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_events.h>
43 #include <linux/ckrm_tsk.h>
44 #include <linux/ckrm_tc.h>
45 #include <linux/ckrm_mem_inline.h>
46 #include <linux/vs_network.h>
47 #include <linux/vs_limit.h>
48 #include <linux/vs_memory.h>
50 #include <asm/pgtable.h>
51 #include <asm/pgalloc.h>
52 #include <asm/uaccess.h>
53 #include <asm/mmu_context.h>
54 #include <asm/cacheflush.h>
55 #include <asm/tlbflush.h>
57 /* The idle threads do not count..
58 * Protected by write_lock_irq(&tasklist_lock)
63 unsigned long total_forks; /* Handle normal Linux uptimes. */
65 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
67 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED; /* outer */
69 EXPORT_SYMBOL(tasklist_lock);
71 int nr_processes(void)
76 for_each_online_cpu(cpu)
77 total += per_cpu(process_counts, cpu);
82 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
83 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
84 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
85 static kmem_cache_t *task_struct_cachep;
88 void free_task(struct task_struct *tsk)
90 free_thread_info(tsk->thread_info);
91 clr_vx_info(&tsk->vx_info);
92 clr_nx_info(&tsk->nx_info);
93 free_task_struct(tsk);
95 EXPORT_SYMBOL(free_task);
97 void __put_task_struct(struct task_struct *tsk)
99 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
100 WARN_ON(atomic_read(&tsk->usage));
101 WARN_ON(tsk == current);
103 if (unlikely(tsk->audit_context))
105 security_task_free(tsk);
107 put_group_info(tsk->group_info);
109 if (!profile_handoff_task(tsk))
113 void __init fork_init(unsigned long mempages)
115 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
116 #ifndef ARCH_MIN_TASKALIGN
117 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
119 /* create a slab on which task_structs can be allocated */
121 kmem_cache_create("task_struct", sizeof(struct task_struct),
122 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
126 * The default maximum number of threads is set to a safe
127 * value: the thread structures can take up at most half
130 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
133 * we need to allow at least 20 threads to boot a system
138 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
139 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
142 static struct task_struct *dup_task_struct(struct task_struct *orig)
144 struct task_struct *tsk;
145 struct thread_info *ti;
147 prepare_to_copy(orig);
149 tsk = alloc_task_struct();
153 ti = alloc_thread_info(tsk);
155 free_task_struct(tsk);
159 *ti = *orig->thread_info;
161 tsk->thread_info = ti;
164 ckrm_cb_newtask(tsk);
165 ckrm_task_mm_init(tsk);
166 /* One for us, one for whoever does the "release_task()" (usually parent) */
167 atomic_set(&tsk->usage,2);
172 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
174 struct vm_area_struct * mpnt, *tmp, **pprev;
175 struct rb_node **rb_link, *rb_parent;
177 unsigned long charge;
178 struct mempolicy *pol;
180 down_write(&oldmm->mmap_sem);
181 flush_cache_mm(current->mm);
184 mm->mmap_cache = NULL;
185 mm->free_area_cache = oldmm->mmap_base;
189 cpus_clear(mm->cpu_vm_mask);
191 rb_link = &mm->mm_rb.rb_node;
195 for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
198 if (mpnt->vm_flags & VM_DONTCOPY) {
199 __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
204 if (mpnt->vm_flags & VM_ACCOUNT) {
205 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
206 if (security_vm_enough_memory(len))
210 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
214 pol = mpol_copy(vma_policy(mpnt));
215 retval = PTR_ERR(pol);
217 goto fail_nomem_policy;
218 vma_set_policy(tmp, pol);
219 tmp->vm_flags &= ~VM_LOCKED;
225 struct inode *inode = file->f_dentry->d_inode;
227 if (tmp->vm_flags & VM_DENYWRITE)
228 atomic_dec(&inode->i_writecount);
230 /* insert tmp into the share list, just after mpnt */
231 spin_lock(&file->f_mapping->i_mmap_lock);
232 flush_dcache_mmap_lock(file->f_mapping);
233 vma_prio_tree_add(tmp, mpnt);
234 flush_dcache_mmap_unlock(file->f_mapping);
235 spin_unlock(&file->f_mapping->i_mmap_lock);
239 * Link in the new vma and copy the page table entries:
240 * link in first so that swapoff can see swap entries,
241 * and try_to_unmap_one's find_vma find the new vma.
243 spin_lock(&mm->page_table_lock);
245 pprev = &tmp->vm_next;
247 __vma_link_rb(mm, tmp, rb_link, rb_parent);
248 rb_link = &tmp->vm_rb.rb_right;
249 rb_parent = &tmp->vm_rb;
252 retval = copy_page_range(mm, current->mm, tmp);
253 spin_unlock(&mm->page_table_lock);
255 if (tmp->vm_ops && tmp->vm_ops->open)
256 tmp->vm_ops->open(tmp);
264 flush_tlb_mm(current->mm);
265 up_write(&oldmm->mmap_sem);
268 kmem_cache_free(vm_area_cachep, tmp);
271 vm_unacct_memory(charge);
275 static inline int mm_alloc_pgd(struct mm_struct * mm)
277 mm->pgd = pgd_alloc(mm);
278 if (unlikely(!mm->pgd))
283 static inline void mm_free_pgd(struct mm_struct * mm)
288 #define dup_mmap(mm, oldmm) (0)
289 #define mm_alloc_pgd(mm) (0)
290 #define mm_free_pgd(mm)
291 #endif /* CONFIG_MMU */
293 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
295 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
296 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
298 #include <linux/init_task.h>
300 static struct mm_struct * mm_init(struct mm_struct * mm)
302 atomic_set(&mm->mm_users, 1);
303 atomic_set(&mm->mm_count, 1);
304 init_rwsem(&mm->mmap_sem);
305 INIT_LIST_HEAD(&mm->mmlist);
306 mm->core_waiters = 0;
308 spin_lock_init(&mm->page_table_lock);
309 rwlock_init(&mm->ioctx_list_lock);
310 mm->ioctx_list = NULL;
311 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
312 mm->free_area_cache = TASK_UNMAPPED_BASE;
315 if (likely(!mm_alloc_pgd(mm))) {
317 set_vx_info(&mm->mm_vx_info, current->vx_info);
325 * Allocate and initialize an mm_struct.
327 struct mm_struct * mm_alloc(void)
329 struct mm_struct * mm;
333 memset(mm, 0, sizeof(*mm));
335 ckrm_mm_setclass(mm, ckrm_get_mem_class(current));
341 * Called when the last reference to the mm
342 * is dropped: either by a lazy thread or by
343 * mmput. Free the page directory and the mm.
345 void fastcall __mmdrop(struct mm_struct *mm)
347 BUG_ON(mm == &init_mm);
350 ckrm_mm_clearclass(mm);
351 clr_vx_info(&mm->mm_vx_info);
356 * Decrement the use count and release all resources for an mm.
358 void mmput(struct mm_struct *mm)
360 if (atomic_dec_and_test(&mm->mm_users)) {
363 if (!list_empty(&mm->mmlist)) {
364 spin_lock(&mmlist_lock);
365 list_del(&mm->mmlist);
366 spin_unlock(&mmlist_lock);
372 EXPORT_SYMBOL_GPL(mmput);
375 * get_task_mm - acquire a reference to the task's mm
377 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
378 * this kernel workthread has transiently adopted a user mm with use_mm,
379 * to do its AIO) is not set and if so returns a reference to it, after
380 * bumping up the use count. User must release the mm via mmput()
381 * after use. Typically used by /proc and ptrace.
383 struct mm_struct *get_task_mm(struct task_struct *task)
385 struct mm_struct *mm;
390 if (task->flags & PF_BORROWED_MM)
393 atomic_inc(&mm->mm_users);
398 EXPORT_SYMBOL_GPL(get_task_mm);
400 /* Please note the differences between mmput and mm_release.
401 * mmput is called whenever we stop holding onto a mm_struct,
402 * error success whatever.
404 * mm_release is called after a mm_struct has been removed
405 * from the current process.
407 * This difference is important for error handling, when we
408 * only half set up a mm_struct for a new process and need to restore
409 * the old one. Because we mmput the new mm_struct before
410 * restoring the old one. . .
411 * Eric Biederman 10 January 1998
413 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
415 struct completion *vfork_done = tsk->vfork_done;
417 /* Get rid of any cached register state */
418 deactivate_mm(tsk, mm);
420 /* notify parent sleeping on vfork() */
422 tsk->vfork_done = NULL;
423 complete(vfork_done);
425 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
426 u32 __user * tidptr = tsk->clear_child_tid;
427 tsk->clear_child_tid = NULL;
430 * We don't check the error code - if userspace has
431 * not set up a proper pointer then tough luck.
434 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
438 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
440 struct mm_struct * mm, *oldmm;
443 tsk->min_flt = tsk->maj_flt = 0;
444 tsk->nvcsw = tsk->nivcsw = 0;
447 tsk->active_mm = NULL;
450 * Are we cloning a kernel thread?
452 * We need to steal a active VM for that..
458 if (clone_flags & CLONE_VM) {
459 atomic_inc(&oldmm->mm_users);
462 * There are cases where the PTL is held to ensure no
463 * new threads start up in user mode using an mm, which
464 * allows optimizing out ipis; the tlb_gather_mmu code
467 spin_unlock_wait(&oldmm->page_table_lock);
476 /* Copy the current MM stuff.. */
477 memcpy(mm, oldmm, sizeof(*mm));
478 mm->mm_vx_info = NULL;
482 if (init_new_context(tsk,mm))
485 retval = dup_mmap(mm, oldmm);
490 ckrm_mm_setclass(mm, oldmm->memclass);
493 ckrm_mm_setclass(mm, oldmm->memclass);
494 ckrm_task_mm_set(mm, tsk);
504 * If init_new_context() failed, we cannot use mmput() to free the mm
505 * because it calls destroy_context()
512 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
514 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
515 /* We don't need to lock fs - think why ;-) */
517 atomic_set(&fs->count, 1);
518 rwlock_init(&fs->lock);
519 fs->umask = old->umask;
520 read_lock(&old->lock);
521 fs->rootmnt = mntget(old->rootmnt);
522 fs->root = dget(old->root);
523 fs->pwdmnt = mntget(old->pwdmnt);
524 fs->pwd = dget(old->pwd);
526 fs->altrootmnt = mntget(old->altrootmnt);
527 fs->altroot = dget(old->altroot);
529 fs->altrootmnt = NULL;
532 read_unlock(&old->lock);
537 struct fs_struct *copy_fs_struct(struct fs_struct *old)
539 return __copy_fs_struct(old);
542 EXPORT_SYMBOL_GPL(copy_fs_struct);
544 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
546 if (clone_flags & CLONE_FS) {
547 atomic_inc(¤t->fs->count);
550 tsk->fs = __copy_fs_struct(current->fs);
556 static int count_open_files(struct files_struct *files, int size)
560 /* Find the last open fd */
561 for (i = size/(8*sizeof(long)); i > 0; ) {
562 if (files->open_fds->fds_bits[--i])
565 i = (i+1) * 8 * sizeof(long);
569 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
571 struct files_struct *oldf, *newf;
572 struct file **old_fds, **new_fds;
573 int open_files, nfds, size, i, error = 0;
576 * A background process may not have any files ...
578 oldf = current->files;
582 if (clone_flags & CLONE_FILES) {
583 atomic_inc(&oldf->count);
588 * Note: we may be using current for both targets (See exec.c)
589 * This works because we cache current->files (old) as oldf. Don't
594 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
598 atomic_set(&newf->count, 1);
600 spin_lock_init(&newf->file_lock);
602 newf->max_fds = NR_OPEN_DEFAULT;
603 newf->max_fdset = __FD_SETSIZE;
604 newf->close_on_exec = &newf->close_on_exec_init;
605 newf->open_fds = &newf->open_fds_init;
606 newf->fd = &newf->fd_array[0];
608 /* We don't yet have the oldf readlock, but even if the old
609 fdset gets grown now, we'll only copy up to "size" fds */
610 size = oldf->max_fdset;
611 if (size > __FD_SETSIZE) {
613 spin_lock(&newf->file_lock);
614 error = expand_fdset(newf, size-1);
615 spin_unlock(&newf->file_lock);
619 spin_lock(&oldf->file_lock);
621 open_files = count_open_files(oldf, size);
624 * Check whether we need to allocate a larger fd array.
625 * Note: we're not a clone task, so the open count won't
628 nfds = NR_OPEN_DEFAULT;
629 if (open_files > nfds) {
630 spin_unlock(&oldf->file_lock);
632 spin_lock(&newf->file_lock);
633 error = expand_fd_array(newf, open_files-1);
634 spin_unlock(&newf->file_lock);
637 nfds = newf->max_fds;
638 spin_lock(&oldf->file_lock);
644 memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
645 memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
647 for (i = open_files; i != 0; i--) {
648 struct file *f = *old_fds++;
653 * The fd may be claimed in the fd bitmap but not yet
654 * instantiated in the files array if a sibling thread
655 * is partway through open(). So make sure that this
656 * fd is available to the new process.
658 FD_CLR(open_files - i, newf->open_fds);
662 spin_unlock(&oldf->file_lock);
664 /* compute the remainder to be cleared */
665 size = (newf->max_fds - open_files) * sizeof(struct file *);
667 /* This is long word aligned thus could use a optimized version */
668 memset(new_fds, 0, size);
670 if (newf->max_fdset > open_files) {
671 int left = (newf->max_fdset-open_files)/8;
672 int start = open_files / (8 * sizeof(unsigned long));
674 memset(&newf->open_fds->fds_bits[start], 0, left);
675 memset(&newf->close_on_exec->fds_bits[start], 0, left);
684 free_fdset (newf->close_on_exec, newf->max_fdset);
685 free_fdset (newf->open_fds, newf->max_fdset);
686 kmem_cache_free(files_cachep, newf);
691 * Helper to unshare the files of the current task.
692 * We don't want to expose copy_files internals to
693 * the exec layer of the kernel.
696 int unshare_files(void)
698 struct files_struct *files = current->files;
704 /* This can race but the race causes us to copy when we don't
705 need to and drop the copy */
706 if(atomic_read(&files->count) == 1)
708 atomic_inc(&files->count);
711 rc = copy_files(0, current);
713 current->files = files;
717 EXPORT_SYMBOL(unshare_files);
719 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
721 struct sighand_struct *sig;
723 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
724 atomic_inc(¤t->sighand->count);
727 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
731 spin_lock_init(&sig->siglock);
732 atomic_set(&sig->count, 1);
733 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
737 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
739 struct signal_struct *sig;
741 if (clone_flags & CLONE_THREAD) {
742 atomic_inc(¤t->signal->count);
743 atomic_inc(¤t->signal->live);
746 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
750 atomic_set(&sig->count, 1);
751 atomic_set(&sig->live, 1);
753 sig->group_exit_code = 0;
754 sig->group_exit_task = NULL;
755 sig->group_stop_count = 0;
757 sig->curr_target = NULL;
758 init_sigpending(&sig->shared_pending);
759 INIT_LIST_HEAD(&sig->posix_timers);
761 sig->tty = current->signal->tty;
762 sig->pgrp = process_group(current);
763 sig->session = current->signal->session;
764 sig->leader = 0; /* session leadership doesn't inherit */
765 sig->tty_old_pgrp = 0;
767 sig->utime = sig->stime = sig->cutime = sig->cstime = 0;
768 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
769 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
771 task_lock(current->group_leader);
772 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
773 task_unlock(current->group_leader);
778 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
780 unsigned long new_flags = p->flags;
782 new_flags &= ~PF_SUPERPRIV;
783 new_flags |= PF_FORKNOEXEC;
784 if (!(clone_flags & CLONE_PTRACE))
786 p->flags = new_flags;
789 asmlinkage long sys_set_tid_address(int __user *tidptr)
791 current->clear_child_tid = tidptr;
797 * This creates a new process as a copy of the old one,
798 * but does not actually start it yet.
800 * It copies the registers, and all the appropriate
801 * parts of the process environment (as per the clone
802 * flags). The actual kick-off is left to the caller.
804 static task_t *copy_process(unsigned long clone_flags,
805 unsigned long stack_start,
806 struct pt_regs *regs,
807 unsigned long stack_size,
808 int __user *parent_tidptr,
809 int __user *child_tidptr,
813 struct task_struct *p = NULL;
816 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
817 return ERR_PTR(-EINVAL);
820 * Thread groups must share signals as well, and detached threads
821 * can only be started up within the thread group.
823 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
824 return ERR_PTR(-EINVAL);
827 * Shared signal handlers imply shared VM. By way of the above,
828 * thread groups also imply shared VM. Blocking this case allows
829 * for various simplifications in other code.
831 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
832 return ERR_PTR(-EINVAL);
834 retval = security_task_create(clone_flags);
839 p = dup_task_struct(current);
845 set_vx_info(&p->vx_info, current->vx_info);
847 set_nx_info(&p->nx_info, current->nx_info);
849 /* check vserver memory */
850 if (p->mm && !(clone_flags & CLONE_VM)) {
851 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
852 vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
856 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
857 if (!vx_rsspages_avail(p->mm, p->mm->rss))
858 goto bad_fork_cleanup_vm;
862 set_vx_info(&p->vx_info, current->vx_info);
864 set_nx_info(&p->nx_info, current->nx_info);
866 /* check vserver memory */
867 if (p->mm && !(clone_flags & CLONE_VM)) {
868 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
869 vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
873 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
874 if (!vx_rsspages_avail(p->mm, p->mm->rss))
875 goto bad_fork_cleanup_vm;
879 if (!vx_nproc_avail(1))
880 goto bad_fork_cleanup_vm;
882 if (atomic_read(&p->user->processes) >=
883 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
884 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
885 p->user != &root_user)
886 goto bad_fork_cleanup_vm;
889 atomic_inc(&p->user->__count);
890 atomic_inc(&p->user->processes);
891 get_group_info(p->group_info);
894 * If multiple threads are within copy_process(), then this check
895 * triggers too late. This doesn't hurt, the check is only there
896 * to stop root fork bombs.
898 if (nr_threads >= max_threads)
899 goto bad_fork_cleanup_count;
901 if (!try_module_get(p->thread_info->exec_domain->module))
902 goto bad_fork_cleanup_count;
904 if (p->binfmt && !try_module_get(p->binfmt->module))
905 goto bad_fork_cleanup_put_domain;
909 copy_flags(clone_flags, p);
912 if (clone_flags & CLONE_PARENT_SETTID)
913 if (put_user(p->pid, parent_tidptr))
914 goto bad_fork_cleanup;
916 p->proc_dentry = NULL;
918 INIT_LIST_HEAD(&p->children);
919 INIT_LIST_HEAD(&p->sibling);
920 init_waitqueue_head(&p->wait_chldexit);
921 p->vfork_done = NULL;
922 spin_lock_init(&p->alloc_lock);
923 spin_lock_init(&p->proc_lock);
925 clear_tsk_thread_flag(p, TIF_SIGPENDING);
926 init_sigpending(&p->pending);
928 p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
929 p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
930 init_timer(&p->real_timer);
931 p->real_timer.data = (unsigned long) p;
933 p->utime = p->stime = 0;
934 p->lock_depth = -1; /* -1 = no lock */
935 do_posix_clock_monotonic_gettime(&p->start_time);
937 p->io_context = NULL;
939 p->audit_context = NULL;
941 p->mempolicy = mpol_copy(p->mempolicy);
942 if (IS_ERR(p->mempolicy)) {
943 retval = PTR_ERR(p->mempolicy);
945 goto bad_fork_cleanup;
950 if (clone_flags & CLONE_THREAD)
951 p->tgid = current->tgid;
953 if ((retval = security_task_alloc(p)))
954 goto bad_fork_cleanup_policy;
955 if ((retval = audit_alloc(p)))
956 goto bad_fork_cleanup_security;
957 /* copy all the process information */
958 if ((retval = copy_semundo(clone_flags, p)))
959 goto bad_fork_cleanup_audit;
960 if ((retval = copy_files(clone_flags, p)))
961 goto bad_fork_cleanup_semundo;
962 if ((retval = copy_fs(clone_flags, p)))
963 goto bad_fork_cleanup_files;
964 if ((retval = copy_sighand(clone_flags, p)))
965 goto bad_fork_cleanup_fs;
966 if ((retval = copy_signal(clone_flags, p)))
967 goto bad_fork_cleanup_sighand;
968 if ((retval = copy_mm(clone_flags, p)))
969 goto bad_fork_cleanup_signal;
970 if ((retval = copy_keys(clone_flags, p)))
971 goto bad_fork_cleanup_mm;
972 if ((retval = copy_namespace(clone_flags, p)))
973 goto bad_fork_cleanup_keys;
974 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
976 goto bad_fork_cleanup_namespace;
978 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
980 * Clear TID on mm_release()?
982 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
985 * Syscall tracing should be turned off in the child regardless
988 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
990 /* Our parent execution domain becomes current domain
991 These must match for thread signalling to apply */
993 p->parent_exec_id = p->self_exec_id;
995 /* ok, now we should be set up.. */
996 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
997 p->pdeath_signal = 0;
1000 /* Perform scheduler related setup */
1004 * Ok, make it visible to the rest of the system.
1005 * We dont wake it up yet.
1007 p->group_leader = p;
1008 INIT_LIST_HEAD(&p->ptrace_children);
1009 INIT_LIST_HEAD(&p->ptrace_list);
1011 /* Need tasklist lock for parent etc handling! */
1012 write_lock_irq(&tasklist_lock);
1015 * The task hasn't been attached yet, so cpus_allowed mask cannot
1016 * have changed. The cpus_allowed mask of the parent may have
1017 * changed after it was copied first time, and it may then move to
1018 * another CPU - so we re-copy it here and set the child's CPU to
1019 * the parent's CPU. This avoids alot of nasty races.
1021 p->cpus_allowed = current->cpus_allowed;
1022 set_task_cpu(p, smp_processor_id());
1025 * Check for pending SIGKILL! The new thread should not be allowed
1026 * to slip out of an OOM kill. (or normal SIGKILL.)
1028 if (sigismember(¤t->pending.signal, SIGKILL)) {
1029 write_unlock_irq(&tasklist_lock);
1031 goto bad_fork_cleanup_namespace;
1034 /* CLONE_PARENT re-uses the old parent */
1035 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1036 p->real_parent = current->real_parent;
1038 p->real_parent = current;
1039 p->parent = p->real_parent;
1041 if (clone_flags & CLONE_THREAD) {
1042 spin_lock(¤t->sighand->siglock);
1044 * Important: if an exit-all has been started then
1045 * do not create this new thread - the whole thread
1046 * group is supposed to exit anyway.
1048 if (current->signal->group_exit) {
1049 spin_unlock(¤t->sighand->siglock);
1050 write_unlock_irq(&tasklist_lock);
1052 goto bad_fork_cleanup_namespace;
1054 p->group_leader = current->group_leader;
1056 if (current->signal->group_stop_count > 0) {
1058 * There is an all-stop in progress for the group.
1059 * We ourselves will stop as soon as we check signals.
1060 * Make the new thread part of that group stop too.
1062 current->signal->group_stop_count++;
1063 set_tsk_thread_flag(p, TIF_SIGPENDING);
1066 spin_unlock(¤t->sighand->siglock);
1070 if (unlikely(p->ptrace & PT_PTRACED))
1071 __ptrace_link(p, current->parent);
1073 attach_pid(p, PIDTYPE_PID, p->pid);
1074 attach_pid(p, PIDTYPE_TGID, p->tgid);
1075 if (thread_group_leader(p)) {
1076 attach_pid(p, PIDTYPE_PGID, process_group(p));
1077 attach_pid(p, PIDTYPE_SID, p->signal->session);
1079 __get_cpu_var(process_counts)++;
1082 p->ioprio = current->ioprio;
1084 /* p is copy of current */
1087 atomic_inc(&vxi->cvirt.nr_threads);
1090 write_unlock_irq(&tasklist_lock);
1095 return ERR_PTR(retval);
1098 bad_fork_cleanup_namespace:
1100 bad_fork_cleanup_keys:
1102 bad_fork_cleanup_mm:
1105 bad_fork_cleanup_signal:
1107 bad_fork_cleanup_sighand:
1109 bad_fork_cleanup_fs:
1110 exit_fs(p); /* blocking */
1111 bad_fork_cleanup_files:
1112 exit_files(p); /* blocking */
1113 bad_fork_cleanup_semundo:
1115 bad_fork_cleanup_audit:
1117 bad_fork_cleanup_security:
1118 security_task_free(p);
1119 bad_fork_cleanup_policy:
1121 mpol_free(p->mempolicy);
1125 module_put(p->binfmt->module);
1126 bad_fork_cleanup_put_domain:
1127 module_put(p->thread_info->exec_domain->module);
1128 bad_fork_cleanup_count:
1129 put_group_info(p->group_info);
1130 atomic_dec(&p->user->processes);
1132 bad_fork_cleanup_vm:
1133 if (p->mm && !(clone_flags & CLONE_VM))
1134 vx_pages_sub(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
1140 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1142 memset(regs, 0, sizeof(struct pt_regs));
1146 task_t * __devinit fork_idle(int cpu)
1149 struct pt_regs regs;
1151 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0);
1153 return ERR_PTR(-ENOMEM);
1154 init_idle(task, cpu);
1155 unhash_process(task);
1159 static inline int fork_traceflag (unsigned clone_flags)
1161 if (clone_flags & CLONE_UNTRACED)
1163 else if (clone_flags & CLONE_VFORK) {
1164 if (current->ptrace & PT_TRACE_VFORK)
1165 return PTRACE_EVENT_VFORK;
1166 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1167 if (current->ptrace & PT_TRACE_CLONE)
1168 return PTRACE_EVENT_CLONE;
1169 } else if (current->ptrace & PT_TRACE_FORK)
1170 return PTRACE_EVENT_FORK;
1176 * Ok, this is the main fork-routine.
1178 * It copies the process, and if successful kick-starts
1179 * it and waits for it to finish using the VM if required.
1181 long do_fork(unsigned long clone_flags,
1182 unsigned long stack_start,
1183 struct pt_regs *regs,
1184 unsigned long stack_size,
1185 int __user *parent_tidptr,
1186 int __user *child_tidptr)
1188 struct task_struct *p;
1190 long pid = alloc_pidmap();
1194 if (unlikely(current->ptrace)) {
1195 trace = fork_traceflag (clone_flags);
1197 clone_flags |= CLONE_PTRACE;
1200 if (numtasks_get_ref(current->taskclass, 0) == 0) {
1203 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1206 * Do this prior waking up the new thread - the thread pointer
1207 * might get invalid after that point, if the thread exits quickly.
1210 struct completion vfork;
1214 if (clone_flags & CLONE_VFORK) {
1215 p->vfork_done = &vfork;
1216 init_completion(&vfork);
1219 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1221 * We'll start up with an immediate SIGSTOP.
1223 sigaddset(&p->pending.signal, SIGSTOP);
1224 set_tsk_thread_flag(p, TIF_SIGPENDING);
1227 if (!(clone_flags & CLONE_STOPPED))
1228 wake_up_new_task(p, clone_flags);
1230 p->state = TASK_STOPPED;
1233 if (unlikely (trace)) {
1234 current->ptrace_message = pid;
1235 ptrace_notify ((trace << 8) | SIGTRAP);
1238 if (clone_flags & CLONE_VFORK) {
1239 wait_for_completion(&vfork);
1240 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1241 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1244 numtasks_put_ref(current->taskclass);
1251 /* SLAB cache for signal_struct structures (tsk->signal) */
1252 kmem_cache_t *signal_cachep;
1254 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1255 kmem_cache_t *sighand_cachep;
1257 /* SLAB cache for files_struct structures (tsk->files) */
1258 kmem_cache_t *files_cachep;
1260 /* SLAB cache for fs_struct structures (tsk->fs) */
1261 kmem_cache_t *fs_cachep;
1263 /* SLAB cache for vm_area_struct structures */
1264 kmem_cache_t *vm_area_cachep;
1266 /* SLAB cache for mm_struct structures (tsk->mm) */
1267 kmem_cache_t *mm_cachep;
1269 void __init proc_caches_init(void)
1271 sighand_cachep = kmem_cache_create("sighand_cache",
1272 sizeof(struct sighand_struct), 0,
1273 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1274 signal_cachep = kmem_cache_create("signal_cache",
1275 sizeof(struct signal_struct), 0,
1276 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1277 files_cachep = kmem_cache_create("files_cache",
1278 sizeof(struct files_struct), 0,
1279 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1280 fs_cachep = kmem_cache_create("fs_cache",
1281 sizeof(struct fs_struct), 0,
1282 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1283 vm_area_cachep = kmem_cache_create("vm_area_struct",
1284 sizeof(struct vm_area_struct), 0,
1285 SLAB_PANIC, NULL, NULL);
1286 mm_cachep = kmem_cache_create("mm_struct",
1287 sizeof(struct mm_struct), 0,
1288 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);