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/binfmts.h>
28 #include <linux/mman.h>
30 #include <linux/cpu.h>
31 #include <linux/security.h>
32 #include <linux/swap.h>
33 #include <linux/syscalls.h>
34 #include <linux/jiffies.h>
35 #include <linux/futex.h>
36 #include <linux/ptrace.h>
37 #include <linux/mount.h>
38 #include <linux/audit.h>
39 #include <linux/rmap.h>
40 #include <linux/vs_network.h>
41 #include <linux/vs_limit.h>
42 #include <linux/vs_memory.h>
43 #include <linux/ckrm.h>
44 #include <linux/ckrm_tsk.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 static 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);
92 void __put_task_struct(struct task_struct *tsk)
94 WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
95 WARN_ON(atomic_read(&tsk->usage));
96 WARN_ON(tsk == current);
98 if (unlikely(tsk->audit_context))
100 security_task_free(tsk);
102 put_group_info(tsk->group_info);
106 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
110 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
111 spin_lock_irqsave(&q->lock, flags);
112 __add_wait_queue(q, wait);
113 spin_unlock_irqrestore(&q->lock, flags);
116 EXPORT_SYMBOL(add_wait_queue);
118 void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
122 wait->flags |= WQ_FLAG_EXCLUSIVE;
123 spin_lock_irqsave(&q->lock, flags);
124 __add_wait_queue_tail(q, wait);
125 spin_unlock_irqrestore(&q->lock, flags);
128 EXPORT_SYMBOL(add_wait_queue_exclusive);
130 void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
134 spin_lock_irqsave(&q->lock, flags);
135 __remove_wait_queue(q, wait);
136 spin_unlock_irqrestore(&q->lock, flags);
139 EXPORT_SYMBOL(remove_wait_queue);
143 * Note: we use "set_current_state()" _after_ the wait-queue add,
144 * because we need a memory barrier there on SMP, so that any
145 * wake-function that tests for the wait-queue being active
146 * will be guaranteed to see waitqueue addition _or_ subsequent
147 * tests in this thread will see the wakeup having taken place.
149 * The spin_unlock() itself is semi-permeable and only protects
150 * one way (it only protects stuff inside the critical region and
151 * stops them from bleeding out - it would still allow subsequent
152 * loads to move into the the critical region).
154 void fastcall prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
158 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
159 spin_lock_irqsave(&q->lock, flags);
160 if (list_empty(&wait->task_list))
161 __add_wait_queue(q, wait);
162 set_current_state(state);
163 spin_unlock_irqrestore(&q->lock, flags);
166 EXPORT_SYMBOL(prepare_to_wait);
169 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
173 wait->flags |= WQ_FLAG_EXCLUSIVE;
174 spin_lock_irqsave(&q->lock, flags);
175 if (list_empty(&wait->task_list))
176 __add_wait_queue_tail(q, wait);
177 set_current_state(state);
178 spin_unlock_irqrestore(&q->lock, flags);
181 EXPORT_SYMBOL(prepare_to_wait_exclusive);
183 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
187 __set_current_state(TASK_RUNNING);
189 * We can check for list emptiness outside the lock
191 * - we use the "careful" check that verifies both
192 * the next and prev pointers, so that there cannot
193 * be any half-pending updates in progress on other
194 * CPU's that we haven't seen yet (and that might
195 * still change the stack area.
197 * - all other users take the lock (ie we can only
198 * have _one_ other CPU that looks at or modifies
201 if (!list_empty_careful(&wait->task_list)) {
202 spin_lock_irqsave(&q->lock, flags);
203 list_del_init(&wait->task_list);
204 spin_unlock_irqrestore(&q->lock, flags);
208 EXPORT_SYMBOL(finish_wait);
210 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
212 int ret = default_wake_function(wait, mode, sync, key);
215 list_del_init(&wait->task_list);
219 EXPORT_SYMBOL(autoremove_wake_function);
221 void __init fork_init(unsigned long mempages)
223 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
224 #ifndef ARCH_MIN_TASKALIGN
225 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
227 /* create a slab on which task_structs can be allocated */
229 kmem_cache_create("task_struct", sizeof(struct task_struct),
230 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
234 * The default maximum number of threads is set to a safe
235 * value: the thread structures can take up at most half
238 max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
240 * we need to allow at least 20 threads to boot a system
245 init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
246 init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
249 static struct task_struct *dup_task_struct(struct task_struct *orig)
251 struct task_struct *tsk;
252 struct thread_info *ti;
254 prepare_to_copy(orig);
256 tsk = alloc_task_struct();
260 ti = alloc_thread_info(tsk);
262 free_task_struct(tsk);
266 *ti = *orig->thread_info;
268 tsk->thread_info = ti;
271 ckrm_cb_newtask(tsk);
272 /* One for us, one for whoever does the "release_task()" (usually parent) */
273 atomic_set(&tsk->usage,2);
278 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
280 struct vm_area_struct * mpnt, *tmp, **pprev;
281 struct rb_node **rb_link, *rb_parent;
283 unsigned long charge;
284 struct mempolicy *pol;
286 down_write(&oldmm->mmap_sem);
287 flush_cache_mm(current->mm);
290 mm->mmap_cache = NULL;
291 mm->free_area_cache = oldmm->mmap_base;
294 cpus_clear(mm->cpu_vm_mask);
296 rb_link = &mm->mm_rb.rb_node;
301 * Add it to the mmlist after the parent.
302 * Doing it this way means that we can order the list,
303 * and fork() won't mess up the ordering significantly.
304 * Add it first so that swapoff can see any swap entries.
306 spin_lock(&mmlist_lock);
307 list_add(&mm->mmlist, ¤t->mm->mmlist);
309 spin_unlock(&mmlist_lock);
311 for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
314 if(mpnt->vm_flags & VM_DONTCOPY)
317 if (mpnt->vm_flags & VM_ACCOUNT) {
318 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
319 if (security_vm_enough_memory(len))
323 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
327 pol = mpol_copy(vma_policy(mpnt));
328 retval = PTR_ERR(pol);
330 goto fail_nomem_policy;
331 vma_set_policy(tmp, pol);
332 tmp->vm_flags &= ~VM_LOCKED;
336 vma_prio_tree_init(tmp);
339 struct inode *inode = file->f_dentry->d_inode;
341 if (tmp->vm_flags & VM_DENYWRITE)
342 atomic_dec(&inode->i_writecount);
344 /* insert tmp into the share list, just after mpnt */
345 spin_lock(&file->f_mapping->i_mmap_lock);
346 flush_dcache_mmap_lock(file->f_mapping);
347 vma_prio_tree_add(tmp, mpnt);
348 flush_dcache_mmap_unlock(file->f_mapping);
349 spin_unlock(&file->f_mapping->i_mmap_lock);
353 * Link in the new vma and copy the page table entries:
354 * link in first so that swapoff can see swap entries,
355 * and try_to_unmap_one's find_vma find the new vma.
357 spin_lock(&mm->page_table_lock);
359 pprev = &tmp->vm_next;
361 __vma_link_rb(mm, tmp, rb_link, rb_parent);
362 rb_link = &tmp->vm_rb.rb_right;
363 rb_parent = &tmp->vm_rb;
366 retval = copy_page_range(mm, current->mm, tmp);
367 spin_unlock(&mm->page_table_lock);
369 if (tmp->vm_ops && tmp->vm_ops->open)
370 tmp->vm_ops->open(tmp);
378 flush_tlb_mm(current->mm);
379 up_write(&oldmm->mmap_sem);
382 kmem_cache_free(vm_area_cachep, tmp);
385 vm_unacct_memory(charge);
389 static inline int mm_alloc_pgd(struct mm_struct * mm)
391 mm->pgd = pgd_alloc(mm);
392 if (unlikely(!mm->pgd))
397 static inline void mm_free_pgd(struct mm_struct * mm)
402 #define dup_mmap(mm, oldmm) (0)
403 #define mm_alloc_pgd(mm) (0)
404 #define mm_free_pgd(mm)
405 #endif /* CONFIG_MMU */
407 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
410 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
411 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
413 #include <linux/init_task.h>
415 static struct mm_struct * mm_init(struct mm_struct * mm)
417 atomic_set(&mm->mm_users, 1);
418 atomic_set(&mm->mm_count, 1);
419 init_rwsem(&mm->mmap_sem);
420 mm->core_waiters = 0;
421 mm->page_table_lock = SPIN_LOCK_UNLOCKED;
422 mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
423 mm->ioctx_list = NULL;
424 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
425 mm->free_area_cache = TASK_UNMAPPED_BASE;
427 if (likely(!mm_alloc_pgd(mm))) {
429 set_vx_info(&mm->mm_vx_info, current->vx_info);
437 * Allocate and initialize an mm_struct.
439 struct mm_struct * mm_alloc(void)
441 struct mm_struct * mm;
445 memset(mm, 0, sizeof(*mm));
452 * Called when the last reference to the mm
453 * is dropped: either by a lazy thread or by
454 * mmput. Free the page directory and the mm.
456 void fastcall __mmdrop(struct mm_struct *mm)
458 BUG_ON(mm == &init_mm);
461 clr_vx_info(&mm->mm_vx_info);
466 * Decrement the use count and release all resources for an mm.
468 void mmput(struct mm_struct *mm)
470 if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
471 list_del(&mm->mmlist);
473 spin_unlock(&mmlist_lock);
482 * Checks if the use count of an mm is non-zero and if so
483 * returns a reference to it after bumping up the use count.
484 * If the use count is zero, it means this mm is going away,
487 struct mm_struct *mmgrab(struct mm_struct *mm)
489 spin_lock(&mmlist_lock);
490 if (!atomic_read(&mm->mm_users))
493 atomic_inc(&mm->mm_users);
494 spin_unlock(&mmlist_lock);
498 /* Please note the differences between mmput and mm_release.
499 * mmput is called whenever we stop holding onto a mm_struct,
500 * error success whatever.
502 * mm_release is called after a mm_struct has been removed
503 * from the current process.
505 * This difference is important for error handling, when we
506 * only half set up a mm_struct for a new process and need to restore
507 * the old one. Because we mmput the new mm_struct before
508 * restoring the old one. . .
509 * Eric Biederman 10 January 1998
511 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
513 struct completion *vfork_done = tsk->vfork_done;
515 /* Get rid of any cached register state */
516 deactivate_mm(tsk, mm);
518 /* notify parent sleeping on vfork() */
520 tsk->vfork_done = NULL;
521 complete(vfork_done);
523 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
524 u32 __user * tidptr = tsk->clear_child_tid;
525 tsk->clear_child_tid = NULL;
528 * We don't check the error code - if userspace has
529 * not set up a proper pointer then tough luck.
532 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
536 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
538 struct mm_struct * mm, *oldmm;
541 tsk->min_flt = tsk->maj_flt = 0;
542 tsk->cmin_flt = tsk->cmaj_flt = 0;
543 tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
546 tsk->active_mm = NULL;
549 * Are we cloning a kernel thread?
551 * We need to steal a active VM for that..
557 if (clone_flags & CLONE_VM) {
558 atomic_inc(&oldmm->mm_users);
561 * There are cases where the PTL is held to ensure no
562 * new threads start up in user mode using an mm, which
563 * allows optimizing out ipis; the tlb_gather_mmu code
566 spin_unlock_wait(&oldmm->page_table_lock);
575 /* Copy the current MM stuff.. */
576 memcpy(mm, oldmm, sizeof(*mm));
577 mm->mm_vx_info = NULL;
581 if (init_new_context(tsk,mm))
584 retval = dup_mmap(mm, oldmm);
600 * If init_new_context() failed, we cannot use mmput() to free the mm
601 * because it calls destroy_context()
608 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
610 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
611 /* We don't need to lock fs - think why ;-) */
613 atomic_set(&fs->count, 1);
614 fs->lock = RW_LOCK_UNLOCKED;
615 fs->umask = old->umask;
616 read_lock(&old->lock);
617 fs->rootmnt = mntget(old->rootmnt);
618 fs->root = dget(old->root);
619 fs->pwdmnt = mntget(old->pwdmnt);
620 fs->pwd = dget(old->pwd);
622 fs->altrootmnt = mntget(old->altrootmnt);
623 fs->altroot = dget(old->altroot);
625 fs->altrootmnt = NULL;
628 read_unlock(&old->lock);
633 struct fs_struct *copy_fs_struct(struct fs_struct *old)
635 return __copy_fs_struct(old);
638 EXPORT_SYMBOL_GPL(copy_fs_struct);
640 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
642 if (clone_flags & CLONE_FS) {
643 atomic_inc(¤t->fs->count);
646 tsk->fs = __copy_fs_struct(current->fs);
652 static int count_open_files(struct files_struct *files, int size)
656 /* Find the last open fd */
657 for (i = size/(8*sizeof(long)); i > 0; ) {
658 if (files->open_fds->fds_bits[--i])
661 i = (i+1) * 8 * sizeof(long);
665 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
667 struct files_struct *oldf, *newf;
668 struct file **old_fds, **new_fds;
669 int open_files, nfds, size, i, error = 0;
672 * A background process may not have any files ...
674 oldf = current->files;
678 if (clone_flags & CLONE_FILES) {
679 atomic_inc(&oldf->count);
684 * Note: we may be using current for both targets (See exec.c)
685 * This works because we cache current->files (old) as oldf. Don't
690 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
694 atomic_set(&newf->count, 1);
696 newf->file_lock = SPIN_LOCK_UNLOCKED;
698 newf->max_fds = NR_OPEN_DEFAULT;
699 newf->max_fdset = __FD_SETSIZE;
700 newf->close_on_exec = &newf->close_on_exec_init;
701 newf->open_fds = &newf->open_fds_init;
702 newf->fd = &newf->fd_array[0];
704 /* We don't yet have the oldf readlock, but even if the old
705 fdset gets grown now, we'll only copy up to "size" fds */
706 size = oldf->max_fdset;
707 if (size > __FD_SETSIZE) {
709 spin_lock(&newf->file_lock);
710 error = expand_fdset(newf, size-1);
711 spin_unlock(&newf->file_lock);
715 spin_lock(&oldf->file_lock);
717 open_files = count_open_files(oldf, size);
720 * Check whether we need to allocate a larger fd array.
721 * Note: we're not a clone task, so the open count won't
724 nfds = NR_OPEN_DEFAULT;
725 if (open_files > nfds) {
726 spin_unlock(&oldf->file_lock);
728 spin_lock(&newf->file_lock);
729 error = expand_fd_array(newf, open_files-1);
730 spin_unlock(&newf->file_lock);
733 nfds = newf->max_fds;
734 spin_lock(&oldf->file_lock);
740 memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
741 memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
743 for (i = open_files; i != 0; i--) {
744 struct file *f = *old_fds++;
749 spin_unlock(&oldf->file_lock);
751 /* compute the remainder to be cleared */
752 size = (newf->max_fds - open_files) * sizeof(struct file *);
754 /* This is long word aligned thus could use a optimized version */
755 memset(new_fds, 0, size);
757 if (newf->max_fdset > open_files) {
758 int left = (newf->max_fdset-open_files)/8;
759 int start = open_files / (8 * sizeof(unsigned long));
761 memset(&newf->open_fds->fds_bits[start], 0, left);
762 memset(&newf->close_on_exec->fds_bits[start], 0, left);
771 free_fdset (newf->close_on_exec, newf->max_fdset);
772 free_fdset (newf->open_fds, newf->max_fdset);
773 kmem_cache_free(files_cachep, newf);
778 * Helper to unshare the files of the current task.
779 * We don't want to expose copy_files internals to
780 * the exec layer of the kernel.
783 int unshare_files(void)
785 struct files_struct *files = current->files;
791 /* This can race but the race causes us to copy when we don't
792 need to and drop the copy */
793 if(atomic_read(&files->count) == 1)
795 atomic_inc(&files->count);
798 rc = copy_files(0, current);
800 current->files = files;
804 EXPORT_SYMBOL(unshare_files);
806 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
808 struct sighand_struct *sig;
810 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
811 atomic_inc(¤t->sighand->count);
814 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
818 spin_lock_init(&sig->siglock);
819 atomic_set(&sig->count, 1);
820 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
824 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
826 struct signal_struct *sig;
828 if (clone_flags & CLONE_THREAD) {
829 atomic_inc(¤t->signal->count);
832 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
836 atomic_set(&sig->count, 1);
838 sig->group_exit_code = 0;
839 sig->group_exit_task = NULL;
840 sig->group_stop_count = 0;
841 sig->curr_target = NULL;
842 init_sigpending(&sig->shared_pending);
843 INIT_LIST_HEAD(&sig->posix_timers);
845 sig->tty = current->signal->tty;
846 sig->pgrp = process_group(current);
847 sig->session = current->signal->session;
848 sig->leader = 0; /* session leadership doesn't inherit */
849 sig->tty_old_pgrp = 0;
854 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
856 unsigned long new_flags = p->flags;
858 new_flags &= ~PF_SUPERPRIV;
859 new_flags |= PF_FORKNOEXEC;
860 if (!(clone_flags & CLONE_PTRACE))
862 p->flags = new_flags;
865 asmlinkage long sys_set_tid_address(int __user *tidptr)
867 current->clear_child_tid = tidptr;
873 * This creates a new process as a copy of the old one,
874 * but does not actually start it yet.
876 * It copies the registers, and all the appropriate
877 * parts of the process environment (as per the clone
878 * flags). The actual kick-off is left to the caller.
880 struct task_struct *copy_process(unsigned long clone_flags,
881 unsigned long stack_start,
882 struct pt_regs *regs,
883 unsigned long stack_size,
884 int __user *parent_tidptr,
885 int __user *child_tidptr)
888 struct task_struct *p = NULL;
891 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
892 return ERR_PTR(-EINVAL);
895 * Thread groups must share signals as well, and detached threads
896 * can only be started up within the thread group.
898 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
899 return ERR_PTR(-EINVAL);
902 * Shared signal handlers imply shared VM. By way of the above,
903 * thread groups also imply shared VM. Blocking this case allows
904 * for various simplifications in other code.
906 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
907 return ERR_PTR(-EINVAL);
909 retval = security_task_create(clone_flags);
914 p = dup_task_struct(current);
920 set_vx_info(&p->vx_info, current->vx_info);
922 set_nx_info(&p->nx_info, current->nx_info);
924 /* check vserver memory */
925 if (p->mm && !(clone_flags & CLONE_VM)) {
926 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
927 vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
931 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
932 if (!vx_rsspages_avail(p->mm, p->mm->rss))
933 goto bad_fork_cleanup_vm;
937 if (!vx_nproc_avail(1))
938 goto bad_fork_cleanup_vm;
940 if (atomic_read(&p->user->processes) >=
941 p->rlim[RLIMIT_NPROC].rlim_cur) {
942 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
943 p->user != &root_user)
944 goto bad_fork_cleanup_vm;
947 atomic_inc(&p->user->__count);
948 atomic_inc(&p->user->processes);
949 get_group_info(p->group_info);
952 * If multiple threads are within copy_process(), then this check
953 * triggers too late. This doesn't hurt, the check is only there
954 * to stop root fork bombs.
956 if (nr_threads >= max_threads)
957 goto bad_fork_cleanup_count;
959 if (!try_module_get(p->thread_info->exec_domain->module))
960 goto bad_fork_cleanup_count;
962 if (p->binfmt && !try_module_get(p->binfmt->module))
963 goto bad_fork_cleanup_put_domain;
967 copy_flags(clone_flags, p);
968 if (clone_flags & CLONE_IDLETASK)
971 p->pid = alloc_pidmap();
973 goto bad_fork_cleanup;
976 if (clone_flags & CLONE_PARENT_SETTID)
977 if (put_user(p->pid, parent_tidptr))
978 goto bad_fork_cleanup;
980 p->proc_dentry = NULL;
982 INIT_LIST_HEAD(&p->children);
983 INIT_LIST_HEAD(&p->sibling);
984 init_waitqueue_head(&p->wait_chldexit);
985 p->vfork_done = NULL;
986 spin_lock_init(&p->alloc_lock);
987 spin_lock_init(&p->proc_lock);
989 clear_tsk_thread_flag(p, TIF_SIGPENDING);
990 init_sigpending(&p->pending);
992 p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
993 p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
994 init_timer(&p->real_timer);
995 p->real_timer.data = (unsigned long) p;
997 p->utime = p->stime = 0;
998 p->cutime = p->cstime = 0;
999 p->lock_depth = -1; /* -1 = no lock */
1000 p->start_time = get_jiffies_64();
1002 p->io_context = NULL;
1003 p->audit_context = NULL;
1005 p->mempolicy = mpol_copy(p->mempolicy);
1006 if (IS_ERR(p->mempolicy)) {
1007 retval = PTR_ERR(p->mempolicy);
1008 p->mempolicy = NULL;
1009 goto bad_fork_cleanup;
1013 if ((retval = security_task_alloc(p)))
1014 goto bad_fork_cleanup_policy;
1015 if ((retval = audit_alloc(p)))
1016 goto bad_fork_cleanup_security;
1017 /* copy all the process information */
1018 if ((retval = copy_semundo(clone_flags, p)))
1019 goto bad_fork_cleanup_audit;
1020 if ((retval = copy_files(clone_flags, p)))
1021 goto bad_fork_cleanup_semundo;
1022 if ((retval = copy_fs(clone_flags, p)))
1023 goto bad_fork_cleanup_files;
1024 if ((retval = copy_sighand(clone_flags, p)))
1025 goto bad_fork_cleanup_fs;
1026 if ((retval = copy_signal(clone_flags, p)))
1027 goto bad_fork_cleanup_sighand;
1028 if ((retval = copy_mm(clone_flags, p)))
1029 goto bad_fork_cleanup_signal;
1030 if ((retval = copy_namespace(clone_flags, p)))
1031 goto bad_fork_cleanup_mm;
1032 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1034 goto bad_fork_cleanup_namespace;
1036 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1038 * Clear TID on mm_release()?
1040 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1043 * Syscall tracing should be turned off in the child regardless
1046 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1048 /* Our parent execution domain becomes current domain
1049 These must match for thread signalling to apply */
1051 p->parent_exec_id = p->self_exec_id;
1053 /* ok, now we should be set up.. */
1054 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1055 p->pdeath_signal = 0;
1057 /* Perform scheduler related setup */
1061 * Ok, make it visible to the rest of the system.
1062 * We dont wake it up yet.
1065 p->group_leader = p;
1066 INIT_LIST_HEAD(&p->ptrace_children);
1067 INIT_LIST_HEAD(&p->ptrace_list);
1069 /* Need tasklist lock for parent etc handling! */
1070 write_lock_irq(&tasklist_lock);
1072 * Check for pending SIGKILL! The new thread should not be allowed
1073 * to slip out of an OOM kill. (or normal SIGKILL.)
1075 if (sigismember(¤t->pending.signal, SIGKILL)) {
1076 write_unlock_irq(&tasklist_lock);
1078 goto bad_fork_cleanup_namespace;
1081 /* CLONE_PARENT re-uses the old parent */
1082 if (clone_flags & CLONE_PARENT)
1083 p->real_parent = current->real_parent;
1085 p->real_parent = current;
1086 p->parent = p->real_parent;
1088 if (clone_flags & CLONE_THREAD) {
1089 spin_lock(¤t->sighand->siglock);
1091 * Important: if an exit-all has been started then
1092 * do not create this new thread - the whole thread
1093 * group is supposed to exit anyway.
1095 if (current->signal->group_exit) {
1096 spin_unlock(¤t->sighand->siglock);
1097 write_unlock_irq(&tasklist_lock);
1099 goto bad_fork_cleanup_namespace;
1101 p->tgid = current->tgid;
1102 p->group_leader = current->group_leader;
1104 if (current->signal->group_stop_count > 0) {
1106 * There is an all-stop in progress for the group.
1107 * We ourselves will stop as soon as we check signals.
1108 * Make the new thread part of that group stop too.
1110 current->signal->group_stop_count++;
1111 set_tsk_thread_flag(p, TIF_SIGPENDING);
1114 spin_unlock(¤t->sighand->siglock);
1118 if (p->ptrace & PT_PTRACED)
1119 __ptrace_link(p, current->parent);
1121 attach_pid(p, PIDTYPE_PID, p->pid);
1122 if (thread_group_leader(p)) {
1123 attach_pid(p, PIDTYPE_TGID, p->tgid);
1124 attach_pid(p, PIDTYPE_PGID, process_group(p));
1125 attach_pid(p, PIDTYPE_SID, p->signal->session);
1127 __get_cpu_var(process_counts)++;
1129 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1132 /* p is copy of current */
1135 atomic_inc(&vxi->cacct.nr_threads);
1136 atomic_inc(&vxi->limit.rcur[RLIMIT_NPROC]);
1138 write_unlock_irq(&tasklist_lock);
1143 return ERR_PTR(retval);
1146 bad_fork_cleanup_namespace:
1148 bad_fork_cleanup_mm:
1151 mmdrop(p->active_mm);
1152 bad_fork_cleanup_signal:
1154 bad_fork_cleanup_sighand:
1156 bad_fork_cleanup_fs:
1157 exit_fs(p); /* blocking */
1158 bad_fork_cleanup_files:
1159 exit_files(p); /* blocking */
1160 bad_fork_cleanup_semundo:
1162 bad_fork_cleanup_audit:
1164 bad_fork_cleanup_security:
1165 security_task_free(p);
1166 bad_fork_cleanup_policy:
1168 mpol_free(p->mempolicy);
1172 free_pidmap(p->pid);
1174 module_put(p->binfmt->module);
1175 bad_fork_cleanup_put_domain:
1176 module_put(p->thread_info->exec_domain->module);
1177 bad_fork_cleanup_count:
1178 put_group_info(p->group_info);
1179 atomic_dec(&p->user->processes);
1181 bad_fork_cleanup_vm:
1182 if (p->mm && !(clone_flags & CLONE_VM))
1183 vx_pages_sub(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
1189 static inline int fork_traceflag (unsigned clone_flags)
1191 if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1193 else if (clone_flags & CLONE_VFORK) {
1194 if (current->ptrace & PT_TRACE_VFORK)
1195 return PTRACE_EVENT_VFORK;
1196 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1197 if (current->ptrace & PT_TRACE_CLONE)
1198 return PTRACE_EVENT_CLONE;
1199 } else if (current->ptrace & PT_TRACE_FORK)
1200 return PTRACE_EVENT_FORK;
1206 * Ok, this is the main fork-routine.
1208 * It copies the process, and if successful kick-starts
1209 * it and waits for it to finish using the VM if required.
1211 long do_fork(unsigned long clone_flags,
1212 unsigned long stack_start,
1213 struct pt_regs *regs,
1214 unsigned long stack_size,
1215 int __user *parent_tidptr,
1216 int __user *child_tidptr)
1218 struct task_struct *p;
1222 if (unlikely(current->ptrace)) {
1223 trace = fork_traceflag (clone_flags);
1225 clone_flags |= CLONE_PTRACE;
1228 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1229 if (numtasks_get_ref(current->taskclass, 0) == 0) {
1234 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1236 * Do this prior waking up the new thread - the thread pointer
1237 * might get invalid after that point, if the thread exits quickly.
1239 pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1242 struct completion vfork;
1246 if (clone_flags & CLONE_VFORK) {
1247 p->vfork_done = &vfork;
1248 init_completion(&vfork);
1251 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1253 * We'll start up with an immediate SIGSTOP.
1255 sigaddset(&p->pending.signal, SIGSTOP);
1256 set_tsk_thread_flag(p, TIF_SIGPENDING);
1259 if (!(clone_flags & CLONE_STOPPED)) {
1261 * Do the wakeup last. On SMP we treat fork() and
1262 * CLONE_VM separately, because fork() has already
1263 * created cache footprint on this CPU (due to
1264 * copying the pagetables), hence migration would
1265 * probably be costy. Threads on the other hand
1266 * have less traction to the current CPU, and if
1267 * there's an imbalance then the scheduler can
1268 * migrate this fresh thread now, before it
1269 * accumulates a larger cache footprint:
1271 if (clone_flags & CLONE_VM)
1272 wake_up_forked_thread(p);
1274 wake_up_forked_process(p);
1276 int cpu = get_cpu();
1278 p->state = TASK_STOPPED;
1279 if (cpu_is_offline(task_cpu(p)))
1280 set_task_cpu(p, cpu);
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);
1297 * Let the child process run first, to avoid most of the
1298 * COW overhead when the child exec()s afterwards.
1302 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1303 numtasks_put_ref(current->taskclass);
1309 /* SLAB cache for signal_struct structures (tsk->signal) */
1310 kmem_cache_t *signal_cachep;
1312 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1313 kmem_cache_t *sighand_cachep;
1315 /* SLAB cache for files_struct structures (tsk->files) */
1316 kmem_cache_t *files_cachep;
1318 /* SLAB cache for fs_struct structures (tsk->fs) */
1319 kmem_cache_t *fs_cachep;
1321 /* SLAB cache for vm_area_struct structures */
1322 kmem_cache_t *vm_area_cachep;
1324 /* SLAB cache for mm_struct structures (tsk->mm) */
1325 kmem_cache_t *mm_cachep;
1327 void __init proc_caches_init(void)
1329 sighand_cachep = kmem_cache_create("sighand_cache",
1330 sizeof(struct sighand_struct), 0,
1331 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1332 signal_cachep = kmem_cache_create("signal_cache",
1333 sizeof(struct signal_struct), 0,
1334 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1335 files_cachep = kmem_cache_create("files_cache",
1336 sizeof(struct files_struct), 0,
1337 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1338 fs_cachep = kmem_cache_create("fs_cache",
1339 sizeof(struct fs_struct), 0,
1340 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1341 vm_area_cachep = kmem_cache_create("vm_area_struct",
1342 sizeof(struct vm_area_struct), 0,
1343 SLAB_PANIC, NULL, NULL);
1344 mm_cachep = kmem_cache_create("mm_struct",
1345 sizeof(struct mm_struct), 0,
1346 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);