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 atomic_set(&sig->count, 1);
806 atomic_set(&sig->live, 1);
807 init_waitqueue_head(&sig->wait_chldexit);
809 sig->group_exit_code = 0;
810 sig->group_exit_task = NULL;
811 sig->group_stop_count = 0;
812 sig->curr_target = tsk;
813 init_sigpending(&sig->shared_pending);
814 INIT_LIST_HEAD(&sig->posix_timers);
816 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
817 sig->it_real_incr.tv64 = 0;
818 sig->real_timer.function = it_real_fn;
820 sig->leader = 0; /* session leadership doesn't inherit */
821 sig->tty_old_pgrp = NULL;
824 sig->cutime = sig->cstime = cputime_zero;
825 sig->gtime = cputime_zero;
826 sig->cgtime = cputime_zero;
827 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
828 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
829 sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
830 task_io_accounting_init(&sig->ioac);
831 taskstats_tgid_init(sig);
833 task_lock(current->group_leader);
834 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
835 task_unlock(current->group_leader);
837 posix_cpu_timers_init_group(sig);
839 acct_init_pacct(&sig->pacct);
846 void __cleanup_signal(struct signal_struct *sig)
848 thread_group_cputime_free(sig);
849 tty_kref_put(sig->tty);
850 kmem_cache_free(signal_cachep, sig);
853 static void cleanup_signal(struct task_struct *tsk)
855 struct signal_struct *sig = tsk->signal;
857 atomic_dec(&sig->live);
859 if (atomic_dec_and_test(&sig->count))
860 __cleanup_signal(sig);
863 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
865 unsigned long new_flags = p->flags;
867 new_flags &= ~PF_SUPERPRIV;
868 new_flags |= PF_FORKNOEXEC;
869 new_flags |= PF_STARTING;
870 p->flags = new_flags;
871 clear_freeze_flag(p);
874 asmlinkage long sys_set_tid_address(int __user *tidptr)
876 current->clear_child_tid = tidptr;
878 return task_pid_vnr(current);
881 static void rt_mutex_init_task(struct task_struct *p)
883 spin_lock_init(&p->pi_lock);
884 #ifdef CONFIG_RT_MUTEXES
885 plist_head_init(&p->pi_waiters, &p->pi_lock);
886 p->pi_blocked_on = NULL;
890 #ifdef CONFIG_MM_OWNER
891 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
895 #endif /* CONFIG_MM_OWNER */
898 * Initialize POSIX timer handling for a single task.
900 static void posix_cpu_timers_init(struct task_struct *tsk)
902 tsk->cputime_expires.prof_exp = cputime_zero;
903 tsk->cputime_expires.virt_exp = cputime_zero;
904 tsk->cputime_expires.sched_exp = 0;
905 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
906 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
907 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
911 * This creates a new process as a copy of the old one,
912 * but does not actually start it yet.
914 * It copies the registers, and all the appropriate
915 * parts of the process environment (as per the clone
916 * flags). The actual kick-off is left to the caller.
918 static struct task_struct *copy_process(unsigned long clone_flags,
919 unsigned long stack_start,
920 struct pt_regs *regs,
921 unsigned long stack_size,
922 int __user *child_tidptr,
927 struct task_struct *p;
928 int cgroup_callbacks_done = 0;
930 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
931 return ERR_PTR(-EINVAL);
934 * Thread groups must share signals as well, and detached threads
935 * can only be started up within the thread group.
937 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
938 return ERR_PTR(-EINVAL);
941 * Shared signal handlers imply shared VM. By way of the above,
942 * thread groups also imply shared VM. Blocking this case allows
943 * for various simplifications in other code.
945 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
946 return ERR_PTR(-EINVAL);
948 retval = security_task_create(clone_flags);
953 p = dup_task_struct(current);
957 rt_mutex_init_task(p);
959 #ifdef CONFIG_PROVE_LOCKING
960 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
961 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
964 if (atomic_read(&p->cred->user->processes) >=
965 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
966 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
967 p->cred->user != current->nsproxy->user_ns->root_user)
971 retval = copy_creds(p, clone_flags);
976 * If multiple threads are within copy_process(), then this check
977 * triggers too late. This doesn't hurt, the check is only there
978 * to stop root fork bombs.
980 if (nr_threads >= max_threads)
981 goto bad_fork_cleanup_count;
983 if (!try_module_get(task_thread_info(p)->exec_domain->module))
984 goto bad_fork_cleanup_count;
986 if (p->binfmt && !try_module_get(p->binfmt->module))
987 goto bad_fork_cleanup_put_domain;
990 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
991 copy_flags(clone_flags, p);
992 INIT_LIST_HEAD(&p->children);
993 INIT_LIST_HEAD(&p->sibling);
994 #ifdef CONFIG_PREEMPT_RCU
995 p->rcu_read_lock_nesting = 0;
996 p->rcu_flipctr_idx = 0;
997 #endif /* #ifdef CONFIG_PREEMPT_RCU */
998 p->vfork_done = NULL;
999 spin_lock_init(&p->alloc_lock);
1001 clear_tsk_thread_flag(p, TIF_SIGPENDING);
1002 init_sigpending(&p->pending);
1004 p->utime = cputime_zero;
1005 p->stime = cputime_zero;
1006 p->gtime = cputime_zero;
1007 p->utimescaled = cputime_zero;
1008 p->stimescaled = cputime_zero;
1009 p->prev_utime = cputime_zero;
1010 p->prev_stime = cputime_zero;
1012 p->default_timer_slack_ns = current->timer_slack_ns;
1014 #ifdef CONFIG_DETECT_SOFTLOCKUP
1015 p->last_switch_count = 0;
1016 p->last_switch_timestamp = 0;
1019 task_io_accounting_init(&p->ioac);
1020 acct_clear_integrals(p);
1022 posix_cpu_timers_init(p);
1024 p->lock_depth = -1; /* -1 = no lock */
1025 do_posix_clock_monotonic_gettime(&p->start_time);
1026 p->real_start_time = p->start_time;
1027 monotonic_to_bootbased(&p->real_start_time);
1028 p->io_context = NULL;
1029 p->audit_context = NULL;
1032 p->mempolicy = mpol_dup(p->mempolicy);
1033 if (IS_ERR(p->mempolicy)) {
1034 retval = PTR_ERR(p->mempolicy);
1035 p->mempolicy = NULL;
1036 goto bad_fork_cleanup_cgroup;
1038 mpol_fix_fork_child_flag(p);
1040 #ifdef CONFIG_TRACE_IRQFLAGS
1042 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1043 p->hardirqs_enabled = 1;
1045 p->hardirqs_enabled = 0;
1047 p->hardirq_enable_ip = 0;
1048 p->hardirq_enable_event = 0;
1049 p->hardirq_disable_ip = _THIS_IP_;
1050 p->hardirq_disable_event = 0;
1051 p->softirqs_enabled = 1;
1052 p->softirq_enable_ip = _THIS_IP_;
1053 p->softirq_enable_event = 0;
1054 p->softirq_disable_ip = 0;
1055 p->softirq_disable_event = 0;
1056 p->hardirq_context = 0;
1057 p->softirq_context = 0;
1059 #ifdef CONFIG_LOCKDEP
1060 p->lockdep_depth = 0; /* no locks held yet */
1061 p->curr_chain_key = 0;
1062 p->lockdep_recursion = 0;
1065 #ifdef CONFIG_DEBUG_MUTEXES
1066 p->blocked_on = NULL; /* not blocked yet */
1069 /* Perform scheduler related setup. Assign this task to a CPU. */
1070 sched_fork(p, clone_flags);
1072 if ((retval = audit_alloc(p)))
1073 goto bad_fork_cleanup_policy;
1074 /* copy all the process information */
1075 if ((retval = copy_semundo(clone_flags, p)))
1076 goto bad_fork_cleanup_audit;
1077 if ((retval = copy_files(clone_flags, p)))
1078 goto bad_fork_cleanup_semundo;
1079 if ((retval = copy_fs(clone_flags, p)))
1080 goto bad_fork_cleanup_files;
1081 if ((retval = copy_sighand(clone_flags, p)))
1082 goto bad_fork_cleanup_fs;
1083 if ((retval = copy_signal(clone_flags, p)))
1084 goto bad_fork_cleanup_sighand;
1085 if ((retval = copy_mm(clone_flags, p)))
1086 goto bad_fork_cleanup_signal;
1087 if ((retval = copy_namespaces(clone_flags, p)))
1088 goto bad_fork_cleanup_mm;
1089 if ((retval = copy_io(clone_flags, p)))
1090 goto bad_fork_cleanup_namespaces;
1091 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1093 goto bad_fork_cleanup_io;
1095 if (pid != &init_struct_pid) {
1097 pid = alloc_pid(task_active_pid_ns(p));
1099 goto bad_fork_cleanup_io;
1101 if (clone_flags & CLONE_NEWPID) {
1102 retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1104 goto bad_fork_free_pid;
1108 p->pid = pid_nr(pid);
1110 if (clone_flags & CLONE_THREAD)
1111 p->tgid = current->tgid;
1113 if (current->nsproxy != p->nsproxy) {
1114 retval = ns_cgroup_clone(p, pid);
1116 goto bad_fork_free_pid;
1119 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1121 * Clear TID on mm_release()?
1123 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1125 p->robust_list = NULL;
1126 #ifdef CONFIG_COMPAT
1127 p->compat_robust_list = NULL;
1129 INIT_LIST_HEAD(&p->pi_state_list);
1130 p->pi_state_cache = NULL;
1133 * sigaltstack should be cleared when sharing the same VM
1135 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1136 p->sas_ss_sp = p->sas_ss_size = 0;
1139 * Syscall tracing should be turned off in the child regardless
1142 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1143 #ifdef TIF_SYSCALL_EMU
1144 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1146 clear_all_latency_tracing(p);
1148 /* Our parent execution domain becomes current domain
1149 These must match for thread signalling to apply */
1150 p->parent_exec_id = p->self_exec_id;
1152 /* ok, now we should be set up.. */
1153 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1154 p->pdeath_signal = 0;
1158 * Ok, make it visible to the rest of the system.
1159 * We dont wake it up yet.
1161 p->group_leader = p;
1162 INIT_LIST_HEAD(&p->thread_group);
1164 /* Now that the task is set up, run cgroup callbacks if
1165 * necessary. We need to run them before the task is visible
1166 * on the tasklist. */
1167 cgroup_fork_callbacks(p);
1168 cgroup_callbacks_done = 1;
1170 /* Need tasklist lock for parent etc handling! */
1171 write_lock_irq(&tasklist_lock);
1174 * The task hasn't been attached yet, so its cpus_allowed mask will
1175 * not be changed, nor will its assigned CPU.
1177 * The cpus_allowed mask of the parent may have changed after it was
1178 * copied first time - so re-copy it here, then check the child's CPU
1179 * to ensure it is on a valid CPU (and if not, just force it back to
1180 * parent's CPU). This avoids alot of nasty races.
1182 p->cpus_allowed = current->cpus_allowed;
1183 p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1184 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1185 !cpu_online(task_cpu(p))))
1186 set_task_cpu(p, smp_processor_id());
1188 /* CLONE_PARENT re-uses the old parent */
1189 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1190 p->real_parent = current->real_parent;
1192 p->real_parent = current;
1194 spin_lock(¤t->sighand->siglock);
1197 * Process group and session signals need to be delivered to just the
1198 * parent before the fork or both the parent and the child after the
1199 * fork. Restart if a signal comes in before we add the new process to
1200 * it's process group.
1201 * A fatal signal pending means that current will exit, so the new
1202 * thread can't slip out of an OOM kill (or normal SIGKILL).
1204 recalc_sigpending();
1205 if (signal_pending(current)) {
1206 spin_unlock(¤t->sighand->siglock);
1207 write_unlock_irq(&tasklist_lock);
1208 retval = -ERESTARTNOINTR;
1209 goto bad_fork_free_pid;
1212 if (clone_flags & CLONE_THREAD) {
1213 p->group_leader = current->group_leader;
1214 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1217 if (likely(p->pid)) {
1218 list_add_tail(&p->sibling, &p->real_parent->children);
1219 tracehook_finish_clone(p, clone_flags, trace);
1221 if (thread_group_leader(p)) {
1222 if (clone_flags & CLONE_NEWPID)
1223 p->nsproxy->pid_ns->child_reaper = p;
1225 p->signal->leader_pid = pid;
1226 tty_kref_put(p->signal->tty);
1227 p->signal->tty = tty_kref_get(current->signal->tty);
1228 set_task_pgrp(p, task_pgrp_nr(current));
1229 set_task_session(p, task_session_nr(current));
1230 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1231 attach_pid(p, PIDTYPE_SID, task_session(current));
1232 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1233 __get_cpu_var(process_counts)++;
1235 attach_pid(p, PIDTYPE_PID, pid);
1240 spin_unlock(¤t->sighand->siglock);
1241 write_unlock_irq(&tasklist_lock);
1242 proc_fork_connector(p);
1243 cgroup_post_fork(p);
1247 if (pid != &init_struct_pid)
1249 bad_fork_cleanup_io:
1250 put_io_context(p->io_context);
1251 bad_fork_cleanup_namespaces:
1252 exit_task_namespaces(p);
1253 bad_fork_cleanup_mm:
1256 bad_fork_cleanup_signal:
1258 bad_fork_cleanup_sighand:
1259 __cleanup_sighand(p->sighand);
1260 bad_fork_cleanup_fs:
1261 exit_fs(p); /* blocking */
1262 bad_fork_cleanup_files:
1263 exit_files(p); /* blocking */
1264 bad_fork_cleanup_semundo:
1266 bad_fork_cleanup_audit:
1268 bad_fork_cleanup_policy:
1270 mpol_put(p->mempolicy);
1271 bad_fork_cleanup_cgroup:
1273 cgroup_exit(p, cgroup_callbacks_done);
1274 delayacct_tsk_free(p);
1276 module_put(p->binfmt->module);
1277 bad_fork_cleanup_put_domain:
1278 module_put(task_thread_info(p)->exec_domain->module);
1279 bad_fork_cleanup_count:
1280 atomic_dec(&p->cred->user->processes);
1285 return ERR_PTR(retval);
1288 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1290 memset(regs, 0, sizeof(struct pt_regs));
1294 struct task_struct * __cpuinit fork_idle(int cpu)
1296 struct task_struct *task;
1297 struct pt_regs regs;
1299 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL,
1300 &init_struct_pid, 0);
1302 init_idle(task, cpu);
1308 * Ok, this is the main fork-routine.
1310 * It copies the process, and if successful kick-starts
1311 * it and waits for it to finish using the VM if required.
1313 long do_fork(unsigned long clone_flags,
1314 unsigned long stack_start,
1315 struct pt_regs *regs,
1316 unsigned long stack_size,
1317 int __user *parent_tidptr,
1318 int __user *child_tidptr)
1320 struct task_struct *p;
1325 * We hope to recycle these flags after 2.6.26
1327 if (unlikely(clone_flags & CLONE_STOPPED)) {
1328 static int __read_mostly count = 100;
1330 if (count > 0 && printk_ratelimit()) {
1331 char comm[TASK_COMM_LEN];
1334 printk(KERN_INFO "fork(): process `%s' used deprecated "
1335 "clone flags 0x%lx\n",
1336 get_task_comm(comm, current),
1337 clone_flags & CLONE_STOPPED);
1342 * When called from kernel_thread, don't do user tracing stuff.
1344 if (likely(user_mode(regs)))
1345 trace = tracehook_prepare_clone(clone_flags);
1347 p = copy_process(clone_flags, stack_start, regs, stack_size,
1348 child_tidptr, NULL, trace);
1350 * Do this prior waking up the new thread - the thread pointer
1351 * might get invalid after that point, if the thread exits quickly.
1354 struct completion vfork;
1356 trace_sched_process_fork(current, p);
1358 nr = task_pid_vnr(p);
1360 if (clone_flags & CLONE_PARENT_SETTID)
1361 put_user(nr, parent_tidptr);
1363 if (clone_flags & CLONE_VFORK) {
1364 p->vfork_done = &vfork;
1365 init_completion(&vfork);
1368 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1371 * We set PF_STARTING at creation in case tracing wants to
1372 * use this to distinguish a fully live task from one that
1373 * hasn't gotten to tracehook_report_clone() yet. Now we
1374 * clear it and set the child going.
1376 p->flags &= ~PF_STARTING;
1378 if (unlikely(clone_flags & CLONE_STOPPED)) {
1380 * We'll start up with an immediate SIGSTOP.
1382 sigaddset(&p->pending.signal, SIGSTOP);
1383 set_tsk_thread_flag(p, TIF_SIGPENDING);
1384 __set_task_state(p, TASK_STOPPED);
1386 wake_up_new_task(p, clone_flags);
1389 tracehook_report_clone_complete(trace, regs,
1390 clone_flags, nr, p);
1392 if (clone_flags & CLONE_VFORK) {
1393 freezer_do_not_count();
1394 wait_for_completion(&vfork);
1396 tracehook_report_vfork_done(p, nr);
1404 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1405 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1408 static void sighand_ctor(void *data)
1410 struct sighand_struct *sighand = data;
1412 spin_lock_init(&sighand->siglock);
1413 init_waitqueue_head(&sighand->signalfd_wqh);
1416 void __init proc_caches_init(void)
1418 sighand_cachep = kmem_cache_create("sighand_cache",
1419 sizeof(struct sighand_struct), 0,
1420 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1422 signal_cachep = kmem_cache_create("signal_cache",
1423 sizeof(struct signal_struct), 0,
1424 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1425 files_cachep = kmem_cache_create("files_cache",
1426 sizeof(struct files_struct), 0,
1427 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1428 fs_cachep = kmem_cache_create("fs_cache",
1429 sizeof(struct fs_struct), 0,
1430 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1431 vm_area_cachep = kmem_cache_create("vm_area_struct",
1432 sizeof(struct vm_area_struct), 0,
1434 mm_cachep = kmem_cache_create("mm_struct",
1435 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1436 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1440 * Check constraints on flags passed to the unshare system call and
1441 * force unsharing of additional process context as appropriate.
1443 static void check_unshare_flags(unsigned long *flags_ptr)
1446 * If unsharing a thread from a thread group, must also
1449 if (*flags_ptr & CLONE_THREAD)
1450 *flags_ptr |= CLONE_VM;
1453 * If unsharing vm, must also unshare signal handlers.
1455 if (*flags_ptr & CLONE_VM)
1456 *flags_ptr |= CLONE_SIGHAND;
1459 * If unsharing signal handlers and the task was created
1460 * using CLONE_THREAD, then must unshare the thread
1462 if ((*flags_ptr & CLONE_SIGHAND) &&
1463 (atomic_read(¤t->signal->count) > 1))
1464 *flags_ptr |= CLONE_THREAD;
1467 * If unsharing namespace, must also unshare filesystem information.
1469 if (*flags_ptr & CLONE_NEWNS)
1470 *flags_ptr |= CLONE_FS;
1474 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1476 static int unshare_thread(unsigned long unshare_flags)
1478 if (unshare_flags & CLONE_THREAD)
1485 * Unshare the filesystem structure if it is being shared
1487 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1489 struct fs_struct *fs = current->fs;
1491 if ((unshare_flags & CLONE_FS) &&
1492 (fs && atomic_read(&fs->count) > 1)) {
1493 *new_fsp = __copy_fs_struct(current->fs);
1502 * Unsharing of sighand is not supported yet
1504 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1506 struct sighand_struct *sigh = current->sighand;
1508 if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1515 * Unshare vm if it is being shared
1517 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1519 struct mm_struct *mm = current->mm;
1521 if ((unshare_flags & CLONE_VM) &&
1522 (mm && atomic_read(&mm->mm_users) > 1)) {
1530 * Unshare file descriptor table if it is being shared
1532 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1534 struct files_struct *fd = current->files;
1537 if ((unshare_flags & CLONE_FILES) &&
1538 (fd && atomic_read(&fd->count) > 1)) {
1539 *new_fdp = dup_fd(fd, &error);
1548 * unshare allows a process to 'unshare' part of the process
1549 * context which was originally shared using clone. copy_*
1550 * functions used by do_fork() cannot be used here directly
1551 * because they modify an inactive task_struct that is being
1552 * constructed. Here we are modifying the current, active,
1555 asmlinkage long sys_unshare(unsigned long unshare_flags)
1558 struct fs_struct *fs, *new_fs = NULL;
1559 struct sighand_struct *new_sigh = NULL;
1560 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1561 struct files_struct *fd, *new_fd = NULL;
1562 struct nsproxy *new_nsproxy = NULL;
1565 check_unshare_flags(&unshare_flags);
1567 /* Return -EINVAL for all unsupported flags */
1569 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1570 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1571 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1573 goto bad_unshare_out;
1576 * CLONE_NEWIPC must also detach from the undolist: after switching
1577 * to a new ipc namespace, the semaphore arrays from the old
1578 * namespace are unreachable.
1580 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1582 if ((err = unshare_thread(unshare_flags)))
1583 goto bad_unshare_out;
1584 if ((err = unshare_fs(unshare_flags, &new_fs)))
1585 goto bad_unshare_cleanup_thread;
1586 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1587 goto bad_unshare_cleanup_fs;
1588 if ((err = unshare_vm(unshare_flags, &new_mm)))
1589 goto bad_unshare_cleanup_sigh;
1590 if ((err = unshare_fd(unshare_flags, &new_fd)))
1591 goto bad_unshare_cleanup_vm;
1592 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1594 goto bad_unshare_cleanup_fd;
1596 if (new_fs || new_mm || new_fd || do_sysvsem || new_nsproxy) {
1599 * CLONE_SYSVSEM is equivalent to sys_exit().
1605 switch_task_namespaces(current, new_nsproxy);
1613 current->fs = new_fs;
1619 active_mm = current->active_mm;
1620 current->mm = new_mm;
1621 current->active_mm = new_mm;
1622 activate_mm(active_mm, new_mm);
1627 fd = current->files;
1628 current->files = new_fd;
1632 task_unlock(current);
1636 put_nsproxy(new_nsproxy);
1638 bad_unshare_cleanup_fd:
1640 put_files_struct(new_fd);
1642 bad_unshare_cleanup_vm:
1646 bad_unshare_cleanup_sigh:
1648 if (atomic_dec_and_test(&new_sigh->count))
1649 kmem_cache_free(sighand_cachep, new_sigh);
1651 bad_unshare_cleanup_fs:
1653 put_fs_struct(new_fs);
1655 bad_unshare_cleanup_thread:
1661 * Helper to unshare the files of the current task.
1662 * We don't want to expose copy_files internals to
1663 * the exec layer of the kernel.
1666 int unshare_files(struct files_struct **displaced)
1668 struct task_struct *task = current;
1669 struct files_struct *copy = NULL;
1672 error = unshare_fd(CLONE_FILES, ©);
1673 if (error || !copy) {
1677 *displaced = task->files;