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/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.h>
31 #include <linux/cpu.h>
32 #include <linux/cpuset.h>
33 #include <linux/security.h>
34 #include <linux/swap.h>
35 #include <linux/syscalls.h>
36 #include <linux/jiffies.h>
37 #include <linux/futex.h>
38 #include <linux/rcupdate.h>
39 #include <linux/ptrace.h>
40 #include <linux/mount.h>
41 #include <linux/audit.h>
42 #include <linux/profile.h>
43 #include <linux/rmap.h>
44 #include <linux/acct.h>
45 #include <linux/cn_proc.h>
47 #include <asm/pgtable.h>
48 #include <asm/pgalloc.h>
49 #include <asm/uaccess.h>
50 #include <asm/mmu_context.h>
51 #include <asm/cacheflush.h>
52 #include <asm/tlbflush.h>
55 * Protected counters by write_lock_irq(&tasklist_lock)
57 unsigned long total_forks; /* Handle normal Linux uptimes. */
58 int nr_threads; /* The idle threads do not count.. */
60 int max_threads; /* tunable limit on nr_threads */
62 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
64 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
66 EXPORT_SYMBOL(tasklist_lock);
68 int nr_processes(void)
73 for_each_online_cpu(cpu)
74 total += per_cpu(process_counts, cpu);
79 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
80 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
81 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
82 static kmem_cache_t *task_struct_cachep;
85 /* SLAB cache for signal_struct structures (tsk->signal) */
86 kmem_cache_t *signal_cachep;
88 /* SLAB cache for sighand_struct structures (tsk->sighand) */
89 kmem_cache_t *sighand_cachep;
91 /* SLAB cache for files_struct structures (tsk->files) */
92 kmem_cache_t *files_cachep;
94 /* SLAB cache for fs_struct structures (tsk->fs) */
95 kmem_cache_t *fs_cachep;
97 /* SLAB cache for vm_area_struct structures */
98 kmem_cache_t *vm_area_cachep;
100 /* SLAB cache for mm_struct structures (tsk->mm) */
101 static kmem_cache_t *mm_cachep;
103 void free_task(struct task_struct *tsk)
105 free_thread_info(tsk->thread_info);
106 free_task_struct(tsk);
108 EXPORT_SYMBOL(free_task);
110 void __put_task_struct(struct task_struct *tsk)
112 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
113 WARN_ON(atomic_read(&tsk->usage));
114 WARN_ON(tsk == current);
116 if (unlikely(tsk->audit_context))
118 security_task_free(tsk);
120 put_group_info(tsk->group_info);
122 if (!profile_handoff_task(tsk))
126 void __init fork_init(unsigned long mempages)
128 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
129 #ifndef ARCH_MIN_TASKALIGN
130 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
132 /* create a slab on which task_structs can be allocated */
134 kmem_cache_create("task_struct", sizeof(struct task_struct),
135 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
139 * The default maximum number of threads is set to a safe
140 * value: the thread structures can take up at most half
143 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
146 * we need to allow at least 20 threads to boot a system
151 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
152 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
153 init_task.signal->rlim[RLIMIT_SIGPENDING] =
154 init_task.signal->rlim[RLIMIT_NPROC];
157 static struct task_struct *dup_task_struct(struct task_struct *orig)
159 struct task_struct *tsk;
160 struct thread_info *ti;
162 prepare_to_copy(orig);
164 tsk = alloc_task_struct();
168 ti = alloc_thread_info(tsk);
170 free_task_struct(tsk);
175 tsk->thread_info = ti;
176 setup_thread_stack(tsk, orig);
178 /* One for us, one for whoever does the "release_task()" (usually parent) */
179 atomic_set(&tsk->usage,2);
180 atomic_set(&tsk->fs_excl, 0);
185 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
187 struct vm_area_struct *mpnt, *tmp, **pprev;
188 struct rb_node **rb_link, *rb_parent;
190 unsigned long charge;
191 struct mempolicy *pol;
193 down_write(&oldmm->mmap_sem);
194 flush_cache_mm(oldmm);
195 down_write(&mm->mmap_sem);
199 mm->mmap_cache = NULL;
200 mm->free_area_cache = oldmm->mmap_base;
201 mm->cached_hole_size = ~0UL;
203 cpus_clear(mm->cpu_vm_mask);
205 rb_link = &mm->mm_rb.rb_node;
209 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
212 if (mpnt->vm_flags & VM_DONTCOPY) {
213 long pages = vma_pages(mpnt);
214 mm->total_vm -= pages;
215 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
220 if (mpnt->vm_flags & VM_ACCOUNT) {
221 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
222 if (security_vm_enough_memory(len))
226 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
230 pol = mpol_copy(vma_policy(mpnt));
231 retval = PTR_ERR(pol);
233 goto fail_nomem_policy;
234 vma_set_policy(tmp, pol);
235 tmp->vm_flags &= ~VM_LOCKED;
241 struct inode *inode = file->f_dentry->d_inode;
243 if (tmp->vm_flags & VM_DENYWRITE)
244 atomic_dec(&inode->i_writecount);
246 /* insert tmp into the share list, just after mpnt */
247 spin_lock(&file->f_mapping->i_mmap_lock);
248 tmp->vm_truncate_count = mpnt->vm_truncate_count;
249 flush_dcache_mmap_lock(file->f_mapping);
250 vma_prio_tree_add(tmp, mpnt);
251 flush_dcache_mmap_unlock(file->f_mapping);
252 spin_unlock(&file->f_mapping->i_mmap_lock);
256 * Link in the new vma and copy the page table entries.
259 pprev = &tmp->vm_next;
261 __vma_link_rb(mm, tmp, rb_link, rb_parent);
262 rb_link = &tmp->vm_rb.rb_right;
263 rb_parent = &tmp->vm_rb;
266 retval = copy_page_range(mm, oldmm, tmp);
268 if (tmp->vm_ops && tmp->vm_ops->open)
269 tmp->vm_ops->open(tmp);
276 up_write(&mm->mmap_sem);
278 up_write(&oldmm->mmap_sem);
281 kmem_cache_free(vm_area_cachep, tmp);
284 vm_unacct_memory(charge);
288 static inline int mm_alloc_pgd(struct mm_struct * mm)
290 mm->pgd = pgd_alloc(mm);
291 if (unlikely(!mm->pgd))
296 static inline void mm_free_pgd(struct mm_struct * mm)
301 #define dup_mmap(mm, oldmm) (0)
302 #define mm_alloc_pgd(mm) (0)
303 #define mm_free_pgd(mm)
304 #endif /* CONFIG_MMU */
306 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
308 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
309 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
311 #include <linux/init_task.h>
313 static struct mm_struct * mm_init(struct mm_struct * mm)
315 atomic_set(&mm->mm_users, 1);
316 atomic_set(&mm->mm_count, 1);
317 init_rwsem(&mm->mmap_sem);
318 INIT_LIST_HEAD(&mm->mmlist);
319 mm->core_waiters = 0;
321 set_mm_counter(mm, file_rss, 0);
322 set_mm_counter(mm, anon_rss, 0);
323 spin_lock_init(&mm->page_table_lock);
324 rwlock_init(&mm->ioctx_list_lock);
325 mm->ioctx_list = NULL;
326 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
327 mm->free_area_cache = TASK_UNMAPPED_BASE;
328 mm->cached_hole_size = ~0UL;
330 if (likely(!mm_alloc_pgd(mm))) {
339 * Allocate and initialize an mm_struct.
341 struct mm_struct * mm_alloc(void)
343 struct mm_struct * mm;
347 memset(mm, 0, sizeof(*mm));
354 * Called when the last reference to the mm
355 * is dropped: either by a lazy thread or by
356 * mmput. Free the page directory and the mm.
358 void fastcall __mmdrop(struct mm_struct *mm)
360 BUG_ON(mm == &init_mm);
367 * Decrement the use count and release all resources for an mm.
369 void mmput(struct mm_struct *mm)
371 if (atomic_dec_and_test(&mm->mm_users)) {
374 if (!list_empty(&mm->mmlist)) {
375 spin_lock(&mmlist_lock);
376 list_del(&mm->mmlist);
377 spin_unlock(&mmlist_lock);
383 EXPORT_SYMBOL_GPL(mmput);
386 * get_task_mm - acquire a reference to the task's mm
388 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
389 * this kernel workthread has transiently adopted a user mm with use_mm,
390 * to do its AIO) is not set and if so returns a reference to it, after
391 * bumping up the use count. User must release the mm via mmput()
392 * after use. Typically used by /proc and ptrace.
394 struct mm_struct *get_task_mm(struct task_struct *task)
396 struct mm_struct *mm;
401 if (task->flags & PF_BORROWED_MM)
404 atomic_inc(&mm->mm_users);
409 EXPORT_SYMBOL_GPL(get_task_mm);
411 /* Please note the differences between mmput and mm_release.
412 * mmput is called whenever we stop holding onto a mm_struct,
413 * error success whatever.
415 * mm_release is called after a mm_struct has been removed
416 * from the current process.
418 * This difference is important for error handling, when we
419 * only half set up a mm_struct for a new process and need to restore
420 * the old one. Because we mmput the new mm_struct before
421 * restoring the old one. . .
422 * Eric Biederman 10 January 1998
424 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
426 struct completion *vfork_done = tsk->vfork_done;
428 /* Get rid of any cached register state */
429 deactivate_mm(tsk, mm);
431 /* notify parent sleeping on vfork() */
433 tsk->vfork_done = NULL;
434 complete(vfork_done);
436 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
437 u32 __user * tidptr = tsk->clear_child_tid;
438 tsk->clear_child_tid = NULL;
441 * We don't check the error code - if userspace has
442 * not set up a proper pointer then tough luck.
445 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
449 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
451 struct mm_struct * mm, *oldmm;
454 tsk->min_flt = tsk->maj_flt = 0;
455 tsk->nvcsw = tsk->nivcsw = 0;
458 tsk->active_mm = NULL;
461 * Are we cloning a kernel thread?
463 * We need to steal a active VM for that..
469 if (clone_flags & CLONE_VM) {
470 atomic_inc(&oldmm->mm_users);
480 /* Copy the current MM stuff.. */
481 memcpy(mm, oldmm, sizeof(*mm));
485 if (init_new_context(tsk,mm))
488 retval = dup_mmap(mm, oldmm);
492 mm->hiwater_rss = get_mm_rss(mm);
493 mm->hiwater_vm = mm->total_vm;
507 * If init_new_context() failed, we cannot use mmput() to free the mm
508 * because it calls destroy_context()
515 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
517 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
518 /* We don't need to lock fs - think why ;-) */
520 atomic_set(&fs->count, 1);
521 rwlock_init(&fs->lock);
522 fs->umask = old->umask;
523 read_lock(&old->lock);
524 fs->rootmnt = mntget(old->rootmnt);
525 fs->root = dget(old->root);
526 fs->pwdmnt = mntget(old->pwdmnt);
527 fs->pwd = dget(old->pwd);
529 fs->altrootmnt = mntget(old->altrootmnt);
530 fs->altroot = dget(old->altroot);
532 fs->altrootmnt = NULL;
535 read_unlock(&old->lock);
540 struct fs_struct *copy_fs_struct(struct fs_struct *old)
542 return __copy_fs_struct(old);
545 EXPORT_SYMBOL_GPL(copy_fs_struct);
547 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
549 if (clone_flags & CLONE_FS) {
550 atomic_inc(¤t->fs->count);
553 tsk->fs = __copy_fs_struct(current->fs);
559 static int count_open_files(struct fdtable *fdt)
561 int size = fdt->max_fdset;
564 /* Find the last open fd */
565 for (i = size/(8*sizeof(long)); i > 0; ) {
566 if (fdt->open_fds->fds_bits[--i])
569 i = (i+1) * 8 * sizeof(long);
573 static struct files_struct *alloc_files(void)
575 struct files_struct *newf;
578 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
582 atomic_set(&newf->count, 1);
584 spin_lock_init(&newf->file_lock);
587 fdt->max_fds = NR_OPEN_DEFAULT;
588 fdt->max_fdset = __FD_SETSIZE;
589 fdt->close_on_exec = &newf->close_on_exec_init;
590 fdt->open_fds = &newf->open_fds_init;
591 fdt->fd = &newf->fd_array[0];
592 INIT_RCU_HEAD(&fdt->rcu);
593 fdt->free_files = NULL;
595 rcu_assign_pointer(newf->fdt, fdt);
600 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
602 struct files_struct *oldf, *newf;
603 struct file **old_fds, **new_fds;
604 int open_files, size, i, error = 0, expand;
605 struct fdtable *old_fdt, *new_fdt;
608 * A background process may not have any files ...
610 oldf = current->files;
614 if (clone_flags & CLONE_FILES) {
615 atomic_inc(&oldf->count);
620 * Note: we may be using current for both targets (See exec.c)
621 * This works because we cache current->files (old) as oldf. Don't
626 newf = alloc_files();
630 spin_lock(&oldf->file_lock);
631 old_fdt = files_fdtable(oldf);
632 new_fdt = files_fdtable(newf);
633 size = old_fdt->max_fdset;
634 open_files = count_open_files(old_fdt);
638 * Check whether we need to allocate a larger fd array or fd set.
639 * Note: we're not a clone task, so the open count won't change.
641 if (open_files > new_fdt->max_fdset) {
642 new_fdt->max_fdset = 0;
645 if (open_files > new_fdt->max_fds) {
646 new_fdt->max_fds = 0;
650 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
652 spin_unlock(&oldf->file_lock);
653 spin_lock(&newf->file_lock);
654 error = expand_files(newf, open_files-1);
655 spin_unlock(&newf->file_lock);
658 new_fdt = files_fdtable(newf);
660 * Reacquire the oldf lock and a pointer to its fd table
661 * who knows it may have a new bigger fd table. We need
662 * the latest pointer.
664 spin_lock(&oldf->file_lock);
665 old_fdt = files_fdtable(oldf);
668 old_fds = old_fdt->fd;
669 new_fds = new_fdt->fd;
671 memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
672 memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
674 for (i = open_files; i != 0; i--) {
675 struct file *f = *old_fds++;
680 * The fd may be claimed in the fd bitmap but not yet
681 * instantiated in the files array if a sibling thread
682 * is partway through open(). So make sure that this
683 * fd is available to the new process.
685 FD_CLR(open_files - i, new_fdt->open_fds);
687 rcu_assign_pointer(*new_fds++, f);
689 spin_unlock(&oldf->file_lock);
691 /* compute the remainder to be cleared */
692 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
694 /* This is long word aligned thus could use a optimized version */
695 memset(new_fds, 0, size);
697 if (new_fdt->max_fdset > open_files) {
698 int left = (new_fdt->max_fdset-open_files)/8;
699 int start = open_files / (8 * sizeof(unsigned long));
701 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
702 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
711 free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
712 free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
713 free_fd_array(new_fdt->fd, new_fdt->max_fds);
714 kmem_cache_free(files_cachep, newf);
719 * Helper to unshare the files of the current task.
720 * We don't want to expose copy_files internals to
721 * the exec layer of the kernel.
724 int unshare_files(void)
726 struct files_struct *files = current->files;
732 /* This can race but the race causes us to copy when we don't
733 need to and drop the copy */
734 if(atomic_read(&files->count) == 1)
736 atomic_inc(&files->count);
739 rc = copy_files(0, current);
741 current->files = files;
745 EXPORT_SYMBOL(unshare_files);
747 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
749 struct sighand_struct *sig;
751 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
752 atomic_inc(¤t->sighand->count);
755 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
759 spin_lock_init(&sig->siglock);
760 atomic_set(&sig->count, 1);
761 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
765 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
767 struct signal_struct *sig;
770 if (clone_flags & CLONE_THREAD) {
771 atomic_inc(¤t->signal->count);
772 atomic_inc(¤t->signal->live);
775 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
780 ret = copy_thread_group_keys(tsk);
782 kmem_cache_free(signal_cachep, sig);
786 atomic_set(&sig->count, 1);
787 atomic_set(&sig->live, 1);
788 init_waitqueue_head(&sig->wait_chldexit);
790 sig->group_exit_code = 0;
791 sig->group_exit_task = NULL;
792 sig->group_stop_count = 0;
793 sig->curr_target = NULL;
794 init_sigpending(&sig->shared_pending);
795 INIT_LIST_HEAD(&sig->posix_timers);
797 sig->it_real_value = sig->it_real_incr = 0;
798 sig->real_timer.function = it_real_fn;
799 sig->real_timer.data = (unsigned long) tsk;
800 init_timer(&sig->real_timer);
802 sig->it_virt_expires = cputime_zero;
803 sig->it_virt_incr = cputime_zero;
804 sig->it_prof_expires = cputime_zero;
805 sig->it_prof_incr = cputime_zero;
807 sig->tty = current->signal->tty;
808 sig->pgrp = process_group(current);
809 sig->session = current->signal->session;
810 sig->leader = 0; /* session leadership doesn't inherit */
811 sig->tty_old_pgrp = 0;
813 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
814 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
815 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
817 INIT_LIST_HEAD(&sig->cpu_timers[0]);
818 INIT_LIST_HEAD(&sig->cpu_timers[1]);
819 INIT_LIST_HEAD(&sig->cpu_timers[2]);
821 task_lock(current->group_leader);
822 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
823 task_unlock(current->group_leader);
825 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
827 * New sole thread in the process gets an expiry time
828 * of the whole CPU time limit.
830 tsk->it_prof_expires =
831 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
837 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
839 unsigned long new_flags = p->flags;
841 new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
842 new_flags |= PF_FORKNOEXEC;
843 if (!(clone_flags & CLONE_PTRACE))
845 p->flags = new_flags;
848 asmlinkage long sys_set_tid_address(int __user *tidptr)
850 current->clear_child_tid = tidptr;
856 * This creates a new process as a copy of the old one,
857 * but does not actually start it yet.
859 * It copies the registers, and all the appropriate
860 * parts of the process environment (as per the clone
861 * flags). The actual kick-off is left to the caller.
863 static task_t *copy_process(unsigned long clone_flags,
864 unsigned long stack_start,
865 struct pt_regs *regs,
866 unsigned long stack_size,
867 int __user *parent_tidptr,
868 int __user *child_tidptr,
872 struct task_struct *p = NULL;
874 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
875 return ERR_PTR(-EINVAL);
878 * Thread groups must share signals as well, and detached threads
879 * can only be started up within the thread group.
881 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
882 return ERR_PTR(-EINVAL);
885 * Shared signal handlers imply shared VM. By way of the above,
886 * thread groups also imply shared VM. Blocking this case allows
887 * for various simplifications in other code.
889 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
890 return ERR_PTR(-EINVAL);
892 retval = security_task_create(clone_flags);
897 p = dup_task_struct(current);
902 if (atomic_read(&p->user->processes) >=
903 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
904 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
905 p->user != &root_user)
909 atomic_inc(&p->user->__count);
910 atomic_inc(&p->user->processes);
911 get_group_info(p->group_info);
914 * If multiple threads are within copy_process(), then this check
915 * triggers too late. This doesn't hurt, the check is only there
916 * to stop root fork bombs.
918 if (nr_threads >= max_threads)
919 goto bad_fork_cleanup_count;
921 if (!try_module_get(task_thread_info(p)->exec_domain->module))
922 goto bad_fork_cleanup_count;
924 if (p->binfmt && !try_module_get(p->binfmt->module))
925 goto bad_fork_cleanup_put_domain;
928 copy_flags(clone_flags, p);
931 if (clone_flags & CLONE_PARENT_SETTID)
932 if (put_user(p->pid, parent_tidptr))
933 goto bad_fork_cleanup;
935 p->proc_dentry = NULL;
937 INIT_LIST_HEAD(&p->children);
938 INIT_LIST_HEAD(&p->sibling);
939 p->vfork_done = NULL;
940 spin_lock_init(&p->alloc_lock);
941 spin_lock_init(&p->proc_lock);
943 clear_tsk_thread_flag(p, TIF_SIGPENDING);
944 init_sigpending(&p->pending);
946 p->utime = cputime_zero;
947 p->stime = cputime_zero;
949 p->rchar = 0; /* I/O counter: bytes read */
950 p->wchar = 0; /* I/O counter: bytes written */
951 p->syscr = 0; /* I/O counter: read syscalls */
952 p->syscw = 0; /* I/O counter: write syscalls */
953 acct_clear_integrals(p);
955 p->it_virt_expires = cputime_zero;
956 p->it_prof_expires = cputime_zero;
957 p->it_sched_expires = 0;
958 INIT_LIST_HEAD(&p->cpu_timers[0]);
959 INIT_LIST_HEAD(&p->cpu_timers[1]);
960 INIT_LIST_HEAD(&p->cpu_timers[2]);
962 p->lock_depth = -1; /* -1 = no lock */
963 do_posix_clock_monotonic_gettime(&p->start_time);
965 p->io_context = NULL;
967 p->audit_context = NULL;
969 p->mempolicy = mpol_copy(p->mempolicy);
970 if (IS_ERR(p->mempolicy)) {
971 retval = PTR_ERR(p->mempolicy);
973 goto bad_fork_cleanup;
978 if (clone_flags & CLONE_THREAD)
979 p->tgid = current->tgid;
981 if ((retval = security_task_alloc(p)))
982 goto bad_fork_cleanup_policy;
983 if ((retval = audit_alloc(p)))
984 goto bad_fork_cleanup_security;
985 /* copy all the process information */
986 if ((retval = copy_semundo(clone_flags, p)))
987 goto bad_fork_cleanup_audit;
988 if ((retval = copy_files(clone_flags, p)))
989 goto bad_fork_cleanup_semundo;
990 if ((retval = copy_fs(clone_flags, p)))
991 goto bad_fork_cleanup_files;
992 if ((retval = copy_sighand(clone_flags, p)))
993 goto bad_fork_cleanup_fs;
994 if ((retval = copy_signal(clone_flags, p)))
995 goto bad_fork_cleanup_sighand;
996 if ((retval = copy_mm(clone_flags, p)))
997 goto bad_fork_cleanup_signal;
998 if ((retval = copy_keys(clone_flags, p)))
999 goto bad_fork_cleanup_mm;
1000 if ((retval = copy_namespace(clone_flags, p)))
1001 goto bad_fork_cleanup_keys;
1002 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1004 goto bad_fork_cleanup_namespace;
1006 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1008 * Clear TID on mm_release()?
1010 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1013 * Syscall tracing should be turned off in the child regardless
1016 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1017 #ifdef TIF_SYSCALL_EMU
1018 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1021 /* Our parent execution domain becomes current domain
1022 These must match for thread signalling to apply */
1024 p->parent_exec_id = p->self_exec_id;
1026 /* ok, now we should be set up.. */
1027 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1028 p->pdeath_signal = 0;
1032 * Ok, make it visible to the rest of the system.
1033 * We dont wake it up yet.
1035 p->group_leader = p;
1036 INIT_LIST_HEAD(&p->ptrace_children);
1037 INIT_LIST_HEAD(&p->ptrace_list);
1039 /* Perform scheduler related setup. Assign this task to a CPU. */
1040 sched_fork(p, clone_flags);
1042 /* Need tasklist lock for parent etc handling! */
1043 write_lock_irq(&tasklist_lock);
1046 * The task hasn't been attached yet, so its cpus_allowed mask will
1047 * not be changed, nor will its assigned CPU.
1049 * The cpus_allowed mask of the parent may have changed after it was
1050 * copied first time - so re-copy it here, then check the child's CPU
1051 * to ensure it is on a valid CPU (and if not, just force it back to
1052 * parent's CPU). This avoids alot of nasty races.
1054 p->cpus_allowed = current->cpus_allowed;
1055 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1056 !cpu_online(task_cpu(p))))
1057 set_task_cpu(p, smp_processor_id());
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|CLONE_THREAD))
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->flags & SIGNAL_GROUP_EXIT) {
1084 spin_unlock(¤t->sighand->siglock);
1085 write_unlock_irq(&tasklist_lock);
1087 goto bad_fork_cleanup_namespace;
1089 p->group_leader = current->group_leader;
1091 if (current->signal->group_stop_count > 0) {
1093 * There is an all-stop in progress for the group.
1094 * We ourselves will stop as soon as we check signals.
1095 * Make the new thread part of that group stop too.
1097 current->signal->group_stop_count++;
1098 set_tsk_thread_flag(p, TIF_SIGPENDING);
1101 if (!cputime_eq(current->signal->it_virt_expires,
1103 !cputime_eq(current->signal->it_prof_expires,
1105 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1106 !list_empty(¤t->signal->cpu_timers[0]) ||
1107 !list_empty(¤t->signal->cpu_timers[1]) ||
1108 !list_empty(¤t->signal->cpu_timers[2])) {
1110 * Have child wake up on its first tick to check
1111 * for process CPU timers.
1113 p->it_prof_expires = jiffies_to_cputime(1);
1116 spin_unlock(¤t->sighand->siglock);
1122 p->ioprio = current->ioprio;
1125 if (unlikely(p->ptrace & PT_PTRACED))
1126 __ptrace_link(p, current->parent);
1130 attach_pid(p, PIDTYPE_PID, p->pid);
1131 attach_pid(p, PIDTYPE_TGID, p->tgid);
1132 if (thread_group_leader(p)) {
1133 attach_pid(p, PIDTYPE_PGID, process_group(p));
1134 attach_pid(p, PIDTYPE_SID, p->signal->session);
1136 __get_cpu_var(process_counts)++;
1139 proc_fork_connector(p);
1140 if (!current->signal->tty && p->signal->tty)
1141 p->signal->tty = NULL;
1145 write_unlock_irq(&tasklist_lock);
1150 return ERR_PTR(retval);
1153 bad_fork_cleanup_namespace:
1155 bad_fork_cleanup_keys:
1157 bad_fork_cleanup_mm:
1160 bad_fork_cleanup_signal:
1162 bad_fork_cleanup_sighand:
1164 bad_fork_cleanup_fs:
1165 exit_fs(p); /* blocking */
1166 bad_fork_cleanup_files:
1167 exit_files(p); /* blocking */
1168 bad_fork_cleanup_semundo:
1170 bad_fork_cleanup_audit:
1172 bad_fork_cleanup_security:
1173 security_task_free(p);
1174 bad_fork_cleanup_policy:
1176 mpol_free(p->mempolicy);
1180 module_put(p->binfmt->module);
1181 bad_fork_cleanup_put_domain:
1182 module_put(task_thread_info(p)->exec_domain->module);
1183 bad_fork_cleanup_count:
1184 put_group_info(p->group_info);
1185 atomic_dec(&p->user->processes);
1192 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1194 memset(regs, 0, sizeof(struct pt_regs));
1198 task_t * __devinit fork_idle(int cpu)
1201 struct pt_regs regs;
1203 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0);
1205 return ERR_PTR(-ENOMEM);
1206 init_idle(task, cpu);
1207 unhash_process(task);
1211 static inline int fork_traceflag (unsigned clone_flags)
1213 if (clone_flags & CLONE_UNTRACED)
1215 else if (clone_flags & CLONE_VFORK) {
1216 if (current->ptrace & PT_TRACE_VFORK)
1217 return PTRACE_EVENT_VFORK;
1218 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1219 if (current->ptrace & PT_TRACE_CLONE)
1220 return PTRACE_EVENT_CLONE;
1221 } else if (current->ptrace & PT_TRACE_FORK)
1222 return PTRACE_EVENT_FORK;
1228 * Ok, this is the main fork-routine.
1230 * It copies the process, and if successful kick-starts
1231 * it and waits for it to finish using the VM if required.
1233 long do_fork(unsigned long clone_flags,
1234 unsigned long stack_start,
1235 struct pt_regs *regs,
1236 unsigned long stack_size,
1237 int __user *parent_tidptr,
1238 int __user *child_tidptr)
1240 struct task_struct *p;
1242 long pid = alloc_pidmap();
1246 if (unlikely(current->ptrace)) {
1247 trace = fork_traceflag (clone_flags);
1249 clone_flags |= CLONE_PTRACE;
1252 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1254 * Do this prior waking up the new thread - the thread pointer
1255 * might get invalid after that point, if the thread exits quickly.
1258 struct completion vfork;
1260 if (clone_flags & CLONE_VFORK) {
1261 p->vfork_done = &vfork;
1262 init_completion(&vfork);
1265 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1267 * We'll start up with an immediate SIGSTOP.
1269 sigaddset(&p->pending.signal, SIGSTOP);
1270 set_tsk_thread_flag(p, TIF_SIGPENDING);
1273 if (!(clone_flags & CLONE_STOPPED))
1274 wake_up_new_task(p, clone_flags);
1276 p->state = TASK_STOPPED;
1278 if (unlikely (trace)) {
1279 current->ptrace_message = pid;
1280 ptrace_notify ((trace << 8) | SIGTRAP);
1283 if (clone_flags & CLONE_VFORK) {
1284 wait_for_completion(&vfork);
1285 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1286 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1295 void __init proc_caches_init(void)
1297 sighand_cachep = kmem_cache_create("sighand_cache",
1298 sizeof(struct sighand_struct), 0,
1299 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1300 signal_cachep = kmem_cache_create("signal_cache",
1301 sizeof(struct signal_struct), 0,
1302 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1303 files_cachep = kmem_cache_create("files_cache",
1304 sizeof(struct files_struct), 0,
1305 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1306 fs_cachep = kmem_cache_create("fs_cache",
1307 sizeof(struct fs_struct), 0,
1308 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1309 vm_area_cachep = kmem_cache_create("vm_area_struct",
1310 sizeof(struct vm_area_struct), 0,
1311 SLAB_PANIC, NULL, NULL);
1312 mm_cachep = kmem_cache_create("mm_struct",
1313 sizeof(struct mm_struct), 0,
1314 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);