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/security.h>
33 #include <linux/swap.h>
34 #include <linux/syscalls.h>
35 #include <linux/jiffies.h>
36 #include <linux/futex.h>
37 #include <linux/ptrace.h>
38 #include <linux/mount.h>
39 #include <linux/audit.h>
40 #include <linux/profile.h>
41 #include <linux/rmap.h>
42 #include <linux/vs_network.h>
43 #include <linux/vs_limit.h>
44 #include <linux/vs_memory.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>
53 /* The idle threads do not count..
54 * Protected by write_lock_irq(&tasklist_lock)
59 unsigned long total_forks; /* Handle normal Linux uptimes. */
61 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
63 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED; /* 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 void free_task(struct task_struct *tsk)
86 free_thread_info(tsk->thread_info);
87 clr_vx_info(&tsk->vx_info);
88 clr_nx_info(&tsk->nx_info);
89 free_task_struct(tsk);
91 EXPORT_SYMBOL(free_task);
93 void __put_task_struct(struct task_struct *tsk)
95 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
96 WARN_ON(atomic_read(&tsk->usage));
97 WARN_ON(tsk == current);
99 if (unlikely(tsk->audit_context))
101 security_task_free(tsk);
103 put_group_info(tsk->group_info);
105 if (!profile_handoff_task(tsk))
109 void __init fork_init(unsigned long mempages)
111 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
112 #ifndef ARCH_MIN_TASKALIGN
113 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
115 /* create a slab on which task_structs can be allocated */
117 kmem_cache_create("task_struct", sizeof(struct task_struct),
118 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
122 * The default maximum number of threads is set to a safe
123 * value: the thread structures can take up at most half
126 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
129 * we need to allow at least 20 threads to boot a system
134 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
135 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
138 static struct task_struct *dup_task_struct(struct task_struct *orig)
140 struct task_struct *tsk;
141 struct thread_info *ti;
143 prepare_to_copy(orig);
145 tsk = alloc_task_struct();
149 ti = alloc_thread_info(tsk);
151 free_task_struct(tsk);
155 *ti = *orig->thread_info;
157 tsk->thread_info = ti;
160 /* One for us, one for whoever does the "release_task()" (usually parent) */
161 atomic_set(&tsk->usage,2);
166 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
168 struct vm_area_struct * mpnt, *tmp, **pprev;
169 struct rb_node **rb_link, *rb_parent;
171 unsigned long charge;
172 struct mempolicy *pol;
174 down_write(&oldmm->mmap_sem);
175 flush_cache_mm(current->mm);
178 mm->mmap_cache = NULL;
179 mm->free_area_cache = oldmm->mmap_base;
183 cpus_clear(mm->cpu_vm_mask);
185 rb_link = &mm->mm_rb.rb_node;
189 for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
192 if (mpnt->vm_flags & VM_DONTCOPY) {
193 __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
198 if (mpnt->vm_flags & VM_ACCOUNT) {
199 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
200 if (security_vm_enough_memory(len))
204 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
208 pol = mpol_copy(vma_policy(mpnt));
209 retval = PTR_ERR(pol);
211 goto fail_nomem_policy;
212 vma_set_policy(tmp, pol);
213 tmp->vm_flags &= ~VM_LOCKED;
219 struct inode *inode = file->f_dentry->d_inode;
221 if (tmp->vm_flags & VM_DENYWRITE)
222 atomic_dec(&inode->i_writecount);
224 /* insert tmp into the share list, just after mpnt */
225 spin_lock(&file->f_mapping->i_mmap_lock);
226 flush_dcache_mmap_lock(file->f_mapping);
227 vma_prio_tree_add(tmp, mpnt);
228 flush_dcache_mmap_unlock(file->f_mapping);
229 spin_unlock(&file->f_mapping->i_mmap_lock);
233 * Link in the new vma and copy the page table entries:
234 * link in first so that swapoff can see swap entries,
235 * and try_to_unmap_one's find_vma find the new vma.
237 spin_lock(&mm->page_table_lock);
239 pprev = &tmp->vm_next;
241 __vma_link_rb(mm, tmp, rb_link, rb_parent);
242 rb_link = &tmp->vm_rb.rb_right;
243 rb_parent = &tmp->vm_rb;
246 retval = copy_page_range(mm, current->mm, tmp);
247 spin_unlock(&mm->page_table_lock);
249 if (tmp->vm_ops && tmp->vm_ops->open)
250 tmp->vm_ops->open(tmp);
258 flush_tlb_mm(current->mm);
259 up_write(&oldmm->mmap_sem);
262 kmem_cache_free(vm_area_cachep, tmp);
265 vm_unacct_memory(charge);
269 static inline int mm_alloc_pgd(struct mm_struct * mm)
271 mm->pgd = pgd_alloc(mm);
272 if (unlikely(!mm->pgd))
277 static inline void mm_free_pgd(struct mm_struct * mm)
282 #define dup_mmap(mm, oldmm) (0)
283 #define mm_alloc_pgd(mm) (0)
284 #define mm_free_pgd(mm)
285 #endif /* CONFIG_MMU */
287 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
289 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
290 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
292 #include <linux/init_task.h>
294 static struct mm_struct * mm_init(struct mm_struct * mm)
296 atomic_set(&mm->mm_users, 1);
297 atomic_set(&mm->mm_count, 1);
298 init_rwsem(&mm->mmap_sem);
299 INIT_LIST_HEAD(&mm->mmlist);
300 mm->core_waiters = 0;
302 spin_lock_init(&mm->page_table_lock);
303 rwlock_init(&mm->ioctx_list_lock);
304 mm->ioctx_list = NULL;
305 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
306 mm->free_area_cache = TASK_UNMAPPED_BASE;
308 if (likely(!mm_alloc_pgd(mm))) {
310 set_vx_info(&mm->mm_vx_info, current->vx_info);
318 * Allocate and initialize an mm_struct.
320 struct mm_struct * mm_alloc(void)
322 struct mm_struct * mm;
326 memset(mm, 0, sizeof(*mm));
333 * Called when the last reference to the mm
334 * is dropped: either by a lazy thread or by
335 * mmput. Free the page directory and the mm.
337 void fastcall __mmdrop(struct mm_struct *mm)
339 BUG_ON(mm == &init_mm);
342 clr_vx_info(&mm->mm_vx_info);
347 * Decrement the use count and release all resources for an mm.
349 void mmput(struct mm_struct *mm)
351 if (atomic_dec_and_test(&mm->mm_users)) {
354 if (!list_empty(&mm->mmlist)) {
355 spin_lock(&mmlist_lock);
356 list_del(&mm->mmlist);
357 spin_unlock(&mmlist_lock);
363 EXPORT_SYMBOL_GPL(mmput);
366 * get_task_mm - acquire a reference to the task's mm
368 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
369 * this kernel workthread has transiently adopted a user mm with use_mm,
370 * to do its AIO) is not set and if so returns a reference to it, after
371 * bumping up the use count. User must release the mm via mmput()
372 * after use. Typically used by /proc and ptrace.
374 struct mm_struct *get_task_mm(struct task_struct *task)
376 struct mm_struct *mm;
381 if (task->flags & PF_BORROWED_MM)
384 atomic_inc(&mm->mm_users);
389 EXPORT_SYMBOL_GPL(get_task_mm);
391 /* Please note the differences between mmput and mm_release.
392 * mmput is called whenever we stop holding onto a mm_struct,
393 * error success whatever.
395 * mm_release is called after a mm_struct has been removed
396 * from the current process.
398 * This difference is important for error handling, when we
399 * only half set up a mm_struct for a new process and need to restore
400 * the old one. Because we mmput the new mm_struct before
401 * restoring the old one. . .
402 * Eric Biederman 10 January 1998
404 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
406 struct completion *vfork_done = tsk->vfork_done;
408 /* Get rid of any cached register state */
409 deactivate_mm(tsk, mm);
411 /* notify parent sleeping on vfork() */
413 tsk->vfork_done = NULL;
414 complete(vfork_done);
416 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
417 u32 __user * tidptr = tsk->clear_child_tid;
418 tsk->clear_child_tid = NULL;
421 * We don't check the error code - if userspace has
422 * not set up a proper pointer then tough luck.
425 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
429 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
431 struct mm_struct * mm, *oldmm;
434 tsk->min_flt = tsk->maj_flt = 0;
435 tsk->nvcsw = tsk->nivcsw = 0;
438 tsk->active_mm = NULL;
441 * Are we cloning a kernel thread?
443 * We need to steal a active VM for that..
449 if (clone_flags & CLONE_VM) {
450 atomic_inc(&oldmm->mm_users);
453 * There are cases where the PTL is held to ensure no
454 * new threads start up in user mode using an mm, which
455 * allows optimizing out ipis; the tlb_gather_mmu code
458 spin_unlock_wait(&oldmm->page_table_lock);
467 /* Copy the current MM stuff.. */
468 memcpy(mm, oldmm, sizeof(*mm));
469 mm->mm_vx_info = NULL;
473 if (init_new_context(tsk,mm))
476 retval = dup_mmap(mm, oldmm);
492 * If init_new_context() failed, we cannot use mmput() to free the mm
493 * because it calls destroy_context()
500 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
502 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
503 /* We don't need to lock fs - think why ;-) */
505 atomic_set(&fs->count, 1);
506 rwlock_init(&fs->lock);
507 fs->umask = old->umask;
508 read_lock(&old->lock);
509 fs->rootmnt = mntget(old->rootmnt);
510 fs->root = dget(old->root);
511 fs->pwdmnt = mntget(old->pwdmnt);
512 fs->pwd = dget(old->pwd);
514 fs->altrootmnt = mntget(old->altrootmnt);
515 fs->altroot = dget(old->altroot);
517 fs->altrootmnt = NULL;
520 read_unlock(&old->lock);
525 struct fs_struct *copy_fs_struct(struct fs_struct *old)
527 return __copy_fs_struct(old);
530 EXPORT_SYMBOL_GPL(copy_fs_struct);
532 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
534 if (clone_flags & CLONE_FS) {
535 atomic_inc(¤t->fs->count);
538 tsk->fs = __copy_fs_struct(current->fs);
544 static int count_open_files(struct files_struct *files, int size)
548 /* Find the last open fd */
549 for (i = size/(8*sizeof(long)); i > 0; ) {
550 if (files->open_fds->fds_bits[--i])
553 i = (i+1) * 8 * sizeof(long);
557 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
559 struct files_struct *oldf, *newf;
560 struct file **old_fds, **new_fds;
561 int open_files, nfds, size, i, error = 0;
564 * A background process may not have any files ...
566 oldf = current->files;
570 if (clone_flags & CLONE_FILES) {
571 atomic_inc(&oldf->count);
576 * Note: we may be using current for both targets (See exec.c)
577 * This works because we cache current->files (old) as oldf. Don't
582 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
586 atomic_set(&newf->count, 1);
588 spin_lock_init(&newf->file_lock);
590 newf->max_fds = NR_OPEN_DEFAULT;
591 newf->max_fdset = __FD_SETSIZE;
592 newf->close_on_exec = &newf->close_on_exec_init;
593 newf->open_fds = &newf->open_fds_init;
594 newf->fd = &newf->fd_array[0];
596 /* We don't yet have the oldf readlock, but even if the old
597 fdset gets grown now, we'll only copy up to "size" fds */
598 size = oldf->max_fdset;
599 if (size > __FD_SETSIZE) {
601 spin_lock(&newf->file_lock);
602 error = expand_fdset(newf, size-1);
603 spin_unlock(&newf->file_lock);
607 spin_lock(&oldf->file_lock);
609 open_files = count_open_files(oldf, size);
612 * Check whether we need to allocate a larger fd array.
613 * Note: we're not a clone task, so the open count won't
616 nfds = NR_OPEN_DEFAULT;
617 if (open_files > nfds) {
618 spin_unlock(&oldf->file_lock);
620 spin_lock(&newf->file_lock);
621 error = expand_fd_array(newf, open_files-1);
622 spin_unlock(&newf->file_lock);
625 nfds = newf->max_fds;
626 spin_lock(&oldf->file_lock);
632 memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
633 memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
635 for (i = open_files; i != 0; i--) {
636 struct file *f = *old_fds++;
641 * The fd may be claimed in the fd bitmap but not yet
642 * instantiated in the files array if a sibling thread
643 * is partway through open(). So make sure that this
644 * fd is available to the new process.
646 FD_CLR(open_files - i, newf->open_fds);
650 spin_unlock(&oldf->file_lock);
652 /* compute the remainder to be cleared */
653 size = (newf->max_fds - open_files) * sizeof(struct file *);
655 /* This is long word aligned thus could use a optimized version */
656 memset(new_fds, 0, size);
658 if (newf->max_fdset > open_files) {
659 int left = (newf->max_fdset-open_files)/8;
660 int start = open_files / (8 * sizeof(unsigned long));
662 memset(&newf->open_fds->fds_bits[start], 0, left);
663 memset(&newf->close_on_exec->fds_bits[start], 0, left);
672 free_fdset (newf->close_on_exec, newf->max_fdset);
673 free_fdset (newf->open_fds, newf->max_fdset);
674 kmem_cache_free(files_cachep, newf);
679 * Helper to unshare the files of the current task.
680 * We don't want to expose copy_files internals to
681 * the exec layer of the kernel.
684 int unshare_files(void)
686 struct files_struct *files = current->files;
692 /* This can race but the race causes us to copy when we don't
693 need to and drop the copy */
694 if(atomic_read(&files->count) == 1)
696 atomic_inc(&files->count);
699 rc = copy_files(0, current);
701 current->files = files;
705 EXPORT_SYMBOL(unshare_files);
707 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
709 struct sighand_struct *sig;
711 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
712 atomic_inc(¤t->sighand->count);
715 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
719 spin_lock_init(&sig->siglock);
720 atomic_set(&sig->count, 1);
721 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
725 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
727 struct signal_struct *sig;
729 if (clone_flags & CLONE_THREAD) {
730 atomic_inc(¤t->signal->count);
731 atomic_inc(¤t->signal->live);
734 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
738 atomic_set(&sig->count, 1);
739 atomic_set(&sig->live, 1);
741 sig->group_exit_code = 0;
742 sig->group_exit_task = NULL;
743 sig->group_stop_count = 0;
745 sig->curr_target = NULL;
746 init_sigpending(&sig->shared_pending);
747 INIT_LIST_HEAD(&sig->posix_timers);
749 sig->tty = current->signal->tty;
750 sig->pgrp = process_group(current);
751 sig->session = current->signal->session;
752 sig->leader = 0; /* session leadership doesn't inherit */
753 sig->tty_old_pgrp = 0;
755 sig->utime = sig->stime = sig->cutime = sig->cstime = 0;
756 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
757 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
759 task_lock(current->group_leader);
760 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
761 task_unlock(current->group_leader);
766 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
768 unsigned long new_flags = p->flags;
770 new_flags &= ~PF_SUPERPRIV;
771 new_flags |= PF_FORKNOEXEC;
772 if (!(clone_flags & CLONE_PTRACE))
774 p->flags = new_flags;
777 asmlinkage long sys_set_tid_address(int __user *tidptr)
779 current->clear_child_tid = tidptr;
785 * This creates a new process as a copy of the old one,
786 * but does not actually start it yet.
788 * It copies the registers, and all the appropriate
789 * parts of the process environment (as per the clone
790 * flags). The actual kick-off is left to the caller.
792 static task_t *copy_process(unsigned long clone_flags,
793 unsigned long stack_start,
794 struct pt_regs *regs,
795 unsigned long stack_size,
796 int __user *parent_tidptr,
797 int __user *child_tidptr,
801 struct task_struct *p = NULL;
804 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
805 return ERR_PTR(-EINVAL);
808 * Thread groups must share signals as well, and detached threads
809 * can only be started up within the thread group.
811 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
812 return ERR_PTR(-EINVAL);
815 * Shared signal handlers imply shared VM. By way of the above,
816 * thread groups also imply shared VM. Blocking this case allows
817 * for various simplifications in other code.
819 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
820 return ERR_PTR(-EINVAL);
822 retval = security_task_create(clone_flags);
827 p = dup_task_struct(current);
833 set_vx_info(&p->vx_info, current->vx_info);
835 set_nx_info(&p->nx_info, current->nx_info);
837 /* check vserver memory */
838 if (p->mm && !(clone_flags & CLONE_VM)) {
839 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
840 vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
844 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
845 if (!vx_rsspages_avail(p->mm, p->mm->rss))
846 goto bad_fork_cleanup_vm;
850 set_vx_info(&p->vx_info, current->vx_info);
852 set_nx_info(&p->nx_info, current->nx_info);
854 /* check vserver memory */
855 if (p->mm && !(clone_flags & CLONE_VM)) {
856 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
857 vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
861 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
862 if (!vx_rsspages_avail(p->mm, p->mm->rss))
863 goto bad_fork_cleanup_vm;
867 if (!vx_nproc_avail(1))
868 goto bad_fork_cleanup_vm;
870 if (atomic_read(&p->user->processes) >=
871 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
872 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
873 p->user != &root_user)
874 goto bad_fork_cleanup_vm;
877 atomic_inc(&p->user->__count);
878 atomic_inc(&p->user->processes);
879 get_group_info(p->group_info);
882 * If multiple threads are within copy_process(), then this check
883 * triggers too late. This doesn't hurt, the check is only there
884 * to stop root fork bombs.
886 if (nr_threads >= max_threads)
887 goto bad_fork_cleanup_count;
889 if (!try_module_get(p->thread_info->exec_domain->module))
890 goto bad_fork_cleanup_count;
892 if (p->binfmt && !try_module_get(p->binfmt->module))
893 goto bad_fork_cleanup_put_domain;
896 copy_flags(clone_flags, p);
899 if (clone_flags & CLONE_PARENT_SETTID)
900 if (put_user(p->pid, parent_tidptr))
901 goto bad_fork_cleanup;
903 p->proc_dentry = NULL;
905 INIT_LIST_HEAD(&p->children);
906 INIT_LIST_HEAD(&p->sibling);
907 init_waitqueue_head(&p->wait_chldexit);
908 p->vfork_done = NULL;
909 spin_lock_init(&p->alloc_lock);
910 spin_lock_init(&p->proc_lock);
912 clear_tsk_thread_flag(p, TIF_SIGPENDING);
913 init_sigpending(&p->pending);
915 p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
916 p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
917 init_timer(&p->real_timer);
918 p->real_timer.data = (unsigned long) p;
920 p->utime = p->stime = 0;
921 p->lock_depth = -1; /* -1 = no lock */
922 do_posix_clock_monotonic_gettime(&p->start_time);
924 p->io_context = NULL;
926 p->audit_context = NULL;
928 p->mempolicy = mpol_copy(p->mempolicy);
929 if (IS_ERR(p->mempolicy)) {
930 retval = PTR_ERR(p->mempolicy);
932 goto bad_fork_cleanup;
937 if (clone_flags & CLONE_THREAD)
938 p->tgid = current->tgid;
940 if ((retval = security_task_alloc(p)))
941 goto bad_fork_cleanup_policy;
942 if ((retval = audit_alloc(p)))
943 goto bad_fork_cleanup_security;
944 /* copy all the process information */
945 if ((retval = copy_semundo(clone_flags, p)))
946 goto bad_fork_cleanup_audit;
947 if ((retval = copy_files(clone_flags, p)))
948 goto bad_fork_cleanup_semundo;
949 if ((retval = copy_fs(clone_flags, p)))
950 goto bad_fork_cleanup_files;
951 if ((retval = copy_sighand(clone_flags, p)))
952 goto bad_fork_cleanup_fs;
953 if ((retval = copy_signal(clone_flags, p)))
954 goto bad_fork_cleanup_sighand;
955 if ((retval = copy_mm(clone_flags, p)))
956 goto bad_fork_cleanup_signal;
957 if ((retval = copy_keys(clone_flags, p)))
958 goto bad_fork_cleanup_mm;
959 if ((retval = copy_namespace(clone_flags, p)))
960 goto bad_fork_cleanup_keys;
961 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
963 goto bad_fork_cleanup_namespace;
965 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
967 * Clear TID on mm_release()?
969 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
972 * Syscall tracing should be turned off in the child regardless
975 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
977 /* Our parent execution domain becomes current domain
978 These must match for thread signalling to apply */
980 p->parent_exec_id = p->self_exec_id;
982 /* ok, now we should be set up.. */
983 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
984 p->pdeath_signal = 0;
987 /* Perform scheduler related setup */
991 * Ok, make it visible to the rest of the system.
992 * We dont wake it up yet.
995 INIT_LIST_HEAD(&p->ptrace_children);
996 INIT_LIST_HEAD(&p->ptrace_list);
998 /* Need tasklist lock for parent etc handling! */
999 write_lock_irq(&tasklist_lock);
1002 * The task hasn't been attached yet, so cpus_allowed mask cannot
1003 * have changed. The cpus_allowed mask of the parent may have
1004 * changed after it was copied first time, and it may then move to
1005 * another CPU - so we re-copy it here and set the child's CPU to
1006 * the parent's CPU. This avoids alot of nasty races.
1008 p->cpus_allowed = current->cpus_allowed;
1009 set_task_cpu(p, smp_processor_id());
1012 * Check for pending SIGKILL! The new thread should not be allowed
1013 * to slip out of an OOM kill. (or normal SIGKILL.)
1015 if (sigismember(¤t->pending.signal, SIGKILL)) {
1016 write_unlock_irq(&tasklist_lock);
1018 goto bad_fork_cleanup_namespace;
1021 /* CLONE_PARENT re-uses the old parent */
1022 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1023 p->real_parent = current->real_parent;
1025 p->real_parent = current;
1026 p->parent = p->real_parent;
1028 if (clone_flags & CLONE_THREAD) {
1029 spin_lock(¤t->sighand->siglock);
1031 * Important: if an exit-all has been started then
1032 * do not create this new thread - the whole thread
1033 * group is supposed to exit anyway.
1035 if (current->signal->group_exit) {
1036 spin_unlock(¤t->sighand->siglock);
1037 write_unlock_irq(&tasklist_lock);
1039 goto bad_fork_cleanup_namespace;
1041 p->group_leader = current->group_leader;
1043 if (current->signal->group_stop_count > 0) {
1045 * There is an all-stop in progress for the group.
1046 * We ourselves will stop as soon as we check signals.
1047 * Make the new thread part of that group stop too.
1049 current->signal->group_stop_count++;
1050 set_tsk_thread_flag(p, TIF_SIGPENDING);
1053 spin_unlock(¤t->sighand->siglock);
1057 if (unlikely(p->ptrace & PT_PTRACED))
1058 __ptrace_link(p, current->parent);
1060 attach_pid(p, PIDTYPE_PID, p->pid);
1061 attach_pid(p, PIDTYPE_TGID, p->tgid);
1062 if (thread_group_leader(p)) {
1063 attach_pid(p, PIDTYPE_PGID, process_group(p));
1064 attach_pid(p, PIDTYPE_SID, p->signal->session);
1066 __get_cpu_var(process_counts)++;
1069 p->ioprio = current->ioprio;
1071 /* p is copy of current */
1074 atomic_inc(&vxi->cvirt.nr_threads);
1077 write_unlock_irq(&tasklist_lock);
1082 return ERR_PTR(retval);
1085 bad_fork_cleanup_namespace:
1087 bad_fork_cleanup_keys:
1089 bad_fork_cleanup_mm:
1092 bad_fork_cleanup_signal:
1094 bad_fork_cleanup_sighand:
1096 bad_fork_cleanup_fs:
1097 exit_fs(p); /* blocking */
1098 bad_fork_cleanup_files:
1099 exit_files(p); /* blocking */
1100 bad_fork_cleanup_semundo:
1102 bad_fork_cleanup_audit:
1104 bad_fork_cleanup_security:
1105 security_task_free(p);
1106 bad_fork_cleanup_policy:
1108 mpol_free(p->mempolicy);
1112 module_put(p->binfmt->module);
1113 bad_fork_cleanup_put_domain:
1114 module_put(p->thread_info->exec_domain->module);
1115 bad_fork_cleanup_count:
1116 put_group_info(p->group_info);
1117 atomic_dec(&p->user->processes);
1119 bad_fork_cleanup_vm:
1120 if (p->mm && !(clone_flags & CLONE_VM))
1121 vx_pages_sub(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
1127 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1129 memset(regs, 0, sizeof(struct pt_regs));
1133 task_t * __devinit fork_idle(int cpu)
1136 struct pt_regs regs;
1138 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0);
1140 return ERR_PTR(-ENOMEM);
1141 init_idle(task, cpu);
1142 unhash_process(task);
1146 static inline int fork_traceflag (unsigned clone_flags)
1148 if (clone_flags & CLONE_UNTRACED)
1150 else if (clone_flags & CLONE_VFORK) {
1151 if (current->ptrace & PT_TRACE_VFORK)
1152 return PTRACE_EVENT_VFORK;
1153 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1154 if (current->ptrace & PT_TRACE_CLONE)
1155 return PTRACE_EVENT_CLONE;
1156 } else if (current->ptrace & PT_TRACE_FORK)
1157 return PTRACE_EVENT_FORK;
1163 * Ok, this is the main fork-routine.
1165 * It copies the process, and if successful kick-starts
1166 * it and waits for it to finish using the VM if required.
1168 long do_fork(unsigned long clone_flags,
1169 unsigned long stack_start,
1170 struct pt_regs *regs,
1171 unsigned long stack_size,
1172 int __user *parent_tidptr,
1173 int __user *child_tidptr)
1175 struct task_struct *p;
1177 long pid = alloc_pidmap();
1181 if (unlikely(current->ptrace)) {
1182 trace = fork_traceflag (clone_flags);
1184 clone_flags |= CLONE_PTRACE;
1187 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1190 * Do this prior waking up the new thread - the thread pointer
1191 * might get invalid after that point, if the thread exits quickly.
1194 struct completion vfork;
1196 if (clone_flags & CLONE_VFORK) {
1197 p->vfork_done = &vfork;
1198 init_completion(&vfork);
1201 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1203 * We'll start up with an immediate SIGSTOP.
1205 sigaddset(&p->pending.signal, SIGSTOP);
1206 set_tsk_thread_flag(p, TIF_SIGPENDING);
1209 if (!(clone_flags & CLONE_STOPPED))
1210 wake_up_new_task(p, clone_flags);
1212 p->state = TASK_STOPPED;
1215 if (unlikely (trace)) {
1216 current->ptrace_message = pid;
1217 ptrace_notify ((trace << 8) | SIGTRAP);
1220 if (clone_flags & CLONE_VFORK) {
1221 wait_for_completion(&vfork);
1222 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1223 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1232 /* SLAB cache for signal_struct structures (tsk->signal) */
1233 kmem_cache_t *signal_cachep;
1235 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1236 kmem_cache_t *sighand_cachep;
1238 /* SLAB cache for files_struct structures (tsk->files) */
1239 kmem_cache_t *files_cachep;
1241 /* SLAB cache for fs_struct structures (tsk->fs) */
1242 kmem_cache_t *fs_cachep;
1244 /* SLAB cache for vm_area_struct structures */
1245 kmem_cache_t *vm_area_cachep;
1247 /* SLAB cache for mm_struct structures (tsk->mm) */
1248 kmem_cache_t *mm_cachep;
1250 void __init proc_caches_init(void)
1252 sighand_cachep = kmem_cache_create("sighand_cache",
1253 sizeof(struct sighand_struct), 0,
1254 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1255 signal_cachep = kmem_cache_create("signal_cache",
1256 sizeof(struct signal_struct), 0,
1257 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1258 files_cachep = kmem_cache_create("files_cache",
1259 sizeof(struct files_struct), 0,
1260 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1261 fs_cachep = kmem_cache_create("fs_cache",
1262 sizeof(struct fs_struct), 0,
1263 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1264 vm_area_cachep = kmem_cache_create("vm_area_struct",
1265 sizeof(struct vm_area_struct), 0,
1266 SLAB_PANIC, NULL, NULL);
1267 mm_cachep = kmem_cache_create("mm_struct",
1268 sizeof(struct mm_struct), 0,
1269 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);