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>
46 #include <asm/pgtable.h>
47 #include <asm/pgalloc.h>
48 #include <asm/uaccess.h>
49 #include <asm/mmu_context.h>
50 #include <asm/cacheflush.h>
51 #include <asm/tlbflush.h>
54 * Protected counters by write_lock_irq(&tasklist_lock)
56 unsigned long total_forks; /* Handle normal Linux uptimes. */
57 int nr_threads; /* The idle threads do not count.. */
59 int max_threads; /* tunable limit on nr_threads */
61 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
63 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
65 EXPORT_SYMBOL(tasklist_lock);
67 int nr_processes(void)
72 for_each_online_cpu(cpu)
73 total += per_cpu(process_counts, cpu);
78 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
79 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
80 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
81 static kmem_cache_t *task_struct_cachep;
84 /* SLAB cache for signal_struct structures (tsk->signal) */
85 kmem_cache_t *signal_cachep;
87 /* SLAB cache for sighand_struct structures (tsk->sighand) */
88 kmem_cache_t *sighand_cachep;
90 /* SLAB cache for files_struct structures (tsk->files) */
91 kmem_cache_t *files_cachep;
93 /* SLAB cache for fs_struct structures (tsk->fs) */
94 kmem_cache_t *fs_cachep;
96 /* SLAB cache for vm_area_struct structures */
97 kmem_cache_t *vm_area_cachep;
99 /* SLAB cache for mm_struct structures (tsk->mm) */
100 static kmem_cache_t *mm_cachep;
102 void free_task(struct task_struct *tsk)
104 free_thread_info(tsk->thread_info);
105 free_task_struct(tsk);
107 EXPORT_SYMBOL(free_task);
109 void __put_task_struct(struct task_struct *tsk)
111 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
112 WARN_ON(atomic_read(&tsk->usage));
113 WARN_ON(tsk == current);
115 if (unlikely(tsk->audit_context))
117 security_task_free(tsk);
119 put_group_info(tsk->group_info);
121 if (!profile_handoff_task(tsk))
125 void __init fork_init(unsigned long mempages)
127 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
128 #ifndef ARCH_MIN_TASKALIGN
129 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
131 /* create a slab on which task_structs can be allocated */
133 kmem_cache_create("task_struct", sizeof(struct task_struct),
134 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
138 * The default maximum number of threads is set to a safe
139 * value: the thread structures can take up at most half
142 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
145 * we need to allow at least 20 threads to boot a system
150 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
151 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
152 init_task.signal->rlim[RLIMIT_SIGPENDING] =
153 init_task.signal->rlim[RLIMIT_NPROC];
156 static struct task_struct *dup_task_struct(struct task_struct *orig)
158 struct task_struct *tsk;
159 struct thread_info *ti;
161 prepare_to_copy(orig);
163 tsk = alloc_task_struct();
167 ti = alloc_thread_info(tsk);
169 free_task_struct(tsk);
173 *ti = *orig->thread_info;
175 tsk->thread_info = ti;
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.
258 spin_lock(&mm->page_table_lock);
260 pprev = &tmp->vm_next;
262 __vma_link_rb(mm, tmp, rb_link, rb_parent);
263 rb_link = &tmp->vm_rb.rb_right;
264 rb_parent = &tmp->vm_rb;
267 retval = copy_page_range(mm, oldmm, tmp);
268 spin_unlock(&mm->page_table_lock);
270 if (tmp->vm_ops && tmp->vm_ops->open)
271 tmp->vm_ops->open(tmp);
278 up_write(&mm->mmap_sem);
280 up_write(&oldmm->mmap_sem);
283 kmem_cache_free(vm_area_cachep, tmp);
286 vm_unacct_memory(charge);
290 static inline int mm_alloc_pgd(struct mm_struct * mm)
292 mm->pgd = pgd_alloc(mm);
293 if (unlikely(!mm->pgd))
298 static inline void mm_free_pgd(struct mm_struct * mm)
303 #define dup_mmap(mm, oldmm) (0)
304 #define mm_alloc_pgd(mm) (0)
305 #define mm_free_pgd(mm)
306 #endif /* CONFIG_MMU */
308 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
310 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
311 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
313 #include <linux/init_task.h>
315 static struct mm_struct * mm_init(struct mm_struct * mm)
317 atomic_set(&mm->mm_users, 1);
318 atomic_set(&mm->mm_count, 1);
319 init_rwsem(&mm->mmap_sem);
320 INIT_LIST_HEAD(&mm->mmlist);
321 mm->core_waiters = 0;
323 set_mm_counter(mm, file_rss, 0);
324 set_mm_counter(mm, anon_rss, 0);
325 spin_lock_init(&mm->page_table_lock);
326 rwlock_init(&mm->ioctx_list_lock);
327 mm->ioctx_list = NULL;
328 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
329 mm->free_area_cache = TASK_UNMAPPED_BASE;
330 mm->cached_hole_size = ~0UL;
332 if (likely(!mm_alloc_pgd(mm))) {
341 * Allocate and initialize an mm_struct.
343 struct mm_struct * mm_alloc(void)
345 struct mm_struct * mm;
349 memset(mm, 0, sizeof(*mm));
356 * Called when the last reference to the mm
357 * is dropped: either by a lazy thread or by
358 * mmput. Free the page directory and the mm.
360 void fastcall __mmdrop(struct mm_struct *mm)
362 BUG_ON(mm == &init_mm);
369 * Decrement the use count and release all resources for an mm.
371 void mmput(struct mm_struct *mm)
373 if (atomic_dec_and_test(&mm->mm_users)) {
376 if (!list_empty(&mm->mmlist)) {
377 spin_lock(&mmlist_lock);
378 list_del(&mm->mmlist);
379 spin_unlock(&mmlist_lock);
385 EXPORT_SYMBOL_GPL(mmput);
388 * get_task_mm - acquire a reference to the task's mm
390 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
391 * this kernel workthread has transiently adopted a user mm with use_mm,
392 * to do its AIO) is not set and if so returns a reference to it, after
393 * bumping up the use count. User must release the mm via mmput()
394 * after use. Typically used by /proc and ptrace.
396 struct mm_struct *get_task_mm(struct task_struct *task)
398 struct mm_struct *mm;
403 if (task->flags & PF_BORROWED_MM)
406 atomic_inc(&mm->mm_users);
411 EXPORT_SYMBOL_GPL(get_task_mm);
413 /* Please note the differences between mmput and mm_release.
414 * mmput is called whenever we stop holding onto a mm_struct,
415 * error success whatever.
417 * mm_release is called after a mm_struct has been removed
418 * from the current process.
420 * This difference is important for error handling, when we
421 * only half set up a mm_struct for a new process and need to restore
422 * the old one. Because we mmput the new mm_struct before
423 * restoring the old one. . .
424 * Eric Biederman 10 January 1998
426 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
428 struct completion *vfork_done = tsk->vfork_done;
430 /* Get rid of any cached register state */
431 deactivate_mm(tsk, mm);
433 /* notify parent sleeping on vfork() */
435 tsk->vfork_done = NULL;
436 complete(vfork_done);
438 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
439 u32 __user * tidptr = tsk->clear_child_tid;
440 tsk->clear_child_tid = NULL;
443 * We don't check the error code - if userspace has
444 * not set up a proper pointer then tough luck.
447 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
451 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
453 struct mm_struct * mm, *oldmm;
456 tsk->min_flt = tsk->maj_flt = 0;
457 tsk->nvcsw = tsk->nivcsw = 0;
460 tsk->active_mm = NULL;
463 * Are we cloning a kernel thread?
465 * We need to steal a active VM for that..
471 if (clone_flags & CLONE_VM) {
472 atomic_inc(&oldmm->mm_users);
475 * There are cases where the PTL is held to ensure no
476 * new threads start up in user mode using an mm, which
477 * allows optimizing out ipis; the tlb_gather_mmu code
480 spin_unlock_wait(&oldmm->page_table_lock);
489 /* Copy the current MM stuff.. */
490 memcpy(mm, oldmm, sizeof(*mm));
494 if (init_new_context(tsk,mm))
497 retval = dup_mmap(mm, oldmm);
501 mm->hiwater_rss = get_mm_rss(mm);
502 mm->hiwater_vm = mm->total_vm;
516 * If init_new_context() failed, we cannot use mmput() to free the mm
517 * because it calls destroy_context()
524 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
526 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
527 /* We don't need to lock fs - think why ;-) */
529 atomic_set(&fs->count, 1);
530 rwlock_init(&fs->lock);
531 fs->umask = old->umask;
532 read_lock(&old->lock);
533 fs->rootmnt = mntget(old->rootmnt);
534 fs->root = dget(old->root);
535 fs->pwdmnt = mntget(old->pwdmnt);
536 fs->pwd = dget(old->pwd);
538 fs->altrootmnt = mntget(old->altrootmnt);
539 fs->altroot = dget(old->altroot);
541 fs->altrootmnt = NULL;
544 read_unlock(&old->lock);
549 struct fs_struct *copy_fs_struct(struct fs_struct *old)
551 return __copy_fs_struct(old);
554 EXPORT_SYMBOL_GPL(copy_fs_struct);
556 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
558 if (clone_flags & CLONE_FS) {
559 atomic_inc(¤t->fs->count);
562 tsk->fs = __copy_fs_struct(current->fs);
568 static int count_open_files(struct fdtable *fdt)
570 int size = fdt->max_fdset;
573 /* Find the last open fd */
574 for (i = size/(8*sizeof(long)); i > 0; ) {
575 if (fdt->open_fds->fds_bits[--i])
578 i = (i+1) * 8 * sizeof(long);
582 static struct files_struct *alloc_files(void)
584 struct files_struct *newf;
587 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
591 atomic_set(&newf->count, 1);
593 spin_lock_init(&newf->file_lock);
596 fdt->max_fds = NR_OPEN_DEFAULT;
597 fdt->max_fdset = __FD_SETSIZE;
598 fdt->close_on_exec = &newf->close_on_exec_init;
599 fdt->open_fds = &newf->open_fds_init;
600 fdt->fd = &newf->fd_array[0];
601 INIT_RCU_HEAD(&fdt->rcu);
602 fdt->free_files = NULL;
604 rcu_assign_pointer(newf->fdt, fdt);
609 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
611 struct files_struct *oldf, *newf;
612 struct file **old_fds, **new_fds;
613 int open_files, size, i, error = 0, expand;
614 struct fdtable *old_fdt, *new_fdt;
617 * A background process may not have any files ...
619 oldf = current->files;
623 if (clone_flags & CLONE_FILES) {
624 atomic_inc(&oldf->count);
629 * Note: we may be using current for both targets (See exec.c)
630 * This works because we cache current->files (old) as oldf. Don't
635 newf = alloc_files();
639 spin_lock(&oldf->file_lock);
640 old_fdt = files_fdtable(oldf);
641 new_fdt = files_fdtable(newf);
642 size = old_fdt->max_fdset;
643 open_files = count_open_files(old_fdt);
647 * Check whether we need to allocate a larger fd array or fd set.
648 * Note: we're not a clone task, so the open count won't change.
650 if (open_files > new_fdt->max_fdset) {
651 new_fdt->max_fdset = 0;
654 if (open_files > new_fdt->max_fds) {
655 new_fdt->max_fds = 0;
659 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
661 spin_unlock(&oldf->file_lock);
662 spin_lock(&newf->file_lock);
663 error = expand_files(newf, open_files-1);
664 spin_unlock(&newf->file_lock);
667 new_fdt = files_fdtable(newf);
669 * Reacquire the oldf lock and a pointer to its fd table
670 * who knows it may have a new bigger fd table. We need
671 * the latest pointer.
673 spin_lock(&oldf->file_lock);
674 old_fdt = files_fdtable(oldf);
677 old_fds = old_fdt->fd;
678 new_fds = new_fdt->fd;
680 memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
681 memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
683 for (i = open_files; i != 0; i--) {
684 struct file *f = *old_fds++;
689 * The fd may be claimed in the fd bitmap but not yet
690 * instantiated in the files array if a sibling thread
691 * is partway through open(). So make sure that this
692 * fd is available to the new process.
694 FD_CLR(open_files - i, new_fdt->open_fds);
696 rcu_assign_pointer(*new_fds++, f);
698 spin_unlock(&oldf->file_lock);
700 /* compute the remainder to be cleared */
701 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
703 /* This is long word aligned thus could use a optimized version */
704 memset(new_fds, 0, size);
706 if (new_fdt->max_fdset > open_files) {
707 int left = (new_fdt->max_fdset-open_files)/8;
708 int start = open_files / (8 * sizeof(unsigned long));
710 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
711 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
720 free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
721 free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
722 free_fd_array(new_fdt->fd, new_fdt->max_fds);
723 kmem_cache_free(files_cachep, newf);
728 * Helper to unshare the files of the current task.
729 * We don't want to expose copy_files internals to
730 * the exec layer of the kernel.
733 int unshare_files(void)
735 struct files_struct *files = current->files;
741 /* This can race but the race causes us to copy when we don't
742 need to and drop the copy */
743 if(atomic_read(&files->count) == 1)
745 atomic_inc(&files->count);
748 rc = copy_files(0, current);
750 current->files = files;
754 EXPORT_SYMBOL(unshare_files);
756 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
758 struct sighand_struct *sig;
760 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
761 atomic_inc(¤t->sighand->count);
764 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
768 spin_lock_init(&sig->siglock);
769 atomic_set(&sig->count, 1);
770 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
774 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
776 struct signal_struct *sig;
779 if (clone_flags & CLONE_THREAD) {
780 atomic_inc(¤t->signal->count);
781 atomic_inc(¤t->signal->live);
784 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
789 ret = copy_thread_group_keys(tsk);
791 kmem_cache_free(signal_cachep, sig);
795 atomic_set(&sig->count, 1);
796 atomic_set(&sig->live, 1);
797 init_waitqueue_head(&sig->wait_chldexit);
799 sig->group_exit_code = 0;
800 sig->group_exit_task = NULL;
801 sig->group_stop_count = 0;
802 sig->curr_target = NULL;
803 init_sigpending(&sig->shared_pending);
804 INIT_LIST_HEAD(&sig->posix_timers);
806 sig->it_real_value = sig->it_real_incr = 0;
807 sig->real_timer.function = it_real_fn;
808 sig->real_timer.data = (unsigned long) tsk;
809 init_timer(&sig->real_timer);
811 sig->it_virt_expires = cputime_zero;
812 sig->it_virt_incr = cputime_zero;
813 sig->it_prof_expires = cputime_zero;
814 sig->it_prof_incr = cputime_zero;
816 sig->tty = current->signal->tty;
817 sig->pgrp = process_group(current);
818 sig->session = current->signal->session;
819 sig->leader = 0; /* session leadership doesn't inherit */
820 sig->tty_old_pgrp = 0;
822 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
823 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
824 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
826 INIT_LIST_HEAD(&sig->cpu_timers[0]);
827 INIT_LIST_HEAD(&sig->cpu_timers[1]);
828 INIT_LIST_HEAD(&sig->cpu_timers[2]);
830 task_lock(current->group_leader);
831 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
832 task_unlock(current->group_leader);
834 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
836 * New sole thread in the process gets an expiry time
837 * of the whole CPU time limit.
839 tsk->it_prof_expires =
840 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
846 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
848 unsigned long new_flags = p->flags;
850 new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
851 new_flags |= PF_FORKNOEXEC;
852 if (!(clone_flags & CLONE_PTRACE))
854 p->flags = new_flags;
857 asmlinkage long sys_set_tid_address(int __user *tidptr)
859 current->clear_child_tid = tidptr;
865 * This creates a new process as a copy of the old one,
866 * but does not actually start it yet.
868 * It copies the registers, and all the appropriate
869 * parts of the process environment (as per the clone
870 * flags). The actual kick-off is left to the caller.
872 static task_t *copy_process(unsigned long clone_flags,
873 unsigned long stack_start,
874 struct pt_regs *regs,
875 unsigned long stack_size,
876 int __user *parent_tidptr,
877 int __user *child_tidptr,
881 struct task_struct *p = NULL;
883 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
884 return ERR_PTR(-EINVAL);
887 * Thread groups must share signals as well, and detached threads
888 * can only be started up within the thread group.
890 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
891 return ERR_PTR(-EINVAL);
894 * Shared signal handlers imply shared VM. By way of the above,
895 * thread groups also imply shared VM. Blocking this case allows
896 * for various simplifications in other code.
898 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
899 return ERR_PTR(-EINVAL);
901 retval = security_task_create(clone_flags);
906 p = dup_task_struct(current);
911 if (atomic_read(&p->user->processes) >=
912 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
913 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
914 p->user != &root_user)
918 atomic_inc(&p->user->__count);
919 atomic_inc(&p->user->processes);
920 get_group_info(p->group_info);
923 * If multiple threads are within copy_process(), then this check
924 * triggers too late. This doesn't hurt, the check is only there
925 * to stop root fork bombs.
927 if (nr_threads >= max_threads)
928 goto bad_fork_cleanup_count;
930 if (!try_module_get(p->thread_info->exec_domain->module))
931 goto bad_fork_cleanup_count;
933 if (p->binfmt && !try_module_get(p->binfmt->module))
934 goto bad_fork_cleanup_put_domain;
937 copy_flags(clone_flags, p);
940 if (clone_flags & CLONE_PARENT_SETTID)
941 if (put_user(p->pid, parent_tidptr))
942 goto bad_fork_cleanup;
944 p->proc_dentry = NULL;
946 INIT_LIST_HEAD(&p->children);
947 INIT_LIST_HEAD(&p->sibling);
948 p->vfork_done = NULL;
949 spin_lock_init(&p->alloc_lock);
950 spin_lock_init(&p->proc_lock);
952 clear_tsk_thread_flag(p, TIF_SIGPENDING);
953 init_sigpending(&p->pending);
955 p->utime = cputime_zero;
956 p->stime = cputime_zero;
958 p->rchar = 0; /* I/O counter: bytes read */
959 p->wchar = 0; /* I/O counter: bytes written */
960 p->syscr = 0; /* I/O counter: read syscalls */
961 p->syscw = 0; /* I/O counter: write syscalls */
962 acct_clear_integrals(p);
964 p->it_virt_expires = cputime_zero;
965 p->it_prof_expires = cputime_zero;
966 p->it_sched_expires = 0;
967 INIT_LIST_HEAD(&p->cpu_timers[0]);
968 INIT_LIST_HEAD(&p->cpu_timers[1]);
969 INIT_LIST_HEAD(&p->cpu_timers[2]);
971 p->lock_depth = -1; /* -1 = no lock */
972 do_posix_clock_monotonic_gettime(&p->start_time);
974 p->io_context = NULL;
976 p->audit_context = NULL;
978 p->mempolicy = mpol_copy(p->mempolicy);
979 if (IS_ERR(p->mempolicy)) {
980 retval = PTR_ERR(p->mempolicy);
982 goto bad_fork_cleanup;
987 if (clone_flags & CLONE_THREAD)
988 p->tgid = current->tgid;
990 if ((retval = security_task_alloc(p)))
991 goto bad_fork_cleanup_policy;
992 if ((retval = audit_alloc(p)))
993 goto bad_fork_cleanup_security;
994 /* copy all the process information */
995 if ((retval = copy_semundo(clone_flags, p)))
996 goto bad_fork_cleanup_audit;
997 if ((retval = copy_files(clone_flags, p)))
998 goto bad_fork_cleanup_semundo;
999 if ((retval = copy_fs(clone_flags, p)))
1000 goto bad_fork_cleanup_files;
1001 if ((retval = copy_sighand(clone_flags, p)))
1002 goto bad_fork_cleanup_fs;
1003 if ((retval = copy_signal(clone_flags, p)))
1004 goto bad_fork_cleanup_sighand;
1005 if ((retval = copy_mm(clone_flags, p)))
1006 goto bad_fork_cleanup_signal;
1007 if ((retval = copy_keys(clone_flags, p)))
1008 goto bad_fork_cleanup_mm;
1009 if ((retval = copy_namespace(clone_flags, p)))
1010 goto bad_fork_cleanup_keys;
1011 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1013 goto bad_fork_cleanup_namespace;
1015 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1017 * Clear TID on mm_release()?
1019 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1022 * Syscall tracing should be turned off in the child regardless
1025 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1026 #ifdef TIF_SYSCALL_EMU
1027 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1030 /* Our parent execution domain becomes current domain
1031 These must match for thread signalling to apply */
1033 p->parent_exec_id = p->self_exec_id;
1035 /* ok, now we should be set up.. */
1036 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1037 p->pdeath_signal = 0;
1041 * Ok, make it visible to the rest of the system.
1042 * We dont wake it up yet.
1044 p->group_leader = p;
1045 INIT_LIST_HEAD(&p->ptrace_children);
1046 INIT_LIST_HEAD(&p->ptrace_list);
1048 /* Perform scheduler related setup. Assign this task to a CPU. */
1049 sched_fork(p, clone_flags);
1051 /* Need tasklist lock for parent etc handling! */
1052 write_lock_irq(&tasklist_lock);
1055 * The task hasn't been attached yet, so its cpus_allowed mask will
1056 * not be changed, nor will its assigned CPU.
1058 * The cpus_allowed mask of the parent may have changed after it was
1059 * copied first time - so re-copy it here, then check the child's CPU
1060 * to ensure it is on a valid CPU (and if not, just force it back to
1061 * parent's CPU). This avoids alot of nasty races.
1063 p->cpus_allowed = current->cpus_allowed;
1064 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1065 !cpu_online(task_cpu(p))))
1066 set_task_cpu(p, smp_processor_id());
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|CLONE_THREAD))
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->flags & SIGNAL_GROUP_EXIT) {
1093 spin_unlock(¤t->sighand->siglock);
1094 write_unlock_irq(&tasklist_lock);
1096 goto bad_fork_cleanup_namespace;
1098 p->group_leader = current->group_leader;
1100 if (current->signal->group_stop_count > 0) {
1102 * There is an all-stop in progress for the group.
1103 * We ourselves will stop as soon as we check signals.
1104 * Make the new thread part of that group stop too.
1106 current->signal->group_stop_count++;
1107 set_tsk_thread_flag(p, TIF_SIGPENDING);
1110 if (!cputime_eq(current->signal->it_virt_expires,
1112 !cputime_eq(current->signal->it_prof_expires,
1114 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1115 !list_empty(¤t->signal->cpu_timers[0]) ||
1116 !list_empty(¤t->signal->cpu_timers[1]) ||
1117 !list_empty(¤t->signal->cpu_timers[2])) {
1119 * Have child wake up on its first tick to check
1120 * for process CPU timers.
1122 p->it_prof_expires = jiffies_to_cputime(1);
1125 spin_unlock(¤t->sighand->siglock);
1131 p->ioprio = current->ioprio;
1134 if (unlikely(p->ptrace & PT_PTRACED))
1135 __ptrace_link(p, current->parent);
1139 attach_pid(p, PIDTYPE_PID, p->pid);
1140 attach_pid(p, PIDTYPE_TGID, p->tgid);
1141 if (thread_group_leader(p)) {
1142 attach_pid(p, PIDTYPE_PGID, process_group(p));
1143 attach_pid(p, PIDTYPE_SID, p->signal->session);
1145 __get_cpu_var(process_counts)++;
1148 if (!current->signal->tty && p->signal->tty)
1149 p->signal->tty = NULL;
1153 write_unlock_irq(&tasklist_lock);
1158 return ERR_PTR(retval);
1161 bad_fork_cleanup_namespace:
1163 bad_fork_cleanup_keys:
1165 bad_fork_cleanup_mm:
1168 bad_fork_cleanup_signal:
1170 bad_fork_cleanup_sighand:
1172 bad_fork_cleanup_fs:
1173 exit_fs(p); /* blocking */
1174 bad_fork_cleanup_files:
1175 exit_files(p); /* blocking */
1176 bad_fork_cleanup_semundo:
1178 bad_fork_cleanup_audit:
1180 bad_fork_cleanup_security:
1181 security_task_free(p);
1182 bad_fork_cleanup_policy:
1184 mpol_free(p->mempolicy);
1188 module_put(p->binfmt->module);
1189 bad_fork_cleanup_put_domain:
1190 module_put(p->thread_info->exec_domain->module);
1191 bad_fork_cleanup_count:
1192 put_group_info(p->group_info);
1193 atomic_dec(&p->user->processes);
1200 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1202 memset(regs, 0, sizeof(struct pt_regs));
1206 task_t * __devinit fork_idle(int cpu)
1209 struct pt_regs regs;
1211 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0);
1213 return ERR_PTR(-ENOMEM);
1214 init_idle(task, cpu);
1215 unhash_process(task);
1219 static inline int fork_traceflag (unsigned clone_flags)
1221 if (clone_flags & CLONE_UNTRACED)
1223 else if (clone_flags & CLONE_VFORK) {
1224 if (current->ptrace & PT_TRACE_VFORK)
1225 return PTRACE_EVENT_VFORK;
1226 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1227 if (current->ptrace & PT_TRACE_CLONE)
1228 return PTRACE_EVENT_CLONE;
1229 } else if (current->ptrace & PT_TRACE_FORK)
1230 return PTRACE_EVENT_FORK;
1236 * Ok, this is the main fork-routine.
1238 * It copies the process, and if successful kick-starts
1239 * it and waits for it to finish using the VM if required.
1241 long do_fork(unsigned long clone_flags,
1242 unsigned long stack_start,
1243 struct pt_regs *regs,
1244 unsigned long stack_size,
1245 int __user *parent_tidptr,
1246 int __user *child_tidptr)
1248 struct task_struct *p;
1250 long pid = alloc_pidmap();
1254 if (unlikely(current->ptrace)) {
1255 trace = fork_traceflag (clone_flags);
1257 clone_flags |= CLONE_PTRACE;
1260 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1262 * Do this prior waking up the new thread - the thread pointer
1263 * might get invalid after that point, if the thread exits quickly.
1266 struct completion vfork;
1268 if (clone_flags & CLONE_VFORK) {
1269 p->vfork_done = &vfork;
1270 init_completion(&vfork);
1273 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1275 * We'll start up with an immediate SIGSTOP.
1277 sigaddset(&p->pending.signal, SIGSTOP);
1278 set_tsk_thread_flag(p, TIF_SIGPENDING);
1281 if (!(clone_flags & CLONE_STOPPED))
1282 wake_up_new_task(p, clone_flags);
1284 p->state = TASK_STOPPED;
1286 if (unlikely (trace)) {
1287 current->ptrace_message = pid;
1288 ptrace_notify ((trace << 8) | SIGTRAP);
1291 if (clone_flags & CLONE_VFORK) {
1292 wait_for_completion(&vfork);
1293 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1294 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1303 void __init proc_caches_init(void)
1305 sighand_cachep = kmem_cache_create("sighand_cache",
1306 sizeof(struct sighand_struct), 0,
1307 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1308 signal_cachep = kmem_cache_create("signal_cache",
1309 sizeof(struct signal_struct), 0,
1310 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1311 files_cachep = kmem_cache_create("files_cache",
1312 sizeof(struct files_struct), 0,
1313 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1314 fs_cachep = kmem_cache_create("fs_cache",
1315 sizeof(struct fs_struct), 0,
1316 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1317 vm_area_cachep = kmem_cache_create("vm_area_struct",
1318 sizeof(struct vm_area_struct), 0,
1319 SLAB_PANIC, NULL, NULL);
1320 mm_cachep = kmem_cache_create("mm_struct",
1321 sizeof(struct mm_struct), 0,
1322 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);