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/ckrm.h>
43 #include <linux/ckrm_tsk.h>
44 #include <linux/ckrm_mem_inline.h>
45 #include <linux/vs_network.h>
46 #include <linux/vs_limit.h>
47 #include <linux/vs_memory.h>
49 #include <asm/pgtable.h>
50 #include <asm/pgalloc.h>
51 #include <asm/uaccess.h>
52 #include <asm/mmu_context.h>
53 #include <asm/cacheflush.h>
54 #include <asm/tlbflush.h>
56 /* The idle threads do not count..
57 * Protected by write_lock_irq(&tasklist_lock)
62 unsigned long total_forks; /* Handle normal Linux uptimes. */
64 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
66 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED; /* outer */
68 EXPORT_SYMBOL(tasklist_lock);
70 int nr_processes(void)
75 for_each_online_cpu(cpu)
76 total += per_cpu(process_counts, cpu);
81 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
82 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
83 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
84 static kmem_cache_t *task_struct_cachep;
87 void free_task(struct task_struct *tsk)
89 free_thread_info(tsk->thread_info);
90 clr_vx_info(&tsk->vx_info);
91 clr_nx_info(&tsk->nx_info);
92 free_task_struct(tsk);
94 EXPORT_SYMBOL(free_task);
96 void __put_task_struct(struct task_struct *tsk)
98 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
99 WARN_ON(atomic_read(&tsk->usage));
100 WARN_ON(tsk == current);
102 if (unlikely(tsk->audit_context))
104 security_task_free(tsk);
106 put_group_info(tsk->group_info);
108 if (!profile_handoff_task(tsk))
112 void __init fork_init(unsigned long mempages)
114 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
115 #ifndef ARCH_MIN_TASKALIGN
116 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
118 /* create a slab on which task_structs can be allocated */
120 kmem_cache_create("task_struct", sizeof(struct task_struct),
121 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
125 * The default maximum number of threads is set to a safe
126 * value: the thread structures can take up at most half
129 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
132 * we need to allow at least 20 threads to boot a system
137 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
138 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
141 static struct task_struct *dup_task_struct(struct task_struct *orig)
143 struct task_struct *tsk;
144 struct thread_info *ti;
146 prepare_to_copy(orig);
148 tsk = alloc_task_struct();
152 ti = alloc_thread_info(tsk);
154 free_task_struct(tsk);
158 *ti = *orig->thread_info;
160 tsk->thread_info = ti;
163 ckrm_cb_newtask(tsk);
164 /* One for us, one for whoever does the "release_task()" (usually parent) */
165 atomic_set(&tsk->usage,2);
166 #ifdef CONFIG_CKRM_RES_MEM
167 INIT_LIST_HEAD(&tsk->mm_peers);
173 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
175 struct vm_area_struct * mpnt, *tmp, **pprev;
176 struct rb_node **rb_link, *rb_parent;
178 unsigned long charge;
179 struct mempolicy *pol;
181 down_write(&oldmm->mmap_sem);
182 flush_cache_mm(current->mm);
185 mm->mmap_cache = NULL;
186 mm->free_area_cache = oldmm->mmap_base;
190 cpus_clear(mm->cpu_vm_mask);
192 rb_link = &mm->mm_rb.rb_node;
196 for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
199 if (mpnt->vm_flags & VM_DONTCOPY) {
200 __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
205 if (mpnt->vm_flags & VM_ACCOUNT) {
206 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
207 if (security_vm_enough_memory(len))
211 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
215 pol = mpol_copy(vma_policy(mpnt));
216 retval = PTR_ERR(pol);
218 goto fail_nomem_policy;
219 vma_set_policy(tmp, pol);
220 tmp->vm_flags &= ~VM_LOCKED;
226 struct inode *inode = file->f_dentry->d_inode;
228 if (tmp->vm_flags & VM_DENYWRITE)
229 atomic_dec(&inode->i_writecount);
231 /* insert tmp into the share list, just after mpnt */
232 spin_lock(&file->f_mapping->i_mmap_lock);
233 flush_dcache_mmap_lock(file->f_mapping);
234 vma_prio_tree_add(tmp, mpnt);
235 flush_dcache_mmap_unlock(file->f_mapping);
236 spin_unlock(&file->f_mapping->i_mmap_lock);
240 * Link in the new vma and copy the page table entries:
241 * link in first so that swapoff can see swap entries,
242 * and try_to_unmap_one's find_vma find the new vma.
244 spin_lock(&mm->page_table_lock);
246 pprev = &tmp->vm_next;
248 __vma_link_rb(mm, tmp, rb_link, rb_parent);
249 rb_link = &tmp->vm_rb.rb_right;
250 rb_parent = &tmp->vm_rb;
253 retval = copy_page_range(mm, current->mm, tmp);
254 spin_unlock(&mm->page_table_lock);
256 if (tmp->vm_ops && tmp->vm_ops->open)
257 tmp->vm_ops->open(tmp);
265 flush_tlb_mm(current->mm);
266 up_write(&oldmm->mmap_sem);
269 kmem_cache_free(vm_area_cachep, tmp);
272 vm_unacct_memory(charge);
276 static inline int mm_alloc_pgd(struct mm_struct * mm)
278 mm->pgd = pgd_alloc(mm);
279 if (unlikely(!mm->pgd))
284 static inline void mm_free_pgd(struct mm_struct * mm)
289 #define dup_mmap(mm, oldmm) (0)
290 #define mm_alloc_pgd(mm) (0)
291 #define mm_free_pgd(mm)
292 #endif /* CONFIG_MMU */
294 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
296 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
297 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
299 #include <linux/init_task.h>
301 static struct mm_struct * mm_init(struct mm_struct * mm)
303 atomic_set(&mm->mm_users, 1);
304 atomic_set(&mm->mm_count, 1);
305 init_rwsem(&mm->mmap_sem);
306 INIT_LIST_HEAD(&mm->mmlist);
307 mm->core_waiters = 0;
309 spin_lock_init(&mm->page_table_lock);
310 rwlock_init(&mm->ioctx_list_lock);
311 mm->ioctx_list = NULL;
312 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
313 mm->free_area_cache = TASK_UNMAPPED_BASE;
314 #ifdef CONFIG_CKRM_RES_MEM
315 INIT_LIST_HEAD(&mm->tasklist);
316 mm->peertask_lock = SPIN_LOCK_UNLOCKED;
319 if (likely(!mm_alloc_pgd(mm))) {
321 set_vx_info(&mm->mm_vx_info, current->vx_info);
329 * Allocate and initialize an mm_struct.
331 struct mm_struct * mm_alloc(void)
333 struct mm_struct * mm;
337 memset(mm, 0, sizeof(*mm));
339 #ifdef CONFIG_CKRM_RES_MEM
340 mm->memclass = GET_MEM_CLASS(current);
341 mem_class_get(mm->memclass);
348 * Called when the last reference to the mm
349 * is dropped: either by a lazy thread or by
350 * mmput. Free the page directory and the mm.
352 void fastcall __mmdrop(struct mm_struct *mm)
354 BUG_ON(mm == &init_mm);
357 #ifdef CONFIG_CKRM_RES_MEM
358 /* class can be null and mm's tasklist can be empty here */
360 mem_class_put(mm->memclass);
364 clr_vx_info(&mm->mm_vx_info);
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));
491 mm->mm_vx_info = NULL;
495 if (init_new_context(tsk,mm))
498 retval = dup_mmap(mm, oldmm);
505 ckrm_init_mm_to_task(mm, tsk);
515 * If init_new_context() failed, we cannot use mmput() to free the mm
516 * because it calls destroy_context()
523 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
525 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
526 /* We don't need to lock fs - think why ;-) */
528 atomic_set(&fs->count, 1);
529 rwlock_init(&fs->lock);
530 fs->umask = old->umask;
531 read_lock(&old->lock);
532 fs->rootmnt = mntget(old->rootmnt);
533 fs->root = dget(old->root);
534 fs->pwdmnt = mntget(old->pwdmnt);
535 fs->pwd = dget(old->pwd);
537 fs->altrootmnt = mntget(old->altrootmnt);
538 fs->altroot = dget(old->altroot);
540 fs->altrootmnt = NULL;
543 read_unlock(&old->lock);
548 struct fs_struct *copy_fs_struct(struct fs_struct *old)
550 return __copy_fs_struct(old);
553 EXPORT_SYMBOL_GPL(copy_fs_struct);
555 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
557 if (clone_flags & CLONE_FS) {
558 atomic_inc(¤t->fs->count);
561 tsk->fs = __copy_fs_struct(current->fs);
567 static int count_open_files(struct files_struct *files, int size)
571 /* Find the last open fd */
572 for (i = size/(8*sizeof(long)); i > 0; ) {
573 if (files->open_fds->fds_bits[--i])
576 i = (i+1) * 8 * sizeof(long);
580 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
582 struct files_struct *oldf, *newf;
583 struct file **old_fds, **new_fds;
584 int open_files, nfds, size, i, error = 0;
587 * A background process may not have any files ...
589 oldf = current->files;
593 if (clone_flags & CLONE_FILES) {
594 atomic_inc(&oldf->count);
599 * Note: we may be using current for both targets (See exec.c)
600 * This works because we cache current->files (old) as oldf. Don't
605 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
609 atomic_set(&newf->count, 1);
611 spin_lock_init(&newf->file_lock);
613 newf->max_fds = NR_OPEN_DEFAULT;
614 newf->max_fdset = __FD_SETSIZE;
615 newf->close_on_exec = &newf->close_on_exec_init;
616 newf->open_fds = &newf->open_fds_init;
617 newf->fd = &newf->fd_array[0];
619 /* We don't yet have the oldf readlock, but even if the old
620 fdset gets grown now, we'll only copy up to "size" fds */
621 size = oldf->max_fdset;
622 if (size > __FD_SETSIZE) {
624 spin_lock(&newf->file_lock);
625 error = expand_fdset(newf, size-1);
626 spin_unlock(&newf->file_lock);
630 spin_lock(&oldf->file_lock);
632 open_files = count_open_files(oldf, size);
635 * Check whether we need to allocate a larger fd array.
636 * Note: we're not a clone task, so the open count won't
639 nfds = NR_OPEN_DEFAULT;
640 if (open_files > nfds) {
641 spin_unlock(&oldf->file_lock);
643 spin_lock(&newf->file_lock);
644 error = expand_fd_array(newf, open_files-1);
645 spin_unlock(&newf->file_lock);
648 nfds = newf->max_fds;
649 spin_lock(&oldf->file_lock);
655 memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
656 memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
658 for (i = open_files; i != 0; i--) {
659 struct file *f = *old_fds++;
664 * The fd may be claimed in the fd bitmap but not yet
665 * instantiated in the files array if a sibling thread
666 * is partway through open(). So make sure that this
667 * fd is available to the new process.
669 FD_CLR(open_files - i, newf->open_fds);
673 spin_unlock(&oldf->file_lock);
675 /* compute the remainder to be cleared */
676 size = (newf->max_fds - open_files) * sizeof(struct file *);
678 /* This is long word aligned thus could use a optimized version */
679 memset(new_fds, 0, size);
681 if (newf->max_fdset > open_files) {
682 int left = (newf->max_fdset-open_files)/8;
683 int start = open_files / (8 * sizeof(unsigned long));
685 memset(&newf->open_fds->fds_bits[start], 0, left);
686 memset(&newf->close_on_exec->fds_bits[start], 0, left);
695 free_fdset (newf->close_on_exec, newf->max_fdset);
696 free_fdset (newf->open_fds, newf->max_fdset);
697 kmem_cache_free(files_cachep, newf);
702 * Helper to unshare the files of the current task.
703 * We don't want to expose copy_files internals to
704 * the exec layer of the kernel.
707 int unshare_files(void)
709 struct files_struct *files = current->files;
715 /* This can race but the race causes us to copy when we don't
716 need to and drop the copy */
717 if(atomic_read(&files->count) == 1)
719 atomic_inc(&files->count);
722 rc = copy_files(0, current);
724 current->files = files;
728 EXPORT_SYMBOL(unshare_files);
730 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
732 struct sighand_struct *sig;
734 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
735 atomic_inc(¤t->sighand->count);
738 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
742 spin_lock_init(&sig->siglock);
743 atomic_set(&sig->count, 1);
744 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
748 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
750 struct signal_struct *sig;
752 if (clone_flags & CLONE_THREAD) {
753 atomic_inc(¤t->signal->count);
754 atomic_inc(¤t->signal->live);
757 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
761 atomic_set(&sig->count, 1);
762 atomic_set(&sig->live, 1);
764 sig->group_exit_code = 0;
765 sig->group_exit_task = NULL;
766 sig->group_stop_count = 0;
768 sig->curr_target = NULL;
769 init_sigpending(&sig->shared_pending);
770 INIT_LIST_HEAD(&sig->posix_timers);
772 sig->tty = current->signal->tty;
773 sig->pgrp = process_group(current);
774 sig->session = current->signal->session;
775 sig->leader = 0; /* session leadership doesn't inherit */
776 sig->tty_old_pgrp = 0;
778 sig->utime = sig->stime = sig->cutime = sig->cstime = 0;
779 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
780 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
782 task_lock(current->group_leader);
783 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
784 task_unlock(current->group_leader);
789 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
791 unsigned long new_flags = p->flags;
793 new_flags &= ~PF_SUPERPRIV;
794 new_flags |= PF_FORKNOEXEC;
795 if (!(clone_flags & CLONE_PTRACE))
797 p->flags = new_flags;
800 asmlinkage long sys_set_tid_address(int __user *tidptr)
802 current->clear_child_tid = tidptr;
808 * This creates a new process as a copy of the old one,
809 * but does not actually start it yet.
811 * It copies the registers, and all the appropriate
812 * parts of the process environment (as per the clone
813 * flags). The actual kick-off is left to the caller.
815 static task_t *copy_process(unsigned long clone_flags,
816 unsigned long stack_start,
817 struct pt_regs *regs,
818 unsigned long stack_size,
819 int __user *parent_tidptr,
820 int __user *child_tidptr,
824 struct task_struct *p = NULL;
827 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
828 return ERR_PTR(-EINVAL);
831 * Thread groups must share signals as well, and detached threads
832 * can only be started up within the thread group.
834 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
835 return ERR_PTR(-EINVAL);
838 * Shared signal handlers imply shared VM. By way of the above,
839 * thread groups also imply shared VM. Blocking this case allows
840 * for various simplifications in other code.
842 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
843 return ERR_PTR(-EINVAL);
845 retval = security_task_create(clone_flags);
850 p = dup_task_struct(current);
856 set_vx_info(&p->vx_info, current->vx_info);
858 set_nx_info(&p->nx_info, current->nx_info);
860 /* check vserver memory */
861 if (p->mm && !(clone_flags & CLONE_VM)) {
862 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
863 vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
867 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
868 if (!vx_rsspages_avail(p->mm, p->mm->rss))
869 goto bad_fork_cleanup_vm;
873 set_vx_info(&p->vx_info, current->vx_info);
875 set_nx_info(&p->nx_info, current->nx_info);
877 /* check vserver memory */
878 if (p->mm && !(clone_flags & CLONE_VM)) {
879 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
880 vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
884 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
885 if (!vx_rsspages_avail(p->mm, p->mm->rss))
886 goto bad_fork_cleanup_vm;
890 if (!vx_nproc_avail(1))
891 goto bad_fork_cleanup_vm;
893 if (atomic_read(&p->user->processes) >=
894 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
895 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
896 p->user != &root_user)
897 goto bad_fork_cleanup_vm;
900 atomic_inc(&p->user->__count);
901 atomic_inc(&p->user->processes);
902 get_group_info(p->group_info);
905 * If multiple threads are within copy_process(), then this check
906 * triggers too late. This doesn't hurt, the check is only there
907 * to stop root fork bombs.
909 if (nr_threads >= max_threads)
910 goto bad_fork_cleanup_count;
912 if (!try_module_get(p->thread_info->exec_domain->module))
913 goto bad_fork_cleanup_count;
915 if (p->binfmt && !try_module_get(p->binfmt->module))
916 goto bad_fork_cleanup_put_domain;
920 copy_flags(clone_flags, p);
923 if (clone_flags & CLONE_PARENT_SETTID)
924 if (put_user(p->pid, parent_tidptr))
925 goto bad_fork_cleanup;
927 p->proc_dentry = NULL;
929 INIT_LIST_HEAD(&p->children);
930 INIT_LIST_HEAD(&p->sibling);
931 init_waitqueue_head(&p->wait_chldexit);
932 p->vfork_done = NULL;
933 spin_lock_init(&p->alloc_lock);
934 spin_lock_init(&p->proc_lock);
936 clear_tsk_thread_flag(p, TIF_SIGPENDING);
937 init_sigpending(&p->pending);
939 p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
940 p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
941 init_timer(&p->real_timer);
942 p->real_timer.data = (unsigned long) p;
944 p->utime = p->stime = 0;
945 p->lock_depth = -1; /* -1 = no lock */
946 do_posix_clock_monotonic_gettime(&p->start_time);
948 p->io_context = NULL;
950 p->audit_context = NULL;
952 p->mempolicy = mpol_copy(p->mempolicy);
953 if (IS_ERR(p->mempolicy)) {
954 retval = PTR_ERR(p->mempolicy);
956 goto bad_fork_cleanup;
961 if (clone_flags & CLONE_THREAD)
962 p->tgid = current->tgid;
964 if ((retval = security_task_alloc(p)))
965 goto bad_fork_cleanup_policy;
966 if ((retval = audit_alloc(p)))
967 goto bad_fork_cleanup_security;
968 /* copy all the process information */
969 if ((retval = copy_semundo(clone_flags, p)))
970 goto bad_fork_cleanup_audit;
971 if ((retval = copy_files(clone_flags, p)))
972 goto bad_fork_cleanup_semundo;
973 if ((retval = copy_fs(clone_flags, p)))
974 goto bad_fork_cleanup_files;
975 if ((retval = copy_sighand(clone_flags, p)))
976 goto bad_fork_cleanup_fs;
977 if ((retval = copy_signal(clone_flags, p)))
978 goto bad_fork_cleanup_sighand;
979 if ((retval = copy_mm(clone_flags, p)))
980 goto bad_fork_cleanup_signal;
981 if ((retval = copy_keys(clone_flags, p)))
982 goto bad_fork_cleanup_mm;
983 if ((retval = copy_namespace(clone_flags, p)))
984 goto bad_fork_cleanup_keys;
985 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
987 goto bad_fork_cleanup_namespace;
989 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
991 * Clear TID on mm_release()?
993 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
996 * Syscall tracing should be turned off in the child regardless
999 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1001 /* Our parent execution domain becomes current domain
1002 These must match for thread signalling to apply */
1004 p->parent_exec_id = p->self_exec_id;
1006 /* ok, now we should be set up.. */
1007 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1008 p->pdeath_signal = 0;
1011 /* Perform scheduler related setup */
1015 * Ok, make it visible to the rest of the system.
1016 * We dont wake it up yet.
1018 p->group_leader = p;
1019 INIT_LIST_HEAD(&p->ptrace_children);
1020 INIT_LIST_HEAD(&p->ptrace_list);
1022 /* Need tasklist lock for parent etc handling! */
1023 write_lock_irq(&tasklist_lock);
1026 * The task hasn't been attached yet, so cpus_allowed mask cannot
1027 * have changed. The cpus_allowed mask of the parent may have
1028 * changed after it was copied first time, and it may then move to
1029 * another CPU - so we re-copy it here and set the child's CPU to
1030 * the parent's CPU. This avoids alot of nasty races.
1032 p->cpus_allowed = current->cpus_allowed;
1033 set_task_cpu(p, smp_processor_id());
1036 * Check for pending SIGKILL! The new thread should not be allowed
1037 * to slip out of an OOM kill. (or normal SIGKILL.)
1039 if (sigismember(¤t->pending.signal, SIGKILL)) {
1040 write_unlock_irq(&tasklist_lock);
1042 goto bad_fork_cleanup_namespace;
1045 /* CLONE_PARENT re-uses the old parent */
1046 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1047 p->real_parent = current->real_parent;
1049 p->real_parent = current;
1050 p->parent = p->real_parent;
1052 if (clone_flags & CLONE_THREAD) {
1053 spin_lock(¤t->sighand->siglock);
1055 * Important: if an exit-all has been started then
1056 * do not create this new thread - the whole thread
1057 * group is supposed to exit anyway.
1059 if (current->signal->group_exit) {
1060 spin_unlock(¤t->sighand->siglock);
1061 write_unlock_irq(&tasklist_lock);
1063 goto bad_fork_cleanup_namespace;
1065 p->group_leader = current->group_leader;
1067 if (current->signal->group_stop_count > 0) {
1069 * There is an all-stop in progress for the group.
1070 * We ourselves will stop as soon as we check signals.
1071 * Make the new thread part of that group stop too.
1073 current->signal->group_stop_count++;
1074 set_tsk_thread_flag(p, TIF_SIGPENDING);
1077 spin_unlock(¤t->sighand->siglock);
1081 if (unlikely(p->ptrace & PT_PTRACED))
1082 __ptrace_link(p, current->parent);
1084 attach_pid(p, PIDTYPE_PID, p->pid);
1085 attach_pid(p, PIDTYPE_TGID, p->tgid);
1086 if (thread_group_leader(p)) {
1087 attach_pid(p, PIDTYPE_PGID, process_group(p));
1088 attach_pid(p, PIDTYPE_SID, p->signal->session);
1090 __get_cpu_var(process_counts)++;
1093 p->ioprio = current->ioprio;
1095 /* p is copy of current */
1098 atomic_inc(&vxi->cvirt.nr_threads);
1101 write_unlock_irq(&tasklist_lock);
1106 return ERR_PTR(retval);
1109 bad_fork_cleanup_namespace:
1111 bad_fork_cleanup_keys:
1113 bad_fork_cleanup_mm:
1116 bad_fork_cleanup_signal:
1118 bad_fork_cleanup_sighand:
1120 bad_fork_cleanup_fs:
1121 exit_fs(p); /* blocking */
1122 bad_fork_cleanup_files:
1123 exit_files(p); /* blocking */
1124 bad_fork_cleanup_semundo:
1126 bad_fork_cleanup_audit:
1128 bad_fork_cleanup_security:
1129 security_task_free(p);
1130 bad_fork_cleanup_policy:
1132 mpol_free(p->mempolicy);
1136 module_put(p->binfmt->module);
1137 bad_fork_cleanup_put_domain:
1138 module_put(p->thread_info->exec_domain->module);
1139 bad_fork_cleanup_count:
1140 put_group_info(p->group_info);
1141 atomic_dec(&p->user->processes);
1143 bad_fork_cleanup_vm:
1144 if (p->mm && !(clone_flags & CLONE_VM))
1145 vx_pages_sub(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
1151 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1153 memset(regs, 0, sizeof(struct pt_regs));
1157 task_t * __devinit fork_idle(int cpu)
1160 struct pt_regs regs;
1162 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0);
1164 return ERR_PTR(-ENOMEM);
1165 init_idle(task, cpu);
1166 unhash_process(task);
1170 static inline int fork_traceflag (unsigned clone_flags)
1172 if (clone_flags & CLONE_UNTRACED)
1174 else if (clone_flags & CLONE_VFORK) {
1175 if (current->ptrace & PT_TRACE_VFORK)
1176 return PTRACE_EVENT_VFORK;
1177 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1178 if (current->ptrace & PT_TRACE_CLONE)
1179 return PTRACE_EVENT_CLONE;
1180 } else if (current->ptrace & PT_TRACE_FORK)
1181 return PTRACE_EVENT_FORK;
1187 * Ok, this is the main fork-routine.
1189 * It copies the process, and if successful kick-starts
1190 * it and waits for it to finish using the VM if required.
1192 long do_fork(unsigned long clone_flags,
1193 unsigned long stack_start,
1194 struct pt_regs *regs,
1195 unsigned long stack_size,
1196 int __user *parent_tidptr,
1197 int __user *child_tidptr)
1199 struct task_struct *p;
1201 long pid = alloc_pidmap();
1205 if (unlikely(current->ptrace)) {
1206 trace = fork_traceflag (clone_flags);
1208 clone_flags |= CLONE_PTRACE;
1211 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1212 if (numtasks_get_ref(current->taskclass, 0) == 0) {
1217 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1219 * Do this prior waking up the new thread - the thread pointer
1220 * might get invalid after that point, if the thread exits quickly.
1223 struct completion vfork;
1227 if (clone_flags & CLONE_VFORK) {
1228 p->vfork_done = &vfork;
1229 init_completion(&vfork);
1232 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1234 * We'll start up with an immediate SIGSTOP.
1236 sigaddset(&p->pending.signal, SIGSTOP);
1237 set_tsk_thread_flag(p, TIF_SIGPENDING);
1240 if (!(clone_flags & CLONE_STOPPED))
1241 wake_up_new_task(p, clone_flags);
1243 p->state = TASK_STOPPED;
1246 if (unlikely (trace)) {
1247 current->ptrace_message = pid;
1248 ptrace_notify ((trace << 8) | SIGTRAP);
1251 if (clone_flags & CLONE_VFORK) {
1252 wait_for_completion(&vfork);
1253 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1254 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1257 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1258 numtasks_put_ref(current->taskclass);
1266 /* SLAB cache for signal_struct structures (tsk->signal) */
1267 kmem_cache_t *signal_cachep;
1269 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1270 kmem_cache_t *sighand_cachep;
1272 /* SLAB cache for files_struct structures (tsk->files) */
1273 kmem_cache_t *files_cachep;
1275 /* SLAB cache for fs_struct structures (tsk->fs) */
1276 kmem_cache_t *fs_cachep;
1278 /* SLAB cache for vm_area_struct structures */
1279 kmem_cache_t *vm_area_cachep;
1281 /* SLAB cache for mm_struct structures (tsk->mm) */
1282 kmem_cache_t *mm_cachep;
1284 void __init proc_caches_init(void)
1286 sighand_cachep = kmem_cache_create("sighand_cache",
1287 sizeof(struct sighand_struct), 0,
1288 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1289 signal_cachep = kmem_cache_create("signal_cache",
1290 sizeof(struct signal_struct), 0,
1291 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1292 files_cachep = kmem_cache_create("files_cache",
1293 sizeof(struct files_struct), 0,
1294 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1295 fs_cachep = kmem_cache_create("fs_cache",
1296 sizeof(struct fs_struct), 0,
1297 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1298 vm_area_cachep = kmem_cache_create("vm_area_struct",
1299 sizeof(struct vm_area_struct), 0,
1300 SLAB_PANIC, NULL, NULL);
1301 mm_cachep = kmem_cache_create("mm_struct",
1302 sizeof(struct mm_struct), 0,
1303 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);