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 vxdprintk("freeing up task %p\n", tsk);
84 clr_vx_info(&tsk->vx_info);
85 clr_nx_info(&tsk->nx_info);
86 free_task_struct(tsk);
89 void __put_task_struct(struct task_struct *tsk)
91 WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
92 WARN_ON(atomic_read(&tsk->usage));
93 WARN_ON(tsk == current);
95 if (unlikely(tsk->audit_context))
97 security_task_free(tsk);
99 put_group_info(tsk->group_info);
103 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
107 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
108 spin_lock_irqsave(&q->lock, flags);
109 __add_wait_queue(q, wait);
110 spin_unlock_irqrestore(&q->lock, flags);
113 EXPORT_SYMBOL(add_wait_queue);
115 void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
119 wait->flags |= WQ_FLAG_EXCLUSIVE;
120 spin_lock_irqsave(&q->lock, flags);
121 __add_wait_queue_tail(q, wait);
122 spin_unlock_irqrestore(&q->lock, flags);
125 EXPORT_SYMBOL(add_wait_queue_exclusive);
127 void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
131 spin_lock_irqsave(&q->lock, flags);
132 __remove_wait_queue(q, wait);
133 spin_unlock_irqrestore(&q->lock, flags);
136 EXPORT_SYMBOL(remove_wait_queue);
140 * Note: we use "set_current_state()" _after_ the wait-queue add,
141 * because we need a memory barrier there on SMP, so that any
142 * wake-function that tests for the wait-queue being active
143 * will be guaranteed to see waitqueue addition _or_ subsequent
144 * tests in this thread will see the wakeup having taken place.
146 * The spin_unlock() itself is semi-permeable and only protects
147 * one way (it only protects stuff inside the critical region and
148 * stops them from bleeding out - it would still allow subsequent
149 * loads to move into the the critical region).
151 void fastcall prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
155 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
156 spin_lock_irqsave(&q->lock, flags);
157 if (list_empty(&wait->task_list))
158 __add_wait_queue(q, 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);
174 set_current_state(state);
175 spin_unlock_irqrestore(&q->lock, flags);
178 EXPORT_SYMBOL(prepare_to_wait_exclusive);
180 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
184 __set_current_state(TASK_RUNNING);
186 * We can check for list emptiness outside the lock
188 * - we use the "careful" check that verifies both
189 * the next and prev pointers, so that there cannot
190 * be any half-pending updates in progress on other
191 * CPU's that we haven't seen yet (and that might
192 * still change the stack area.
194 * - all other users take the lock (ie we can only
195 * have _one_ other CPU that looks at or modifies
198 if (!list_empty_careful(&wait->task_list)) {
199 spin_lock_irqsave(&q->lock, flags);
200 list_del_init(&wait->task_list);
201 spin_unlock_irqrestore(&q->lock, flags);
205 EXPORT_SYMBOL(finish_wait);
207 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
209 int ret = default_wake_function(wait, mode, sync, key);
212 list_del_init(&wait->task_list);
216 EXPORT_SYMBOL(autoremove_wake_function);
218 void __init fork_init(unsigned long mempages)
220 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
221 #ifndef ARCH_MIN_TASKALIGN
222 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
224 /* create a slab on which task_structs can be allocated */
226 kmem_cache_create("task_struct", sizeof(struct task_struct),
227 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
231 * The default maximum number of threads is set to a safe
232 * value: the thread structures can take up at most half
235 max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
237 * we need to allow at least 20 threads to boot a system
242 init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
243 init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
246 static struct task_struct *dup_task_struct(struct task_struct *orig)
248 struct task_struct *tsk;
249 struct thread_info *ti;
251 prepare_to_copy(orig);
253 tsk = alloc_task_struct();
257 ti = alloc_thread_info(tsk);
259 free_task_struct(tsk);
263 *ti = *orig->thread_info;
265 tsk->thread_info = ti;
268 /* One for us, one for whoever does the "release_task()" (usually parent) */
269 atomic_set(&tsk->usage,2);
274 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
276 struct vm_area_struct * mpnt, *tmp, **pprev;
277 struct rb_node **rb_link, *rb_parent;
279 unsigned long charge = 0;
280 struct mempolicy *pol;
282 down_write(&oldmm->mmap_sem);
283 flush_cache_mm(current->mm);
286 mm->mmap_cache = NULL;
287 mm->free_area_cache = TASK_UNMAPPED_BASE;
290 cpus_clear(mm->cpu_vm_mask);
292 rb_link = &mm->mm_rb.rb_node;
297 * Add it to the mmlist after the parent.
298 * Doing it this way means that we can order the list,
299 * and fork() won't mess up the ordering significantly.
300 * Add it first so that swapoff can see any swap entries.
302 spin_lock(&mmlist_lock);
303 list_add(&mm->mmlist, ¤t->mm->mmlist);
305 spin_unlock(&mmlist_lock);
307 for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
310 if(mpnt->vm_flags & VM_DONTCOPY)
312 if (mpnt->vm_flags & VM_ACCOUNT) {
313 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
314 if (security_vm_enough_memory(len))
318 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
322 pol = mpol_copy(vma_policy(mpnt));
323 retval = PTR_ERR(pol);
325 goto fail_nomem_policy;
326 vma_set_policy(tmp, pol);
327 tmp->vm_flags &= ~VM_LOCKED;
331 vma_prio_tree_init(tmp);
334 struct inode *inode = file->f_dentry->d_inode;
336 if (tmp->vm_flags & VM_DENYWRITE)
337 atomic_dec(&inode->i_writecount);
339 /* insert tmp into the share list, just after mpnt */
340 spin_lock(&file->f_mapping->i_mmap_lock);
341 flush_dcache_mmap_lock(file->f_mapping);
342 vma_prio_tree_add(tmp, mpnt);
343 flush_dcache_mmap_unlock(file->f_mapping);
344 spin_unlock(&file->f_mapping->i_mmap_lock);
348 * Link in the new vma and copy the page table entries:
349 * link in first so that swapoff can see swap entries,
350 * and try_to_unmap_one's find_vma find the new vma.
352 spin_lock(&mm->page_table_lock);
354 pprev = &tmp->vm_next;
356 __vma_link_rb(mm, tmp, rb_link, rb_parent);
357 rb_link = &tmp->vm_rb.rb_right;
358 rb_parent = &tmp->vm_rb;
361 retval = copy_page_range(mm, current->mm, tmp);
362 spin_unlock(&mm->page_table_lock);
364 if (tmp->vm_ops && tmp->vm_ops->open)
365 tmp->vm_ops->open(tmp);
373 flush_tlb_mm(current->mm);
374 up_write(&oldmm->mmap_sem);
377 kmem_cache_free(vm_area_cachep, tmp);
381 vm_unacct_memory(charge);
384 static inline int mm_alloc_pgd(struct mm_struct * mm)
386 mm->pgd = pgd_alloc(mm);
387 if (unlikely(!mm->pgd))
392 static inline void mm_free_pgd(struct mm_struct * mm)
397 #define dup_mmap(mm, oldmm) (0)
398 #define mm_alloc_pgd(mm) (0)
399 #define mm_free_pgd(mm)
400 #endif /* CONFIG_MMU */
402 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
405 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
406 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
408 #include <linux/init_task.h>
410 static struct mm_struct * mm_init(struct mm_struct * mm)
412 atomic_set(&mm->mm_users, 1);
413 atomic_set(&mm->mm_count, 1);
414 init_rwsem(&mm->mmap_sem);
415 mm->core_waiters = 0;
416 mm->page_table_lock = SPIN_LOCK_UNLOCKED;
417 mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
418 mm->ioctx_list = NULL;
419 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
420 mm->free_area_cache = TASK_UNMAPPED_BASE;
422 if (likely(!mm_alloc_pgd(mm))) {
424 #ifdef __HAVE_ARCH_MMAP_TOP
425 mm->mmap_top = mmap_top();
427 set_vx_info(&mm->mm_vx_info, current->vx_info);
435 * Allocate and initialize an mm_struct.
437 struct mm_struct * mm_alloc(void)
439 struct mm_struct * mm;
443 memset(mm, 0, sizeof(*mm));
450 * Called when the last reference to the mm
451 * is dropped: either by a lazy thread or by
452 * mmput. Free the page directory and the mm.
454 void fastcall __mmdrop(struct mm_struct *mm)
456 BUG_ON(mm == &init_mm);
459 clr_vx_info(&mm->mm_vx_info);
464 * Decrement the use count and release all resources for an mm.
466 void mmput(struct mm_struct *mm)
468 if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
469 list_del(&mm->mmlist);
471 spin_unlock(&mmlist_lock);
479 * Checks if the use count of an mm is non-zero and if so
480 * returns a reference to it after bumping up the use count.
481 * If the use count is zero, it means this mm is going away,
484 struct mm_struct *mmgrab(struct mm_struct *mm)
486 spin_lock(&mmlist_lock);
487 if (!atomic_read(&mm->mm_users))
490 atomic_inc(&mm->mm_users);
491 spin_unlock(&mmlist_lock);
495 /* Please note the differences between mmput and mm_release.
496 * mmput is called whenever we stop holding onto a mm_struct,
497 * error success whatever.
499 * mm_release is called after a mm_struct has been removed
500 * from the current process.
502 * This difference is important for error handling, when we
503 * only half set up a mm_struct for a new process and need to restore
504 * the old one. Because we mmput the new mm_struct before
505 * restoring the old one. . .
506 * Eric Biederman 10 January 1998
508 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
510 struct completion *vfork_done = tsk->vfork_done;
512 /* Get rid of any cached register state */
513 deactivate_mm(tsk, mm);
515 /* notify parent sleeping on vfork() */
517 tsk->vfork_done = NULL;
518 complete(vfork_done);
520 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
521 u32 __user * tidptr = tsk->clear_child_tid;
522 tsk->clear_child_tid = NULL;
525 * We don't check the error code - if userspace has
526 * not set up a proper pointer then tough luck.
529 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL);
533 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
535 struct mm_struct * mm, *oldmm;
538 tsk->min_flt = tsk->maj_flt = 0;
539 tsk->cmin_flt = tsk->cmaj_flt = 0;
540 tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
543 tsk->active_mm = NULL;
546 * Are we cloning a kernel thread?
548 * We need to steal a active VM for that..
554 if (clone_flags & CLONE_VM) {
555 atomic_inc(&oldmm->mm_users);
558 * There are cases where the PTL is held to ensure no
559 * new threads start up in user mode using an mm, which
560 * allows optimizing out ipis; the tlb_gather_mmu code
563 spin_unlock_wait(&oldmm->page_table_lock);
572 /* Copy the current MM stuff.. */
573 memcpy(mm, oldmm, sizeof(*mm));
574 mm->mm_vx_info = NULL;
578 if (init_new_context(tsk,mm))
581 retval = dup_mmap(mm, oldmm);
597 * If init_new_context() failed, we cannot use mmput() to free the mm
598 * because it calls destroy_context()
605 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
607 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
608 /* We don't need to lock fs - think why ;-) */
610 atomic_set(&fs->count, 1);
611 fs->lock = RW_LOCK_UNLOCKED;
612 fs->umask = old->umask;
613 read_lock(&old->lock);
614 fs->rootmnt = mntget(old->rootmnt);
615 fs->root = dget(old->root);
616 fs->pwdmnt = mntget(old->pwdmnt);
617 fs->pwd = dget(old->pwd);
619 fs->altrootmnt = mntget(old->altrootmnt);
620 fs->altroot = dget(old->altroot);
622 fs->altrootmnt = NULL;
625 read_unlock(&old->lock);
630 struct fs_struct *copy_fs_struct(struct fs_struct *old)
632 return __copy_fs_struct(old);
635 EXPORT_SYMBOL_GPL(copy_fs_struct);
637 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
639 if (clone_flags & CLONE_FS) {
640 atomic_inc(¤t->fs->count);
643 tsk->fs = __copy_fs_struct(current->fs);
649 static int count_open_files(struct files_struct *files, int size)
653 /* Find the last open fd */
654 for (i = size/(8*sizeof(long)); i > 0; ) {
655 if (files->open_fds->fds_bits[--i])
658 i = (i+1) * 8 * sizeof(long);
662 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
664 struct files_struct *oldf, *newf;
665 struct file **old_fds, **new_fds;
666 int open_files, nfds, size, i, error = 0;
669 * A background process may not have any files ...
671 oldf = current->files;
675 if (clone_flags & CLONE_FILES) {
676 atomic_inc(&oldf->count);
681 * Note: we may be using current for both targets (See exec.c)
682 * This works because we cache current->files (old) as oldf. Don't
687 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
691 atomic_set(&newf->count, 1);
693 newf->file_lock = SPIN_LOCK_UNLOCKED;
695 newf->max_fds = NR_OPEN_DEFAULT;
696 newf->max_fdset = __FD_SETSIZE;
697 newf->close_on_exec = &newf->close_on_exec_init;
698 newf->open_fds = &newf->open_fds_init;
699 newf->fd = &newf->fd_array[0];
701 /* We don't yet have the oldf readlock, but even if the old
702 fdset gets grown now, we'll only copy up to "size" fds */
703 size = oldf->max_fdset;
704 if (size > __FD_SETSIZE) {
706 spin_lock(&newf->file_lock);
707 error = expand_fdset(newf, size-1);
708 spin_unlock(&newf->file_lock);
712 spin_lock(&oldf->file_lock);
714 open_files = count_open_files(oldf, size);
717 * Check whether we need to allocate a larger fd array.
718 * Note: we're not a clone task, so the open count won't
721 nfds = NR_OPEN_DEFAULT;
722 if (open_files > nfds) {
723 spin_unlock(&oldf->file_lock);
725 spin_lock(&newf->file_lock);
726 error = expand_fd_array(newf, open_files-1);
727 spin_unlock(&newf->file_lock);
730 nfds = newf->max_fds;
731 spin_lock(&oldf->file_lock);
737 memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
738 memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
740 for (i = open_files; i != 0; i--) {
741 struct file *f = *old_fds++;
746 spin_unlock(&oldf->file_lock);
748 /* compute the remainder to be cleared */
749 size = (newf->max_fds - open_files) * sizeof(struct file *);
751 /* This is long word aligned thus could use a optimized version */
752 memset(new_fds, 0, size);
754 if (newf->max_fdset > open_files) {
755 int left = (newf->max_fdset-open_files)/8;
756 int start = open_files / (8 * sizeof(unsigned long));
758 memset(&newf->open_fds->fds_bits[start], 0, left);
759 memset(&newf->close_on_exec->fds_bits[start], 0, left);
768 free_fdset (newf->close_on_exec, newf->max_fdset);
769 free_fdset (newf->open_fds, newf->max_fdset);
770 kmem_cache_free(files_cachep, newf);
775 * Helper to unshare the files of the current task.
776 * We don't want to expose copy_files internals to
777 * the exec layer of the kernel.
780 int unshare_files(void)
782 struct files_struct *files = current->files;
788 /* This can race but the race causes us to copy when we don't
789 need to and drop the copy */
790 if(atomic_read(&files->count) == 1)
792 atomic_inc(&files->count);
795 rc = copy_files(0, current);
797 current->files = files;
801 EXPORT_SYMBOL(unshare_files);
803 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
805 struct sighand_struct *sig;
807 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
808 atomic_inc(¤t->sighand->count);
811 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
815 spin_lock_init(&sig->siglock);
816 atomic_set(&sig->count, 1);
817 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
821 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
823 struct signal_struct *sig;
825 if (clone_flags & CLONE_THREAD) {
826 atomic_inc(¤t->signal->count);
829 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
833 atomic_set(&sig->count, 1);
835 sig->group_exit_code = 0;
836 sig->group_exit_task = NULL;
837 sig->group_stop_count = 0;
838 sig->curr_target = NULL;
839 init_sigpending(&sig->shared_pending);
840 INIT_LIST_HEAD(&sig->posix_timers);
842 sig->tty = current->signal->tty;
843 sig->pgrp = process_group(current);
844 sig->session = current->signal->session;
845 sig->leader = 0; /* session leadership doesn't inherit */
846 sig->tty_old_pgrp = 0;
851 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
853 unsigned long new_flags = p->flags;
855 new_flags &= ~PF_SUPERPRIV;
856 new_flags |= PF_FORKNOEXEC;
857 if (!(clone_flags & CLONE_PTRACE))
859 p->flags = new_flags;
862 asmlinkage long sys_set_tid_address(int __user *tidptr)
864 current->clear_child_tid = tidptr;
870 * This creates a new process as a copy of the old one,
871 * but does not actually start it yet.
873 * It copies the registers, and all the appropriate
874 * parts of the process environment (as per the clone
875 * flags). The actual kick-off is left to the caller.
877 struct task_struct *copy_process(unsigned long clone_flags,
878 unsigned long stack_start,
879 struct pt_regs *regs,
880 unsigned long stack_size,
881 int __user *parent_tidptr,
882 int __user *child_tidptr)
885 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);
918 set_vx_info(&p->vx_info, current->vx_info);
920 set_nx_info(&p->nx_info, current->nx_info);
922 /* check vserver memory */
923 if (p->mm && !(clone_flags & CLONE_VM)) {
924 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
925 vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
929 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
930 if (!vx_rsspages_avail(p->mm, p->mm->rss))
935 vxi = current->vx_info;
936 if (vxi && (atomic_read(&vxi->limit.res[RLIMIT_NPROC])
937 >= vxi->limit.rlim[RLIMIT_NPROC]))
940 if (atomic_read(&p->user->processes) >=
941 p->rlim[RLIMIT_NPROC].rlim_cur) {
942 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
943 p->user != &root_user)
947 atomic_inc(&p->user->__count);
948 atomic_inc(&p->user->processes);
949 get_group_info(p->group_info);
952 * If multiple threads are within copy_process(), then this check
953 * triggers too late. This doesn't hurt, the check is only there
954 * to stop root fork bombs.
956 if (nr_threads >= max_threads)
957 goto bad_fork_cleanup_count;
959 if (!try_module_get(p->thread_info->exec_domain->module))
960 goto bad_fork_cleanup_count;
962 if (p->binfmt && !try_module_get(p->binfmt->module))
963 goto bad_fork_cleanup_put_domain;
966 copy_flags(clone_flags, p);
967 if (clone_flags & CLONE_IDLETASK)
970 p->pid = alloc_pidmap();
972 goto bad_fork_cleanup;
975 if (clone_flags & CLONE_PARENT_SETTID)
976 if (put_user(p->pid, parent_tidptr))
977 goto bad_fork_cleanup;
979 p->proc_dentry = NULL;
981 INIT_LIST_HEAD(&p->children);
982 INIT_LIST_HEAD(&p->sibling);
983 init_waitqueue_head(&p->wait_chldexit);
984 p->vfork_done = NULL;
985 spin_lock_init(&p->alloc_lock);
986 spin_lock_init(&p->proc_lock);
988 clear_tsk_thread_flag(p, TIF_SIGPENDING);
989 init_sigpending(&p->pending);
991 p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
992 p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
993 init_timer(&p->real_timer);
994 p->real_timer.data = (unsigned long) p;
996 p->utime = p->stime = 0;
997 p->cutime = p->cstime = 0;
998 p->lock_depth = -1; /* -1 = no lock */
999 p->start_time = get_jiffies_64();
1001 p->io_context = NULL;
1002 p->audit_context = NULL;
1004 p->mempolicy = mpol_copy(p->mempolicy);
1005 if (IS_ERR(p->mempolicy)) {
1006 retval = PTR_ERR(p->mempolicy);
1007 p->mempolicy = NULL;
1008 goto bad_fork_cleanup;
1013 if ((retval = security_task_alloc(p)))
1014 goto bad_fork_cleanup_policy;
1015 if ((retval = audit_alloc(p)))
1016 goto bad_fork_cleanup_security;
1017 /* copy all the process information */
1018 if ((retval = copy_semundo(clone_flags, p)))
1019 goto bad_fork_cleanup_audit;
1020 if ((retval = copy_files(clone_flags, p)))
1021 goto bad_fork_cleanup_semundo;
1022 if ((retval = copy_fs(clone_flags, p)))
1023 goto bad_fork_cleanup_files;
1024 if ((retval = copy_sighand(clone_flags, p)))
1025 goto bad_fork_cleanup_fs;
1026 if ((retval = copy_signal(clone_flags, p)))
1027 goto bad_fork_cleanup_sighand;
1028 if ((retval = copy_mm(clone_flags, p)))
1029 goto bad_fork_cleanup_signal;
1030 if ((retval = copy_namespace(clone_flags, p)))
1031 goto bad_fork_cleanup_mm;
1032 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1034 goto bad_fork_cleanup_namespace;
1036 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1038 * Clear TID on mm_release()?
1040 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1043 * Syscall tracing should be turned off in the child regardless
1046 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1048 /* Our parent execution domain becomes current domain
1049 These must match for thread signalling to apply */
1051 p->parent_exec_id = p->self_exec_id;
1053 /* ok, now we should be set up.. */
1054 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1055 p->pdeath_signal = 0;
1057 /* Perform scheduler related setup */
1061 * Ok, make it visible to the rest of the system.
1062 * We dont wake it up yet.
1065 p->group_leader = p;
1066 INIT_LIST_HEAD(&p->ptrace_children);
1067 INIT_LIST_HEAD(&p->ptrace_list);
1069 /* Need tasklist lock for parent etc handling! */
1070 write_lock_irq(&tasklist_lock);
1072 * Check for pending SIGKILL! The new thread should not be allowed
1073 * to slip out of an OOM kill. (or normal SIGKILL.)
1075 if (sigismember(¤t->pending.signal, SIGKILL)) {
1076 write_unlock_irq(&tasklist_lock);
1078 goto bad_fork_cleanup_namespace;
1081 /* CLONE_PARENT re-uses the old parent */
1082 if (clone_flags & CLONE_PARENT)
1083 p->real_parent = current->real_parent;
1085 p->real_parent = current;
1086 p->parent = p->real_parent;
1088 if (clone_flags & CLONE_THREAD) {
1089 spin_lock(¤t->sighand->siglock);
1091 * Important: if an exit-all has been started then
1092 * do not create this new thread - the whole thread
1093 * group is supposed to exit anyway.
1095 if (current->signal->group_exit) {
1096 spin_unlock(¤t->sighand->siglock);
1097 write_unlock_irq(&tasklist_lock);
1099 goto bad_fork_cleanup_namespace;
1101 p->tgid = current->tgid;
1102 p->group_leader = current->group_leader;
1104 if (current->signal->group_stop_count > 0) {
1106 * There is an all-stop in progress for the group.
1107 * We ourselves will stop as soon as we check signals.
1108 * Make the new thread part of that group stop too.
1110 current->signal->group_stop_count++;
1111 set_tsk_thread_flag(p, TIF_SIGPENDING);
1114 spin_unlock(¤t->sighand->siglock);
1118 if (p->ptrace & PT_PTRACED)
1119 __ptrace_link(p, current->parent);
1121 attach_pid(p, PIDTYPE_PID, p->pid);
1122 if (thread_group_leader(p)) {
1123 attach_pid(p, PIDTYPE_TGID, p->tgid);
1124 attach_pid(p, PIDTYPE_PGID, process_group(p));
1125 attach_pid(p, PIDTYPE_SID, p->signal->session);
1127 __get_cpu_var(process_counts)++;
1129 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1133 atomic_inc(&vxi->cacct.nr_threads);
1134 atomic_inc(&vxi->limit.res[RLIMIT_NPROC]);
1136 write_unlock_irq(&tasklist_lock);
1141 return ERR_PTR(retval);
1144 bad_fork_cleanup_namespace:
1146 bad_fork_cleanup_mm:
1149 mmdrop(p->active_mm);
1150 bad_fork_cleanup_signal:
1152 bad_fork_cleanup_sighand:
1154 bad_fork_cleanup_fs:
1155 exit_fs(p); /* blocking */
1156 bad_fork_cleanup_files:
1157 exit_files(p); /* blocking */
1158 bad_fork_cleanup_semundo:
1160 bad_fork_cleanup_audit:
1162 bad_fork_cleanup_security:
1163 security_task_free(p);
1164 bad_fork_cleanup_policy:
1166 mpol_free(p->mempolicy);
1170 free_pidmap(p->pid);
1172 module_put(p->binfmt->module);
1173 bad_fork_cleanup_put_domain:
1174 module_put(p->thread_info->exec_domain->module);
1175 bad_fork_cleanup_count:
1176 put_group_info(p->group_info);
1177 atomic_dec(&p->user->processes);
1184 static inline int fork_traceflag (unsigned clone_flags)
1186 if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1188 else if (clone_flags & CLONE_VFORK) {
1189 if (current->ptrace & PT_TRACE_VFORK)
1190 return PTRACE_EVENT_VFORK;
1191 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1192 if (current->ptrace & PT_TRACE_CLONE)
1193 return PTRACE_EVENT_CLONE;
1194 } else if (current->ptrace & PT_TRACE_FORK)
1195 return PTRACE_EVENT_FORK;
1201 * Ok, this is the main fork-routine.
1203 * It copies the process, and if successful kick-starts
1204 * it and waits for it to finish using the VM if required.
1206 long do_fork(unsigned long clone_flags,
1207 unsigned long stack_start,
1208 struct pt_regs *regs,
1209 unsigned long stack_size,
1210 int __user *parent_tidptr,
1211 int __user *child_tidptr)
1213 struct task_struct *p;
1217 if (unlikely(current->ptrace)) {
1218 trace = fork_traceflag (clone_flags);
1220 clone_flags |= CLONE_PTRACE;
1223 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1225 * Do this prior waking up the new thread - the thread pointer
1226 * might get invalid after that point, if the thread exits quickly.
1228 pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1231 struct completion vfork;
1233 if (clone_flags & CLONE_VFORK) {
1234 p->vfork_done = &vfork;
1235 init_completion(&vfork);
1238 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1240 * We'll start up with an immediate SIGSTOP.
1242 sigaddset(&p->pending.signal, SIGSTOP);
1243 set_tsk_thread_flag(p, TIF_SIGPENDING);
1246 if (!(clone_flags & CLONE_STOPPED)) {
1248 * Do the wakeup last. On SMP we treat fork() and
1249 * CLONE_VM separately, because fork() has already
1250 * created cache footprint on this CPU (due to
1251 * copying the pagetables), hence migration would
1252 * probably be costy. Threads on the other hand
1253 * have less traction to the current CPU, and if
1254 * there's an imbalance then the scheduler can
1255 * migrate this fresh thread now, before it
1256 * accumulates a larger cache footprint:
1258 if (clone_flags & CLONE_VM)
1259 wake_up_forked_thread(p);
1261 wake_up_forked_process(p);
1263 int cpu = get_cpu();
1265 p->state = TASK_STOPPED;
1266 if (cpu_is_offline(task_cpu(p)))
1267 set_task_cpu(p, cpu);
1273 if (unlikely (trace)) {
1274 current->ptrace_message = pid;
1275 ptrace_notify ((trace << 8) | SIGTRAP);
1278 if (clone_flags & CLONE_VFORK) {
1279 wait_for_completion(&vfork);
1280 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1281 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1284 * Let the child process run first, to avoid most of the
1285 * COW overhead when the child exec()s afterwards.
1292 /* SLAB cache for signal_struct structures (tsk->signal) */
1293 kmem_cache_t *signal_cachep;
1295 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1296 kmem_cache_t *sighand_cachep;
1298 /* SLAB cache for files_struct structures (tsk->files) */
1299 kmem_cache_t *files_cachep;
1301 /* SLAB cache for fs_struct structures (tsk->fs) */
1302 kmem_cache_t *fs_cachep;
1304 /* SLAB cache for vm_area_struct structures */
1305 kmem_cache_t *vm_area_cachep;
1307 /* SLAB cache for mm_struct structures (tsk->mm) */
1308 kmem_cache_t *mm_cachep;
1310 void __init proc_caches_init(void)
1312 sighand_cachep = kmem_cache_create("sighand_cache",
1313 sizeof(struct sighand_struct), 0,
1314 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1315 signal_cachep = kmem_cache_create("signal_cache",
1316 sizeof(struct signal_struct), 0,
1317 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1318 files_cachep = kmem_cache_create("files_cache",
1319 sizeof(struct files_struct), 0,
1320 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1321 fs_cachep = kmem_cache_create("fs_cache",
1322 sizeof(struct fs_struct), 0,
1323 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1324 vm_area_cachep = kmem_cache_create("vm_area_struct",
1325 sizeof(struct vm_area_struct), 0,
1326 SLAB_PANIC, NULL, NULL);
1327 mm_cachep = kmem_cache_create("mm_struct",
1328 sizeof(struct mm_struct), 0,
1329 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);