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>
39 #include <linux/ckrm.h>
40 #include <linux/ckrm_tsk.h>
41 #include <linux/ckrm_mem_inline.h>
43 #include <asm/pgtable.h>
44 #include <asm/pgalloc.h>
45 #include <asm/uaccess.h>
46 #include <asm/mmu_context.h>
47 #include <asm/cacheflush.h>
48 #include <asm/tlbflush.h>
50 /* The idle threads do not count..
51 * Protected by write_lock_irq(&tasklist_lock)
56 unsigned long total_forks; /* Handle normal Linux uptimes. */
58 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
60 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED; /* outer */
62 EXPORT_SYMBOL(tasklist_lock);
64 int nr_processes(void)
69 for_each_online_cpu(cpu)
70 total += per_cpu(process_counts, cpu);
75 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
76 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
77 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
78 static kmem_cache_t *task_struct_cachep;
81 static void free_task(struct task_struct *tsk)
83 free_thread_info(tsk->thread_info);
84 free_task_struct(tsk);
87 void __put_task_struct(struct task_struct *tsk)
89 WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
90 WARN_ON(atomic_read(&tsk->usage));
91 WARN_ON(tsk == current);
93 if (unlikely(tsk->audit_context))
95 security_task_free(tsk);
97 put_group_info(tsk->group_info);
101 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
105 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
106 spin_lock_irqsave(&q->lock, flags);
107 __add_wait_queue(q, wait);
108 spin_unlock_irqrestore(&q->lock, flags);
111 EXPORT_SYMBOL(add_wait_queue);
113 void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
117 wait->flags |= WQ_FLAG_EXCLUSIVE;
118 spin_lock_irqsave(&q->lock, flags);
119 __add_wait_queue_tail(q, wait);
120 spin_unlock_irqrestore(&q->lock, flags);
123 EXPORT_SYMBOL(add_wait_queue_exclusive);
125 void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
129 spin_lock_irqsave(&q->lock, flags);
130 __remove_wait_queue(q, wait);
131 spin_unlock_irqrestore(&q->lock, flags);
134 EXPORT_SYMBOL(remove_wait_queue);
138 * Note: we use "set_current_state()" _after_ the wait-queue add,
139 * because we need a memory barrier there on SMP, so that any
140 * wake-function that tests for the wait-queue being active
141 * will be guaranteed to see waitqueue addition _or_ subsequent
142 * tests in this thread will see the wakeup having taken place.
144 * The spin_unlock() itself is semi-permeable and only protects
145 * one way (it only protects stuff inside the critical region and
146 * stops them from bleeding out - it would still allow subsequent
147 * loads to move into the the critical region).
149 void fastcall prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
153 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
154 spin_lock_irqsave(&q->lock, flags);
155 if (list_empty(&wait->task_list))
156 __add_wait_queue(q, wait);
157 set_current_state(state);
158 spin_unlock_irqrestore(&q->lock, flags);
161 EXPORT_SYMBOL(prepare_to_wait);
164 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
168 wait->flags |= WQ_FLAG_EXCLUSIVE;
169 spin_lock_irqsave(&q->lock, flags);
170 if (list_empty(&wait->task_list))
171 __add_wait_queue_tail(q, wait);
172 set_current_state(state);
173 spin_unlock_irqrestore(&q->lock, flags);
176 EXPORT_SYMBOL(prepare_to_wait_exclusive);
178 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
182 __set_current_state(TASK_RUNNING);
184 * We can check for list emptiness outside the lock
186 * - we use the "careful" check that verifies both
187 * the next and prev pointers, so that there cannot
188 * be any half-pending updates in progress on other
189 * CPU's that we haven't seen yet (and that might
190 * still change the stack area.
192 * - all other users take the lock (ie we can only
193 * have _one_ other CPU that looks at or modifies
196 if (!list_empty_careful(&wait->task_list)) {
197 spin_lock_irqsave(&q->lock, flags);
198 list_del_init(&wait->task_list);
199 spin_unlock_irqrestore(&q->lock, flags);
203 EXPORT_SYMBOL(finish_wait);
205 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
207 int ret = default_wake_function(wait, mode, sync, key);
210 list_del_init(&wait->task_list);
214 EXPORT_SYMBOL(autoremove_wake_function);
216 void __init fork_init(unsigned long mempages)
218 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
219 #ifndef ARCH_MIN_TASKALIGN
220 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
222 /* create a slab on which task_structs can be allocated */
224 kmem_cache_create("task_struct", sizeof(struct task_struct),
225 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
229 * The default maximum number of threads is set to a safe
230 * value: the thread structures can take up at most half
233 max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
235 * we need to allow at least 20 threads to boot a system
240 init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
241 init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
244 static struct task_struct *dup_task_struct(struct task_struct *orig)
246 struct task_struct *tsk;
247 struct thread_info *ti;
249 prepare_to_copy(orig);
251 tsk = alloc_task_struct();
255 ti = alloc_thread_info(tsk);
257 free_task_struct(tsk);
261 *ti = *orig->thread_info;
263 tsk->thread_info = ti;
266 ckrm_cb_newtask(tsk);
267 /* One for us, one for whoever does the "release_task()" (usually parent) */
268 atomic_set(&tsk->usage,2);
269 #ifdef CONFIG_CKRM_RES_MEM
270 INIT_LIST_HEAD(&tsk->mm_peers);
276 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
278 struct vm_area_struct * mpnt, *tmp, **pprev;
279 struct rb_node **rb_link, *rb_parent;
281 unsigned long charge;
282 struct mempolicy *pol;
284 down_write(&oldmm->mmap_sem);
285 flush_cache_mm(current->mm);
288 mm->mmap_cache = NULL;
289 mm->free_area_cache = TASK_UNMAPPED_BASE;
292 cpus_clear(mm->cpu_vm_mask);
294 rb_link = &mm->mm_rb.rb_node;
299 * Add it to the mmlist after the parent.
300 * Doing it this way means that we can order the list,
301 * and fork() won't mess up the ordering significantly.
302 * Add it first so that swapoff can see any swap entries.
304 spin_lock(&mmlist_lock);
305 list_add(&mm->mmlist, ¤t->mm->mmlist);
307 spin_unlock(&mmlist_lock);
309 for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
312 if(mpnt->vm_flags & VM_DONTCOPY)
315 if (mpnt->vm_flags & VM_ACCOUNT) {
316 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
317 if (security_vm_enough_memory(len))
321 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
325 pol = mpol_copy(vma_policy(mpnt));
326 retval = PTR_ERR(pol);
328 goto fail_nomem_policy;
329 vma_set_policy(tmp, pol);
330 tmp->vm_flags &= ~VM_LOCKED;
334 vma_prio_tree_init(tmp);
337 struct inode *inode = file->f_dentry->d_inode;
339 if (tmp->vm_flags & VM_DENYWRITE)
340 atomic_dec(&inode->i_writecount);
342 /* insert tmp into the share list, just after mpnt */
343 spin_lock(&file->f_mapping->i_mmap_lock);
344 flush_dcache_mmap_lock(file->f_mapping);
345 vma_prio_tree_add(tmp, mpnt);
346 flush_dcache_mmap_unlock(file->f_mapping);
347 spin_unlock(&file->f_mapping->i_mmap_lock);
351 * Link in the new vma and copy the page table entries:
352 * link in first so that swapoff can see swap entries,
353 * and try_to_unmap_one's find_vma find the new vma.
355 spin_lock(&mm->page_table_lock);
357 pprev = &tmp->vm_next;
359 __vma_link_rb(mm, tmp, rb_link, rb_parent);
360 rb_link = &tmp->vm_rb.rb_right;
361 rb_parent = &tmp->vm_rb;
364 retval = copy_page_range(mm, current->mm, tmp);
365 spin_unlock(&mm->page_table_lock);
367 if (tmp->vm_ops && tmp->vm_ops->open)
368 tmp->vm_ops->open(tmp);
376 flush_tlb_mm(current->mm);
377 up_write(&oldmm->mmap_sem);
380 kmem_cache_free(vm_area_cachep, tmp);
383 vm_unacct_memory(charge);
387 static inline int mm_alloc_pgd(struct mm_struct * mm)
389 mm->pgd = pgd_alloc(mm);
390 if (unlikely(!mm->pgd))
395 static inline void mm_free_pgd(struct mm_struct * mm)
400 #define dup_mmap(mm, oldmm) (0)
401 #define mm_alloc_pgd(mm) (0)
402 #define mm_free_pgd(mm)
403 #endif /* CONFIG_MMU */
405 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
408 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
409 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
411 #include <linux/init_task.h>
413 static struct mm_struct * mm_init(struct mm_struct * mm)
415 atomic_set(&mm->mm_users, 1);
416 atomic_set(&mm->mm_count, 1);
417 init_rwsem(&mm->mmap_sem);
418 mm->core_waiters = 0;
419 mm->page_table_lock = SPIN_LOCK_UNLOCKED;
420 mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
421 mm->ioctx_list = NULL;
422 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
423 mm->free_area_cache = TASK_UNMAPPED_BASE;
424 #ifdef CONFIG_CKRM_RES_MEM
425 INIT_LIST_HEAD(&mm->tasklist);
426 mm->peertask_lock = SPIN_LOCK_UNLOCKED;
429 if (likely(!mm_alloc_pgd(mm))) {
438 * Allocate and initialize an mm_struct.
440 struct mm_struct * mm_alloc(void)
442 struct mm_struct * mm;
446 memset(mm, 0, sizeof(*mm));
448 #ifdef CONFIG_CKRM_RES_MEM
449 mm->memclass = GET_MEM_CLASS(current);
450 mem_class_get(mm->memclass);
457 * Called when the last reference to the mm
458 * is dropped: either by a lazy thread or by
459 * mmput. Free the page directory and the mm.
461 void fastcall __mmdrop(struct mm_struct *mm)
463 BUG_ON(mm == &init_mm);
466 #ifdef CONFIG_CKRM_RES_MEM
467 /* class can be null and mm's tasklist can be empty here */
469 mem_class_put(mm->memclass);
477 * Decrement the use count and release all resources for an mm.
479 void mmput(struct mm_struct *mm)
481 if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
482 list_del(&mm->mmlist);
484 spin_unlock(&mmlist_lock);
492 * Checks if the use count of an mm is non-zero and if so
493 * returns a reference to it after bumping up the use count.
494 * If the use count is zero, it means this mm is going away,
497 struct mm_struct *mmgrab(struct mm_struct *mm)
499 spin_lock(&mmlist_lock);
500 if (!atomic_read(&mm->mm_users))
503 atomic_inc(&mm->mm_users);
504 spin_unlock(&mmlist_lock);
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 cached register state */
526 deactivate_mm(tsk, mm);
528 /* notify parent sleeping on vfork() */
530 tsk->vfork_done = NULL;
531 complete(vfork_done);
533 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
534 u32 __user * tidptr = tsk->clear_child_tid;
535 tsk->clear_child_tid = NULL;
538 * We don't check the error code - if userspace has
539 * not set up a proper pointer then tough luck.
542 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
546 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
548 struct mm_struct * mm, *oldmm;
551 tsk->min_flt = tsk->maj_flt = 0;
552 tsk->cmin_flt = tsk->cmaj_flt = 0;
553 tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
556 tsk->active_mm = NULL;
559 * Are we cloning a kernel thread?
561 * We need to steal a active VM for that..
567 if (clone_flags & CLONE_VM) {
568 atomic_inc(&oldmm->mm_users);
571 * There are cases where the PTL is held to ensure no
572 * new threads start up in user mode using an mm, which
573 * allows optimizing out ipis; the tlb_gather_mmu code
576 spin_unlock_wait(&oldmm->page_table_lock);
585 /* Copy the current MM stuff.. */
586 memcpy(mm, oldmm, sizeof(*mm));
590 if (init_new_context(tsk,mm))
593 retval = dup_mmap(mm, oldmm);
600 ckrm_init_mm_to_task(mm, tsk);
610 * If init_new_context() failed, we cannot use mmput() to free the mm
611 * because it calls destroy_context()
618 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
620 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
621 /* We don't need to lock fs - think why ;-) */
623 atomic_set(&fs->count, 1);
624 fs->lock = RW_LOCK_UNLOCKED;
625 fs->umask = old->umask;
626 read_lock(&old->lock);
627 fs->rootmnt = mntget(old->rootmnt);
628 fs->root = dget(old->root);
629 fs->pwdmnt = mntget(old->pwdmnt);
630 fs->pwd = dget(old->pwd);
632 fs->altrootmnt = mntget(old->altrootmnt);
633 fs->altroot = dget(old->altroot);
635 fs->altrootmnt = NULL;
638 read_unlock(&old->lock);
643 struct fs_struct *copy_fs_struct(struct fs_struct *old)
645 return __copy_fs_struct(old);
648 EXPORT_SYMBOL_GPL(copy_fs_struct);
650 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
652 if (clone_flags & CLONE_FS) {
653 atomic_inc(¤t->fs->count);
656 tsk->fs = __copy_fs_struct(current->fs);
662 static int count_open_files(struct files_struct *files, int size)
666 /* Find the last open fd */
667 for (i = size/(8*sizeof(long)); i > 0; ) {
668 if (files->open_fds->fds_bits[--i])
671 i = (i+1) * 8 * sizeof(long);
675 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
677 struct files_struct *oldf, *newf;
678 struct file **old_fds, **new_fds;
679 int open_files, nfds, size, i, error = 0;
682 * A background process may not have any files ...
684 oldf = current->files;
688 if (clone_flags & CLONE_FILES) {
689 atomic_inc(&oldf->count);
694 * Note: we may be using current for both targets (See exec.c)
695 * This works because we cache current->files (old) as oldf. Don't
700 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
704 atomic_set(&newf->count, 1);
706 newf->file_lock = SPIN_LOCK_UNLOCKED;
708 newf->max_fds = NR_OPEN_DEFAULT;
709 newf->max_fdset = __FD_SETSIZE;
710 newf->close_on_exec = &newf->close_on_exec_init;
711 newf->open_fds = &newf->open_fds_init;
712 newf->fd = &newf->fd_array[0];
714 /* We don't yet have the oldf readlock, but even if the old
715 fdset gets grown now, we'll only copy up to "size" fds */
716 size = oldf->max_fdset;
717 if (size > __FD_SETSIZE) {
719 spin_lock(&newf->file_lock);
720 error = expand_fdset(newf, size-1);
721 spin_unlock(&newf->file_lock);
725 spin_lock(&oldf->file_lock);
727 open_files = count_open_files(oldf, size);
730 * Check whether we need to allocate a larger fd array.
731 * Note: we're not a clone task, so the open count won't
734 nfds = NR_OPEN_DEFAULT;
735 if (open_files > nfds) {
736 spin_unlock(&oldf->file_lock);
738 spin_lock(&newf->file_lock);
739 error = expand_fd_array(newf, open_files-1);
740 spin_unlock(&newf->file_lock);
743 nfds = newf->max_fds;
744 spin_lock(&oldf->file_lock);
750 memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
751 memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
753 for (i = open_files; i != 0; i--) {
754 struct file *f = *old_fds++;
759 spin_unlock(&oldf->file_lock);
761 /* compute the remainder to be cleared */
762 size = (newf->max_fds - open_files) * sizeof(struct file *);
764 /* This is long word aligned thus could use a optimized version */
765 memset(new_fds, 0, size);
767 if (newf->max_fdset > open_files) {
768 int left = (newf->max_fdset-open_files)/8;
769 int start = open_files / (8 * sizeof(unsigned long));
771 memset(&newf->open_fds->fds_bits[start], 0, left);
772 memset(&newf->close_on_exec->fds_bits[start], 0, left);
781 free_fdset (newf->close_on_exec, newf->max_fdset);
782 free_fdset (newf->open_fds, newf->max_fdset);
783 kmem_cache_free(files_cachep, newf);
788 * Helper to unshare the files of the current task.
789 * We don't want to expose copy_files internals to
790 * the exec layer of the kernel.
793 int unshare_files(void)
795 struct files_struct *files = current->files;
801 /* This can race but the race causes us to copy when we don't
802 need to and drop the copy */
803 if(atomic_read(&files->count) == 1)
805 atomic_inc(&files->count);
808 rc = copy_files(0, current);
810 current->files = files;
814 EXPORT_SYMBOL(unshare_files);
816 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
818 struct sighand_struct *sig;
820 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
821 atomic_inc(¤t->sighand->count);
824 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
828 spin_lock_init(&sig->siglock);
829 atomic_set(&sig->count, 1);
830 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
834 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
836 struct signal_struct *sig;
838 if (clone_flags & CLONE_THREAD) {
839 atomic_inc(¤t->signal->count);
842 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
846 atomic_set(&sig->count, 1);
848 sig->group_exit_code = 0;
849 sig->group_exit_task = NULL;
850 sig->group_stop_count = 0;
851 sig->curr_target = NULL;
852 init_sigpending(&sig->shared_pending);
853 INIT_LIST_HEAD(&sig->posix_timers);
855 sig->tty = current->signal->tty;
856 sig->pgrp = process_group(current);
857 sig->session = current->signal->session;
858 sig->leader = 0; /* session leadership doesn't inherit */
859 sig->tty_old_pgrp = 0;
864 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
866 unsigned long new_flags = p->flags;
868 new_flags &= ~PF_SUPERPRIV;
869 new_flags |= PF_FORKNOEXEC;
870 if (!(clone_flags & CLONE_PTRACE))
872 p->flags = new_flags;
875 asmlinkage long sys_set_tid_address(int __user *tidptr)
877 current->clear_child_tid = tidptr;
883 * This creates a new process as a copy of the old one,
884 * but does not actually start it yet.
886 * It copies the registers, and all the appropriate
887 * parts of the process environment (as per the clone
888 * flags). The actual kick-off is left to the caller.
890 struct task_struct *copy_process(unsigned long clone_flags,
891 unsigned long stack_start,
892 struct pt_regs *regs,
893 unsigned long stack_size,
894 int __user *parent_tidptr,
895 int __user *child_tidptr)
898 struct task_struct *p = NULL;
900 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
901 return ERR_PTR(-EINVAL);
904 * Thread groups must share signals as well, and detached threads
905 * can only be started up within the thread group.
907 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
908 return ERR_PTR(-EINVAL);
911 * Shared signal handlers imply shared VM. By way of the above,
912 * thread groups also imply shared VM. Blocking this case allows
913 * for various simplifications in other code.
915 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
916 return ERR_PTR(-EINVAL);
918 retval = security_task_create(clone_flags);
923 p = dup_task_struct(current);
928 if (atomic_read(&p->user->processes) >=
929 p->rlim[RLIMIT_NPROC].rlim_cur) {
930 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
931 p->user != &root_user)
935 atomic_inc(&p->user->__count);
936 atomic_inc(&p->user->processes);
937 get_group_info(p->group_info);
940 * If multiple threads are within copy_process(), then this check
941 * triggers too late. This doesn't hurt, the check is only there
942 * to stop root fork bombs.
944 if (nr_threads >= max_threads)
945 goto bad_fork_cleanup_count;
947 if (!try_module_get(p->thread_info->exec_domain->module))
948 goto bad_fork_cleanup_count;
950 if (p->binfmt && !try_module_get(p->binfmt->module))
951 goto bad_fork_cleanup_put_domain;
955 copy_flags(clone_flags, p);
956 if (clone_flags & CLONE_IDLETASK)
959 p->pid = alloc_pidmap();
961 goto bad_fork_cleanup;
964 if (clone_flags & CLONE_PARENT_SETTID)
965 if (put_user(p->pid, parent_tidptr))
966 goto bad_fork_cleanup;
968 p->proc_dentry = NULL;
970 INIT_LIST_HEAD(&p->children);
971 INIT_LIST_HEAD(&p->sibling);
972 init_waitqueue_head(&p->wait_chldexit);
973 p->vfork_done = NULL;
974 spin_lock_init(&p->alloc_lock);
975 spin_lock_init(&p->proc_lock);
977 clear_tsk_thread_flag(p, TIF_SIGPENDING);
978 init_sigpending(&p->pending);
980 p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
981 p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
982 init_timer(&p->real_timer);
983 p->real_timer.data = (unsigned long) p;
985 p->utime = p->stime = 0;
986 p->cutime = p->cstime = 0;
987 p->lock_depth = -1; /* -1 = no lock */
988 p->start_time = get_jiffies_64();
990 p->io_context = NULL;
991 p->audit_context = NULL;
993 p->mempolicy = mpol_copy(p->mempolicy);
994 if (IS_ERR(p->mempolicy)) {
995 retval = PTR_ERR(p->mempolicy);
997 goto bad_fork_cleanup;
1001 if ((retval = security_task_alloc(p)))
1002 goto bad_fork_cleanup_policy;
1003 if ((retval = audit_alloc(p)))
1004 goto bad_fork_cleanup_security;
1005 /* copy all the process information */
1006 if ((retval = copy_semundo(clone_flags, p)))
1007 goto bad_fork_cleanup_audit;
1008 if ((retval = copy_files(clone_flags, p)))
1009 goto bad_fork_cleanup_semundo;
1010 if ((retval = copy_fs(clone_flags, p)))
1011 goto bad_fork_cleanup_files;
1012 if ((retval = copy_sighand(clone_flags, p)))
1013 goto bad_fork_cleanup_fs;
1014 if ((retval = copy_signal(clone_flags, p)))
1015 goto bad_fork_cleanup_sighand;
1016 if ((retval = copy_mm(clone_flags, p)))
1017 goto bad_fork_cleanup_signal;
1018 if ((retval = copy_namespace(clone_flags, p)))
1019 goto bad_fork_cleanup_mm;
1020 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1022 goto bad_fork_cleanup_namespace;
1024 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1026 * Clear TID on mm_release()?
1028 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1031 * Syscall tracing should be turned off in the child regardless
1034 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1036 /* Our parent execution domain becomes current domain
1037 These must match for thread signalling to apply */
1039 p->parent_exec_id = p->self_exec_id;
1041 /* ok, now we should be set up.. */
1042 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1043 p->pdeath_signal = 0;
1045 /* Perform scheduler related setup */
1049 * Ok, make it visible to the rest of the system.
1050 * We dont wake it up yet.
1053 p->group_leader = p;
1054 INIT_LIST_HEAD(&p->ptrace_children);
1055 INIT_LIST_HEAD(&p->ptrace_list);
1057 /* Need tasklist lock for parent etc handling! */
1058 write_lock_irq(&tasklist_lock);
1060 * Check for pending SIGKILL! The new thread should not be allowed
1061 * to slip out of an OOM kill. (or normal SIGKILL.)
1063 if (sigismember(¤t->pending.signal, SIGKILL)) {
1064 write_unlock_irq(&tasklist_lock);
1066 goto bad_fork_cleanup_namespace;
1069 /* CLONE_PARENT re-uses the old parent */
1070 if (clone_flags & CLONE_PARENT)
1071 p->real_parent = current->real_parent;
1073 p->real_parent = current;
1074 p->parent = p->real_parent;
1076 if (clone_flags & CLONE_THREAD) {
1077 spin_lock(¤t->sighand->siglock);
1079 * Important: if an exit-all has been started then
1080 * do not create this new thread - the whole thread
1081 * group is supposed to exit anyway.
1083 if (current->signal->group_exit) {
1084 spin_unlock(¤t->sighand->siglock);
1085 write_unlock_irq(&tasklist_lock);
1087 goto bad_fork_cleanup_namespace;
1089 p->tgid = current->tgid;
1090 p->group_leader = current->group_leader;
1092 if (current->signal->group_stop_count > 0) {
1094 * There is an all-stop in progress for the group.
1095 * We ourselves will stop as soon as we check signals.
1096 * Make the new thread part of that group stop too.
1098 current->signal->group_stop_count++;
1099 set_tsk_thread_flag(p, TIF_SIGPENDING);
1102 spin_unlock(¤t->sighand->siglock);
1106 if (p->ptrace & PT_PTRACED)
1107 __ptrace_link(p, current->parent);
1109 attach_pid(p, PIDTYPE_PID, p->pid);
1110 if (thread_group_leader(p)) {
1111 attach_pid(p, PIDTYPE_TGID, p->tgid);
1112 attach_pid(p, PIDTYPE_PGID, process_group(p));
1113 attach_pid(p, PIDTYPE_SID, p->signal->session);
1115 __get_cpu_var(process_counts)++;
1117 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1120 write_unlock_irq(&tasklist_lock);
1125 return ERR_PTR(retval);
1128 bad_fork_cleanup_namespace:
1130 bad_fork_cleanup_mm:
1133 mmdrop(p->active_mm);
1134 bad_fork_cleanup_signal:
1136 bad_fork_cleanup_sighand:
1138 bad_fork_cleanup_fs:
1139 exit_fs(p); /* blocking */
1140 bad_fork_cleanup_files:
1141 exit_files(p); /* blocking */
1142 bad_fork_cleanup_semundo:
1144 bad_fork_cleanup_audit:
1146 bad_fork_cleanup_security:
1147 security_task_free(p);
1148 bad_fork_cleanup_policy:
1150 mpol_free(p->mempolicy);
1154 free_pidmap(p->pid);
1156 module_put(p->binfmt->module);
1157 bad_fork_cleanup_put_domain:
1158 module_put(p->thread_info->exec_domain->module);
1159 bad_fork_cleanup_count:
1160 put_group_info(p->group_info);
1161 atomic_dec(&p->user->processes);
1168 static inline int fork_traceflag (unsigned clone_flags)
1170 if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1172 else if (clone_flags & CLONE_VFORK) {
1173 if (current->ptrace & PT_TRACE_VFORK)
1174 return PTRACE_EVENT_VFORK;
1175 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1176 if (current->ptrace & PT_TRACE_CLONE)
1177 return PTRACE_EVENT_CLONE;
1178 } else if (current->ptrace & PT_TRACE_FORK)
1179 return PTRACE_EVENT_FORK;
1185 * Ok, this is the main fork-routine.
1187 * It copies the process, and if successful kick-starts
1188 * it and waits for it to finish using the VM if required.
1190 long do_fork(unsigned long clone_flags,
1191 unsigned long stack_start,
1192 struct pt_regs *regs,
1193 unsigned long stack_size,
1194 int __user *parent_tidptr,
1195 int __user *child_tidptr)
1197 struct task_struct *p;
1201 if (unlikely(current->ptrace)) {
1202 trace = fork_traceflag (clone_flags);
1204 clone_flags |= CLONE_PTRACE;
1207 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1208 if (numtasks_get_ref(current->taskclass, 0) == 0) {
1213 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1215 * Do this prior waking up the new thread - the thread pointer
1216 * might get invalid after that point, if the thread exits quickly.
1218 pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1221 struct completion vfork;
1225 if (clone_flags & CLONE_VFORK) {
1226 p->vfork_done = &vfork;
1227 init_completion(&vfork);
1230 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1232 * We'll start up with an immediate SIGSTOP.
1234 sigaddset(&p->pending.signal, SIGSTOP);
1235 set_tsk_thread_flag(p, TIF_SIGPENDING);
1238 if (!(clone_flags & CLONE_STOPPED)) {
1240 * Do the wakeup last. On SMP we treat fork() and
1241 * CLONE_VM separately, because fork() has already
1242 * created cache footprint on this CPU (due to
1243 * copying the pagetables), hence migration would
1244 * probably be costy. Threads on the other hand
1245 * have less traction to the current CPU, and if
1246 * there's an imbalance then the scheduler can
1247 * migrate this fresh thread now, before it
1248 * accumulates a larger cache footprint:
1250 if (clone_flags & CLONE_VM)
1251 wake_up_forked_thread(p);
1253 wake_up_forked_process(p);
1255 int cpu = get_cpu();
1257 p->state = TASK_STOPPED;
1258 if (cpu_is_offline(task_cpu(p)))
1259 set_task_cpu(p, cpu);
1265 if (unlikely (trace)) {
1266 current->ptrace_message = pid;
1267 ptrace_notify ((trace << 8) | SIGTRAP);
1270 if (clone_flags & CLONE_VFORK) {
1271 wait_for_completion(&vfork);
1272 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1273 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1276 * Let the child process run first, to avoid most of the
1277 * COW overhead when the child exec()s afterwards.
1281 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1282 numtasks_put_ref(current->taskclass);
1288 /* SLAB cache for signal_struct structures (tsk->signal) */
1289 kmem_cache_t *signal_cachep;
1291 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1292 kmem_cache_t *sighand_cachep;
1294 /* SLAB cache for files_struct structures (tsk->files) */
1295 kmem_cache_t *files_cachep;
1297 /* SLAB cache for fs_struct structures (tsk->fs) */
1298 kmem_cache_t *fs_cachep;
1300 /* SLAB cache for vm_area_struct structures */
1301 kmem_cache_t *vm_area_cachep;
1303 /* SLAB cache for mm_struct structures (tsk->mm) */
1304 kmem_cache_t *mm_cachep;
1306 void __init proc_caches_init(void)
1308 sighand_cachep = kmem_cache_create("sighand_cache",
1309 sizeof(struct sighand_struct), 0,
1310 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1311 signal_cachep = kmem_cache_create("signal_cache",
1312 sizeof(struct signal_struct), 0,
1313 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1314 files_cachep = kmem_cache_create("files_cache",
1315 sizeof(struct files_struct), 0,
1316 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1317 fs_cachep = kmem_cache_create("fs_cache",
1318 sizeof(struct fs_struct), 0,
1319 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1320 vm_area_cachep = kmem_cache_create("vm_area_struct",
1321 sizeof(struct vm_area_struct), 0,
1322 SLAB_PANIC, NULL, NULL);
1323 mm_cachep = kmem_cache_create("mm_struct",
1324 sizeof(struct mm_struct), 0,
1325 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);