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 <linux/magic.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);
149 security_task_free(tsk);
151 put_group_info(tsk->group_info);
152 delayacct_tsk_free(tsk);
154 if (!profile_handoff_task(tsk))
159 * macro override instead of weak attribute alias, to workaround
160 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
162 #ifndef arch_task_cache_init
163 #define arch_task_cache_init()
166 void __init fork_init(unsigned long mempages)
168 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
169 #ifndef ARCH_MIN_TASKALIGN
170 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
172 /* create a slab on which task_structs can be allocated */
174 kmem_cache_create("task_struct", sizeof(struct task_struct),
175 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
178 /* do the arch specific task caches init */
179 arch_task_cache_init();
182 * The default maximum number of threads is set to a safe
183 * value: the thread structures can take up at most half
186 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
189 * we need to allow at least 20 threads to boot a system
194 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
195 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
196 init_task.signal->rlim[RLIMIT_SIGPENDING] =
197 init_task.signal->rlim[RLIMIT_NPROC];
200 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
201 struct task_struct *src)
207 static struct task_struct *dup_task_struct(struct task_struct *orig)
209 struct task_struct *tsk;
210 struct thread_info *ti;
211 unsigned long *stackend;
215 prepare_to_copy(orig);
217 tsk = alloc_task_struct();
221 ti = alloc_thread_info(tsk);
223 free_task_struct(tsk);
227 err = arch_dup_task_struct(tsk, orig);
233 err = prop_local_init_single(&tsk->dirties);
237 setup_thread_stack(tsk, orig);
238 stackend = end_of_stack(tsk);
239 *stackend = STACK_END_MAGIC; /* for overflow detection */
241 #ifdef CONFIG_CC_STACKPROTECTOR
242 tsk->stack_canary = get_random_int();
245 /* One for us, one for whoever does the "release_task()" (usually parent) */
246 atomic_set(&tsk->usage,2);
247 atomic_set(&tsk->fs_excl, 0);
248 #ifdef CONFIG_BLK_DEV_IO_TRACE
251 tsk->splice_pipe = NULL;
255 free_thread_info(ti);
256 free_task_struct(tsk);
261 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
263 struct vm_area_struct *mpnt, *tmp, **pprev;
264 struct rb_node **rb_link, *rb_parent;
266 unsigned long charge;
267 struct mempolicy *pol;
269 down_write(&oldmm->mmap_sem);
270 flush_cache_dup_mm(oldmm);
272 * Not linked in yet - no deadlock potential:
274 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
278 mm->mmap_cache = NULL;
279 mm->free_area_cache = oldmm->mmap_base;
280 mm->cached_hole_size = ~0UL;
282 cpus_clear(mm->cpu_vm_mask);
284 rb_link = &mm->mm_rb.rb_node;
288 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
291 if (mpnt->vm_flags & VM_DONTCOPY) {
292 long pages = vma_pages(mpnt);
293 mm->total_vm -= pages;
294 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
299 if (mpnt->vm_flags & VM_ACCOUNT) {
300 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
301 if (security_vm_enough_memory(len))
305 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
309 pol = mpol_dup(vma_policy(mpnt));
310 retval = PTR_ERR(pol);
312 goto fail_nomem_policy;
313 vma_set_policy(tmp, pol);
314 tmp->vm_flags &= ~VM_LOCKED;
320 struct inode *inode = file->f_path.dentry->d_inode;
322 if (tmp->vm_flags & VM_DENYWRITE)
323 atomic_dec(&inode->i_writecount);
325 /* insert tmp into the share list, just after mpnt */
326 spin_lock(&file->f_mapping->i_mmap_lock);
327 tmp->vm_truncate_count = mpnt->vm_truncate_count;
328 flush_dcache_mmap_lock(file->f_mapping);
329 vma_prio_tree_add(tmp, mpnt);
330 flush_dcache_mmap_unlock(file->f_mapping);
331 spin_unlock(&file->f_mapping->i_mmap_lock);
335 * Clear hugetlb-related page reserves for children. This only
336 * affects MAP_PRIVATE mappings. Faults generated by the child
337 * are not guaranteed to succeed, even if read-only
339 if (is_vm_hugetlb_page(tmp))
340 reset_vma_resv_huge_pages(tmp);
343 * Link in the new vma and copy the page table entries.
346 pprev = &tmp->vm_next;
348 __vma_link_rb(mm, tmp, rb_link, rb_parent);
349 rb_link = &tmp->vm_rb.rb_right;
350 rb_parent = &tmp->vm_rb;
353 retval = copy_page_range(mm, oldmm, mpnt);
355 if (tmp->vm_ops && tmp->vm_ops->open)
356 tmp->vm_ops->open(tmp);
361 /* a new mm has just been created */
362 arch_dup_mmap(oldmm, mm);
365 up_write(&mm->mmap_sem);
367 up_write(&oldmm->mmap_sem);
370 kmem_cache_free(vm_area_cachep, tmp);
373 vm_unacct_memory(charge);
377 static inline int mm_alloc_pgd(struct mm_struct * mm)
379 mm->pgd = pgd_alloc(mm);
380 if (unlikely(!mm->pgd))
385 static inline void mm_free_pgd(struct mm_struct * mm)
387 pgd_free(mm, mm->pgd);
390 #define dup_mmap(mm, oldmm) (0)
391 #define mm_alloc_pgd(mm) (0)
392 #define mm_free_pgd(mm)
393 #endif /* CONFIG_MMU */
395 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
397 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
398 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
400 #include <linux/init_task.h>
402 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
404 atomic_set(&mm->mm_users, 1);
405 atomic_set(&mm->mm_count, 1);
406 init_rwsem(&mm->mmap_sem);
407 INIT_LIST_HEAD(&mm->mmlist);
408 mm->flags = (current->mm) ? current->mm->flags
409 : MMF_DUMP_FILTER_DEFAULT;
410 mm->core_state = NULL;
412 set_mm_counter(mm, file_rss, 0);
413 set_mm_counter(mm, anon_rss, 0);
414 spin_lock_init(&mm->page_table_lock);
415 rwlock_init(&mm->ioctx_list_lock);
416 mm->ioctx_list = NULL;
417 mm->free_area_cache = TASK_UNMAPPED_BASE;
418 mm->cached_hole_size = ~0UL;
419 mm_init_owner(mm, p);
421 if (likely(!mm_alloc_pgd(mm))) {
423 mmu_notifier_mm_init(mm);
432 * Allocate and initialize an mm_struct.
434 struct mm_struct * mm_alloc(void)
436 struct mm_struct * mm;
440 memset(mm, 0, sizeof(*mm));
441 mm = mm_init(mm, current);
447 * Called when the last reference to the mm
448 * is dropped: either by a lazy thread or by
449 * mmput. Free the page directory and the mm.
451 void __mmdrop(struct mm_struct *mm)
453 BUG_ON(mm == &init_mm);
456 mmu_notifier_mm_destroy(mm);
459 EXPORT_SYMBOL_GPL(__mmdrop);
462 * Decrement the use count and release all resources for an mm.
464 void mmput(struct mm_struct *mm)
468 if (atomic_dec_and_test(&mm->mm_users)) {
471 set_mm_exe_file(mm, NULL);
472 if (!list_empty(&mm->mmlist)) {
473 spin_lock(&mmlist_lock);
474 list_del(&mm->mmlist);
475 spin_unlock(&mmlist_lock);
481 EXPORT_SYMBOL_GPL(mmput);
484 * get_task_mm - acquire a reference to the task's mm
486 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
487 * this kernel workthread has transiently adopted a user mm with use_mm,
488 * to do its AIO) is not set and if so returns a reference to it, after
489 * bumping up the use count. User must release the mm via mmput()
490 * after use. Typically used by /proc and ptrace.
492 struct mm_struct *get_task_mm(struct task_struct *task)
494 struct mm_struct *mm;
499 if (task->flags & PF_KTHREAD)
502 atomic_inc(&mm->mm_users);
507 EXPORT_SYMBOL_GPL(get_task_mm);
509 /* Please note the differences between mmput and mm_release.
510 * mmput is called whenever we stop holding onto a mm_struct,
511 * error success whatever.
513 * mm_release is called after a mm_struct has been removed
514 * from the current process.
516 * This difference is important for error handling, when we
517 * only half set up a mm_struct for a new process and need to restore
518 * the old one. Because we mmput the new mm_struct before
519 * restoring the old one. . .
520 * Eric Biederman 10 January 1998
522 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
524 struct completion *vfork_done = tsk->vfork_done;
526 /* Get rid of any cached register state */
527 deactivate_mm(tsk, mm);
529 /* notify parent sleeping on vfork() */
531 tsk->vfork_done = NULL;
532 complete(vfork_done);
536 * If we're exiting normally, clear a user-space tid field if
537 * requested. We leave this alone when dying by signal, to leave
538 * the value intact in a core dump, and to save the unnecessary
539 * trouble otherwise. Userland only wants this done for a sys_exit.
541 if (tsk->clear_child_tid
542 && !(tsk->flags & PF_SIGNALED)
543 && atomic_read(&mm->mm_users) > 1) {
544 u32 __user * tidptr = tsk->clear_child_tid;
545 tsk->clear_child_tid = NULL;
548 * We don't check the error code - if userspace has
549 * not set up a proper pointer then tough luck.
552 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
557 * Allocate a new mm structure and copy contents from the
558 * mm structure of the passed in task structure.
560 struct mm_struct *dup_mm(struct task_struct *tsk)
562 struct mm_struct *mm, *oldmm = current->mm;
572 memcpy(mm, oldmm, sizeof(*mm));
574 /* Initializing for Swap token stuff */
575 mm->token_priority = 0;
576 mm->last_interval = 0;
578 if (!mm_init(mm, tsk))
581 if (init_new_context(tsk, mm))
584 dup_mm_exe_file(oldmm, mm);
586 err = dup_mmap(mm, oldmm);
590 mm->hiwater_rss = get_mm_rss(mm);
591 mm->hiwater_vm = mm->total_vm;
603 * If init_new_context() failed, we cannot use mmput() to free the mm
604 * because it calls destroy_context()
611 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
613 struct mm_struct * mm, *oldmm;
616 tsk->min_flt = tsk->maj_flt = 0;
617 tsk->nvcsw = tsk->nivcsw = 0;
620 tsk->active_mm = NULL;
623 * Are we cloning a kernel thread?
625 * We need to steal a active VM for that..
631 if (clone_flags & CLONE_VM) {
632 atomic_inc(&oldmm->mm_users);
643 /* Initializing for Swap token stuff */
644 mm->token_priority = 0;
645 mm->last_interval = 0;
655 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
657 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
658 /* We don't need to lock fs - think why ;-) */
660 atomic_set(&fs->count, 1);
661 rwlock_init(&fs->lock);
662 fs->umask = old->umask;
663 read_lock(&old->lock);
664 fs->root = old->root;
665 path_get(&old->root);
668 read_unlock(&old->lock);
673 struct fs_struct *copy_fs_struct(struct fs_struct *old)
675 return __copy_fs_struct(old);
678 EXPORT_SYMBOL_GPL(copy_fs_struct);
680 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
682 if (clone_flags & CLONE_FS) {
683 atomic_inc(¤t->fs->count);
686 tsk->fs = __copy_fs_struct(current->fs);
692 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
694 struct files_struct *oldf, *newf;
698 * A background process may not have any files ...
700 oldf = current->files;
704 if (clone_flags & CLONE_FILES) {
705 atomic_inc(&oldf->count);
709 newf = dup_fd(oldf, &error);
719 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
722 struct io_context *ioc = current->io_context;
727 * Share io context with parent, if CLONE_IO is set
729 if (clone_flags & CLONE_IO) {
730 tsk->io_context = ioc_task_link(ioc);
731 if (unlikely(!tsk->io_context))
733 } else if (ioprio_valid(ioc->ioprio)) {
734 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
735 if (unlikely(!tsk->io_context))
738 tsk->io_context->ioprio = ioc->ioprio;
744 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
746 struct sighand_struct *sig;
748 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
749 atomic_inc(¤t->sighand->count);
752 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
753 rcu_assign_pointer(tsk->sighand, sig);
756 atomic_set(&sig->count, 1);
757 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
761 void __cleanup_sighand(struct sighand_struct *sighand)
763 if (atomic_dec_and_test(&sighand->count))
764 kmem_cache_free(sighand_cachep, sighand);
767 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
769 struct signal_struct *sig;
772 if (clone_flags & CLONE_THREAD) {
773 atomic_inc(¤t->signal->count);
774 atomic_inc(¤t->signal->live);
777 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
782 ret = copy_thread_group_keys(tsk);
784 kmem_cache_free(signal_cachep, sig);
788 atomic_set(&sig->count, 1);
789 atomic_set(&sig->live, 1);
790 init_waitqueue_head(&sig->wait_chldexit);
792 sig->group_exit_code = 0;
793 sig->group_exit_task = NULL;
794 sig->group_stop_count = 0;
795 sig->curr_target = tsk;
796 init_sigpending(&sig->shared_pending);
797 INIT_LIST_HEAD(&sig->posix_timers);
799 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
800 sig->it_real_incr.tv64 = 0;
801 sig->real_timer.function = it_real_fn;
803 sig->it_virt_expires = cputime_zero;
804 sig->it_virt_incr = cputime_zero;
805 sig->it_prof_expires = cputime_zero;
806 sig->it_prof_incr = cputime_zero;
808 sig->leader = 0; /* session leadership doesn't inherit */
809 sig->tty_old_pgrp = NULL;
812 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
813 sig->gtime = cputime_zero;
814 sig->cgtime = cputime_zero;
815 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
816 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
817 sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
818 task_io_accounting_init(&sig->ioac);
819 sig->sum_sched_runtime = 0;
820 INIT_LIST_HEAD(&sig->cpu_timers[0]);
821 INIT_LIST_HEAD(&sig->cpu_timers[1]);
822 INIT_LIST_HEAD(&sig->cpu_timers[2]);
823 taskstats_tgid_init(sig);
825 task_lock(current->group_leader);
826 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
827 task_unlock(current->group_leader);
829 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
831 * New sole thread in the process gets an expiry time
832 * of the whole CPU time limit.
834 tsk->it_prof_expires =
835 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
837 acct_init_pacct(&sig->pacct);
844 void __cleanup_signal(struct signal_struct *sig)
846 exit_thread_group_keys(sig);
847 tty_kref_put(sig->tty);
848 kmem_cache_free(signal_cachep, sig);
851 static void cleanup_signal(struct task_struct *tsk)
853 struct signal_struct *sig = tsk->signal;
855 atomic_dec(&sig->live);
857 if (atomic_dec_and_test(&sig->count))
858 __cleanup_signal(sig);
861 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
863 unsigned long new_flags = p->flags;
865 new_flags &= ~PF_SUPERPRIV;
866 new_flags |= PF_FORKNOEXEC;
867 new_flags |= PF_STARTING;
868 p->flags = new_flags;
869 clear_freeze_flag(p);
872 asmlinkage long sys_set_tid_address(int __user *tidptr)
874 current->clear_child_tid = tidptr;
876 return task_pid_vnr(current);
879 static void rt_mutex_init_task(struct task_struct *p)
881 spin_lock_init(&p->pi_lock);
882 #ifdef CONFIG_RT_MUTEXES
883 plist_head_init(&p->pi_waiters, &p->pi_lock);
884 p->pi_blocked_on = NULL;
888 #ifdef CONFIG_MM_OWNER
889 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
893 #endif /* CONFIG_MM_OWNER */
896 * This creates a new process as a copy of the old one,
897 * but does not actually start it yet.
899 * It copies the registers, and all the appropriate
900 * parts of the process environment (as per the clone
901 * flags). The actual kick-off is left to the caller.
903 static struct task_struct *copy_process(unsigned long clone_flags,
904 unsigned long stack_start,
905 struct pt_regs *regs,
906 unsigned long stack_size,
907 int __user *child_tidptr,
912 struct task_struct *p;
913 int cgroup_callbacks_done = 0;
915 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
916 return ERR_PTR(-EINVAL);
919 * Thread groups must share signals as well, and detached threads
920 * can only be started up within the thread group.
922 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
923 return ERR_PTR(-EINVAL);
926 * Shared signal handlers imply shared VM. By way of the above,
927 * thread groups also imply shared VM. Blocking this case allows
928 * for various simplifications in other code.
930 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
931 return ERR_PTR(-EINVAL);
933 retval = security_task_create(clone_flags);
938 p = dup_task_struct(current);
942 rt_mutex_init_task(p);
944 #ifdef CONFIG_PROVE_LOCKING
945 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
946 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
949 if (atomic_read(&p->user->processes) >=
950 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
951 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
952 p->user != current->nsproxy->user_ns->root_user)
956 atomic_inc(&p->user->__count);
957 atomic_inc(&p->user->processes);
958 get_group_info(p->group_info);
961 * If multiple threads are within copy_process(), then this check
962 * triggers too late. This doesn't hurt, the check is only there
963 * to stop root fork bombs.
965 if (nr_threads >= max_threads)
966 goto bad_fork_cleanup_count;
968 if (!try_module_get(task_thread_info(p)->exec_domain->module))
969 goto bad_fork_cleanup_count;
971 if (p->binfmt && !try_module_get(p->binfmt->module))
972 goto bad_fork_cleanup_put_domain;
975 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
976 copy_flags(clone_flags, p);
977 INIT_LIST_HEAD(&p->children);
978 INIT_LIST_HEAD(&p->sibling);
979 #ifdef CONFIG_PREEMPT_RCU
980 p->rcu_read_lock_nesting = 0;
981 p->rcu_flipctr_idx = 0;
982 #endif /* #ifdef CONFIG_PREEMPT_RCU */
983 p->vfork_done = NULL;
984 spin_lock_init(&p->alloc_lock);
986 clear_tsk_thread_flag(p, TIF_SIGPENDING);
987 init_sigpending(&p->pending);
989 p->utime = cputime_zero;
990 p->stime = cputime_zero;
991 p->gtime = cputime_zero;
992 p->utimescaled = cputime_zero;
993 p->stimescaled = cputime_zero;
994 p->prev_utime = cputime_zero;
995 p->prev_stime = cputime_zero;
997 #ifdef CONFIG_DETECT_SOFTLOCKUP
998 p->last_switch_count = 0;
999 p->last_switch_timestamp = 0;
1002 task_io_accounting_init(&p->ioac);
1003 acct_clear_integrals(p);
1005 p->it_virt_expires = cputime_zero;
1006 p->it_prof_expires = cputime_zero;
1007 p->it_sched_expires = 0;
1008 INIT_LIST_HEAD(&p->cpu_timers[0]);
1009 INIT_LIST_HEAD(&p->cpu_timers[1]);
1010 INIT_LIST_HEAD(&p->cpu_timers[2]);
1012 p->lock_depth = -1; /* -1 = no lock */
1013 do_posix_clock_monotonic_gettime(&p->start_time);
1014 p->real_start_time = p->start_time;
1015 monotonic_to_bootbased(&p->real_start_time);
1016 #ifdef CONFIG_SECURITY
1019 p->cap_bset = current->cap_bset;
1020 p->io_context = NULL;
1021 p->audit_context = NULL;
1024 p->mempolicy = mpol_dup(p->mempolicy);
1025 if (IS_ERR(p->mempolicy)) {
1026 retval = PTR_ERR(p->mempolicy);
1027 p->mempolicy = NULL;
1028 goto bad_fork_cleanup_cgroup;
1030 mpol_fix_fork_child_flag(p);
1032 #ifdef CONFIG_TRACE_IRQFLAGS
1034 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1035 p->hardirqs_enabled = 1;
1037 p->hardirqs_enabled = 0;
1039 p->hardirq_enable_ip = 0;
1040 p->hardirq_enable_event = 0;
1041 p->hardirq_disable_ip = _THIS_IP_;
1042 p->hardirq_disable_event = 0;
1043 p->softirqs_enabled = 1;
1044 p->softirq_enable_ip = _THIS_IP_;
1045 p->softirq_enable_event = 0;
1046 p->softirq_disable_ip = 0;
1047 p->softirq_disable_event = 0;
1048 p->hardirq_context = 0;
1049 p->softirq_context = 0;
1051 #ifdef CONFIG_LOCKDEP
1052 p->lockdep_depth = 0; /* no locks held yet */
1053 p->curr_chain_key = 0;
1054 p->lockdep_recursion = 0;
1057 #ifdef CONFIG_DEBUG_MUTEXES
1058 p->blocked_on = NULL; /* not blocked yet */
1061 /* Perform scheduler related setup. Assign this task to a CPU. */
1062 sched_fork(p, clone_flags);
1064 if ((retval = security_task_alloc(p)))
1065 goto bad_fork_cleanup_policy;
1066 if ((retval = audit_alloc(p)))
1067 goto bad_fork_cleanup_security;
1068 /* copy all the process information */
1069 if ((retval = copy_semundo(clone_flags, p)))
1070 goto bad_fork_cleanup_audit;
1071 if ((retval = copy_files(clone_flags, p)))
1072 goto bad_fork_cleanup_semundo;
1073 if ((retval = copy_fs(clone_flags, p)))
1074 goto bad_fork_cleanup_files;
1075 if ((retval = copy_sighand(clone_flags, p)))
1076 goto bad_fork_cleanup_fs;
1077 if ((retval = copy_signal(clone_flags, p)))
1078 goto bad_fork_cleanup_sighand;
1079 if ((retval = copy_mm(clone_flags, p)))
1080 goto bad_fork_cleanup_signal;
1081 if ((retval = copy_keys(clone_flags, p)))
1082 goto bad_fork_cleanup_mm;
1083 if ((retval = copy_namespaces(clone_flags, p)))
1084 goto bad_fork_cleanup_keys;
1085 if ((retval = copy_io(clone_flags, p)))
1086 goto bad_fork_cleanup_namespaces;
1087 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1089 goto bad_fork_cleanup_io;
1091 if (pid != &init_struct_pid) {
1093 pid = alloc_pid(task_active_pid_ns(p));
1095 goto bad_fork_cleanup_io;
1097 if (clone_flags & CLONE_NEWPID) {
1098 retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1100 goto bad_fork_free_pid;
1104 p->pid = pid_nr(pid);
1106 if (clone_flags & CLONE_THREAD)
1107 p->tgid = current->tgid;
1109 if (current->nsproxy != p->nsproxy) {
1110 retval = ns_cgroup_clone(p, pid);
1112 goto bad_fork_free_pid;
1115 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1117 * Clear TID on mm_release()?
1119 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1121 p->robust_list = NULL;
1122 #ifdef CONFIG_COMPAT
1123 p->compat_robust_list = NULL;
1125 INIT_LIST_HEAD(&p->pi_state_list);
1126 p->pi_state_cache = NULL;
1129 * sigaltstack should be cleared when sharing the same VM
1131 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1132 p->sas_ss_sp = p->sas_ss_size = 0;
1135 * Syscall tracing should be turned off in the child regardless
1138 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1139 #ifdef TIF_SYSCALL_EMU
1140 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1142 clear_all_latency_tracing(p);
1144 /* Our parent execution domain becomes current domain
1145 These must match for thread signalling to apply */
1146 p->parent_exec_id = p->self_exec_id;
1148 /* ok, now we should be set up.. */
1149 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1150 p->pdeath_signal = 0;
1154 * Ok, make it visible to the rest of the system.
1155 * We dont wake it up yet.
1157 p->group_leader = p;
1158 INIT_LIST_HEAD(&p->thread_group);
1160 /* Now that the task is set up, run cgroup callbacks if
1161 * necessary. We need to run them before the task is visible
1162 * on the tasklist. */
1163 cgroup_fork_callbacks(p);
1164 cgroup_callbacks_done = 1;
1166 /* Need tasklist lock for parent etc handling! */
1167 write_lock_irq(&tasklist_lock);
1170 * The task hasn't been attached yet, so its cpus_allowed mask will
1171 * not be changed, nor will its assigned CPU.
1173 * The cpus_allowed mask of the parent may have changed after it was
1174 * copied first time - so re-copy it here, then check the child's CPU
1175 * to ensure it is on a valid CPU (and if not, just force it back to
1176 * parent's CPU). This avoids alot of nasty races.
1178 p->cpus_allowed = current->cpus_allowed;
1179 p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1180 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1181 !cpu_online(task_cpu(p))))
1182 set_task_cpu(p, smp_processor_id());
1184 /* CLONE_PARENT re-uses the old parent */
1185 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1186 p->real_parent = current->real_parent;
1188 p->real_parent = current;
1190 spin_lock(¤t->sighand->siglock);
1193 * Process group and session signals need to be delivered to just the
1194 * parent before the fork or both the parent and the child after the
1195 * fork. Restart if a signal comes in before we add the new process to
1196 * it's process group.
1197 * A fatal signal pending means that current will exit, so the new
1198 * thread can't slip out of an OOM kill (or normal SIGKILL).
1200 recalc_sigpending();
1201 if (signal_pending(current)) {
1202 spin_unlock(¤t->sighand->siglock);
1203 write_unlock_irq(&tasklist_lock);
1204 retval = -ERESTARTNOINTR;
1205 goto bad_fork_free_pid;
1208 if (clone_flags & CLONE_THREAD) {
1209 p->group_leader = current->group_leader;
1210 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1212 if (!cputime_eq(current->signal->it_virt_expires,
1214 !cputime_eq(current->signal->it_prof_expires,
1216 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1217 !list_empty(¤t->signal->cpu_timers[0]) ||
1218 !list_empty(¤t->signal->cpu_timers[1]) ||
1219 !list_empty(¤t->signal->cpu_timers[2])) {
1221 * Have child wake up on its first tick to check
1222 * for process CPU timers.
1224 p->it_prof_expires = jiffies_to_cputime(1);
1228 if (likely(p->pid)) {
1229 list_add_tail(&p->sibling, &p->real_parent->children);
1230 tracehook_finish_clone(p, clone_flags, trace);
1232 if (thread_group_leader(p)) {
1233 if (clone_flags & CLONE_NEWPID)
1234 p->nsproxy->pid_ns->child_reaper = p;
1236 p->signal->leader_pid = pid;
1237 tty_kref_put(p->signal->tty);
1238 p->signal->tty = tty_kref_get(current->signal->tty);
1239 set_task_pgrp(p, task_pgrp_nr(current));
1240 set_task_session(p, task_session_nr(current));
1241 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1242 attach_pid(p, PIDTYPE_SID, task_session(current));
1243 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1244 __get_cpu_var(process_counts)++;
1246 attach_pid(p, PIDTYPE_PID, pid);
1251 spin_unlock(¤t->sighand->siglock);
1252 write_unlock_irq(&tasklist_lock);
1253 proc_fork_connector(p);
1254 cgroup_post_fork(p);
1258 if (pid != &init_struct_pid)
1260 bad_fork_cleanup_io:
1261 put_io_context(p->io_context);
1262 bad_fork_cleanup_namespaces:
1263 exit_task_namespaces(p);
1264 bad_fork_cleanup_keys:
1266 bad_fork_cleanup_mm:
1269 bad_fork_cleanup_signal:
1271 bad_fork_cleanup_sighand:
1272 __cleanup_sighand(p->sighand);
1273 bad_fork_cleanup_fs:
1274 exit_fs(p); /* blocking */
1275 bad_fork_cleanup_files:
1276 exit_files(p); /* blocking */
1277 bad_fork_cleanup_semundo:
1279 bad_fork_cleanup_audit:
1281 bad_fork_cleanup_security:
1282 security_task_free(p);
1283 bad_fork_cleanup_policy:
1285 mpol_put(p->mempolicy);
1286 bad_fork_cleanup_cgroup:
1288 cgroup_exit(p, cgroup_callbacks_done);
1289 delayacct_tsk_free(p);
1291 module_put(p->binfmt->module);
1292 bad_fork_cleanup_put_domain:
1293 module_put(task_thread_info(p)->exec_domain->module);
1294 bad_fork_cleanup_count:
1295 put_group_info(p->group_info);
1296 atomic_dec(&p->user->processes);
1301 return ERR_PTR(retval);
1304 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1306 memset(regs, 0, sizeof(struct pt_regs));
1310 struct task_struct * __cpuinit fork_idle(int cpu)
1312 struct task_struct *task;
1313 struct pt_regs regs;
1315 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL,
1316 &init_struct_pid, 0);
1318 init_idle(task, cpu);
1324 * Ok, this is the main fork-routine.
1326 * It copies the process, and if successful kick-starts
1327 * it and waits for it to finish using the VM if required.
1329 long do_fork(unsigned long clone_flags,
1330 unsigned long stack_start,
1331 struct pt_regs *regs,
1332 unsigned long stack_size,
1333 int __user *parent_tidptr,
1334 int __user *child_tidptr)
1336 struct task_struct *p;
1341 * We hope to recycle these flags after 2.6.26
1343 if (unlikely(clone_flags & CLONE_STOPPED)) {
1344 static int __read_mostly count = 100;
1346 if (count > 0 && printk_ratelimit()) {
1347 char comm[TASK_COMM_LEN];
1350 printk(KERN_INFO "fork(): process `%s' used deprecated "
1351 "clone flags 0x%lx\n",
1352 get_task_comm(comm, current),
1353 clone_flags & CLONE_STOPPED);
1358 * When called from kernel_thread, don't do user tracing stuff.
1360 if (likely(user_mode(regs)))
1361 trace = tracehook_prepare_clone(clone_flags);
1363 p = copy_process(clone_flags, stack_start, regs, stack_size,
1364 child_tidptr, NULL, trace);
1366 * Do this prior waking up the new thread - the thread pointer
1367 * might get invalid after that point, if the thread exits quickly.
1370 struct completion vfork;
1372 nr = task_pid_vnr(p);
1374 if (clone_flags & CLONE_PARENT_SETTID)
1375 put_user(nr, parent_tidptr);
1377 if (clone_flags & CLONE_VFORK) {
1378 p->vfork_done = &vfork;
1379 init_completion(&vfork);
1382 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1385 * We set PF_STARTING at creation in case tracing wants to
1386 * use this to distinguish a fully live task from one that
1387 * hasn't gotten to tracehook_report_clone() yet. Now we
1388 * clear it and set the child going.
1390 p->flags &= ~PF_STARTING;
1392 if (unlikely(clone_flags & CLONE_STOPPED)) {
1394 * We'll start up with an immediate SIGSTOP.
1396 sigaddset(&p->pending.signal, SIGSTOP);
1397 set_tsk_thread_flag(p, TIF_SIGPENDING);
1398 __set_task_state(p, TASK_STOPPED);
1400 wake_up_new_task(p, clone_flags);
1403 tracehook_report_clone_complete(trace, regs,
1404 clone_flags, nr, p);
1406 if (clone_flags & CLONE_VFORK) {
1407 freezer_do_not_count();
1408 wait_for_completion(&vfork);
1410 tracehook_report_vfork_done(p, nr);
1418 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1419 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1422 static void sighand_ctor(void *data)
1424 struct sighand_struct *sighand = data;
1426 spin_lock_init(&sighand->siglock);
1427 init_waitqueue_head(&sighand->signalfd_wqh);
1430 void __init proc_caches_init(void)
1432 sighand_cachep = kmem_cache_create("sighand_cache",
1433 sizeof(struct sighand_struct), 0,
1434 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1436 signal_cachep = kmem_cache_create("signal_cache",
1437 sizeof(struct signal_struct), 0,
1438 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1439 files_cachep = kmem_cache_create("files_cache",
1440 sizeof(struct files_struct), 0,
1441 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1442 fs_cachep = kmem_cache_create("fs_cache",
1443 sizeof(struct fs_struct), 0,
1444 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1445 vm_area_cachep = kmem_cache_create("vm_area_struct",
1446 sizeof(struct vm_area_struct), 0,
1448 mm_cachep = kmem_cache_create("mm_struct",
1449 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1450 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1454 * Check constraints on flags passed to the unshare system call and
1455 * force unsharing of additional process context as appropriate.
1457 static void check_unshare_flags(unsigned long *flags_ptr)
1460 * If unsharing a thread from a thread group, must also
1463 if (*flags_ptr & CLONE_THREAD)
1464 *flags_ptr |= CLONE_VM;
1467 * If unsharing vm, must also unshare signal handlers.
1469 if (*flags_ptr & CLONE_VM)
1470 *flags_ptr |= CLONE_SIGHAND;
1473 * If unsharing signal handlers and the task was created
1474 * using CLONE_THREAD, then must unshare the thread
1476 if ((*flags_ptr & CLONE_SIGHAND) &&
1477 (atomic_read(¤t->signal->count) > 1))
1478 *flags_ptr |= CLONE_THREAD;
1481 * If unsharing namespace, must also unshare filesystem information.
1483 if (*flags_ptr & CLONE_NEWNS)
1484 *flags_ptr |= CLONE_FS;
1488 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1490 static int unshare_thread(unsigned long unshare_flags)
1492 if (unshare_flags & CLONE_THREAD)
1499 * Unshare the filesystem structure if it is being shared
1501 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1503 struct fs_struct *fs = current->fs;
1505 if ((unshare_flags & CLONE_FS) &&
1506 (fs && atomic_read(&fs->count) > 1)) {
1507 *new_fsp = __copy_fs_struct(current->fs);
1516 * Unsharing of sighand is not supported yet
1518 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1520 struct sighand_struct *sigh = current->sighand;
1522 if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1529 * Unshare vm if it is being shared
1531 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1533 struct mm_struct *mm = current->mm;
1535 if ((unshare_flags & CLONE_VM) &&
1536 (mm && atomic_read(&mm->mm_users) > 1)) {
1544 * Unshare file descriptor table if it is being shared
1546 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1548 struct files_struct *fd = current->files;
1551 if ((unshare_flags & CLONE_FILES) &&
1552 (fd && atomic_read(&fd->count) > 1)) {
1553 *new_fdp = dup_fd(fd, &error);
1562 * unshare allows a process to 'unshare' part of the process
1563 * context which was originally shared using clone. copy_*
1564 * functions used by do_fork() cannot be used here directly
1565 * because they modify an inactive task_struct that is being
1566 * constructed. Here we are modifying the current, active,
1569 asmlinkage long sys_unshare(unsigned long unshare_flags)
1572 struct fs_struct *fs, *new_fs = NULL;
1573 struct sighand_struct *new_sigh = NULL;
1574 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1575 struct files_struct *fd, *new_fd = NULL;
1576 struct nsproxy *new_nsproxy = NULL;
1579 check_unshare_flags(&unshare_flags);
1581 /* Return -EINVAL for all unsupported flags */
1583 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1584 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1585 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1587 goto bad_unshare_out;
1590 * CLONE_NEWIPC must also detach from the undolist: after switching
1591 * to a new ipc namespace, the semaphore arrays from the old
1592 * namespace are unreachable.
1594 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1596 if ((err = unshare_thread(unshare_flags)))
1597 goto bad_unshare_out;
1598 if ((err = unshare_fs(unshare_flags, &new_fs)))
1599 goto bad_unshare_cleanup_thread;
1600 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1601 goto bad_unshare_cleanup_fs;
1602 if ((err = unshare_vm(unshare_flags, &new_mm)))
1603 goto bad_unshare_cleanup_sigh;
1604 if ((err = unshare_fd(unshare_flags, &new_fd)))
1605 goto bad_unshare_cleanup_vm;
1606 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1608 goto bad_unshare_cleanup_fd;
1610 if (new_fs || new_mm || new_fd || do_sysvsem || new_nsproxy) {
1613 * CLONE_SYSVSEM is equivalent to sys_exit().
1619 switch_task_namespaces(current, new_nsproxy);
1627 current->fs = new_fs;
1633 active_mm = current->active_mm;
1634 current->mm = new_mm;
1635 current->active_mm = new_mm;
1636 activate_mm(active_mm, new_mm);
1641 fd = current->files;
1642 current->files = new_fd;
1646 task_unlock(current);
1650 put_nsproxy(new_nsproxy);
1652 bad_unshare_cleanup_fd:
1654 put_files_struct(new_fd);
1656 bad_unshare_cleanup_vm:
1660 bad_unshare_cleanup_sigh:
1662 if (atomic_dec_and_test(&new_sigh->count))
1663 kmem_cache_free(sighand_cachep, new_sigh);
1665 bad_unshare_cleanup_fs:
1667 put_fs_struct(new_fs);
1669 bad_unshare_cleanup_thread:
1675 * Helper to unshare the files of the current task.
1676 * We don't want to expose copy_files internals to
1677 * the exec layer of the kernel.
1680 int unshare_files(struct files_struct **displaced)
1682 struct task_struct *task = current;
1683 struct files_struct *copy = NULL;
1686 error = unshare_fd(CLONE_FILES, ©);
1687 if (error || !copy) {
1691 *displaced = task->files;