oldmm, mpnt);
349 if (tmp->vm_ops && tmp->vm_ops->open)
350 tmp->vm_ops->open(tmp);
355 /* a new mm has just been created */
356 arch_dup_mmap(oldmm, mm);
359 up_write(&mm->mmap_sem);
361 up_write(&oldmm->mmap_sem);
364 kmem_cache_free(vm_area_cachep, tmp);
367 vm_unacct_memory(charge);
371 static inline int mm_alloc_pgd(struct mm_struct * mm)
373 mm->pgd = pgd_alloc(mm);
374 if (unlikely(!mm->pgd))
379 static inline void mm_free_pgd(struct mm_struct * mm)
381 pgd_free(mm, mm->pgd);
384 #define dup_mmap(mm, oldmm) (0)
385 #define mm_alloc_pgd(mm) (0)
386 #define mm_free_pgd(mm)
387 #endif /* CONFIG_MMU */
389 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
391 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
392 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
394 #include <linux/init_task.h>
396 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
398 atomic_set(&mm->mm_users, 1);
399 atomic_set(&mm->mm_count, 1);
400 init_rwsem(&mm->mmap_sem);
401 INIT_LIST_HEAD(&mm->mmlist);
402 mm->flags = (current->mm) ? current->mm->flags
403 : MMF_DUMP_FILTER_DEFAULT;
404 mm->core_state = NULL;
406 set_mm_counter(mm, file_rss, 0);
407 set_mm_counter(mm, anon_rss, 0);
408 spin_lock_init(&mm->page_table_lock);
409 rwlock_init(&mm->ioctx_list_lock);
410 mm->ioctx_list = NULL;
411 mm->free_area_cache = TASK_UNMAPPED_BASE;
412 mm->cached_hole_size = ~0UL;
413 mm_init_owner(mm, p);
415 if (likely(!mm_alloc_pgd(mm))) {
417 mmu_notifier_mm_init(mm);
426 * Allocate and initialize an mm_struct.
428 struct mm_struct * mm_alloc(void)
430 struct mm_struct * mm;
434 memset(mm, 0, sizeof(*mm));
435 mm = mm_init(mm, current);
441 * Called when the last reference to the mm
442 * is dropped: either by a lazy thread or by
443 * mmput. Free the page directory and the mm.
445 void __mmdrop(struct mm_struct *mm)
447 BUG_ON(mm == &init_mm);
450 mmu_notifier_mm_destroy(mm);
453 EXPORT_SYMBOL_GPL(__mmdrop);
456 * Decrement the use count and release all resources for an mm.
458 void mmput(struct mm_struct *mm)
462 if (atomic_dec_and_test(&mm->mm_users)) {
465 set_mm_exe_file(mm, NULL);
466 if (!list_empty(&mm->mmlist)) {
467 spin_lock(&mmlist_lock);
468 list_del(&mm->mmlist);
469 spin_unlock(&mmlist_lock);
475 EXPORT_SYMBOL_GPL(mmput);
478 * get_task_mm - acquire a reference to the task's mm
480 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
481 * this kernel workthread has transiently adopted a user mm with use_mm,
482 * to do its AIO) is not set and if so returns a reference to it, after
483 * bumping up the use count. User must release the mm via mmput()
484 * after use. Typically used by /proc and ptrace.
486 struct mm_struct *get_task_mm(struct task_struct *task)
488 struct mm_struct *mm;
493 if (task->flags & PF_KTHREAD)
496 atomic_inc(&mm->mm_users);
501 EXPORT_SYMBOL_GPL(get_task_mm);
503 /* Please note the differences between mmput and mm_release.
504 * mmput is called whenever we stop holding onto a mm_struct,
505 * error success whatever.
507 * mm_release is called after a mm_struct has been removed
508 * from the current process.
510 * This difference is important for error handling, when we
511 * only half set up a mm_struct for a new process and need to restore
512 * the old one. Because we mmput the new mm_struct before
513 * restoring the old one. . .
514 * Eric Biederman 10 January 1998
516 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
518 struct completion *vfork_done = tsk->vfork_done;
520 /* Get rid of any cached register state */
521 deactivate_mm(tsk, mm);
523 /* notify parent sleeping on vfork() */
525 tsk->vfork_done = NULL;
526 complete(vfork_done);
530 * If we're exiting normally, clear a user-space tid field if
531 * requested. We leave this alone when dying by signal, to leave
532 * the value intact in a core dump, and to save the unnecessary
533 * trouble otherwise. Userland only wants this done for a sys_exit.
535 if (tsk->clear_child_tid
536 && !(tsk->flags & PF_SIGNALED)
537 && atomic_read(&mm->mm_users) > 1) {
538 u32 __user * tidptr = tsk->clear_child_tid;
539 tsk->clear_child_tid = NULL;
542 * We don't check the error code - if userspace has
543 * not set up a proper pointer then tough luck.
546 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
551 * Allocate a new mm structure and copy contents from the
552 * mm structure of the passed in task structure.
554 struct mm_struct *dup_mm(struct task_struct *tsk)
556 struct mm_struct *mm, *oldmm = current->mm;
566 memcpy(mm, oldmm, sizeof(*mm));
568 /* Initializing for Swap token stuff */
569 mm->token_priority = 0;
570 mm->last_interval = 0;
572 if (!mm_init(mm, tsk))
575 if (init_new_context(tsk, mm))
578 dup_mm_exe_file(oldmm, mm);
580 err = dup_mmap(mm, oldmm);
584 mm->hiwater_rss = get_mm_rss(mm);
585 mm->hiwater_vm = mm->total_vm;
597 * If init_new_context() failed, we cannot use mmput() to free the mm
598 * because it calls destroy_context()
605 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
607 struct mm_struct * mm, *oldmm;
610 tsk->min_flt = tsk->maj_flt = 0;
611 tsk->nvcsw = tsk->nivcsw = 0;
614 tsk->active_mm = NULL;
617 * Are we cloning a kernel thread?
619 * We need to steal a active VM for that..
625 if (clone_flags & CLONE_VM) {
626 atomic_inc(&oldmm->mm_users);
637 /* Initializing for Swap token stuff */
638 mm->token_priority = 0;
639 mm->last_interval = 0;
649 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
651 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
652 /* We don't need to lock fs - think why ;-) */
654 atomic_set(&fs->count, 1);
655 rwlock_init(&fs->lock);
656 fs->umask = old->umask;
657 read_lock(&old->lock);
658 fs->root = old->root;
659 path_get(&old->root);
662 read_unlock(&old->lock);
667 struct fs_struct *copy_fs_struct(struct fs_struct *old)
669 return __copy_fs_struct(old);
672 EXPORT_SYMBOL_GPL(copy_fs_struct);
674 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
676 if (clone_flags & CLONE_FS) {
677 atomic_inc(¤t->fs->count);
680 tsk->fs = __copy_fs_struct(current->fs);
686 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
688 struct files_struct *oldf, *newf;
692 * A background process may not have any files ...
694 oldf = current->files;
698 if (clone_flags & CLONE_FILES) {
699 atomic_inc(&oldf->count);
703 newf = dup_fd(oldf, &error);
713 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
716 struct io_context *ioc = current->io_context;
721 * Share io context with parent, if CLONE_IO is set
723 if (clone_flags & CLONE_IO) {
724 tsk->io_context = ioc_task_link(ioc);
725 if (unlikely(!tsk->io_context))
727 } else if (ioprio_valid(ioc->ioprio)) {
728 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
729 if (unlikely(!tsk->io_context))
732 tsk->io_context->ioprio = ioc->ioprio;
738 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
740 struct sighand_struct *sig;
742 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
743 atomic_inc(¤t->sighand->count);
746 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
747 rcu_assign_pointer(tsk->sighand, sig);
750 atomic_set(&sig->count, 1);
751 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
755 void __cleanup_sighand(struct sighand_struct *sighand)
757 if (atomic_dec_and_test(&sighand->count))
758 kmem_cache_free(sighand_cachep, sighand);
763 * Initialize POSIX timer handling for a thread group.
765 static void posix_cpu_timers_init_group(struct signal_struct *sig)
767 /* Thread group counters. */
768 thread_group_cputime_init(sig);
770 /* Expiration times and increments. */
771 sig->it_virt_expires = cputime_zero;
772 sig->it_virt_incr = cputime_zero;
773 sig->it_prof_expires = cputime_zero;
774 sig->it_prof_incr = cputime_zero;
776 /* Cached expiration times. */
777 sig->cputime_expires.prof_exp = cputime_zero;
778 sig->cputime_expires.virt_exp = cputime_zero;
779 sig->cputime_expires.sched_exp = 0;
781 /* The timer lists. */
782 INIT_LIST_HEAD(&sig->cpu_timers[0]);
783 INIT_LIST_HEAD(&sig->cpu_timers[1]);
784 INIT_LIST_HEAD(&sig->cpu_timers[2]);
787 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
789 struct signal_struct *sig;
792 if (clone_flags & CLONE_THREAD) {
793 ret = thread_group_cputime_clone_thread(current);
795 atomic_inc(¤t->signal->count);
796 atomic_inc(¤t->signal->live);
800 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
805 ret = copy_thread_group_keys(tsk);
807 kmem_cache_free(signal_cachep, sig);
811 atomic_set(&sig->count, 1);
812 atomic_set(&sig->live, 1);
813 init_waitqueue_head(&sig->wait_chldexit);
815 sig->group_exit_code = 0;
816 sig->group_exit_task = NULL;
817 sig->group_stop_count = 0;
818 sig->curr_target = tsk;
819 init_sigpending(&sig->shared_pending);
820 INIT_LIST_HEAD(&sig->posix_timers);
822 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
823 sig->it_real_incr.tv64 = 0;
824 sig->real_timer.function = it_real_fn;
826 sig->leader = 0; /* session leadership doesn't inherit */
827 sig->tty_old_pgrp = NULL;
830 sig->cutime = sig->cstime = cputime_zero;
831 sig->gtime = cputime_zero;
832 sig->cgtime = cputime_zero;
833 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
834 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
835 sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
836 task_io_accounting_init(&sig->ioac);
837 taskstats_tgid_init(sig);
839 task_lock(current->group_leader);
840 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
841 task_unlock(current->group_leader);
843 posix_cpu_timers_init_group(sig);
845 acct_init_pacct(&sig->pacct);
852 void __cleanup_signal(struct signal_struct *sig)
854 thread_group_cputime_free(sig);
855 exit_thread_group_keys(sig);
856 tty_kref_put(sig->tty);
857 kmem_cache_free(signal_cachep, sig);
860 static void cleanup_signal(struct task_struct *tsk)
862 struct signal_struct *sig = tsk->signal;
864 atomic_dec(&sig->live);
866 if (atomic_dec_and_test(&sig->count))
867 __cleanup_signal(sig);
870 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
872 unsigned long new_flags = p->flags;
874 new_flags &= ~PF_SUPERPRIV;
875 new_flags |= PF_FORKNOEXEC;
876 new_flags |= PF_STARTING;
877 p->flags = new_flags;
878 clear_freeze_flag(p);
881 asmlinkage long sys_set_tid_address(int __user *tidptr)
883 current->clear_child_tid = tidptr;
885 return task_pid_vnr(current);
888 static void rt_mutex_init_task(struct task_struct *p)
890 spin_lock_init(&p->pi_lock);
891 #ifdef CONFIG_RT_MUTEXES
892 plist_head_init(&p->pi_waiters, &p->pi_lock);
893 p->pi_blocked_on = NULL;
897 #ifdef CONFIG_MM_OWNER
898 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
902 #endif /* CONFIG_MM_OWNER */
905 * Initialize POSIX timer handling for a single task.
907 static void posix_cpu_timers_init(struct task_struct *tsk)
909 tsk->cputime_expires.prof_exp = cputime_zero;
910 tsk->cputime_expires.virt_exp = cputime_zero;
911 tsk->cputime_expires.sched_exp = 0;
912 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
913 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
914 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
918 * This creates a new process as a copy of the old one,
919 * but does not actually start it yet.
921 * It copies the registers, and all the appropriate
922 * parts of the process environment (as per the clone
923 * flags). The actual kick-off is left to the caller.
925 static struct task_struct *copy_process(unsigned long clone_flags,
926 unsigned long stack_start,
927 struct pt_regs *regs,
928 unsigned long stack_size,
929 int __user *child_tidptr,
934 struct task_struct *p;
935 int cgroup_callbacks_done = 0;
937 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
938 return ERR_PTR(-EINVAL);
941 * Thread groups must share signals as well, and detached threads
942 * can only be started up within the thread group.
944 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
945 return ERR_PTR(-EINVAL);
948 * Shared signal handlers imply shared VM. By way of the above,
949 * thread groups also imply shared VM. Blocking this case allows
950 * for various simplifications in other code.
952 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
953 return ERR_PTR(-EINVAL);
955 retval = security_task_create(clone_flags);
960 p = dup_task_struct(current);
964 rt_mutex_init_task(p);
966 #ifdef CONFIG_PROVE_LOCKING
967 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
968 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
971 if (atomic_read(&p->cred->user->processes) >=
972 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
973 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
974 p->cred->user != current->nsproxy->user_ns->root_user)
978 retval = copy_creds(p, clone_flags);
983 * If multiple threads are within copy_process(), then this check
984 * triggers too late. This doesn't hurt, the check is only there
985 * to stop root fork bombs.
987 if (nr_threads >= max_threads)
988 goto bad_fork_cleanup_count;
990 if (!try_module_get(task_thread_info(p)->exec_domain->module))
991 goto bad_fork_cleanup_count;
993 if (p->binfmt && !try_module_get(p->binfmt->module))
994 goto bad_fork_cleanup_put_domain;
997 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
998 copy_flags(clone_flags, p);
999 INIT_LIST_HEAD(&p->children);
1000 INIT_LIST_HEAD(&p->sibling);
1001 #ifdef CONFIG_PREEMPT_RCU
1002 p->rcu_read_lock_nesting = 0;
1003 p->rcu_flipctr_idx = 0;
1004 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1005 p->vfork_done = NULL;
1006 spin_lock_init(&p->alloc_lock);
1008 clear_tsk_thread_flag(p, TIF_SIGPENDING);
1009 init_sigpending(&p->pending);
1011 p->utime = cputime_zero;
1012 p->stime = cputime_zero;
1013 p->gtime = cputime_zero;
1014 p->utimescaled = cputime_zero;
1015 p->stimescaled = cputime_zero;
1016 p->prev_utime = cputime_zero;
1017 p->prev_stime = cputime_zero;
1019 p->default_timer_slack_ns = current->timer_slack_ns;
1021 #ifdef CONFIG_DETECT_SOFTLOCKUP
1022 p->last_switch_count = 0;
1023 p->last_switch_timestamp = 0;
1026 task_io_accounting_init(&p->ioac);
1027 acct_clear_integrals(p);
1029 posix_cpu_timers_init(p);
1031 p->lock_depth = -1; /* -1 = no lock */
1032 do_posix_clock_monotonic_gettime(&p->start_time);
1033 p->real_start_time = p->start_time;
1034 monotonic_to_bootbased(&p->real_start_time);
1035 p->io_context = NULL;
1036 p->audit_context = NULL;
1039 p->mempolicy = mpol_dup(p->mempolicy);
1040 if (IS_ERR(p->mempolicy)) {
1041 retval = PTR_ERR(p->mempolicy);
1042 p->mempolicy = NULL;
1043 goto bad_fork_cleanup_cgroup;
1045 mpol_fix_fork_child_flag(p);
1047 #ifdef CONFIG_TRACE_IRQFLAGS
1049 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1050 p->hardirqs_enabled = 1;
1052 p->hardirqs_enabled = 0;
1054 p->hardirq_enable_ip = 0;
1055 p->hardirq_enable_event = 0;
1056 p->hardirq_disable_ip = _THIS_IP_;
1057 p->hardirq_disable_event = 0;
1058 p->softirqs_enabled = 1;
1059 p->softirq_enable_ip = _THIS_IP_;
1060 p->softirq_enable_event = 0;
1061 p->softirq_disable_ip = 0;
1062 p->softirq_disable_event = 0;
1063 p->hardirq_context = 0;
1064 p->softirq_context = 0;
1066 #ifdef CONFIG_LOCKDEP
1067 p->lockdep_depth = 0; /* no locks held yet */
1068 p->curr_chain_key = 0;
1069 p->lockdep_recursion = 0;
1072 #ifdef CONFIG_DEBUG_MUTEXES
1073 p->blocked_on = NULL; /* not blocked yet */
1076 /* Perform scheduler related setup. Assign this task to a CPU. */
1077 sched_fork(p, clone_flags);
1079 if ((retval = audit_alloc(p)))
1080 goto bad_fork_cleanup_policy;
1081 /* copy all the process information */
1082 if ((retval = copy_semundo(clone_flags, p)))
1083 goto bad_fork_cleanup_audit;
1084 if ((retval = copy_files(clone_flags, p)))
1085 goto bad_fork_cleanup_semundo;
1086 if ((retval = copy_fs(clone_flags, p)))
1087 goto bad_fork_cleanup_files;
1088 if ((retval = copy_sighand(clone_flags, p)))
1089 goto bad_fork_cleanup_fs;
1090 if ((retval = copy_signal(clone_flags, p)))
1091 goto bad_fork_cleanup_sighand;
1092 if ((retval = copy_mm(clone_flags, p)))
1093 goto bad_fork_cleanup_signal;
1094 if ((retval = copy_keys(clone_flags, p)))
1095 goto bad_fork_cleanup_mm;
1096 if ((retval = copy_namespaces(clone_flags, p)))
1097 goto bad_fork_cleanup_keys;
1098 if ((retval = copy_io(clone_flags, p)))
1099 goto bad_fork_cleanup_namespaces;
1100 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1102 goto bad_fork_cleanup_io;
1104 if (pid != &init_struct_pid) {
1106 pid = alloc_pid(task_active_pid_ns(p));
1108 goto bad_fork_cleanup_io;
1110 if (clone_flags & CLONE_NEWPID) {
1111 retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1113 goto bad_fork_free_pid;
1117 p->pid = pid_nr(pid);
1119 if (clone_flags & CLONE_THREAD)
1120 p->tgid = current->tgid;
1122 if (current->nsproxy != p->nsproxy) {
1123 retval = ns_cgroup_clone(p, pid);
1125 goto bad_fork_free_pid;
1128 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1130 * Clear TID on mm_release()?
1132 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1134 p->robust_list = NULL;
1135 #ifdef CONFIG_COMPAT
1136 p->compat_robust_list = NULL;
1138 INIT_LIST_HEAD(&p->pi_state_list);
1139 p->pi_state_cache = NULL;
1142 * sigaltstack should be cleared when sharing the same VM
1144 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1145 p->sas_ss_sp = p->sas_ss_size = 0;
1148 * Syscall tracing should be turned off in the child regardless
1151 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1152 #ifdef TIF_SYSCALL_EMU
1153 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1155 clear_all_latency_tracing(p);
1157 /* Our parent execution domain becomes current domain
1158 These must match for thread signalling to apply */
1159 p->parent_exec_id = p->self_exec_id;
1161 /* ok, now we should be set up.. */
1162 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1163 p->pdeath_signal = 0;
1167 * Ok, make it visible to the rest of the system.
1168 * We dont wake it up yet.
1170 p->group_leader = p;
1171 INIT_LIST_HEAD(&p->thread_group);
1173 /* Now that the task is set up, run cgroup callbacks if
1174 * necessary. We need to run them before the task is visible
1175 * on the tasklist. */
1176 cgroup_fork_callbacks(p);
1177 cgroup_callbacks_done = 1;
1179 /* Need tasklist lock for parent etc handling! */
1180 write_lock_irq(&tasklist_lock);
1183 * The task hasn't been attached yet, so its cpus_allowed mask will
1184 * not be changed, nor will its assigned CPU.
1186 * The cpus_allowed mask of the parent may have changed after it was
1187 * copied first time - so re-copy it here, then check the child's CPU
1188 * to ensure it is on a valid CPU (and if not, just force it back to
1189 * parent's CPU). This avoids alot of nasty races.
1191 p->cpus_allowed = current->cpus_allowed;
1192 p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1193 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1194 !cpu_online(task_cpu(p))))
1195 set_task_cpu(p, smp_processor_id());
1197 /* CLONE_PARENT re-uses the old parent */
1198 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1199 p->real_parent = current->real_parent;
1201 p->real_parent = current;
1203 spin_lock(¤t->sighand->siglock);
1206 * Process group and session signals need to be delivered to just the
1207 * parent before the fork or both the parent and the child after the
1208 * fork. Restart if a signal comes in before we add the new process to
1209 * it's process group.
1210 * A fatal signal pending means that current will exit, so the new
1211 * thread can't slip out of an OOM kill (or normal SIGKILL).
1213 recalc_sigpending();
1214 if (signal_pending(current)) {
1215 spin_unlock(¤t->sighand->siglock);
1216 write_unlock_irq(&tasklist_lock);
1217 retval = -ERESTARTNOINTR;
1218 goto bad_fork_free_pid;
1221 if (clone_flags & CLONE_THREAD) {
1222 p->group_leader = current->group_leader;
1223 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1226 if (likely(p->pid)) {
1227 list_add_tail(&p->sibling, &p->real_parent->children);
1228 tracehook_finish_clone(p, clone_flags, trace);
1230 if (thread_group_leader(p)) {
1231 if (clone_flags & CLONE_NEWPID)
1232 p->nsproxy->pid_ns->child_reaper = p;
1234 p->signal->leader_pid = pid;
1235 tty_kref_put(p->signal->tty);
1236 p->signal->tty = tty_kref_get(current->signal->tty);
1237 set_task_pgrp(p, task_pgrp_nr(current));
1238 set_task_session(p, task_session_nr(current));
1239 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1240 attach_pid(p, PIDTYPE_SID, task_session(current));
1241 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1242 __get_cpu_var(process_counts)++;
1244 attach_pid(p, PIDTYPE_PID, pid);
1249 spin_unlock(¤t->sighand->siglock);
1250 write_unlock_irq(&tasklist_lock);
1251 proc_fork_connector(p);
1252 cgroup_post_fork(p);
1256 if (pid != &init_struct_pid)
1258 bad_fork_cleanup_io:
1259 put_io_context(p->io_context);
1260 bad_fork_cleanup_namespaces:
1261 exit_task_namespaces(p);
1262 bad_fork_cleanup_keys:
1264 bad_fork_cleanup_mm:
1267 bad_fork_cleanup_signal:
1269 bad_fork_cleanup_sighand:
1270 __cleanup_sighand(p->sighand);
1271 bad_fork_cleanup_fs:
1272 exit_fs(p); /* blocking */
1273 bad_fork_cleanup_files:
1274 exit_files(p); /* blocking */
1275 bad_fork_cleanup_semundo:
1277 bad_fork_cleanup_audit:
1279 bad_fork_cleanup_policy:
1281 mpol_put(p->mempolicy);
1282 bad_fork_cleanup_cgroup:
1284 cgroup_exit(p, cgroup_callbacks_done);
1285 delayacct_tsk_free(p);
1287 module_put(p->binfmt->module);
1288 bad_fork_cleanup_put_domain:
1289 module_put(task_thread_info(p)->exec_domain->module);
1290 bad_fork_cleanup_count:
1295 return ERR_PTR(retval);
1298 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1300 memset(regs, 0, sizeof(struct pt_regs));
1304 struct task_struct * __cpuinit fork_idle(int cpu)
1306 struct task_struct *task;
1307 struct pt_regs regs;
1309 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL,
1310 &init_struct_pid, 0);
1312 init_idle(task, cpu);
1318 * Ok, this is the main fork-routine.
1320 * It copies the process, and if successful kick-starts
1321 * it and waits for it to finish using the VM if required.
1323 long do_fork(unsigned long clone_flags,
1324 unsigned long stack_start,
1325 struct pt_regs *regs,
1326 unsigned long stack_size,
1327 int __user *parent_tidptr,
1328 int __user *child_tidptr)
1330 struct task_struct *p;
1335 * We hope to recycle these flags after 2.6.26
1337 if (unlikely(clone_flags & CLONE_STOPPED)) {
1338 static int __read_mostly count = 100;
1340 if (count > 0 && printk_ratelimit()) {
1341 char comm[TASK_COMM_LEN];
1344 printk(KERN_INFO "fork(): process `%s' used deprecated "
1345 "clone flags 0x%lx\n",
1346 get_task_comm(comm, current),
1347 clone_flags & CLONE_STOPPED);
1352 * When called from kernel_thread, don't do user tracing stuff.
1354 if (likely(user_mode(regs)))
1355 trace = tracehook_prepare_clone(clone_flags);
1357 p = copy_process(clone_flags, stack_start, regs, stack_size,
1358 child_tidptr, NULL, trace);
1360 * Do this prior waking up the new thread - the thread pointer
1361 * might get invalid after that point, if the thread exits quickly.
1364 struct completion vfork;
1366 trace_sched_process_fork(current, p);
1368 nr = task_pid_vnr(p);
1370 if (clone_flags & CLONE_PARENT_SETTID)
1371 put_user(nr, parent_tidptr);
1373 if (clone_flags & CLONE_VFORK) {
1374 p->vfork_done = &vfork;
1375 init_completion(&vfork);
1378 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1381 * We set PF_STARTING at creation in case tracing wants to
1382 * use this to distinguish a fully live task from one that
1383 * hasn't gotten to tracehook_report_clone() yet. Now we
1384 * clear it and set the child going.
1386 p->flags &= ~PF_STARTING;
1388 if (unlikely(clone_flags & CLONE_STOPPED)) {
1390 * We'll start up with an immediate SIGSTOP.
1392 sigaddset(&p->pending.signal, SIGSTOP);
1393 set_tsk_thread_flag(p, TIF_SIGPENDING);
1394 __set_task_state(p, TASK_STOPPED);
1396 wake_up_new_task(p, clone_flags);
1399 tracehook_report_clone_complete(trace, regs,
1400 clone_flags, nr, p);
1402 if (clone_flags & CLONE_VFORK) {
1403 freezer_do_not_count();
1404 wait_for_completion(&vfork);
1406 tracehook_report_vfork_done(p, nr);
1414 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1415 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1418 static void sighand_ctor(void *data)
1420 struct sighand_struct *sighand = data;
1422 spin_lock_init(&sighand->siglock);
1423 init_waitqueue_head(&sighand->signalfd_wqh);
1426 void __init proc_caches_init(void)
1428 sighand_cachep = kmem_cache_create("sighand_cache",
1429 sizeof(struct sighand_struct), 0,
1430 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1432 signal_cachep = kmem_cache_create("signal_cache",
1433 sizeof(struct signal_struct), 0,
1434 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1435 files_cachep = kmem_cache_create("files_cache",
1436 sizeof(struct files_struct), 0,
1437 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1438 fs_cachep = kmem_cache_create("fs_cache",
1439 sizeof(struct fs_struct), 0,
1440 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1441 vm_area_cachep = kmem_cache_create("vm_area_struct",
1442 sizeof(struct vm_area_struct), 0,
1444 mm_cachep = kmem_cache_create("mm_struct",
1445 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1446 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1450 * Check constraints on flags passed to the unshare system call and
1451 * force unsharing of additional process context as appropriate.
1453 static void check_unshare_flags(unsigned long *flags_ptr)
1456 * If unsharing a thread from a thread group, must also
1459 if (*flags_ptr & CLONE_THREAD)
1460 *flags_ptr |= CLONE_VM;
1463 * If unsharing vm, must also unshare signal handlers.
1465 if (*flags_ptr & CLONE_VM)
1466 *flags_ptr |= CLONE_SIGHAND;
1469 * If unsharing signal handlers and the task was created
1470 * using CLONE_THREAD, then must unshare the thread
1472 if ((*flags_ptr & CLONE_SIGHAND) &&
1473 (atomic_read(¤t->signal->count) > 1))
1474 *flags_ptr |= CLONE_THREAD;
1477 * If unsharing namespace, must also unshare filesystem information.
1479 if (*flags_ptr & CLONE_NEWNS)
1480 *flags_ptr |= CLONE_FS;
1484 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1486 static int unshare_thread(unsigned long unshare_flags)
1488 if (unshare_flags & CLONE_THREAD)
1495 * Unshare the filesystem structure if it is being shared
1497 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1499 struct fs_struct *fs = current->fs;
1501 if ((unshare_flags & CLONE_FS) &&
1502 (fs && atomic_read(&fs->count) > 1)) {
1503 *new_fsp = __copy_fs_struct(current->fs);
1512 * Unsharing of sighand is not supported yet
1514 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1516 struct sighand_struct *sigh = current->sighand;
1518 if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1525 * Unshare vm if it is being shared
1527 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1529 struct mm_struct *mm = current->mm;
1531 if ((unshare_flags & CLONE_VM) &&
1532 (mm && atomic_read(&mm->mm_users) > 1)) {
1540 * Unshare file descriptor table if it is being shared
1542 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1544 struct files_struct *fd = current->files;
1547 if ((unshare_flags & CLONE_FILES) &&
1548 (fd && atomic_read(&fd->count) > 1)) {
1549 *new_fdp = dup_fd(fd, &error);
1558 * unshare allows a process to 'unshare' part of the process
1559 * context which was originally shared using clone. copy_*
1560 * functions used by do_fork() cannot be used here directly
1561 * because they modify an inactive task_struct that is being
1562 * constructed. Here we are modifying the current, active,
1565 asmlinkage long sys_unshare(unsigned long unshare_flags)
1568 struct fs_struct *fs, *new_fs = NULL;
1569 struct sighand_struct *new_sigh = NULL;
1570 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1571 struct files_struct *fd, *new_fd = NULL;
1572 struct nsproxy *new_nsproxy = NULL;
1575 check_unshare_flags(&unshare_flags);
1577 /* Return -EINVAL for all unsupported flags */
1579 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1580 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1581 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1583 goto bad_unshare_out;
1586 * CLONE_NEWIPC must also detach from the undolist: after switching
1587 * to a new ipc namespace, the semaphore arrays from the old
1588 * namespace are unreachable.
1590 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1592 if ((err = unshare_thread(unshare_flags)))
1593 goto bad_unshare_out;
1594 if ((err = unshare_fs(unshare_flags, &new_fs)))
1595 goto bad_unshare_cleanup_thread;
1596 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1597 goto bad_unshare_cleanup_fs;
1598 if ((err = unshare_vm(unshare_flags, &new_mm)))
1599 goto bad_unshare_cleanup_sigh;
1600 if ((err = unshare_fd(unshare_flags, &new_fd)))
1601 goto bad_unshare_cleanup_vm;
1602 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1604 goto bad_unshare_cleanup_fd;
1606 if (new_fs || new_mm || new_fd || do_sysvsem || new_nsproxy) {
1609 * CLONE_SYSVSEM is equivalent to sys_exit().
1615 switch_task_namespaces(current, new_nsproxy);
1623 current->fs = new_fs;
1629 active_mm = current->active_mm;
1630 current->mm = new_mm;
1631 current->active_mm = new_mm;
1632 activate_mm(active_mm, new_mm);
1637 fd = current->files;
1638 current->files = new_fd;
1642 task_unlock(current);
1646 put_nsproxy(new_nsproxy);
1648 bad_unshare_cleanup_fd:
1650 put_files_struct(new_fd);
1652 bad_unshare_cleanup_vm:
1656 bad_unshare_cleanup_sigh:
1658 if (atomic_dec_and_test(&new_sigh->count))
1659 kmem_cache_free(sighand_cachep, new_sigh);
1661 bad_unshare_cleanup_fs:
1663 put_fs_struct(new_fs);
1665 bad_unshare_cleanup_thread:
1671 * Helper to unshare the files of the current task.
1672 * We don't want to expose copy_files internals to
1673 * the exec layer of the kernel.
1676 int unshare_files(struct files_struct **displaced)
1678 struct task_struct *task = current;
1679 struct files_struct *copy = NULL;
1682 error = unshare_fd(CLONE_FILES, ©);
1683 if (error || !copy) {
1687 *displaced = task->files;