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/compat.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/acct.h>
53 #include <linux/tsacct_kern.h>
54 #include <linux/cn_proc.h>
55 #include <linux/freezer.h>
56 #include <linux/delayacct.h>
57 #include <linux/taskstats_kern.h>
58 #include <linux/random.h>
59 #include <linux/tty.h>
60 #include <linux/proc_fs.h>
61 #include <linux/blkdev.h>
62 #include <trace/sched.h>
64 #include <asm/pgtable.h>
65 #include <asm/pgalloc.h>
66 #include <asm/uaccess.h>
67 #include <asm/mmu_context.h>
68 #include <asm/cacheflush.h>
69 #include <asm/tlbflush.h>
72 * Protected counters by write_lock_irq(&tasklist_lock)
74 unsigned long total_forks; /* Handle normal Linux uptimes. */
75 int nr_threads; /* The idle threads do not count.. */
77 int max_threads; /* tunable limit on nr_threads */
79 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
81 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
83 int nr_processes(void)
88 for_each_online_cpu(cpu)
89 total += per_cpu(process_counts, cpu);
94 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
95 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
96 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
97 static struct kmem_cache *task_struct_cachep;
100 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
101 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
103 #ifdef CONFIG_DEBUG_STACK_USAGE
104 gfp_t mask = GFP_KERNEL | __GFP_ZERO;
106 gfp_t mask = GFP_KERNEL;
108 return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
111 static inline void free_thread_info(struct thread_info *ti)
113 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
117 /* SLAB cache for signal_struct structures (tsk->signal) */
118 static struct kmem_cache *signal_cachep;
120 /* SLAB cache for sighand_struct structures (tsk->sighand) */
121 struct kmem_cache *sighand_cachep;
123 /* SLAB cache for files_struct structures (tsk->files) */
124 struct kmem_cache *files_cachep;
126 /* SLAB cache for fs_struct structures (tsk->fs) */
127 struct kmem_cache *fs_cachep;
129 /* SLAB cache for vm_area_struct structures */
130 struct kmem_cache *vm_area_cachep;
132 /* SLAB cache for mm_struct structures (tsk->mm) */
133 static struct kmem_cache *mm_cachep;
135 void free_task(struct task_struct *tsk)
137 prop_local_destroy_single(&tsk->dirties);
138 free_thread_info(tsk->stack);
139 rt_mutex_debug_task_free(tsk);
140 free_task_struct(tsk);
142 EXPORT_SYMBOL(free_task);
144 void __put_task_struct(struct task_struct *tsk)
146 WARN_ON(!tsk->exit_state);
147 WARN_ON(atomic_read(&tsk->usage));
148 WARN_ON(tsk == current);
150 put_cred(tsk->real_cred);
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;
213 prepare_to_copy(orig);
215 tsk = alloc_task_struct();
219 ti = alloc_thread_info(tsk);
221 free_task_struct(tsk);
225 err = arch_dup_task_struct(tsk, orig);
231 err = prop_local_init_single(&tsk->dirties);
235 setup_thread_stack(tsk, orig);
237 #ifdef CONFIG_CC_STACKPROTECTOR
238 tsk->stack_canary = get_random_int();
241 /* One for us, one for whoever does the "release_task()" (usually parent) */
242 atomic_set(&tsk->usage,2);
243 atomic_set(&tsk->fs_excl, 0);
244 #ifdef CONFIG_BLK_DEV_IO_TRACE
247 tsk->splice_pipe = NULL;
251 free_thread_info(ti);
252 free_task_struct(tsk);
257 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
259 struct vm_area_struct *mpnt, *tmp, **pprev;
260 struct rb_node **rb_link, *rb_parent;
262 unsigned long charge;
263 struct mempolicy *pol;
265 down_write(&oldmm->mmap_sem);
266 flush_cache_dup_mm(oldmm);
268 * Not linked in yet - no deadlock potential:
270 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
274 mm->mmap_cache = NULL;
275 mm->free_area_cache = oldmm->mmap_base;
276 mm->cached_hole_size = ~0UL;
278 cpus_clear(mm->cpu_vm_mask);
280 rb_link = &mm->mm_rb.rb_node;
284 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
287 if (mpnt->vm_flags & VM_DONTCOPY) {
288 long pages = vma_pages(mpnt);
289 mm->total_vm -= pages;
290 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
295 if (mpnt->vm_flags & VM_ACCOUNT) {
296 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
297 if (security_vm_enough_memory(len))
301 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
305 pol = mpol_dup(vma_policy(mpnt));
306 retval = PTR_ERR(pol);
308 goto fail_nomem_policy;
309 vma_set_policy(tmp, pol);
310 tmp->vm_flags &= ~VM_LOCKED;
316 struct inode *inode = file->f_path.dentry->d_inode;
317 struct address_space *mapping = file->f_mapping;
320 if (tmp->vm_flags & VM_DENYWRITE)
321 atomic_dec(&inode->i_writecount);
322 spin_lock(&mapping->i_mmap_lock);
323 if (tmp->vm_flags & VM_SHARED)
324 mapping->i_mmap_writable++;
325 tmp->vm_truncate_count = mpnt->vm_truncate_count;
326 flush_dcache_mmap_lock(mapping);
327 /* insert tmp into the share list, just after mpnt */
328 vma_prio_tree_add(tmp, mpnt);
329 flush_dcache_mmap_unlock(mapping);
330 spin_unlock(&mapping->i_mmap_lock);
334 * Clear hugetlb-related page reserves for children. This only
335 * affects MAP_PRIVATE mappings. Faults generated by the child
336 * are not guaranteed to succeed, even if read-only
338 if (is_vm_hugetlb_page(tmp))
339 reset_vma_resv_huge_pages(tmp);
342 * Link in the new vma and copy the page table entries.
345 pprev = &tmp->vm_next;
347 __vma_link_rb(mm, tmp, rb_link, rb_parent);
348 rb_link = &tmp->vm_rb.rb_right;
349 rb_parent = &tmp->vm_rb;
352 retval = copy_page_range(mm, oldmm, mpnt);
354 if (tmp->vm_ops && tmp->vm_ops->open)
355 tmp->vm_ops->open(tmp);
360 /* a new mm has just been created */
361 arch_dup_mmap(oldmm, mm);
364 up_write(&mm->mmap_sem);
366 up_write(&oldmm->mmap_sem);
369 kmem_cache_free(vm_area_cachep, tmp);
372 vm_unacct_memory(charge);
376 static inline int mm_alloc_pgd(struct mm_struct * mm)
378 mm->pgd = pgd_alloc(mm);
379 if (unlikely(!mm->pgd))
384 static inline void mm_free_pgd(struct mm_struct * mm)
386 pgd_free(mm, mm->pgd);
389 #define dup_mmap(mm, oldmm) (0)
390 #define mm_alloc_pgd(mm) (0)
391 #define mm_free_pgd(mm)
392 #endif /* CONFIG_MMU */
394 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
396 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
397 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
399 #include <linux/init_task.h>
401 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
403 atomic_set(&mm->mm_users, 1);
404 atomic_set(&mm->mm_count, 1);
405 init_rwsem(&mm->mmap_sem);
406 INIT_LIST_HEAD(&mm->mmlist);
407 mm->flags = (current->mm) ? current->mm->flags
408 : MMF_DUMP_FILTER_DEFAULT;
409 mm->core_state = NULL;
411 set_mm_counter(mm, file_rss, 0);
412 set_mm_counter(mm, anon_rss, 0);
413 spin_lock_init(&mm->page_table_lock);
414 rwlock_init(&mm->ioctx_list_lock);
415 mm->ioctx_list = NULL;
416 mm->free_area_cache = TASK_UNMAPPED_BASE;
417 mm->cached_hole_size = ~0UL;
418 mm_init_owner(mm, p);
420 if (likely(!mm_alloc_pgd(mm))) {
422 mmu_notifier_mm_init(mm);
431 * Allocate and initialize an mm_struct.
433 struct mm_struct * mm_alloc(void)
435 struct mm_struct * mm;
439 memset(mm, 0, sizeof(*mm));
440 mm = mm_init(mm, current);
446 * Called when the last reference to the mm
447 * is dropped: either by a lazy thread or by
448 * mmput. Free the page directory and the mm.
450 void __mmdrop(struct mm_struct *mm)
452 BUG_ON(mm == &init_mm);
455 mmu_notifier_mm_destroy(mm);
458 EXPORT_SYMBOL_GPL(__mmdrop);
461 * Decrement the use count and release all resources for an mm.
463 void mmput(struct mm_struct *mm)
467 if (atomic_dec_and_test(&mm->mm_users)) {
470 set_mm_exe_file(mm, NULL);
471 if (!list_empty(&mm->mmlist)) {
472 spin_lock(&mmlist_lock);
473 list_del(&mm->mmlist);
474 spin_unlock(&mmlist_lock);
480 EXPORT_SYMBOL_GPL(mmput);
483 * get_task_mm - acquire a reference to the task's mm
485 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
486 * this kernel workthread has transiently adopted a user mm with use_mm,
487 * to do its AIO) is not set and if so returns a reference to it, after
488 * bumping up the use count. User must release the mm via mmput()
489 * after use. Typically used by /proc and ptrace.
491 struct mm_struct *get_task_mm(struct task_struct *task)
493 struct mm_struct *mm;
498 if (task->flags & PF_KTHREAD)
501 atomic_inc(&mm->mm_users);
506 EXPORT_SYMBOL_GPL(get_task_mm);
508 /* Please note the differences between mmput and mm_release.
509 * mmput is called whenever we stop holding onto a mm_struct,
510 * error success whatever.
512 * mm_release is called after a mm_struct has been removed
513 * from the current process.
515 * This difference is important for error handling, when we
516 * only half set up a mm_struct for a new process and need to restore
517 * the old one. Because we mmput the new mm_struct before
518 * restoring the old one. . .
519 * Eric Biederman 10 January 1998
521 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
523 struct completion *vfork_done = tsk->vfork_done;
525 /* Get rid of any futexes when releasing the mm */
527 if (unlikely(tsk->robust_list))
528 exit_robust_list(tsk);
530 if (unlikely(tsk->compat_robust_list))
531 compat_exit_robust_list(tsk);
535 /* Get rid of any cached register state */
536 deactivate_mm(tsk, mm);
538 /* notify parent sleeping on vfork() */
540 tsk->vfork_done = NULL;
541 complete(vfork_done);
545 * If we're exiting normally, clear a user-space tid field if
546 * requested. We leave this alone when dying by signal, to leave
547 * the value intact in a core dump, and to save the unnecessary
548 * trouble otherwise. Userland only wants this done for a sys_exit.
550 if (tsk->clear_child_tid
551 && !(tsk->flags & PF_SIGNALED)
552 && atomic_read(&mm->mm_users) > 1) {
553 u32 __user * tidptr = tsk->clear_child_tid;
554 tsk->clear_child_tid = NULL;
557 * We don't check the error code - if userspace has
558 * not set up a proper pointer then tough luck.
561 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
566 * Allocate a new mm structure and copy contents from the
567 * mm structure of the passed in task structure.
569 struct mm_struct *dup_mm(struct task_struct *tsk)
571 struct mm_struct *mm, *oldmm = current->mm;
581 memcpy(mm, oldmm, sizeof(*mm));
583 /* Initializing for Swap token stuff */
584 mm->token_priority = 0;
585 mm->last_interval = 0;
587 if (!mm_init(mm, tsk))
590 if (init_new_context(tsk, mm))
593 dup_mm_exe_file(oldmm, mm);
595 err = dup_mmap(mm, oldmm);
599 mm->hiwater_rss = get_mm_rss(mm);
600 mm->hiwater_vm = mm->total_vm;
612 * If init_new_context() failed, we cannot use mmput() to free the mm
613 * because it calls destroy_context()
620 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
622 struct mm_struct * mm, *oldmm;
625 tsk->min_flt = tsk->maj_flt = 0;
626 tsk->nvcsw = tsk->nivcsw = 0;
629 tsk->active_mm = NULL;
632 * Are we cloning a kernel thread?
634 * We need to steal a active VM for that..
640 if (clone_flags & CLONE_VM) {
641 atomic_inc(&oldmm->mm_users);
652 /* Initializing for Swap token stuff */
653 mm->token_priority = 0;
654 mm->last_interval = 0;
664 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
666 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
667 /* We don't need to lock fs - think why ;-) */
669 atomic_set(&fs->count, 1);
670 rwlock_init(&fs->lock);
671 fs->umask = old->umask;
672 read_lock(&old->lock);
673 fs->root = old->root;
674 path_get(&old->root);
677 read_unlock(&old->lock);
682 struct fs_struct *copy_fs_struct(struct fs_struct *old)
684 return __copy_fs_struct(old);
687 EXPORT_SYMBOL_GPL(copy_fs_struct);
689 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
691 if (clone_flags & CLONE_FS) {
692 atomic_inc(¤t->fs->count);
695 tsk->fs = __copy_fs_struct(current->fs);
701 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
703 struct files_struct *oldf, *newf;
707 * A background process may not have any files ...
709 oldf = current->files;
713 if (clone_flags & CLONE_FILES) {
714 atomic_inc(&oldf->count);
718 newf = dup_fd(oldf, &error);
728 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
731 struct io_context *ioc = current->io_context;
736 * Share io context with parent, if CLONE_IO is set
738 if (clone_flags & CLONE_IO) {
739 tsk->io_context = ioc_task_link(ioc);
740 if (unlikely(!tsk->io_context))
742 } else if (ioprio_valid(ioc->ioprio)) {
743 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
744 if (unlikely(!tsk->io_context))
747 tsk->io_context->ioprio = ioc->ioprio;
753 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
755 struct sighand_struct *sig;
757 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
758 atomic_inc(¤t->sighand->count);
761 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
762 rcu_assign_pointer(tsk->sighand, sig);
765 atomic_set(&sig->count, 1);
766 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
770 void __cleanup_sighand(struct sighand_struct *sighand)
772 if (atomic_dec_and_test(&sighand->count))
773 kmem_cache_free(sighand_cachep, sighand);
778 * Initialize POSIX timer handling for a thread group.
780 static void posix_cpu_timers_init_group(struct signal_struct *sig)
782 /* Thread group counters. */
783 thread_group_cputime_init(sig);
785 /* Expiration times and increments. */
786 sig->it_virt_expires = cputime_zero;
787 sig->it_virt_incr = cputime_zero;
788 sig->it_prof_expires = cputime_zero;
789 sig->it_prof_incr = cputime_zero;
791 /* Cached expiration times. */
792 sig->cputime_expires.prof_exp = cputime_zero;
793 sig->cputime_expires.virt_exp = cputime_zero;
794 sig->cputime_expires.sched_exp = 0;
796 /* The timer lists. */
797 INIT_LIST_HEAD(&sig->cpu_timers[0]);
798 INIT_LIST_HEAD(&sig->cpu_timers[1]);
799 INIT_LIST_HEAD(&sig->cpu_timers[2]);
802 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
804 struct signal_struct *sig;
807 if (clone_flags & CLONE_THREAD) {
808 ret = thread_group_cputime_clone_thread(current);
810 atomic_inc(¤t->signal->count);
811 atomic_inc(¤t->signal->live);
815 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
820 atomic_set(&sig->count, 1);
821 atomic_set(&sig->live, 1);
822 init_waitqueue_head(&sig->wait_chldexit);
824 sig->group_exit_code = 0;
825 sig->group_exit_task = NULL;
826 sig->group_stop_count = 0;
827 sig->curr_target = tsk;
828 init_sigpending(&sig->shared_pending);
829 INIT_LIST_HEAD(&sig->posix_timers);
831 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
832 sig->it_real_incr.tv64 = 0;
833 sig->real_timer.function = it_real_fn;
835 sig->leader = 0; /* session leadership doesn't inherit */
836 sig->tty_old_pgrp = NULL;
839 sig->cutime = sig->cstime = cputime_zero;
840 sig->gtime = cputime_zero;
841 sig->cgtime = cputime_zero;
842 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
843 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
844 sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
845 task_io_accounting_init(&sig->ioac);
846 taskstats_tgid_init(sig);
848 task_lock(current->group_leader);
849 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
850 task_unlock(current->group_leader);
852 posix_cpu_timers_init_group(sig);
854 acct_init_pacct(&sig->pacct);
861 void __cleanup_signal(struct signal_struct *sig)
863 thread_group_cputime_free(sig);
864 tty_kref_put(sig->tty);
865 kmem_cache_free(signal_cachep, sig);
868 static void cleanup_signal(struct task_struct *tsk)
870 struct signal_struct *sig = tsk->signal;
872 atomic_dec(&sig->live);
874 if (atomic_dec_and_test(&sig->count))
875 __cleanup_signal(sig);
878 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
880 unsigned long new_flags = p->flags;
882 new_flags &= ~PF_SUPERPRIV;
883 new_flags |= PF_FORKNOEXEC;
884 new_flags |= PF_STARTING;
885 p->flags = new_flags;
886 clear_freeze_flag(p);
889 asmlinkage long sys_set_tid_address(int __user *tidptr)
891 current->clear_child_tid = tidptr;
893 return task_pid_vnr(current);
896 static void rt_mutex_init_task(struct task_struct *p)
898 spin_lock_init(&p->pi_lock);
899 #ifdef CONFIG_RT_MUTEXES
900 plist_head_init(&p->pi_waiters, &p->pi_lock);
901 p->pi_blocked_on = NULL;
905 #ifdef CONFIG_MM_OWNER
906 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
910 #endif /* CONFIG_MM_OWNER */
913 * Initialize POSIX timer handling for a single task.
915 static void posix_cpu_timers_init(struct task_struct *tsk)
917 tsk->cputime_expires.prof_exp = cputime_zero;
918 tsk->cputime_expires.virt_exp = cputime_zero;
919 tsk->cputime_expires.sched_exp = 0;
920 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
921 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
922 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
926 * This creates a new process as a copy of the old one,
927 * but does not actually start it yet.
929 * It copies the registers, and all the appropriate
930 * parts of the process environment (as per the clone
931 * flags). The actual kick-off is left to the caller.
933 static struct task_struct *copy_process(unsigned long clone_flags,
934 unsigned long stack_start,
935 struct pt_regs *regs,
936 unsigned long stack_size,
937 int __user *child_tidptr,
942 struct task_struct *p;
943 int cgroup_callbacks_done = 0;
945 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
946 return ERR_PTR(-EINVAL);
949 * Thread groups must share signals as well, and detached threads
950 * can only be started up within the thread group.
952 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
953 return ERR_PTR(-EINVAL);
956 * Shared signal handlers imply shared VM. By way of the above,
957 * thread groups also imply shared VM. Blocking this case allows
958 * for various simplifications in other code.
960 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
961 return ERR_PTR(-EINVAL);
963 retval = security_task_create(clone_flags);
968 p = dup_task_struct(current);
972 rt_mutex_init_task(p);
974 #ifdef CONFIG_PROVE_LOCKING
975 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
976 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
979 if (atomic_read(&p->real_cred->user->processes) >=
980 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
981 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
982 p->real_cred->user != INIT_USER)
986 retval = copy_creds(p, clone_flags);
991 * If multiple threads are within copy_process(), then this check
992 * triggers too late. This doesn't hurt, the check is only there
993 * to stop root fork bombs.
995 if (nr_threads >= max_threads)
996 goto bad_fork_cleanup_count;
998 if (!try_module_get(task_thread_info(p)->exec_domain->module))
999 goto bad_fork_cleanup_count;
1001 if (p->binfmt && !try_module_get(p->binfmt->module))
1002 goto bad_fork_cleanup_put_domain;
1005 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1006 copy_flags(clone_flags, p);
1007 INIT_LIST_HEAD(&p->children);
1008 INIT_LIST_HEAD(&p->sibling);
1009 #ifdef CONFIG_PREEMPT_RCU
1010 p->rcu_read_lock_nesting = 0;
1011 p->rcu_flipctr_idx = 0;
1012 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1013 p->vfork_done = NULL;
1014 spin_lock_init(&p->alloc_lock);
1016 clear_tsk_thread_flag(p, TIF_SIGPENDING);
1017 init_sigpending(&p->pending);
1019 p->utime = cputime_zero;
1020 p->stime = cputime_zero;
1021 p->gtime = cputime_zero;
1022 p->utimescaled = cputime_zero;
1023 p->stimescaled = cputime_zero;
1024 p->prev_utime = cputime_zero;
1025 p->prev_stime = cputime_zero;
1027 p->default_timer_slack_ns = current->timer_slack_ns;
1029 #ifdef CONFIG_DETECT_SOFTLOCKUP
1030 p->last_switch_count = 0;
1031 p->last_switch_timestamp = 0;
1034 task_io_accounting_init(&p->ioac);
1035 acct_clear_integrals(p);
1037 posix_cpu_timers_init(p);
1039 p->lock_depth = -1; /* -1 = no lock */
1040 do_posix_clock_monotonic_gettime(&p->start_time);
1041 p->real_start_time = p->start_time;
1042 monotonic_to_bootbased(&p->real_start_time);
1043 p->io_context = NULL;
1044 p->audit_context = NULL;
1047 p->mempolicy = mpol_dup(p->mempolicy);
1048 if (IS_ERR(p->mempolicy)) {
1049 retval = PTR_ERR(p->mempolicy);
1050 p->mempolicy = NULL;
1051 goto bad_fork_cleanup_cgroup;
1053 mpol_fix_fork_child_flag(p);
1055 #ifdef CONFIG_TRACE_IRQFLAGS
1057 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1058 p->hardirqs_enabled = 1;
1060 p->hardirqs_enabled = 0;
1062 p->hardirq_enable_ip = 0;
1063 p->hardirq_enable_event = 0;
1064 p->hardirq_disable_ip = _THIS_IP_;
1065 p->hardirq_disable_event = 0;
1066 p->softirqs_enabled = 1;
1067 p->softirq_enable_ip = _THIS_IP_;
1068 p->softirq_enable_event = 0;
1069 p->softirq_disable_ip = 0;
1070 p->softirq_disable_event = 0;
1071 p->hardirq_context = 0;
1072 p->softirq_context = 0;
1074 #ifdef CONFIG_LOCKDEP
1075 p->lockdep_depth = 0; /* no locks held yet */
1076 p->curr_chain_key = 0;
1077 p->lockdep_recursion = 0;
1080 #ifdef CONFIG_DEBUG_MUTEXES
1081 p->blocked_on = NULL; /* not blocked yet */
1084 /* Perform scheduler related setup. Assign this task to a CPU. */
1085 sched_fork(p, clone_flags);
1087 if ((retval = audit_alloc(p)))
1088 goto bad_fork_cleanup_policy;
1089 /* copy all the process information */
1090 if ((retval = copy_semundo(clone_flags, p)))
1091 goto bad_fork_cleanup_audit;
1092 if ((retval = copy_files(clone_flags, p)))
1093 goto bad_fork_cleanup_semundo;
1094 if ((retval = copy_fs(clone_flags, p)))
1095 goto bad_fork_cleanup_files;
1096 if ((retval = copy_sighand(clone_flags, p)))
1097 goto bad_fork_cleanup_fs;
1098 if ((retval = copy_signal(clone_flags, p)))
1099 goto bad_fork_cleanup_sighand;
1100 if ((retval = copy_mm(clone_flags, p)))
1101 goto bad_fork_cleanup_signal;
1102 if ((retval = copy_namespaces(clone_flags, p)))
1103 goto bad_fork_cleanup_mm;
1104 if ((retval = copy_io(clone_flags, p)))
1105 goto bad_fork_cleanup_namespaces;
1106 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1108 goto bad_fork_cleanup_io;
1110 if (pid != &init_struct_pid) {
1112 pid = alloc_pid(task_active_pid_ns(p));
1114 goto bad_fork_cleanup_io;
1116 if (clone_flags & CLONE_NEWPID) {
1117 retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1119 goto bad_fork_free_pid;
1123 p->pid = pid_nr(pid);
1125 if (clone_flags & CLONE_THREAD)
1126 p->tgid = current->tgid;
1128 if (current->nsproxy != p->nsproxy) {
1129 retval = ns_cgroup_clone(p, pid);
1131 goto bad_fork_free_pid;
1134 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1136 * Clear TID on mm_release()?
1138 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1140 p->robust_list = NULL;
1141 #ifdef CONFIG_COMPAT
1142 p->compat_robust_list = NULL;
1144 INIT_LIST_HEAD(&p->pi_state_list);
1145 p->pi_state_cache = NULL;
1148 * sigaltstack should be cleared when sharing the same VM
1150 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1151 p->sas_ss_sp = p->sas_ss_size = 0;
1154 * Syscall tracing should be turned off in the child regardless
1157 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1158 #ifdef TIF_SYSCALL_EMU
1159 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1161 clear_all_latency_tracing(p);
1163 /* Our parent execution domain becomes current domain
1164 These must match for thread signalling to apply */
1165 p->parent_exec_id = p->self_exec_id;
1167 /* ok, now we should be set up.. */
1168 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1169 p->pdeath_signal = 0;
1173 * Ok, make it visible to the rest of the system.
1174 * We dont wake it up yet.
1176 p->group_leader = p;
1177 INIT_LIST_HEAD(&p->thread_group);
1179 /* Now that the task is set up, run cgroup callbacks if
1180 * necessary. We need to run them before the task is visible
1181 * on the tasklist. */
1182 cgroup_fork_callbacks(p);
1183 cgroup_callbacks_done = 1;
1185 /* Need tasklist lock for parent etc handling! */
1186 write_lock_irq(&tasklist_lock);
1189 * The task hasn't been attached yet, so its cpus_allowed mask will
1190 * not be changed, nor will its assigned CPU.
1192 * The cpus_allowed mask of the parent may have changed after it was
1193 * copied first time - so re-copy it here, then check the child's CPU
1194 * to ensure it is on a valid CPU (and if not, just force it back to
1195 * parent's CPU). This avoids alot of nasty races.
1197 p->cpus_allowed = current->cpus_allowed;
1198 p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1199 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1200 !cpu_online(task_cpu(p))))
1201 set_task_cpu(p, smp_processor_id());
1203 /* CLONE_PARENT re-uses the old parent */
1204 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1205 p->real_parent = current->real_parent;
1207 p->real_parent = current;
1209 spin_lock(¤t->sighand->siglock);
1212 * Process group and session signals need to be delivered to just the
1213 * parent before the fork or both the parent and the child after the
1214 * fork. Restart if a signal comes in before we add the new process to
1215 * it's process group.
1216 * A fatal signal pending means that current will exit, so the new
1217 * thread can't slip out of an OOM kill (or normal SIGKILL).
1219 recalc_sigpending();
1220 if (signal_pending(current)) {
1221 spin_unlock(¤t->sighand->siglock);
1222 write_unlock_irq(&tasklist_lock);
1223 retval = -ERESTARTNOINTR;
1224 goto bad_fork_free_pid;
1227 if (clone_flags & CLONE_THREAD) {
1228 p->group_leader = current->group_leader;
1229 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1232 if (likely(p->pid)) {
1233 list_add_tail(&p->sibling, &p->real_parent->children);
1234 tracehook_finish_clone(p, clone_flags, trace);
1236 if (thread_group_leader(p)) {
1237 if (clone_flags & CLONE_NEWPID)
1238 p->nsproxy->pid_ns->child_reaper = p;
1240 p->signal->leader_pid = pid;
1241 tty_kref_put(p->signal->tty);
1242 p->signal->tty = tty_kref_get(current->signal->tty);
1243 set_task_pgrp(p, task_pgrp_nr(current));
1244 set_task_session(p, task_session_nr(current));
1245 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1246 attach_pid(p, PIDTYPE_SID, task_session(current));
1247 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1248 __get_cpu_var(process_counts)++;
1250 attach_pid(p, PIDTYPE_PID, pid);
1255 spin_unlock(¤t->sighand->siglock);
1256 write_unlock_irq(&tasklist_lock);
1257 proc_fork_connector(p);
1258 cgroup_post_fork(p);
1262 if (pid != &init_struct_pid)
1264 bad_fork_cleanup_io:
1265 put_io_context(p->io_context);
1266 bad_fork_cleanup_namespaces:
1267 exit_task_namespaces(p);
1268 bad_fork_cleanup_mm:
1271 bad_fork_cleanup_signal:
1273 bad_fork_cleanup_sighand:
1274 __cleanup_sighand(p->sighand);
1275 bad_fork_cleanup_fs:
1276 exit_fs(p); /* blocking */
1277 bad_fork_cleanup_files:
1278 exit_files(p); /* blocking */
1279 bad_fork_cleanup_semundo:
1281 bad_fork_cleanup_audit:
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 atomic_dec(&p->cred->user->processes);
1296 put_cred(p->real_cred);
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 * Do some preliminary argument and permissions checking before we
1342 * actually start allocating stuff
1344 if (clone_flags & CLONE_NEWUSER) {
1345 if (clone_flags & CLONE_THREAD)
1347 /* hopefully this check will go away when userns support is
1350 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1351 !capable(CAP_SETGID))
1356 * We hope to recycle these flags after 2.6.26
1358 if (unlikely(clone_flags & CLONE_STOPPED)) {
1359 static int __read_mostly count = 100;
1361 if (count > 0 && printk_ratelimit()) {
1362 char comm[TASK_COMM_LEN];
1365 printk(KERN_INFO "fork(): process `%s' used deprecated "
1366 "clone flags 0x%lx\n",
1367 get_task_comm(comm, current),
1368 clone_flags & CLONE_STOPPED);
1373 * When called from kernel_thread, don't do user tracing stuff.
1375 if (likely(user_mode(regs)))
1376 trace = tracehook_prepare_clone(clone_flags);
1378 p = copy_process(clone_flags, stack_start, regs, stack_size,
1379 child_tidptr, NULL, trace);
1381 * Do this prior waking up the new thread - the thread pointer
1382 * might get invalid after that point, if the thread exits quickly.
1385 struct completion vfork;
1387 trace_sched_process_fork(current, p);
1389 nr = task_pid_vnr(p);
1391 if (clone_flags & CLONE_PARENT_SETTID)
1392 put_user(nr, parent_tidptr);
1394 if (clone_flags & CLONE_VFORK) {
1395 p->vfork_done = &vfork;
1396 init_completion(&vfork);
1399 audit_finish_fork(p);
1400 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1403 * We set PF_STARTING at creation in case tracing wants to
1404 * use this to distinguish a fully live task from one that
1405 * hasn't gotten to tracehook_report_clone() yet. Now we
1406 * clear it and set the child going.
1408 p->flags &= ~PF_STARTING;
1410 if (unlikely(clone_flags & CLONE_STOPPED)) {
1412 * We'll start up with an immediate SIGSTOP.
1414 sigaddset(&p->pending.signal, SIGSTOP);
1415 set_tsk_thread_flag(p, TIF_SIGPENDING);
1416 __set_task_state(p, TASK_STOPPED);
1418 wake_up_new_task(p, clone_flags);
1421 tracehook_report_clone_complete(trace, regs,
1422 clone_flags, nr, p);
1424 if (clone_flags & CLONE_VFORK) {
1425 freezer_do_not_count();
1426 wait_for_completion(&vfork);
1428 tracehook_report_vfork_done(p, nr);
1436 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1437 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1440 static void sighand_ctor(void *data)
1442 struct sighand_struct *sighand = data;
1444 spin_lock_init(&sighand->siglock);
1445 init_waitqueue_head(&sighand->signalfd_wqh);
1448 void __init proc_caches_init(void)
1450 sighand_cachep = kmem_cache_create("sighand_cache",
1451 sizeof(struct sighand_struct), 0,
1452 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1454 signal_cachep = kmem_cache_create("signal_cache",
1455 sizeof(struct signal_struct), 0,
1456 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1457 files_cachep = kmem_cache_create("files_cache",
1458 sizeof(struct files_struct), 0,
1459 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1460 fs_cachep = kmem_cache_create("fs_cache",
1461 sizeof(struct fs_struct), 0,
1462 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1463 vm_area_cachep = kmem_cache_create("vm_area_struct",
1464 sizeof(struct vm_area_struct), 0,
1466 mm_cachep = kmem_cache_create("mm_struct",
1467 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1468 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1472 * Check constraints on flags passed to the unshare system call and
1473 * force unsharing of additional process context as appropriate.
1475 static void check_unshare_flags(unsigned long *flags_ptr)
1478 * If unsharing a thread from a thread group, must also
1481 if (*flags_ptr & CLONE_THREAD)
1482 *flags_ptr |= CLONE_VM;
1485 * If unsharing vm, must also unshare signal handlers.
1487 if (*flags_ptr & CLONE_VM)
1488 *flags_ptr |= CLONE_SIGHAND;
1491 * If unsharing signal handlers and the task was created
1492 * using CLONE_THREAD, then must unshare the thread
1494 if ((*flags_ptr & CLONE_SIGHAND) &&
1495 (atomic_read(¤t->signal->count) > 1))
1496 *flags_ptr |= CLONE_THREAD;
1499 * If unsharing namespace, must also unshare filesystem information.
1501 if (*flags_ptr & CLONE_NEWNS)
1502 *flags_ptr |= CLONE_FS;
1506 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1508 static int unshare_thread(unsigned long unshare_flags)
1510 if (unshare_flags & CLONE_THREAD)
1517 * Unshare the filesystem structure if it is being shared
1519 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1521 struct fs_struct *fs = current->fs;
1523 if ((unshare_flags & CLONE_FS) &&
1524 (fs && atomic_read(&fs->count) > 1)) {
1525 *new_fsp = __copy_fs_struct(current->fs);
1534 * Unsharing of sighand is not supported yet
1536 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1538 struct sighand_struct *sigh = current->sighand;
1540 if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1547 * Unshare vm if it is being shared
1549 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1551 struct mm_struct *mm = current->mm;
1553 if ((unshare_flags & CLONE_VM) &&
1554 (mm && atomic_read(&mm->mm_users) > 1)) {
1562 * Unshare file descriptor table if it is being shared
1564 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1566 struct files_struct *fd = current->files;
1569 if ((unshare_flags & CLONE_FILES) &&
1570 (fd && atomic_read(&fd->count) > 1)) {
1571 *new_fdp = dup_fd(fd, &error);
1580 * unshare allows a process to 'unshare' part of the process
1581 * context which was originally shared using clone. copy_*
1582 * functions used by do_fork() cannot be used here directly
1583 * because they modify an inactive task_struct that is being
1584 * constructed. Here we are modifying the current, active,
1587 asmlinkage long sys_unshare(unsigned long unshare_flags)
1590 struct fs_struct *fs, *new_fs = NULL;
1591 struct sighand_struct *new_sigh = NULL;
1592 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1593 struct files_struct *fd, *new_fd = NULL;
1594 struct nsproxy *new_nsproxy = NULL;
1597 check_unshare_flags(&unshare_flags);
1599 /* Return -EINVAL for all unsupported flags */
1601 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1602 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1603 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1604 goto bad_unshare_out;
1607 * CLONE_NEWIPC must also detach from the undolist: after switching
1608 * to a new ipc namespace, the semaphore arrays from the old
1609 * namespace are unreachable.
1611 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1613 if ((err = unshare_thread(unshare_flags)))
1614 goto bad_unshare_out;
1615 if ((err = unshare_fs(unshare_flags, &new_fs)))
1616 goto bad_unshare_cleanup_thread;
1617 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1618 goto bad_unshare_cleanup_fs;
1619 if ((err = unshare_vm(unshare_flags, &new_mm)))
1620 goto bad_unshare_cleanup_sigh;
1621 if ((err = unshare_fd(unshare_flags, &new_fd)))
1622 goto bad_unshare_cleanup_vm;
1623 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1625 goto bad_unshare_cleanup_fd;
1627 if (new_fs || new_mm || new_fd || do_sysvsem || new_nsproxy) {
1630 * CLONE_SYSVSEM is equivalent to sys_exit().
1636 switch_task_namespaces(current, new_nsproxy);
1644 current->fs = new_fs;
1650 active_mm = current->active_mm;
1651 current->mm = new_mm;
1652 current->active_mm = new_mm;
1653 activate_mm(active_mm, new_mm);
1658 fd = current->files;
1659 current->files = new_fd;
1663 task_unlock(current);
1667 put_nsproxy(new_nsproxy);
1669 bad_unshare_cleanup_fd:
1671 put_files_struct(new_fd);
1673 bad_unshare_cleanup_vm:
1677 bad_unshare_cleanup_sigh:
1679 if (atomic_dec_and_test(&new_sigh->count))
1680 kmem_cache_free(sighand_cachep, new_sigh);
1682 bad_unshare_cleanup_fs:
1684 put_fs_struct(new_fs);
1686 bad_unshare_cleanup_thread:
1692 * Helper to unshare the files of the current task.
1693 * We don't want to expose copy_files internals to
1694 * the exec layer of the kernel.
1697 int unshare_files(struct files_struct **displaced)
1699 struct task_struct *task = current;
1700 struct files_struct *copy = NULL;
1703 error = unshare_fd(CLONE_FILES, ©);
1704 if (error || !copy) {
1708 *displaced = task->files;