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/config.h>
15 #include <linux/slab.h>
16 #include <linux/init.h>
17 #include <linux/unistd.h>
18 #include <linux/smp_lock.h>
19 #include <linux/module.h>
20 #include <linux/vmalloc.h>
21 #include <linux/completion.h>
22 #include <linux/namespace.h>
23 #include <linux/personality.h>
24 #include <linux/mempolicy.h>
25 #include <linux/sem.h>
26 #include <linux/file.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
30 #include <linux/cpu.h>
31 #include <linux/security.h>
32 #include <linux/syscalls.h>
33 #include <linux/jiffies.h>
34 #include <linux/futex.h>
35 #include <linux/ptrace.h>
36 #include <linux/mount.h>
37 #include <linux/audit.h>
38 #include <linux/rmap.h>
40 #include <asm/pgtable.h>
41 #include <asm/pgalloc.h>
42 #include <asm/uaccess.h>
43 #include <asm/mmu_context.h>
44 #include <asm/cacheflush.h>
45 #include <asm/tlbflush.h>
47 /* The idle threads do not count..
48 * Protected by write_lock_irq(&tasklist_lock)
53 unsigned long total_forks; /* Handle normal Linux uptimes. */
55 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
57 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED; /* outer */
59 EXPORT_SYMBOL(tasklist_lock);
61 int nr_processes(void)
66 for_each_online_cpu(cpu)
67 total += per_cpu(process_counts, cpu);
72 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
73 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
74 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
75 static kmem_cache_t *task_struct_cachep;
78 static void free_task(struct task_struct *tsk)
80 free_thread_info(tsk->thread_info);
81 free_task_struct(tsk);
84 void __put_task_struct(struct task_struct *tsk)
86 WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
87 WARN_ON(atomic_read(&tsk->usage));
88 WARN_ON(tsk == current);
90 if (unlikely(tsk->audit_context))
92 security_task_free(tsk);
94 put_group_info(tsk->group_info);
98 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
102 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
103 spin_lock_irqsave(&q->lock, flags);
104 __add_wait_queue(q, wait);
105 spin_unlock_irqrestore(&q->lock, flags);
108 EXPORT_SYMBOL(add_wait_queue);
110 void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
114 wait->flags |= WQ_FLAG_EXCLUSIVE;
115 spin_lock_irqsave(&q->lock, flags);
116 __add_wait_queue_tail(q, wait);
117 spin_unlock_irqrestore(&q->lock, flags);
120 EXPORT_SYMBOL(add_wait_queue_exclusive);
122 void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
126 spin_lock_irqsave(&q->lock, flags);
127 __remove_wait_queue(q, wait);
128 spin_unlock_irqrestore(&q->lock, flags);
131 EXPORT_SYMBOL(remove_wait_queue);
135 * Note: we use "set_current_state()" _after_ the wait-queue add,
136 * because we need a memory barrier there on SMP, so that any
137 * wake-function that tests for the wait-queue being active
138 * will be guaranteed to see waitqueue addition _or_ subsequent
139 * tests in this thread will see the wakeup having taken place.
141 * The spin_unlock() itself is semi-permeable and only protects
142 * one way (it only protects stuff inside the critical region and
143 * stops them from bleeding out - it would still allow subsequent
144 * loads to move into the the critical region).
146 void fastcall prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
150 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
151 spin_lock_irqsave(&q->lock, flags);
152 if (list_empty(&wait->task_list))
153 __add_wait_queue(q, wait);
155 * don't alter the task state if this is just going to
156 * queue an async wait queue callback
158 if (is_sync_wait(wait))
159 set_current_state(state);
160 spin_unlock_irqrestore(&q->lock, flags);
163 EXPORT_SYMBOL(prepare_to_wait);
166 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
170 wait->flags |= WQ_FLAG_EXCLUSIVE;
171 spin_lock_irqsave(&q->lock, flags);
172 if (list_empty(&wait->task_list))
173 __add_wait_queue_tail(q, wait);
175 * don't alter the task state if this is just going to
176 * queue an async wait queue callback
178 if (is_sync_wait(wait))
179 set_current_state(state);
180 spin_unlock_irqrestore(&q->lock, flags);
183 EXPORT_SYMBOL(prepare_to_wait_exclusive);
185 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
189 __set_current_state(TASK_RUNNING);
191 * We can check for list emptiness outside the lock
193 * - we use the "careful" check that verifies both
194 * the next and prev pointers, so that there cannot
195 * be any half-pending updates in progress on other
196 * CPU's that we haven't seen yet (and that might
197 * still change the stack area.
199 * - all other users take the lock (ie we can only
200 * have _one_ other CPU that looks at or modifies
203 if (!list_empty_careful(&wait->task_list)) {
204 spin_lock_irqsave(&q->lock, flags);
205 list_del_init(&wait->task_list);
206 spin_unlock_irqrestore(&q->lock, flags);
210 EXPORT_SYMBOL(finish_wait);
212 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
214 int ret = default_wake_function(wait, mode, sync, key);
217 list_del_init(&wait->task_list);
221 EXPORT_SYMBOL(autoremove_wake_function);
223 void __init fork_init(unsigned long mempages)
225 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
226 #ifndef ARCH_MIN_TASKALIGN
227 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
229 /* create a slab on which task_structs can be allocated */
231 kmem_cache_create("task_struct", sizeof(struct task_struct),
232 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
236 * The default maximum number of threads is set to a safe
237 * value: the thread structures can take up at most half
240 max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
242 * we need to allow at least 20 threads to boot a system
247 init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
248 init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
251 static struct task_struct *dup_task_struct(struct task_struct *orig)
253 struct task_struct *tsk;
254 struct thread_info *ti;
256 prepare_to_copy(orig);
258 tsk = alloc_task_struct();
262 ti = alloc_thread_info(tsk);
264 free_task_struct(tsk);
268 *ti = *orig->thread_info;
270 tsk->thread_info = ti;
273 /* initialize list of pages privately reserved by process */
274 INIT_LIST_HEAD(&tsk->private_pages);
275 tsk->private_pages_count = 0;
277 /* One for us, one for whoever does the "release_task()" (usually parent) */
278 atomic_set(&tsk->usage,2);
283 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
285 struct vm_area_struct * mpnt, *tmp, **pprev;
286 struct rb_node **rb_link, *rb_parent;
288 unsigned long charge;
289 struct mempolicy *pol;
291 down_write(&oldmm->mmap_sem);
292 flush_cache_mm(current->mm);
295 mm->mmap_cache = NULL;
296 mm->free_area_cache = TASK_UNMAPPED_BASE;
299 cpus_clear(mm->cpu_vm_mask);
301 rb_link = &mm->mm_rb.rb_node;
306 * Add it to the mmlist after the parent.
307 * Doing it this way means that we can order the list,
308 * and fork() won't mess up the ordering significantly.
309 * Add it first so that swapoff can see any swap entries.
311 spin_lock(&mmlist_lock);
312 list_add(&mm->mmlist, ¤t->mm->mmlist);
314 spin_unlock(&mmlist_lock);
316 for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
319 if(mpnt->vm_flags & VM_DONTCOPY)
322 if (mpnt->vm_flags & VM_ACCOUNT) {
323 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
324 if (security_vm_enough_memory(len))
328 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
332 pol = mpol_copy(vma_policy(mpnt));
333 retval = PTR_ERR(pol);
335 goto fail_nomem_policy;
336 vma_set_policy(tmp, pol);
337 tmp->vm_flags &= ~VM_LOCKED;
341 vma_prio_tree_init(tmp);
344 struct inode *inode = file->f_dentry->d_inode;
346 if (tmp->vm_flags & VM_DENYWRITE)
347 atomic_dec(&inode->i_writecount);
349 /* insert tmp into the share list, just after mpnt */
350 spin_lock(&file->f_mapping->i_mmap_lock);
351 flush_dcache_mmap_lock(file->f_mapping);
352 vma_prio_tree_add(tmp, mpnt);
353 flush_dcache_mmap_unlock(file->f_mapping);
354 spin_unlock(&file->f_mapping->i_mmap_lock);
358 * Link in the new vma and copy the page table entries:
359 * link in first so that swapoff can see swap entries,
360 * and try_to_unmap_one's find_vma find the new vma.
362 spin_lock(&mm->page_table_lock);
364 pprev = &tmp->vm_next;
366 __vma_link_rb(mm, tmp, rb_link, rb_parent);
367 rb_link = &tmp->vm_rb.rb_right;
368 rb_parent = &tmp->vm_rb;
371 retval = copy_page_range(mm, current->mm, tmp);
372 spin_unlock(&mm->page_table_lock);
374 if (tmp->vm_ops && tmp->vm_ops->open)
375 tmp->vm_ops->open(tmp);
383 flush_tlb_mm(current->mm);
384 up_write(&oldmm->mmap_sem);
387 kmem_cache_free(vm_area_cachep, tmp);
390 vm_unacct_memory(charge);
394 static inline int mm_alloc_pgd(struct mm_struct * mm)
396 mm->pgd = pgd_alloc(mm);
397 if (unlikely(!mm->pgd))
402 static inline void mm_free_pgd(struct mm_struct * mm)
407 #define dup_mmap(mm, oldmm) (0)
408 #define mm_alloc_pgd(mm) (0)
409 #define mm_free_pgd(mm)
410 #endif /* CONFIG_MMU */
412 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
415 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
416 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
418 #include <linux/init_task.h>
420 static struct mm_struct * mm_init(struct mm_struct * mm)
422 atomic_set(&mm->mm_users, 1);
423 atomic_set(&mm->mm_count, 1);
424 init_rwsem(&mm->mmap_sem);
425 mm->core_waiters = 0;
426 mm->page_table_lock = SPIN_LOCK_UNLOCKED;
427 mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
428 mm->ioctx_list = NULL;
429 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
430 mm->free_area_cache = TASK_UNMAPPED_BASE;
432 if (likely(!mm_alloc_pgd(mm))) {
441 * Allocate and initialize an mm_struct.
443 struct mm_struct * mm_alloc(void)
445 struct mm_struct * mm;
449 memset(mm, 0, sizeof(*mm));
456 * Called when the last reference to the mm
457 * is dropped: either by a lazy thread or by
458 * mmput. Free the page directory and the mm.
460 void fastcall __mmdrop(struct mm_struct *mm)
462 BUG_ON(mm == &init_mm);
469 * Decrement the use count and release all resources for an mm.
471 void mmput(struct mm_struct *mm)
473 if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
474 list_del(&mm->mmlist);
476 spin_unlock(&mmlist_lock);
484 * Checks if the use count of an mm is non-zero and if so
485 * returns a reference to it after bumping up the use count.
486 * If the use count is zero, it means this mm is going away,
489 struct mm_struct *mmgrab(struct mm_struct *mm)
491 spin_lock(&mmlist_lock);
492 if (!atomic_read(&mm->mm_users))
495 atomic_inc(&mm->mm_users);
496 spin_unlock(&mmlist_lock);
500 /* Please note the differences between mmput and mm_release.
501 * mmput is called whenever we stop holding onto a mm_struct,
502 * error success whatever.
504 * mm_release is called after a mm_struct has been removed
505 * from the current process.
507 * This difference is important for error handling, when we
508 * only half set up a mm_struct for a new process and need to restore
509 * the old one. Because we mmput the new mm_struct before
510 * restoring the old one. . .
511 * Eric Biederman 10 January 1998
513 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
515 struct completion *vfork_done = tsk->vfork_done;
517 /* Get rid of any cached register state */
518 deactivate_mm(tsk, mm);
520 /* notify parent sleeping on vfork() */
522 tsk->vfork_done = NULL;
523 complete(vfork_done);
525 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
526 u32 __user * tidptr = tsk->clear_child_tid;
527 tsk->clear_child_tid = NULL;
530 * We don't check the error code - if userspace has
531 * not set up a proper pointer then tough luck.
534 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
538 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
540 struct mm_struct * mm, *oldmm;
543 tsk->min_flt = tsk->maj_flt = 0;
544 tsk->cmin_flt = tsk->cmaj_flt = 0;
545 tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
548 tsk->active_mm = NULL;
551 * Are we cloning a kernel thread?
553 * We need to steal a active VM for that..
559 if (clone_flags & CLONE_VM) {
560 atomic_inc(&oldmm->mm_users);
563 * There are cases where the PTL is held to ensure no
564 * new threads start up in user mode using an mm, which
565 * allows optimizing out ipis; the tlb_gather_mmu code
568 spin_unlock_wait(&oldmm->page_table_lock);
577 /* Copy the current MM stuff.. */
578 memcpy(mm, oldmm, sizeof(*mm));
582 if (init_new_context(tsk,mm))
585 retval = dup_mmap(mm, oldmm);
601 * If init_new_context() failed, we cannot use mmput() to free the mm
602 * because it calls destroy_context()
609 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
611 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
612 /* We don't need to lock fs - think why ;-) */
614 atomic_set(&fs->count, 1);
615 fs->lock = RW_LOCK_UNLOCKED;
616 fs->umask = old->umask;
617 read_lock(&old->lock);
618 fs->rootmnt = mntget(old->rootmnt);
619 fs->root = dget(old->root);
620 fs->pwdmnt = mntget(old->pwdmnt);
621 fs->pwd = dget(old->pwd);
623 fs->altrootmnt = mntget(old->altrootmnt);
624 fs->altroot = dget(old->altroot);
626 fs->altrootmnt = NULL;
629 read_unlock(&old->lock);
634 struct fs_struct *copy_fs_struct(struct fs_struct *old)
636 return __copy_fs_struct(old);
639 EXPORT_SYMBOL_GPL(copy_fs_struct);
641 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
643 if (clone_flags & CLONE_FS) {
644 atomic_inc(¤t->fs->count);
647 tsk->fs = __copy_fs_struct(current->fs);
653 static int count_open_files(struct files_struct *files, int size)
657 /* Find the last open fd */
658 for (i = size/(8*sizeof(long)); i > 0; ) {
659 if (files->open_fds->fds_bits[--i])
662 i = (i+1) * 8 * sizeof(long);
666 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
668 struct files_struct *oldf, *newf;
669 struct file **old_fds, **new_fds;
670 int open_files, nfds, size, i, error = 0;
673 * A background process may not have any files ...
675 oldf = current->files;
679 if (clone_flags & CLONE_FILES) {
680 atomic_inc(&oldf->count);
685 * Note: we may be using current for both targets (See exec.c)
686 * This works because we cache current->files (old) as oldf. Don't
691 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
695 atomic_set(&newf->count, 1);
697 newf->file_lock = SPIN_LOCK_UNLOCKED;
699 newf->max_fds = NR_OPEN_DEFAULT;
700 newf->max_fdset = __FD_SETSIZE;
701 newf->close_on_exec = &newf->close_on_exec_init;
702 newf->open_fds = &newf->open_fds_init;
703 newf->fd = &newf->fd_array[0];
705 /* We don't yet have the oldf readlock, but even if the old
706 fdset gets grown now, we'll only copy up to "size" fds */
707 size = oldf->max_fdset;
708 if (size > __FD_SETSIZE) {
710 spin_lock(&newf->file_lock);
711 error = expand_fdset(newf, size-1);
712 spin_unlock(&newf->file_lock);
716 spin_lock(&oldf->file_lock);
718 open_files = count_open_files(oldf, size);
721 * Check whether we need to allocate a larger fd array.
722 * Note: we're not a clone task, so the open count won't
725 nfds = NR_OPEN_DEFAULT;
726 if (open_files > nfds) {
727 spin_unlock(&oldf->file_lock);
729 spin_lock(&newf->file_lock);
730 error = expand_fd_array(newf, open_files-1);
731 spin_unlock(&newf->file_lock);
734 nfds = newf->max_fds;
735 spin_lock(&oldf->file_lock);
741 memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
742 memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
744 for (i = open_files; i != 0; i--) {
745 struct file *f = *old_fds++;
750 spin_unlock(&oldf->file_lock);
752 /* compute the remainder to be cleared */
753 size = (newf->max_fds - open_files) * sizeof(struct file *);
755 /* This is long word aligned thus could use a optimized version */
756 memset(new_fds, 0, size);
758 if (newf->max_fdset > open_files) {
759 int left = (newf->max_fdset-open_files)/8;
760 int start = open_files / (8 * sizeof(unsigned long));
762 memset(&newf->open_fds->fds_bits[start], 0, left);
763 memset(&newf->close_on_exec->fds_bits[start], 0, left);
772 free_fdset (newf->close_on_exec, newf->max_fdset);
773 free_fdset (newf->open_fds, newf->max_fdset);
774 kmem_cache_free(files_cachep, newf);
779 * Helper to unshare the files of the current task.
780 * We don't want to expose copy_files internals to
781 * the exec layer of the kernel.
784 int unshare_files(void)
786 struct files_struct *files = current->files;
792 /* This can race but the race causes us to copy when we don't
793 need to and drop the copy */
794 if(atomic_read(&files->count) == 1)
796 atomic_inc(&files->count);
799 rc = copy_files(0, current);
801 current->files = files;
805 EXPORT_SYMBOL(unshare_files);
807 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
809 struct sighand_struct *sig;
811 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
812 atomic_inc(¤t->sighand->count);
815 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
819 spin_lock_init(&sig->siglock);
820 atomic_set(&sig->count, 1);
821 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
825 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
827 struct signal_struct *sig;
829 if (clone_flags & CLONE_THREAD) {
830 atomic_inc(¤t->signal->count);
833 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
837 atomic_set(&sig->count, 1);
839 sig->group_exit_code = 0;
840 sig->group_exit_task = NULL;
841 sig->group_stop_count = 0;
842 sig->curr_target = NULL;
843 init_sigpending(&sig->shared_pending);
844 INIT_LIST_HEAD(&sig->posix_timers);
846 sig->tty = current->signal->tty;
847 sig->pgrp = process_group(current);
848 sig->session = current->signal->session;
849 sig->leader = 0; /* session leadership doesn't inherit */
850 sig->tty_old_pgrp = 0;
855 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
857 unsigned long new_flags = p->flags;
859 new_flags &= ~PF_SUPERPRIV;
860 new_flags |= PF_FORKNOEXEC;
861 if (!(clone_flags & CLONE_PTRACE))
863 p->flags = new_flags;
866 asmlinkage long sys_set_tid_address(int __user *tidptr)
868 current->clear_child_tid = tidptr;
874 * This creates a new process as a copy of the old one,
875 * but does not actually start it yet.
877 * It copies the registers, and all the appropriate
878 * parts of the process environment (as per the clone
879 * flags). The actual kick-off is left to the caller.
881 struct task_struct *copy_process(unsigned long clone_flags,
882 unsigned long stack_start,
883 struct pt_regs *regs,
884 unsigned long stack_size,
885 int __user *parent_tidptr,
886 int __user *child_tidptr)
889 struct task_struct *p = NULL;
891 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
892 return ERR_PTR(-EINVAL);
895 * Thread groups must share signals as well, and detached threads
896 * can only be started up within the thread group.
898 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
899 return ERR_PTR(-EINVAL);
902 * Shared signal handlers imply shared VM. By way of the above,
903 * thread groups also imply shared VM. Blocking this case allows
904 * for various simplifications in other code.
906 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
907 return ERR_PTR(-EINVAL);
909 retval = security_task_create(clone_flags);
914 p = dup_task_struct(current);
919 if (atomic_read(&p->user->processes) >=
920 p->rlim[RLIMIT_NPROC].rlim_cur) {
921 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
922 p->user != &root_user)
926 atomic_inc(&p->user->__count);
927 atomic_inc(&p->user->processes);
928 get_group_info(p->group_info);
931 * If multiple threads are within copy_process(), then this check
932 * triggers too late. This doesn't hurt, the check is only there
933 * to stop root fork bombs.
935 if (nr_threads >= max_threads)
936 goto bad_fork_cleanup_count;
938 if (!try_module_get(p->thread_info->exec_domain->module))
939 goto bad_fork_cleanup_count;
941 if (p->binfmt && !try_module_get(p->binfmt->module))
942 goto bad_fork_cleanup_put_domain;
945 copy_flags(clone_flags, p);
946 if (clone_flags & CLONE_IDLETASK)
949 p->pid = alloc_pidmap();
951 goto bad_fork_cleanup;
954 if (clone_flags & CLONE_PARENT_SETTID)
955 if (put_user(p->pid, parent_tidptr))
956 goto bad_fork_cleanup;
958 p->proc_dentry = NULL;
960 INIT_LIST_HEAD(&p->children);
961 INIT_LIST_HEAD(&p->sibling);
962 init_waitqueue_head(&p->wait_chldexit);
963 p->vfork_done = NULL;
964 spin_lock_init(&p->alloc_lock);
965 spin_lock_init(&p->proc_lock);
967 clear_tsk_thread_flag(p, TIF_SIGPENDING);
968 init_sigpending(&p->pending);
970 p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
971 p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
972 init_timer(&p->real_timer);
973 p->real_timer.data = (unsigned long) p;
975 p->utime = p->stime = 0;
976 p->cutime = p->cstime = 0;
977 p->lock_depth = -1; /* -1 = no lock */
978 p->start_time = get_jiffies_64();
980 p->io_context = NULL;
982 p->audit_context = NULL;
984 p->mempolicy = mpol_copy(p->mempolicy);
985 if (IS_ERR(p->mempolicy)) {
986 retval = PTR_ERR(p->mempolicy);
988 goto bad_fork_cleanup;
992 if ((retval = security_task_alloc(p)))
993 goto bad_fork_cleanup_policy;
994 if ((retval = audit_alloc(p)))
995 goto bad_fork_cleanup_security;
996 /* copy all the process information */
997 if ((retval = copy_semundo(clone_flags, p)))
998 goto bad_fork_cleanup_audit;
999 if ((retval = copy_files(clone_flags, p)))
1000 goto bad_fork_cleanup_semundo;
1001 if ((retval = copy_fs(clone_flags, p)))
1002 goto bad_fork_cleanup_files;
1003 if ((retval = copy_sighand(clone_flags, p)))
1004 goto bad_fork_cleanup_fs;
1005 if ((retval = copy_signal(clone_flags, p)))
1006 goto bad_fork_cleanup_sighand;
1007 if ((retval = copy_mm(clone_flags, p)))
1008 goto bad_fork_cleanup_signal;
1009 if ((retval = copy_namespace(clone_flags, p)))
1010 goto bad_fork_cleanup_mm;
1011 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1013 goto bad_fork_cleanup_namespace;
1015 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1017 * Clear TID on mm_release()?
1019 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1022 * Syscall tracing should be turned off in the child regardless
1025 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1027 /* Our parent execution domain becomes current domain
1028 These must match for thread signalling to apply */
1030 p->parent_exec_id = p->self_exec_id;
1032 /* ok, now we should be set up.. */
1033 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1034 p->pdeath_signal = 0;
1036 /* Perform scheduler related setup */
1040 * Ok, make it visible to the rest of the system.
1041 * We dont wake it up yet.
1044 p->group_leader = p;
1045 INIT_LIST_HEAD(&p->ptrace_children);
1046 INIT_LIST_HEAD(&p->ptrace_list);
1048 /* Need tasklist lock for parent etc handling! */
1049 write_lock_irq(&tasklist_lock);
1051 * Check for pending SIGKILL! The new thread should not be allowed
1052 * to slip out of an OOM kill. (or normal SIGKILL.)
1054 if (sigismember(¤t->pending.signal, SIGKILL)) {
1055 write_unlock_irq(&tasklist_lock);
1057 goto bad_fork_cleanup_namespace;
1060 /* CLONE_PARENT re-uses the old parent */
1061 if (clone_flags & CLONE_PARENT)
1062 p->real_parent = current->real_parent;
1064 p->real_parent = current;
1065 p->parent = p->real_parent;
1067 if (clone_flags & CLONE_THREAD) {
1068 spin_lock(¤t->sighand->siglock);
1070 * Important: if an exit-all has been started then
1071 * do not create this new thread - the whole thread
1072 * group is supposed to exit anyway.
1074 if (current->signal->group_exit) {
1075 spin_unlock(¤t->sighand->siglock);
1076 write_unlock_irq(&tasklist_lock);
1078 goto bad_fork_cleanup_namespace;
1080 p->tgid = current->tgid;
1081 p->group_leader = current->group_leader;
1083 if (current->signal->group_stop_count > 0) {
1085 * There is an all-stop in progress for the group.
1086 * We ourselves will stop as soon as we check signals.
1087 * Make the new thread part of that group stop too.
1089 current->signal->group_stop_count++;
1090 set_tsk_thread_flag(p, TIF_SIGPENDING);
1093 spin_unlock(¤t->sighand->siglock);
1097 if (p->ptrace & PT_PTRACED)
1098 __ptrace_link(p, current->parent);
1100 attach_pid(p, PIDTYPE_PID, p->pid);
1101 if (thread_group_leader(p)) {
1102 attach_pid(p, PIDTYPE_TGID, p->tgid);
1103 attach_pid(p, PIDTYPE_PGID, process_group(p));
1104 attach_pid(p, PIDTYPE_SID, p->signal->session);
1106 __get_cpu_var(process_counts)++;
1108 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1111 write_unlock_irq(&tasklist_lock);
1116 return ERR_PTR(retval);
1119 bad_fork_cleanup_namespace:
1121 bad_fork_cleanup_mm:
1124 mmdrop(p->active_mm);
1125 bad_fork_cleanup_signal:
1127 bad_fork_cleanup_sighand:
1129 bad_fork_cleanup_fs:
1130 exit_fs(p); /* blocking */
1131 bad_fork_cleanup_files:
1132 exit_files(p); /* blocking */
1133 bad_fork_cleanup_semundo:
1135 bad_fork_cleanup_audit:
1137 bad_fork_cleanup_security:
1138 security_task_free(p);
1139 bad_fork_cleanup_policy:
1141 mpol_free(p->mempolicy);
1145 free_pidmap(p->pid);
1147 module_put(p->binfmt->module);
1148 bad_fork_cleanup_put_domain:
1149 module_put(p->thread_info->exec_domain->module);
1150 bad_fork_cleanup_count:
1151 put_group_info(p->group_info);
1152 atomic_dec(&p->user->processes);
1159 static inline int fork_traceflag (unsigned clone_flags)
1161 if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1163 else if (clone_flags & CLONE_VFORK) {
1164 if (current->ptrace & PT_TRACE_VFORK)
1165 return PTRACE_EVENT_VFORK;
1166 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1167 if (current->ptrace & PT_TRACE_CLONE)
1168 return PTRACE_EVENT_CLONE;
1169 } else if (current->ptrace & PT_TRACE_FORK)
1170 return PTRACE_EVENT_FORK;
1176 * Ok, this is the main fork-routine.
1178 * It copies the process, and if successful kick-starts
1179 * it and waits for it to finish using the VM if required.
1181 long do_fork(unsigned long clone_flags,
1182 unsigned long stack_start,
1183 struct pt_regs *regs,
1184 unsigned long stack_size,
1185 int __user *parent_tidptr,
1186 int __user *child_tidptr)
1188 struct task_struct *p;
1192 if (unlikely(current->ptrace)) {
1193 trace = fork_traceflag (clone_flags);
1195 clone_flags |= CLONE_PTRACE;
1198 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1200 * Do this prior waking up the new thread - the thread pointer
1201 * might get invalid after that point, if the thread exits quickly.
1203 pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1206 struct completion vfork;
1208 if (clone_flags & CLONE_VFORK) {
1209 p->vfork_done = &vfork;
1210 init_completion(&vfork);
1213 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1215 * We'll start up with an immediate SIGSTOP.
1217 sigaddset(&p->pending.signal, SIGSTOP);
1218 set_tsk_thread_flag(p, TIF_SIGPENDING);
1221 if (!(clone_flags & CLONE_STOPPED)) {
1223 * Do the wakeup last. On SMP we treat fork() and
1224 * CLONE_VM separately, because fork() has already
1225 * created cache footprint on this CPU (due to
1226 * copying the pagetables), hence migration would
1227 * probably be costy. Threads on the other hand
1228 * have less traction to the current CPU, and if
1229 * there's an imbalance then the scheduler can
1230 * migrate this fresh thread now, before it
1231 * accumulates a larger cache footprint:
1233 if (clone_flags & CLONE_VM)
1234 wake_up_forked_thread(p);
1236 wake_up_forked_process(p);
1238 int cpu = get_cpu();
1240 p->state = TASK_STOPPED;
1241 if (cpu_is_offline(task_cpu(p)))
1242 set_task_cpu(p, cpu);
1248 if (unlikely (trace)) {
1249 current->ptrace_message = pid;
1250 ptrace_notify ((trace << 8) | SIGTRAP);
1253 if (clone_flags & CLONE_VFORK) {
1254 wait_for_completion(&vfork);
1255 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1256 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1259 * Let the child process run first, to avoid most of the
1260 * COW overhead when the child exec()s afterwards.
1267 /* SLAB cache for signal_struct structures (tsk->signal) */
1268 kmem_cache_t *signal_cachep;
1270 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1271 kmem_cache_t *sighand_cachep;
1273 /* SLAB cache for files_struct structures (tsk->files) */
1274 kmem_cache_t *files_cachep;
1276 /* SLAB cache for fs_struct structures (tsk->fs) */
1277 kmem_cache_t *fs_cachep;
1279 /* SLAB cache for vm_area_struct structures */
1280 kmem_cache_t *vm_area_cachep;
1282 /* SLAB cache for mm_struct structures (tsk->mm) */
1283 kmem_cache_t *mm_cachep;
1285 void __init proc_caches_init(void)
1287 sighand_cachep = kmem_cache_create("sighand_cache",
1288 sizeof(struct sighand_struct), 0,
1289 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1290 signal_cachep = kmem_cache_create("signal_cache",
1291 sizeof(struct signal_struct), 0,
1292 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1293 files_cachep = kmem_cache_create("files_cache",
1294 sizeof(struct files_struct), 0,
1295 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1296 fs_cachep = kmem_cache_create("fs_cache",
1297 sizeof(struct fs_struct), 0,
1298 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1299 vm_area_cachep = kmem_cache_create("vm_area_struct",
1300 sizeof(struct vm_area_struct), 0,
1301 SLAB_PANIC, NULL, NULL);
1302 mm_cachep = kmem_cache_create("mm_struct",
1303 sizeof(struct mm_struct), 0,
1304 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);