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/vinline.h>
40 #include <linux/ninline.h>
42 #include <asm/pgtable.h>
43 #include <asm/pgalloc.h>
44 #include <asm/uaccess.h>
45 #include <asm/mmu_context.h>
46 #include <asm/cacheflush.h>
47 #include <asm/tlbflush.h>
49 /* The idle threads do not count..
50 * Protected by write_lock_irq(&tasklist_lock)
55 unsigned long total_forks; /* Handle normal Linux uptimes. */
57 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
59 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED; /* outer */
61 EXPORT_SYMBOL(tasklist_lock);
63 int nr_processes(void)
68 for_each_online_cpu(cpu)
69 total += per_cpu(process_counts, cpu);
74 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
75 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
76 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
77 static kmem_cache_t *task_struct_cachep;
80 static void free_task(struct task_struct *tsk)
82 free_thread_info(tsk->thread_info);
83 clr_vx_info(&tsk->vx_info);
84 clr_nx_info(&tsk->nx_info);
85 free_task_struct(tsk);
88 void __put_task_struct(struct task_struct *tsk)
90 WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
91 WARN_ON(atomic_read(&tsk->usage));
92 WARN_ON(tsk == current);
94 if (unlikely(tsk->audit_context))
96 security_task_free(tsk);
98 put_group_info(tsk->group_info);
102 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
106 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
107 spin_lock_irqsave(&q->lock, flags);
108 __add_wait_queue(q, wait);
109 spin_unlock_irqrestore(&q->lock, flags);
112 EXPORT_SYMBOL(add_wait_queue);
114 void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
118 wait->flags |= WQ_FLAG_EXCLUSIVE;
119 spin_lock_irqsave(&q->lock, flags);
120 __add_wait_queue_tail(q, wait);
121 spin_unlock_irqrestore(&q->lock, flags);
124 EXPORT_SYMBOL(add_wait_queue_exclusive);
126 void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
130 spin_lock_irqsave(&q->lock, flags);
131 __remove_wait_queue(q, wait);
132 spin_unlock_irqrestore(&q->lock, flags);
135 EXPORT_SYMBOL(remove_wait_queue);
139 * Note: we use "set_current_state()" _after_ the wait-queue add,
140 * because we need a memory barrier there on SMP, so that any
141 * wake-function that tests for the wait-queue being active
142 * will be guaranteed to see waitqueue addition _or_ subsequent
143 * tests in this thread will see the wakeup having taken place.
145 * The spin_unlock() itself is semi-permeable and only protects
146 * one way (it only protects stuff inside the critical region and
147 * stops them from bleeding out - it would still allow subsequent
148 * loads to move into the the critical region).
150 void fastcall prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
154 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
155 spin_lock_irqsave(&q->lock, flags);
156 if (list_empty(&wait->task_list))
157 __add_wait_queue(q, wait);
158 set_current_state(state);
159 spin_unlock_irqrestore(&q->lock, flags);
162 EXPORT_SYMBOL(prepare_to_wait);
165 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
169 wait->flags |= WQ_FLAG_EXCLUSIVE;
170 spin_lock_irqsave(&q->lock, flags);
171 if (list_empty(&wait->task_list))
172 __add_wait_queue_tail(q, wait);
173 set_current_state(state);
174 spin_unlock_irqrestore(&q->lock, flags);
177 EXPORT_SYMBOL(prepare_to_wait_exclusive);
179 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
183 __set_current_state(TASK_RUNNING);
185 * We can check for list emptiness outside the lock
187 * - we use the "careful" check that verifies both
188 * the next and prev pointers, so that there cannot
189 * be any half-pending updates in progress on other
190 * CPU's that we haven't seen yet (and that might
191 * still change the stack area.
193 * - all other users take the lock (ie we can only
194 * have _one_ other CPU that looks at or modifies
197 if (!list_empty_careful(&wait->task_list)) {
198 spin_lock_irqsave(&q->lock, flags);
199 list_del_init(&wait->task_list);
200 spin_unlock_irqrestore(&q->lock, flags);
204 EXPORT_SYMBOL(finish_wait);
206 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
208 int ret = default_wake_function(wait, mode, sync, key);
211 list_del_init(&wait->task_list);
215 EXPORT_SYMBOL(autoremove_wake_function);
217 void __init fork_init(unsigned long mempages)
219 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
220 #ifndef ARCH_MIN_TASKALIGN
221 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
223 /* create a slab on which task_structs can be allocated */
225 kmem_cache_create("task_struct", sizeof(struct task_struct),
226 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
230 * The default maximum number of threads is set to a safe
231 * value: the thread structures can take up at most half
234 max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
236 * we need to allow at least 20 threads to boot a system
241 init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
242 init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
245 static struct task_struct *dup_task_struct(struct task_struct *orig)
247 struct task_struct *tsk;
248 struct thread_info *ti;
250 prepare_to_copy(orig);
252 tsk = alloc_task_struct();
256 ti = alloc_thread_info(tsk);
258 free_task_struct(tsk);
262 *ti = *orig->thread_info;
264 tsk->thread_info = ti;
267 /* One for us, one for whoever does the "release_task()" (usually parent) */
268 atomic_set(&tsk->usage,2);
273 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
275 struct vm_area_struct * mpnt, *tmp, **pprev;
276 struct rb_node **rb_link, *rb_parent;
278 unsigned long charge = 0;
279 struct mempolicy *pol;
281 down_write(&oldmm->mmap_sem);
282 flush_cache_mm(current->mm);
285 mm->mmap_cache = NULL;
286 mm->free_area_cache = TASK_UNMAPPED_BASE;
289 cpus_clear(mm->cpu_vm_mask);
291 rb_link = &mm->mm_rb.rb_node;
296 * Add it to the mmlist after the parent.
297 * Doing it this way means that we can order the list,
298 * and fork() won't mess up the ordering significantly.
299 * Add it first so that swapoff can see any swap entries.
301 spin_lock(&mmlist_lock);
302 list_add(&mm->mmlist, ¤t->mm->mmlist);
304 spin_unlock(&mmlist_lock);
306 for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
309 if(mpnt->vm_flags & VM_DONTCOPY)
311 if (mpnt->vm_flags & VM_ACCOUNT) {
312 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
313 if (security_vm_enough_memory(len))
317 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
321 pol = mpol_copy(vma_policy(mpnt));
322 retval = PTR_ERR(pol);
324 goto fail_nomem_policy;
325 vma_set_policy(tmp, pol);
326 tmp->vm_flags &= ~VM_LOCKED;
330 vma_prio_tree_init(tmp);
333 struct inode *inode = file->f_dentry->d_inode;
335 if (tmp->vm_flags & VM_DENYWRITE)
336 atomic_dec(&inode->i_writecount);
338 /* insert tmp into the share list, just after mpnt */
339 spin_lock(&file->f_mapping->i_mmap_lock);
340 flush_dcache_mmap_lock(file->f_mapping);
341 vma_prio_tree_add(tmp, mpnt);
342 flush_dcache_mmap_unlock(file->f_mapping);
343 spin_unlock(&file->f_mapping->i_mmap_lock);
347 * Link in the new vma and copy the page table entries:
348 * link in first so that swapoff can see swap entries,
349 * and try_to_unmap_one's find_vma find the new vma.
351 spin_lock(&mm->page_table_lock);
353 pprev = &tmp->vm_next;
355 __vma_link_rb(mm, tmp, rb_link, rb_parent);
356 rb_link = &tmp->vm_rb.rb_right;
357 rb_parent = &tmp->vm_rb;
360 retval = copy_page_range(mm, current->mm, tmp);
361 spin_unlock(&mm->page_table_lock);
363 if (tmp->vm_ops && tmp->vm_ops->open)
364 tmp->vm_ops->open(tmp);
372 flush_tlb_mm(current->mm);
373 up_write(&oldmm->mmap_sem);
376 kmem_cache_free(vm_area_cachep, tmp);
380 vm_unacct_memory(charge);
383 static inline int mm_alloc_pgd(struct mm_struct * mm)
385 mm->pgd = pgd_alloc(mm);
386 if (unlikely(!mm->pgd))
391 static inline void mm_free_pgd(struct mm_struct * mm)
396 #define dup_mmap(mm, oldmm) (0)
397 #define mm_alloc_pgd(mm) (0)
398 #define mm_free_pgd(mm)
399 #endif /* CONFIG_MMU */
401 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
404 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
405 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
407 #include <linux/init_task.h>
409 static struct mm_struct * mm_init(struct mm_struct * mm)
411 atomic_set(&mm->mm_users, 1);
412 atomic_set(&mm->mm_count, 1);
413 init_rwsem(&mm->mmap_sem);
414 mm->core_waiters = 0;
415 mm->page_table_lock = SPIN_LOCK_UNLOCKED;
416 mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
417 mm->ioctx_list = NULL;
418 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
419 mm->free_area_cache = TASK_UNMAPPED_BASE;
421 if (likely(!mm_alloc_pgd(mm))) {
423 #ifdef __HAVE_ARCH_MMAP_TOP
424 mm->mmap_top = mmap_top();
426 set_vx_info(&mm->mm_vx_info, current->vx_info);
434 * Allocate and initialize an mm_struct.
436 struct mm_struct * mm_alloc(void)
438 struct mm_struct * mm;
442 memset(mm, 0, sizeof(*mm));
449 * Called when the last reference to the mm
450 * is dropped: either by a lazy thread or by
451 * mmput. Free the page directory and the mm.
453 void fastcall __mmdrop(struct mm_struct *mm)
455 BUG_ON(mm == &init_mm);
458 clr_vx_info(&mm->mm_vx_info);
463 * Decrement the use count and release all resources for an mm.
465 void mmput(struct mm_struct *mm)
467 if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
468 list_del(&mm->mmlist);
470 spin_unlock(&mmlist_lock);
478 * Checks if the use count of an mm is non-zero and if so
479 * returns a reference to it after bumping up the use count.
480 * If the use count is zero, it means this mm is going away,
483 struct mm_struct *mmgrab(struct mm_struct *mm)
485 spin_lock(&mmlist_lock);
486 if (!atomic_read(&mm->mm_users))
489 atomic_inc(&mm->mm_users);
490 spin_unlock(&mmlist_lock);
494 /* Please note the differences between mmput and mm_release.
495 * mmput is called whenever we stop holding onto a mm_struct,
496 * error success whatever.
498 * mm_release is called after a mm_struct has been removed
499 * from the current process.
501 * This difference is important for error handling, when we
502 * only half set up a mm_struct for a new process and need to restore
503 * the old one. Because we mmput the new mm_struct before
504 * restoring the old one. . .
505 * Eric Biederman 10 January 1998
507 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
509 struct completion *vfork_done = tsk->vfork_done;
511 /* Get rid of any cached register state */
512 deactivate_mm(tsk, mm);
514 /* notify parent sleeping on vfork() */
516 tsk->vfork_done = NULL;
517 complete(vfork_done);
519 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
520 u32 __user * tidptr = tsk->clear_child_tid;
521 tsk->clear_child_tid = NULL;
524 * We don't check the error code - if userspace has
525 * not set up a proper pointer then tough luck.
528 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL);
532 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
534 struct mm_struct * mm, *oldmm;
537 tsk->min_flt = tsk->maj_flt = 0;
538 tsk->cmin_flt = tsk->cmaj_flt = 0;
539 tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
542 tsk->active_mm = NULL;
545 * Are we cloning a kernel thread?
547 * We need to steal a active VM for that..
553 if (clone_flags & CLONE_VM) {
554 atomic_inc(&oldmm->mm_users);
557 * There are cases where the PTL is held to ensure no
558 * new threads start up in user mode using an mm, which
559 * allows optimizing out ipis; the tlb_gather_mmu code
562 spin_unlock_wait(&oldmm->page_table_lock);
571 /* Copy the current MM stuff.. */
572 memcpy(mm, oldmm, sizeof(*mm));
573 mm->mm_vx_info = NULL;
577 if (init_new_context(tsk,mm))
580 retval = dup_mmap(mm, oldmm);
596 * If init_new_context() failed, we cannot use mmput() to free the mm
597 * because it calls destroy_context()
604 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
606 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
607 /* We don't need to lock fs - think why ;-) */
609 atomic_set(&fs->count, 1);
610 fs->lock = RW_LOCK_UNLOCKED;
611 fs->umask = old->umask;
612 read_lock(&old->lock);
613 fs->rootmnt = mntget(old->rootmnt);
614 fs->root = dget(old->root);
615 fs->pwdmnt = mntget(old->pwdmnt);
616 fs->pwd = dget(old->pwd);
618 fs->altrootmnt = mntget(old->altrootmnt);
619 fs->altroot = dget(old->altroot);
621 fs->altrootmnt = NULL;
624 read_unlock(&old->lock);
629 struct fs_struct *copy_fs_struct(struct fs_struct *old)
631 return __copy_fs_struct(old);
634 EXPORT_SYMBOL_GPL(copy_fs_struct);
636 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
638 if (clone_flags & CLONE_FS) {
639 atomic_inc(¤t->fs->count);
642 tsk->fs = __copy_fs_struct(current->fs);
648 static int count_open_files(struct files_struct *files, int size)
652 /* Find the last open fd */
653 for (i = size/(8*sizeof(long)); i > 0; ) {
654 if (files->open_fds->fds_bits[--i])
657 i = (i+1) * 8 * sizeof(long);
661 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
663 struct files_struct *oldf, *newf;
664 struct file **old_fds, **new_fds;
665 int open_files, nfds, size, i, error = 0;
668 * A background process may not have any files ...
670 oldf = current->files;
674 if (clone_flags & CLONE_FILES) {
675 atomic_inc(&oldf->count);
680 * Note: we may be using current for both targets (See exec.c)
681 * This works because we cache current->files (old) as oldf. Don't
686 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
690 atomic_set(&newf->count, 1);
692 newf->file_lock = SPIN_LOCK_UNLOCKED;
694 newf->max_fds = NR_OPEN_DEFAULT;
695 newf->max_fdset = __FD_SETSIZE;
696 newf->close_on_exec = &newf->close_on_exec_init;
697 newf->open_fds = &newf->open_fds_init;
698 newf->fd = &newf->fd_array[0];
700 /* We don't yet have the oldf readlock, but even if the old
701 fdset gets grown now, we'll only copy up to "size" fds */
702 size = oldf->max_fdset;
703 if (size > __FD_SETSIZE) {
705 spin_lock(&newf->file_lock);
706 error = expand_fdset(newf, size-1);
707 spin_unlock(&newf->file_lock);
711 spin_lock(&oldf->file_lock);
713 open_files = count_open_files(oldf, size);
716 * Check whether we need to allocate a larger fd array.
717 * Note: we're not a clone task, so the open count won't
720 nfds = NR_OPEN_DEFAULT;
721 if (open_files > nfds) {
722 spin_unlock(&oldf->file_lock);
724 spin_lock(&newf->file_lock);
725 error = expand_fd_array(newf, open_files-1);
726 spin_unlock(&newf->file_lock);
729 nfds = newf->max_fds;
730 spin_lock(&oldf->file_lock);
736 memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
737 memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
739 for (i = open_files; i != 0; i--) {
740 struct file *f = *old_fds++;
745 spin_unlock(&oldf->file_lock);
747 /* compute the remainder to be cleared */
748 size = (newf->max_fds - open_files) * sizeof(struct file *);
750 /* This is long word aligned thus could use a optimized version */
751 memset(new_fds, 0, size);
753 if (newf->max_fdset > open_files) {
754 int left = (newf->max_fdset-open_files)/8;
755 int start = open_files / (8 * sizeof(unsigned long));
757 memset(&newf->open_fds->fds_bits[start], 0, left);
758 memset(&newf->close_on_exec->fds_bits[start], 0, left);
767 free_fdset (newf->close_on_exec, newf->max_fdset);
768 free_fdset (newf->open_fds, newf->max_fdset);
769 kmem_cache_free(files_cachep, newf);
774 * Helper to unshare the files of the current task.
775 * We don't want to expose copy_files internals to
776 * the exec layer of the kernel.
779 int unshare_files(void)
781 struct files_struct *files = current->files;
787 /* This can race but the race causes us to copy when we don't
788 need to and drop the copy */
789 if(atomic_read(&files->count) == 1)
791 atomic_inc(&files->count);
794 rc = copy_files(0, current);
796 current->files = files;
800 EXPORT_SYMBOL(unshare_files);
802 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
804 struct sighand_struct *sig;
806 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
807 atomic_inc(¤t->sighand->count);
810 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
814 spin_lock_init(&sig->siglock);
815 atomic_set(&sig->count, 1);
816 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
820 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
822 struct signal_struct *sig;
824 if (clone_flags & CLONE_THREAD) {
825 atomic_inc(¤t->signal->count);
828 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
832 atomic_set(&sig->count, 1);
834 sig->group_exit_code = 0;
835 sig->group_exit_task = NULL;
836 sig->group_stop_count = 0;
837 sig->curr_target = NULL;
838 init_sigpending(&sig->shared_pending);
839 INIT_LIST_HEAD(&sig->posix_timers);
841 sig->tty = current->signal->tty;
842 sig->pgrp = process_group(current);
843 sig->session = current->signal->session;
844 sig->leader = 0; /* session leadership doesn't inherit */
845 sig->tty_old_pgrp = 0;
850 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
852 unsigned long new_flags = p->flags;
854 new_flags &= ~PF_SUPERPRIV;
855 new_flags |= PF_FORKNOEXEC;
856 if (!(clone_flags & CLONE_PTRACE))
858 p->flags = new_flags;
861 asmlinkage long sys_set_tid_address(int __user *tidptr)
863 current->clear_child_tid = tidptr;
869 * This creates a new process as a copy of the old one,
870 * but does not actually start it yet.
872 * It copies the registers, and all the appropriate
873 * parts of the process environment (as per the clone
874 * flags). The actual kick-off is left to the caller.
876 struct task_struct *copy_process(unsigned long clone_flags,
877 unsigned long stack_start,
878 struct pt_regs *regs,
879 unsigned long stack_size,
880 int __user *parent_tidptr,
881 int __user *child_tidptr)
884 struct task_struct *p = NULL;
888 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
889 return ERR_PTR(-EINVAL);
892 * Thread groups must share signals as well, and detached threads
893 * can only be started up within the thread group.
895 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
896 return ERR_PTR(-EINVAL);
899 * Shared signal handlers imply shared VM. By way of the above,
900 * thread groups also imply shared VM. Blocking this case allows
901 * for various simplifications in other code.
903 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
904 return ERR_PTR(-EINVAL);
906 retval = security_task_create(clone_flags);
912 p = dup_task_struct(current);
917 vxi = get_vx_info(current->vx_info);
918 nxi = get_nx_info(current->nx_info);
920 /* check vserver memory */
921 if (p->mm && !(clone_flags & CLONE_VM)) {
922 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
923 vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
927 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
928 if (!vx_rsspages_avail(p->mm, p->mm->rss))
933 if (vxi && (atomic_read(&vxi->limit.res[RLIMIT_NPROC])
934 >= vxi->limit.rlim[RLIMIT_NPROC]))
937 if (atomic_read(&p->user->processes) >=
938 p->rlim[RLIMIT_NPROC].rlim_cur) {
939 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
940 p->user != &root_user)
944 atomic_inc(&p->user->__count);
945 atomic_inc(&p->user->processes);
946 get_group_info(p->group_info);
949 * If multiple threads are within copy_process(), then this check
950 * triggers too late. This doesn't hurt, the check is only there
951 * to stop root fork bombs.
953 if (nr_threads >= max_threads)
954 goto bad_fork_cleanup_count;
956 if (!try_module_get(p->thread_info->exec_domain->module))
957 goto bad_fork_cleanup_count;
959 if (p->binfmt && !try_module_get(p->binfmt->module))
960 goto bad_fork_cleanup_put_domain;
963 copy_flags(clone_flags, p);
964 if (clone_flags & CLONE_IDLETASK)
967 p->pid = alloc_pidmap();
969 goto bad_fork_cleanup;
972 if (clone_flags & CLONE_PARENT_SETTID)
973 if (put_user(p->pid, parent_tidptr))
974 goto bad_fork_cleanup;
976 p->proc_dentry = NULL;
978 INIT_LIST_HEAD(&p->children);
979 INIT_LIST_HEAD(&p->sibling);
980 init_waitqueue_head(&p->wait_chldexit);
981 p->vfork_done = NULL;
982 spin_lock_init(&p->alloc_lock);
983 spin_lock_init(&p->proc_lock);
985 clear_tsk_thread_flag(p, TIF_SIGPENDING);
986 init_sigpending(&p->pending);
988 p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
989 p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
990 init_timer(&p->real_timer);
991 p->real_timer.data = (unsigned long) p;
993 p->utime = p->stime = 0;
994 p->cutime = p->cstime = 0;
995 p->lock_depth = -1; /* -1 = no lock */
996 p->start_time = get_jiffies_64();
998 p->io_context = NULL;
999 p->audit_context = NULL;
1001 p->mempolicy = mpol_copy(p->mempolicy);
1002 if (IS_ERR(p->mempolicy)) {
1003 retval = PTR_ERR(p->mempolicy);
1004 p->mempolicy = NULL;
1005 goto bad_fork_cleanup;
1010 if ((retval = security_task_alloc(p)))
1011 goto bad_fork_cleanup_policy;
1012 if ((retval = audit_alloc(p)))
1013 goto bad_fork_cleanup_security;
1014 /* copy all the process information */
1015 if ((retval = copy_semundo(clone_flags, p)))
1016 goto bad_fork_cleanup_audit;
1017 if ((retval = copy_files(clone_flags, p)))
1018 goto bad_fork_cleanup_semundo;
1019 if ((retval = copy_fs(clone_flags, p)))
1020 goto bad_fork_cleanup_files;
1021 if ((retval = copy_sighand(clone_flags, p)))
1022 goto bad_fork_cleanup_fs;
1023 if ((retval = copy_signal(clone_flags, p)))
1024 goto bad_fork_cleanup_sighand;
1025 if ((retval = copy_mm(clone_flags, p)))
1026 goto bad_fork_cleanup_signal;
1027 if ((retval = copy_namespace(clone_flags, p)))
1028 goto bad_fork_cleanup_mm;
1029 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1031 goto bad_fork_cleanup_namespace;
1033 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1035 * Clear TID on mm_release()?
1037 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1040 * Syscall tracing should be turned off in the child regardless
1043 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1045 /* Our parent execution domain becomes current domain
1046 These must match for thread signalling to apply */
1048 p->parent_exec_id = p->self_exec_id;
1050 /* ok, now we should be set up.. */
1051 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1052 p->pdeath_signal = 0;
1054 /* Perform scheduler related setup */
1058 * Ok, make it visible to the rest of the system.
1059 * We dont wake it up yet.
1062 p->group_leader = p;
1063 INIT_LIST_HEAD(&p->ptrace_children);
1064 INIT_LIST_HEAD(&p->ptrace_list);
1066 /* Need tasklist lock for parent etc handling! */
1067 write_lock_irq(&tasklist_lock);
1069 * Check for pending SIGKILL! The new thread should not be allowed
1070 * to slip out of an OOM kill. (or normal SIGKILL.)
1072 if (sigismember(¤t->pending.signal, SIGKILL)) {
1073 write_unlock_irq(&tasklist_lock);
1075 goto bad_fork_cleanup_namespace;
1078 /* CLONE_PARENT re-uses the old parent */
1079 if (clone_flags & CLONE_PARENT)
1080 p->real_parent = current->real_parent;
1082 p->real_parent = current;
1083 p->parent = p->real_parent;
1085 if (clone_flags & CLONE_THREAD) {
1086 spin_lock(¤t->sighand->siglock);
1088 * Important: if an exit-all has been started then
1089 * do not create this new thread - the whole thread
1090 * group is supposed to exit anyway.
1092 if (current->signal->group_exit) {
1093 spin_unlock(¤t->sighand->siglock);
1094 write_unlock_irq(&tasklist_lock);
1096 goto bad_fork_cleanup_namespace;
1098 p->tgid = current->tgid;
1099 p->group_leader = current->group_leader;
1101 if (current->signal->group_stop_count > 0) {
1103 * There is an all-stop in progress for the group.
1104 * We ourselves will stop as soon as we check signals.
1105 * Make the new thread part of that group stop too.
1107 current->signal->group_stop_count++;
1108 set_tsk_thread_flag(p, TIF_SIGPENDING);
1111 spin_unlock(¤t->sighand->siglock);
1115 if (p->ptrace & PT_PTRACED)
1116 __ptrace_link(p, current->parent);
1118 attach_pid(p, PIDTYPE_PID, p->pid);
1119 if (thread_group_leader(p)) {
1120 attach_pid(p, PIDTYPE_TGID, p->tgid);
1121 attach_pid(p, PIDTYPE_PGID, process_group(p));
1122 attach_pid(p, PIDTYPE_SID, p->signal->session);
1124 __get_cpu_var(process_counts)++;
1126 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1130 atomic_inc(&vxi->cacct.nr_threads);
1131 atomic_inc(&vxi->limit.res[RLIMIT_NPROC]);
1133 write_unlock_irq(&tasklist_lock);
1138 return ERR_PTR(retval);
1141 bad_fork_cleanup_namespace:
1143 bad_fork_cleanup_mm:
1146 mmdrop(p->active_mm);
1147 bad_fork_cleanup_signal:
1149 bad_fork_cleanup_sighand:
1151 bad_fork_cleanup_fs:
1152 exit_fs(p); /* blocking */
1153 bad_fork_cleanup_files:
1154 exit_files(p); /* blocking */
1155 bad_fork_cleanup_semundo:
1157 bad_fork_cleanup_audit:
1159 bad_fork_cleanup_security:
1160 security_task_free(p);
1161 bad_fork_cleanup_policy:
1163 mpol_free(p->mempolicy);
1167 free_pidmap(p->pid);
1169 module_put(p->binfmt->module);
1170 bad_fork_cleanup_put_domain:
1171 module_put(p->thread_info->exec_domain->module);
1172 bad_fork_cleanup_count:
1173 put_group_info(p->group_info);
1174 atomic_dec(&p->user->processes);
1181 static inline int fork_traceflag (unsigned clone_flags)
1183 if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1185 else if (clone_flags & CLONE_VFORK) {
1186 if (current->ptrace & PT_TRACE_VFORK)
1187 return PTRACE_EVENT_VFORK;
1188 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1189 if (current->ptrace & PT_TRACE_CLONE)
1190 return PTRACE_EVENT_CLONE;
1191 } else if (current->ptrace & PT_TRACE_FORK)
1192 return PTRACE_EVENT_FORK;
1198 * Ok, this is the main fork-routine.
1200 * It copies the process, and if successful kick-starts
1201 * it and waits for it to finish using the VM if required.
1203 long do_fork(unsigned long clone_flags,
1204 unsigned long stack_start,
1205 struct pt_regs *regs,
1206 unsigned long stack_size,
1207 int __user *parent_tidptr,
1208 int __user *child_tidptr)
1210 struct task_struct *p;
1214 if (unlikely(current->ptrace)) {
1215 trace = fork_traceflag (clone_flags);
1217 clone_flags |= CLONE_PTRACE;
1220 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1222 * Do this prior waking up the new thread - the thread pointer
1223 * might get invalid after that point, if the thread exits quickly.
1225 pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1228 struct completion vfork;
1230 if (clone_flags & CLONE_VFORK) {
1231 p->vfork_done = &vfork;
1232 init_completion(&vfork);
1235 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1237 * We'll start up with an immediate SIGSTOP.
1239 sigaddset(&p->pending.signal, SIGSTOP);
1240 set_tsk_thread_flag(p, TIF_SIGPENDING);
1243 if (!(clone_flags & CLONE_STOPPED)) {
1245 * Do the wakeup last. On SMP we treat fork() and
1246 * CLONE_VM separately, because fork() has already
1247 * created cache footprint on this CPU (due to
1248 * copying the pagetables), hence migration would
1249 * probably be costy. Threads on the other hand
1250 * have less traction to the current CPU, and if
1251 * there's an imbalance then the scheduler can
1252 * migrate this fresh thread now, before it
1253 * accumulates a larger cache footprint:
1255 if (clone_flags & CLONE_VM)
1256 wake_up_forked_thread(p);
1258 wake_up_forked_process(p);
1260 int cpu = get_cpu();
1262 p->state = TASK_STOPPED;
1263 if (cpu_is_offline(task_cpu(p)))
1264 set_task_cpu(p, cpu);
1270 if (unlikely (trace)) {
1271 current->ptrace_message = pid;
1272 ptrace_notify ((trace << 8) | SIGTRAP);
1275 if (clone_flags & CLONE_VFORK) {
1276 wait_for_completion(&vfork);
1277 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1278 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1281 * Let the child process run first, to avoid most of the
1282 * COW overhead when the child exec()s afterwards.
1289 /* SLAB cache for signal_struct structures (tsk->signal) */
1290 kmem_cache_t *signal_cachep;
1292 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1293 kmem_cache_t *sighand_cachep;
1295 /* SLAB cache for files_struct structures (tsk->files) */
1296 kmem_cache_t *files_cachep;
1298 /* SLAB cache for fs_struct structures (tsk->fs) */
1299 kmem_cache_t *fs_cachep;
1301 /* SLAB cache for vm_area_struct structures */
1302 kmem_cache_t *vm_area_cachep;
1304 /* SLAB cache for mm_struct structures (tsk->mm) */
1305 kmem_cache_t *mm_cachep;
1307 void __init proc_caches_init(void)
1309 sighand_cachep = kmem_cache_create("sighand_cache",
1310 sizeof(struct sighand_struct), 0,
1311 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1312 signal_cachep = kmem_cache_create("signal_cache",
1313 sizeof(struct signal_struct), 0,
1314 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1315 files_cachep = kmem_cache_create("files_cache",
1316 sizeof(struct files_struct), 0,
1317 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1318 fs_cachep = kmem_cache_create("fs_cache",
1319 sizeof(struct fs_struct), 0,
1320 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1321 vm_area_cachep = kmem_cache_create("vm_area_struct",
1322 sizeof(struct vm_area_struct), 0,
1323 SLAB_PANIC, NULL, NULL);
1324 mm_cachep = kmem_cache_create("mm_struct",
1325 sizeof(struct mm_struct), 0,
1326 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);