4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 'fork.c' contains the help-routines for the 'fork' system call
9 * (see also entry.S and others).
10 * Fork is rather simple, once you get the hang of it, but the memory
11 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
14 #include <linux/config.h>
15 #include <linux/slab.h>
16 #include <linux/init.h>
17 #include <linux/unistd.h>
18 #include <linux/smp_lock.h>
19 #include <linux/module.h>
20 #include <linux/vmalloc.h>
21 #include <linux/completion.h>
22 #include <linux/namespace.h>
23 #include <linux/personality.h>
24 #include <linux/mempolicy.h>
25 #include <linux/sem.h>
26 #include <linux/file.h>
27 #include <linux/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.h>
31 #include <linux/cpu.h>
32 #include <linux/cpuset.h>
33 #include <linux/security.h>
34 #include <linux/swap.h>
35 #include <linux/syscalls.h>
36 #include <linux/jiffies.h>
37 #include <linux/futex.h>
38 #include <linux/ptrace.h>
39 #include <linux/mount.h>
40 #include <linux/audit.h>
41 #include <linux/profile.h>
42 #include <linux/rmap.h>
43 #include <linux/acct.h>
44 #include <linux/vs_network.h>
45 #include <linux/vs_limit.h>
46 #include <linux/vs_memory.h>
48 #include <asm/pgtable.h>
49 #include <asm/pgalloc.h>
50 #include <asm/uaccess.h>
51 #include <asm/mmu_context.h>
52 #include <asm/cacheflush.h>
53 #include <asm/tlbflush.h>
56 * Protected counters by write_lock_irq(&tasklist_lock)
58 unsigned long total_forks; /* Handle normal Linux uptimes. */
59 int nr_threads; /* The idle threads do not count.. */
61 int max_threads; /* tunable limit on nr_threads */
63 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
65 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
67 EXPORT_SYMBOL(tasklist_lock);
69 int nr_processes(void)
74 for_each_online_cpu(cpu)
75 total += per_cpu(process_counts, cpu);
80 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
81 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
82 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
83 static kmem_cache_t *task_struct_cachep;
86 /* SLAB cache for signal_struct structures (tsk->signal) */
87 kmem_cache_t *signal_cachep;
89 /* SLAB cache for sighand_struct structures (tsk->sighand) */
90 kmem_cache_t *sighand_cachep;
92 /* SLAB cache for files_struct structures (tsk->files) */
93 kmem_cache_t *files_cachep;
95 /* SLAB cache for fs_struct structures (tsk->fs) */
96 kmem_cache_t *fs_cachep;
98 /* SLAB cache for vm_area_struct structures */
99 kmem_cache_t *vm_area_cachep;
101 /* SLAB cache for mm_struct structures (tsk->mm) */
102 static kmem_cache_t *mm_cachep;
104 void free_task(struct task_struct *tsk)
106 free_thread_info(tsk->thread_info);
107 clr_vx_info(&tsk->vx_info);
108 clr_nx_info(&tsk->nx_info);
109 free_task_struct(tsk);
111 EXPORT_SYMBOL(free_task);
113 void __put_task_struct(struct task_struct *tsk)
115 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
116 WARN_ON(atomic_read(&tsk->usage));
117 WARN_ON(tsk == current);
119 if (unlikely(tsk->audit_context))
121 security_task_free(tsk);
123 put_group_info(tsk->group_info);
125 if (!profile_handoff_task(tsk))
129 void __init fork_init(unsigned long mempages)
131 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
132 #ifndef ARCH_MIN_TASKALIGN
133 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
135 /* create a slab on which task_structs can be allocated */
137 kmem_cache_create("task_struct", sizeof(struct task_struct),
138 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
142 * The default maximum number of threads is set to a safe
143 * value: the thread structures can take up at most half
146 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
149 * we need to allow at least 20 threads to boot a system
154 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
155 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
156 init_task.signal->rlim[RLIMIT_SIGPENDING] =
157 init_task.signal->rlim[RLIMIT_NPROC];
160 static struct task_struct *dup_task_struct(struct task_struct *orig)
162 struct task_struct *tsk;
163 struct thread_info *ti;
165 prepare_to_copy(orig);
167 tsk = alloc_task_struct();
171 ti = alloc_thread_info(tsk);
173 free_task_struct(tsk);
177 *ti = *orig->thread_info;
179 tsk->thread_info = ti;
182 /* One for us, one for whoever does the "release_task()" (usually parent) */
183 atomic_set(&tsk->usage,2);
188 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
190 struct vm_area_struct * mpnt, *tmp, **pprev;
191 struct rb_node **rb_link, *rb_parent;
193 unsigned long charge;
194 struct mempolicy *pol;
196 down_write(&oldmm->mmap_sem);
197 flush_cache_mm(current->mm);
200 mm->mmap_cache = NULL;
201 mm->free_area_cache = oldmm->mmap_base;
203 __set_mm_counter(mm, rss, 0);
204 __set_mm_counter(mm, anon_rss, 0);
205 cpus_clear(mm->cpu_vm_mask);
207 rb_link = &mm->mm_rb.rb_node;
211 for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
214 if (mpnt->vm_flags & VM_DONTCOPY) {
215 __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
220 if (mpnt->vm_flags & VM_ACCOUNT) {
221 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
222 if (security_vm_enough_memory(len))
226 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
230 pol = mpol_copy(vma_policy(mpnt));
231 retval = PTR_ERR(pol);
233 goto fail_nomem_policy;
234 vma_set_policy(tmp, pol);
235 tmp->vm_flags &= ~VM_LOCKED;
241 struct inode *inode = file->f_dentry->d_inode;
243 if (tmp->vm_flags & VM_DENYWRITE)
244 atomic_dec(&inode->i_writecount);
246 /* insert tmp into the share list, just after mpnt */
247 spin_lock(&file->f_mapping->i_mmap_lock);
248 tmp->vm_truncate_count = mpnt->vm_truncate_count;
249 flush_dcache_mmap_lock(file->f_mapping);
250 vma_prio_tree_add(tmp, mpnt);
251 flush_dcache_mmap_unlock(file->f_mapping);
252 spin_unlock(&file->f_mapping->i_mmap_lock);
256 * Link in the new vma and copy the page table entries:
257 * link in first so that swapoff can see swap entries,
258 * and try_to_unmap_one's find_vma find the new vma.
260 spin_lock(&mm->page_table_lock);
262 pprev = &tmp->vm_next;
264 __vma_link_rb(mm, tmp, rb_link, rb_parent);
265 rb_link = &tmp->vm_rb.rb_right;
266 rb_parent = &tmp->vm_rb;
269 retval = copy_page_range(mm, current->mm, tmp);
270 spin_unlock(&mm->page_table_lock);
272 if (tmp->vm_ops && tmp->vm_ops->open)
273 tmp->vm_ops->open(tmp);
281 flush_tlb_mm(current->mm);
282 up_write(&oldmm->mmap_sem);
285 kmem_cache_free(vm_area_cachep, tmp);
288 vm_unacct_memory(charge);
292 static inline int mm_alloc_pgd(struct mm_struct * mm)
294 mm->pgd = pgd_alloc(mm);
295 if (unlikely(!mm->pgd))
300 static inline void mm_free_pgd(struct mm_struct * mm)
305 #define dup_mmap(mm, oldmm) (0)
306 #define mm_alloc_pgd(mm) (0)
307 #define mm_free_pgd(mm)
308 #endif /* CONFIG_MMU */
310 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
312 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
313 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
315 #include <linux/init_task.h>
317 static struct mm_struct * mm_init(struct mm_struct * mm)
319 atomic_set(&mm->mm_users, 1);
320 atomic_set(&mm->mm_count, 1);
321 init_rwsem(&mm->mmap_sem);
322 INIT_LIST_HEAD(&mm->mmlist);
323 mm->core_waiters = 0;
325 spin_lock_init(&mm->page_table_lock);
326 rwlock_init(&mm->ioctx_list_lock);
327 mm->ioctx_list = NULL;
328 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
329 mm->free_area_cache = TASK_UNMAPPED_BASE;
331 if (likely(!mm_alloc_pgd(mm))) {
333 set_vx_info(&mm->mm_vx_info, current->vx_info);
341 * Allocate and initialize an mm_struct.
343 struct mm_struct * mm_alloc(void)
345 struct mm_struct * mm;
349 memset(mm, 0, sizeof(*mm));
356 * Called when the last reference to the mm
357 * is dropped: either by a lazy thread or by
358 * mmput. Free the page directory and the mm.
360 void fastcall __mmdrop(struct mm_struct *mm)
362 BUG_ON(mm == &init_mm);
365 clr_vx_info(&mm->mm_vx_info);
370 * Decrement the use count and release all resources for an mm.
372 void mmput(struct mm_struct *mm)
374 if (atomic_dec_and_test(&mm->mm_users)) {
377 if (!list_empty(&mm->mmlist)) {
378 spin_lock(&mmlist_lock);
379 list_del(&mm->mmlist);
380 spin_unlock(&mmlist_lock);
386 EXPORT_SYMBOL_GPL(mmput);
389 * get_task_mm - acquire a reference to the task's mm
391 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
392 * this kernel workthread has transiently adopted a user mm with use_mm,
393 * to do its AIO) is not set and if so returns a reference to it, after
394 * bumping up the use count. User must release the mm via mmput()
395 * after use. Typically used by /proc and ptrace.
397 struct mm_struct *get_task_mm(struct task_struct *task)
399 struct mm_struct *mm;
404 if (task->flags & PF_BORROWED_MM)
407 atomic_inc(&mm->mm_users);
412 EXPORT_SYMBOL_GPL(get_task_mm);
414 /* Please note the differences between mmput and mm_release.
415 * mmput is called whenever we stop holding onto a mm_struct,
416 * error success whatever.
418 * mm_release is called after a mm_struct has been removed
419 * from the current process.
421 * This difference is important for error handling, when we
422 * only half set up a mm_struct for a new process and need to restore
423 * the old one. Because we mmput the new mm_struct before
424 * restoring the old one. . .
425 * Eric Biederman 10 January 1998
427 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
429 struct completion *vfork_done = tsk->vfork_done;
431 /* Get rid of any cached register state */
432 deactivate_mm(tsk, mm);
434 /* notify parent sleeping on vfork() */
436 tsk->vfork_done = NULL;
437 complete(vfork_done);
439 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
440 u32 __user * tidptr = tsk->clear_child_tid;
441 tsk->clear_child_tid = NULL;
444 * We don't check the error code - if userspace has
445 * not set up a proper pointer then tough luck.
448 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
452 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
454 struct mm_struct * mm, *oldmm;
457 tsk->min_flt = tsk->maj_flt = 0;
458 tsk->nvcsw = tsk->nivcsw = 0;
461 tsk->active_mm = NULL;
464 * Are we cloning a kernel thread?
466 * We need to steal a active VM for that..
472 if (clone_flags & CLONE_VM) {
473 atomic_inc(&oldmm->mm_users);
476 * There are cases where the PTL is held to ensure no
477 * new threads start up in user mode using an mm, which
478 * allows optimizing out ipis; the tlb_gather_mmu code
481 spin_unlock_wait(&oldmm->page_table_lock);
490 /* Copy the current MM stuff.. */
491 memcpy(mm, oldmm, sizeof(*mm));
492 mm->mm_vx_info = NULL;
496 if (init_new_context(tsk,mm))
499 retval = dup_mmap(mm, oldmm);
503 mm->hiwater_rss = get_mm_counter(mm,rss);
504 mm->hiwater_vm = mm->total_vm;
518 * If init_new_context() failed, we cannot use mmput() to free the mm
519 * because it calls destroy_context()
521 clr_vx_info(&mm->mm_vx_info);
527 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
529 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
530 /* We don't need to lock fs - think why ;-) */
532 atomic_set(&fs->count, 1);
533 rwlock_init(&fs->lock);
534 fs->umask = old->umask;
535 read_lock(&old->lock);
536 fs->rootmnt = mntget(old->rootmnt);
537 fs->root = dget(old->root);
538 fs->pwdmnt = mntget(old->pwdmnt);
539 fs->pwd = dget(old->pwd);
541 fs->altrootmnt = mntget(old->altrootmnt);
542 fs->altroot = dget(old->altroot);
544 fs->altrootmnt = NULL;
547 read_unlock(&old->lock);
552 struct fs_struct *copy_fs_struct(struct fs_struct *old)
554 return __copy_fs_struct(old);
557 EXPORT_SYMBOL_GPL(copy_fs_struct);
559 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
561 if (clone_flags & CLONE_FS) {
562 atomic_inc(¤t->fs->count);
565 tsk->fs = __copy_fs_struct(current->fs);
571 static int count_open_files(struct files_struct *files, int size)
575 /* Find the last open fd */
576 for (i = size/(8*sizeof(long)); i > 0; ) {
577 if (files->open_fds->fds_bits[--i])
580 i = (i+1) * 8 * sizeof(long);
584 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
586 struct files_struct *oldf, *newf;
587 struct file **old_fds, **new_fds;
588 int open_files, size, i, error = 0, expand;
591 * A background process may not have any files ...
593 oldf = current->files;
597 if (clone_flags & CLONE_FILES) {
598 atomic_inc(&oldf->count);
603 * Note: we may be using current for both targets (See exec.c)
604 * This works because we cache current->files (old) as oldf. Don't
609 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
613 atomic_set(&newf->count, 1);
615 spin_lock_init(&newf->file_lock);
617 newf->max_fds = NR_OPEN_DEFAULT;
618 newf->max_fdset = __FD_SETSIZE;
619 newf->close_on_exec = &newf->close_on_exec_init;
620 newf->open_fds = &newf->open_fds_init;
621 newf->fd = &newf->fd_array[0];
623 spin_lock(&oldf->file_lock);
625 open_files = count_open_files(oldf, oldf->max_fdset);
629 * Check whether we need to allocate a larger fd array or fd set.
630 * Note: we're not a clone task, so the open count won't change.
632 if (open_files > newf->max_fdset) {
636 if (open_files > newf->max_fds) {
641 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
643 spin_unlock(&oldf->file_lock);
644 spin_lock(&newf->file_lock);
645 error = expand_files(newf, open_files-1);
646 spin_unlock(&newf->file_lock);
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++;
662 /* FIXME sum it first for check and performance */
663 vx_openfd_inc(open_files - i);
666 * The fd may be claimed in the fd bitmap but not yet
667 * instantiated in the files array if a sibling thread
668 * is partway through open(). So make sure that this
669 * fd is available to the new process.
671 FD_CLR(open_files - i, newf->open_fds);
675 spin_unlock(&oldf->file_lock);
677 /* compute the remainder to be cleared */
678 size = (newf->max_fds - open_files) * sizeof(struct file *);
680 /* This is long word aligned thus could use a optimized version */
681 memset(new_fds, 0, size);
683 if (newf->max_fdset > open_files) {
684 int left = (newf->max_fdset-open_files)/8;
685 int start = open_files / (8 * sizeof(unsigned long));
687 memset(&newf->open_fds->fds_bits[start], 0, left);
688 memset(&newf->close_on_exec->fds_bits[start], 0, left);
697 free_fdset (newf->close_on_exec, newf->max_fdset);
698 free_fdset (newf->open_fds, newf->max_fdset);
699 free_fd_array(newf->fd, newf->max_fds);
700 kmem_cache_free(files_cachep, newf);
705 * Helper to unshare the files of the current task.
706 * We don't want to expose copy_files internals to
707 * the exec layer of the kernel.
710 int unshare_files(void)
712 struct files_struct *files = current->files;
718 /* This can race but the race causes us to copy when we don't
719 need to and drop the copy */
720 if(atomic_read(&files->count) == 1)
722 atomic_inc(&files->count);
725 rc = copy_files(0, current);
727 current->files = files;
731 EXPORT_SYMBOL(unshare_files);
733 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
735 struct sighand_struct *sig;
737 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
738 atomic_inc(¤t->sighand->count);
741 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
745 spin_lock_init(&sig->siglock);
746 atomic_set(&sig->count, 1);
747 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
751 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
753 struct signal_struct *sig;
756 if (clone_flags & CLONE_THREAD) {
757 atomic_inc(¤t->signal->count);
758 atomic_inc(¤t->signal->live);
761 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
766 ret = copy_thread_group_keys(tsk);
768 kmem_cache_free(signal_cachep, sig);
772 atomic_set(&sig->count, 1);
773 atomic_set(&sig->live, 1);
774 init_waitqueue_head(&sig->wait_chldexit);
776 sig->group_exit_code = 0;
777 sig->group_exit_task = NULL;
778 sig->group_stop_count = 0;
779 sig->curr_target = NULL;
780 init_sigpending(&sig->shared_pending);
781 INIT_LIST_HEAD(&sig->posix_timers);
783 sig->it_real_value = sig->it_real_incr = 0;
784 sig->real_timer.function = it_real_fn;
785 sig->real_timer.data = (unsigned long) tsk;
786 init_timer(&sig->real_timer);
788 sig->it_virt_expires = cputime_zero;
789 sig->it_virt_incr = cputime_zero;
790 sig->it_prof_expires = cputime_zero;
791 sig->it_prof_incr = cputime_zero;
793 sig->tty = current->signal->tty;
794 sig->pgrp = process_group(current);
795 sig->session = current->signal->session;
796 sig->leader = 0; /* session leadership doesn't inherit */
797 sig->tty_old_pgrp = 0;
799 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
800 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
801 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
803 INIT_LIST_HEAD(&sig->cpu_timers[0]);
804 INIT_LIST_HEAD(&sig->cpu_timers[1]);
805 INIT_LIST_HEAD(&sig->cpu_timers[2]);
807 task_lock(current->group_leader);
808 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
809 task_unlock(current->group_leader);
811 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
813 * New sole thread in the process gets an expiry time
814 * of the whole CPU time limit.
816 tsk->it_prof_expires =
817 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
823 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
825 unsigned long new_flags = p->flags;
827 new_flags &= ~PF_SUPERPRIV;
828 new_flags |= PF_FORKNOEXEC;
829 if (!(clone_flags & CLONE_PTRACE))
831 p->flags = new_flags;
834 asmlinkage long sys_set_tid_address(int __user *tidptr)
836 current->clear_child_tid = tidptr;
842 * This creates a new process as a copy of the old one,
843 * but does not actually start it yet.
845 * It copies the registers, and all the appropriate
846 * parts of the process environment (as per the clone
847 * flags). The actual kick-off is left to the caller.
849 static task_t *copy_process(unsigned long clone_flags,
850 unsigned long stack_start,
851 struct pt_regs *regs,
852 unsigned long stack_size,
853 int __user *parent_tidptr,
854 int __user *child_tidptr,
858 struct task_struct *p = NULL;
862 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
863 return ERR_PTR(-EINVAL);
866 * Thread groups must share signals as well, and detached threads
867 * can only be started up within the thread group.
869 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
870 return ERR_PTR(-EINVAL);
873 * Shared signal handlers imply shared VM. By way of the above,
874 * thread groups also imply shared VM. Blocking this case allows
875 * for various simplifications in other code.
877 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
878 return ERR_PTR(-EINVAL);
880 retval = security_task_create(clone_flags);
885 p = dup_task_struct(current);
889 init_vx_info(&p->vx_info, current->vx_info);
890 init_nx_info(&p->nx_info, current->nx_info);
892 /* check vserver memory */
893 if (p->mm && !(clone_flags & CLONE_VM)) {
894 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
895 vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
899 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
900 if (!vx_rsspages_avail(p->mm, get_mm_counter(p->mm, rss)))
901 goto bad_fork_cleanup_vm;
905 if (!vx_nproc_avail(1))
906 goto bad_fork_cleanup_vm;
908 if (atomic_read(&p->user->processes) >=
909 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
910 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
911 p->user != &root_user)
912 goto bad_fork_cleanup_vm;
915 atomic_inc(&p->user->__count);
916 atomic_inc(&p->user->processes);
917 get_group_info(p->group_info);
920 * If multiple threads are within copy_process(), then this check
921 * triggers too late. This doesn't hurt, the check is only there
922 * to stop root fork bombs.
924 if (nr_threads >= max_threads)
925 goto bad_fork_cleanup_count;
927 if (!try_module_get(p->thread_info->exec_domain->module))
928 goto bad_fork_cleanup_count;
930 if (p->binfmt && !try_module_get(p->binfmt->module))
931 goto bad_fork_cleanup_put_domain;
934 copy_flags(clone_flags, p);
937 if (clone_flags & CLONE_PARENT_SETTID)
938 if (put_user(p->pid, parent_tidptr))
939 goto bad_fork_cleanup;
941 p->proc_dentry = NULL;
943 INIT_LIST_HEAD(&p->children);
944 INIT_LIST_HEAD(&p->sibling);
945 p->vfork_done = NULL;
946 spin_lock_init(&p->alloc_lock);
947 spin_lock_init(&p->proc_lock);
949 clear_tsk_thread_flag(p, TIF_SIGPENDING);
950 init_sigpending(&p->pending);
952 p->utime = cputime_zero;
953 p->stime = cputime_zero;
955 p->rchar = 0; /* I/O counter: bytes read */
956 p->wchar = 0; /* I/O counter: bytes written */
957 p->syscr = 0; /* I/O counter: read syscalls */
958 p->syscw = 0; /* I/O counter: write syscalls */
959 acct_clear_integrals(p);
961 p->it_virt_expires = cputime_zero;
962 p->it_prof_expires = cputime_zero;
963 p->it_sched_expires = 0;
964 INIT_LIST_HEAD(&p->cpu_timers[0]);
965 INIT_LIST_HEAD(&p->cpu_timers[1]);
966 INIT_LIST_HEAD(&p->cpu_timers[2]);
968 p->lock_depth = -1; /* -1 = no lock */
969 do_posix_clock_monotonic_gettime(&p->start_time);
971 p->io_context = NULL;
973 p->audit_context = NULL;
975 p->mempolicy = mpol_copy(p->mempolicy);
976 if (IS_ERR(p->mempolicy)) {
977 retval = PTR_ERR(p->mempolicy);
979 goto bad_fork_cleanup;
984 if (clone_flags & CLONE_THREAD)
985 p->tgid = current->tgid;
987 if ((retval = security_task_alloc(p)))
988 goto bad_fork_cleanup_policy;
989 if ((retval = audit_alloc(p)))
990 goto bad_fork_cleanup_security;
991 /* copy all the process information */
992 if ((retval = copy_semundo(clone_flags, p)))
993 goto bad_fork_cleanup_audit;
994 if ((retval = copy_files(clone_flags, p)))
995 goto bad_fork_cleanup_semundo;
996 if ((retval = copy_fs(clone_flags, p)))
997 goto bad_fork_cleanup_files;
998 if ((retval = copy_sighand(clone_flags, p)))
999 goto bad_fork_cleanup_fs;
1000 if ((retval = copy_signal(clone_flags, p)))
1001 goto bad_fork_cleanup_sighand;
1002 if ((retval = copy_mm(clone_flags, p)))
1003 goto bad_fork_cleanup_signal;
1004 if ((retval = copy_keys(clone_flags, p)))
1005 goto bad_fork_cleanup_mm;
1006 if ((retval = copy_namespace(clone_flags, p)))
1007 goto bad_fork_cleanup_keys;
1008 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1010 goto bad_fork_cleanup_namespace;
1012 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1014 * Clear TID on mm_release()?
1016 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1019 * Syscall tracing should be turned off in the child regardless
1022 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1024 /* Our parent execution domain becomes current domain
1025 These must match for thread signalling to apply */
1027 p->parent_exec_id = p->self_exec_id;
1029 /* ok, now we should be set up.. */
1030 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1031 p->pdeath_signal = 0;
1034 /* Perform scheduler related setup */
1038 * Ok, make it visible to the rest of the system.
1039 * We dont wake it up yet.
1041 p->group_leader = p;
1042 INIT_LIST_HEAD(&p->ptrace_children);
1043 INIT_LIST_HEAD(&p->ptrace_list);
1045 /* Need tasklist lock for parent etc handling! */
1046 write_lock_irq(&tasklist_lock);
1049 * The task hasn't been attached yet, so cpus_allowed mask cannot
1050 * have changed. The cpus_allowed mask of the parent may have
1051 * changed after it was copied first time, and it may then move to
1052 * another CPU - so we re-copy it here and set the child's CPU to
1053 * the parent's CPU. This avoids alot of nasty races.
1055 p->cpus_allowed = current->cpus_allowed;
1056 set_task_cpu(p, smp_processor_id());
1059 * Check for pending SIGKILL! The new thread should not be allowed
1060 * to slip out of an OOM kill. (or normal SIGKILL.)
1062 if (sigismember(¤t->pending.signal, SIGKILL)) {
1063 write_unlock_irq(&tasklist_lock);
1065 goto bad_fork_cleanup_namespace;
1068 /* CLONE_PARENT re-uses the old parent */
1069 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1070 p->real_parent = current->real_parent;
1072 p->real_parent = current;
1073 p->parent = p->real_parent;
1075 if (clone_flags & CLONE_THREAD) {
1076 spin_lock(¤t->sighand->siglock);
1078 * Important: if an exit-all has been started then
1079 * do not create this new thread - the whole thread
1080 * group is supposed to exit anyway.
1082 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1083 spin_unlock(¤t->sighand->siglock);
1084 write_unlock_irq(&tasklist_lock);
1086 goto bad_fork_cleanup_namespace;
1088 p->group_leader = current->group_leader;
1090 if (current->signal->group_stop_count > 0) {
1092 * There is an all-stop in progress for the group.
1093 * We ourselves will stop as soon as we check signals.
1094 * Make the new thread part of that group stop too.
1096 current->signal->group_stop_count++;
1097 set_tsk_thread_flag(p, TIF_SIGPENDING);
1100 if (!cputime_eq(current->signal->it_virt_expires,
1102 !cputime_eq(current->signal->it_prof_expires,
1104 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1105 !list_empty(¤t->signal->cpu_timers[0]) ||
1106 !list_empty(¤t->signal->cpu_timers[1]) ||
1107 !list_empty(¤t->signal->cpu_timers[2])) {
1109 * Have child wake up on its first tick to check
1110 * for process CPU timers.
1112 p->it_prof_expires = jiffies_to_cputime(1);
1115 spin_unlock(¤t->sighand->siglock);
1119 if (unlikely(p->ptrace & PT_PTRACED))
1120 __ptrace_link(p, current->parent);
1124 attach_pid(p, PIDTYPE_PID, p->pid);
1125 attach_pid(p, PIDTYPE_TGID, p->tgid);
1126 if (thread_group_leader(p)) {
1127 attach_pid(p, PIDTYPE_PGID, process_group(p));
1128 attach_pid(p, PIDTYPE_SID, p->signal->session);
1130 __get_cpu_var(process_counts)++;
1136 /* p is copy of current */
1139 claim_vx_info(vxi, p);
1140 atomic_inc(&vxi->cvirt.nr_threads);
1141 atomic_inc(&vxi->cvirt.total_forks);
1146 claim_nx_info(nxi, p);
1147 write_unlock_irq(&tasklist_lock);
1152 return ERR_PTR(retval);
1155 bad_fork_cleanup_namespace:
1157 bad_fork_cleanup_keys:
1159 bad_fork_cleanup_mm:
1162 bad_fork_cleanup_signal:
1164 bad_fork_cleanup_sighand:
1166 bad_fork_cleanup_fs:
1167 exit_fs(p); /* blocking */
1168 bad_fork_cleanup_files:
1169 exit_files(p); /* blocking */
1170 bad_fork_cleanup_semundo:
1172 bad_fork_cleanup_audit:
1174 bad_fork_cleanup_security:
1175 security_task_free(p);
1176 bad_fork_cleanup_policy:
1178 mpol_free(p->mempolicy);
1182 module_put(p->binfmt->module);
1183 bad_fork_cleanup_put_domain:
1184 module_put(p->thread_info->exec_domain->module);
1185 bad_fork_cleanup_count:
1186 put_group_info(p->group_info);
1187 atomic_dec(&p->user->processes);
1189 bad_fork_cleanup_vm:
1190 if (p->mm && !(clone_flags & CLONE_VM))
1191 vx_pages_sub(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
1197 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1199 memset(regs, 0, sizeof(struct pt_regs));
1203 task_t * __devinit fork_idle(int cpu)
1206 struct pt_regs regs;
1208 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0);
1210 return ERR_PTR(-ENOMEM);
1211 init_idle(task, cpu);
1212 unhash_process(task);
1216 static inline int fork_traceflag (unsigned clone_flags)
1218 if (clone_flags & CLONE_UNTRACED)
1220 else if (clone_flags & CLONE_VFORK) {
1221 if (current->ptrace & PT_TRACE_VFORK)
1222 return PTRACE_EVENT_VFORK;
1223 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1224 if (current->ptrace & PT_TRACE_CLONE)
1225 return PTRACE_EVENT_CLONE;
1226 } else if (current->ptrace & PT_TRACE_FORK)
1227 return PTRACE_EVENT_FORK;
1233 * Ok, this is the main fork-routine.
1235 * It copies the process, and if successful kick-starts
1236 * it and waits for it to finish using the VM if required.
1238 long do_fork(unsigned long clone_flags,
1239 unsigned long stack_start,
1240 struct pt_regs *regs,
1241 unsigned long stack_size,
1242 int __user *parent_tidptr,
1243 int __user *child_tidptr)
1245 struct task_struct *p;
1247 long pid = alloc_pidmap();
1251 if (unlikely(current->ptrace)) {
1252 trace = fork_traceflag (clone_flags);
1254 clone_flags |= CLONE_PTRACE;
1257 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1259 * Do this prior waking up the new thread - the thread pointer
1260 * might get invalid after that point, if the thread exits quickly.
1263 struct completion vfork;
1265 if (clone_flags & CLONE_VFORK) {
1266 p->vfork_done = &vfork;
1267 init_completion(&vfork);
1270 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1272 * We'll start up with an immediate SIGSTOP.
1274 sigaddset(&p->pending.signal, SIGSTOP);
1275 set_tsk_thread_flag(p, TIF_SIGPENDING);
1278 if (!(clone_flags & CLONE_STOPPED))
1279 wake_up_new_task(p, clone_flags);
1281 p->state = TASK_STOPPED;
1283 if (unlikely (trace)) {
1284 current->ptrace_message = pid;
1285 ptrace_notify ((trace << 8) | SIGTRAP);
1288 if (clone_flags & CLONE_VFORK) {
1289 wait_for_completion(&vfork);
1290 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1291 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1300 void __init proc_caches_init(void)
1302 sighand_cachep = kmem_cache_create("sighand_cache",
1303 sizeof(struct sighand_struct), 0,
1304 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1305 signal_cachep = kmem_cache_create("signal_cache",
1306 sizeof(struct signal_struct), 0,
1307 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1308 files_cachep = kmem_cache_create("files_cache",
1309 sizeof(struct files_struct), 0,
1310 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1311 fs_cachep = kmem_cache_create("fs_cache",
1312 sizeof(struct fs_struct), 0,
1313 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1314 vm_area_cachep = kmem_cache_create("vm_area_struct",
1315 sizeof(struct vm_area_struct), 0,
1316 SLAB_PANIC, NULL, NULL);
1317 mm_cachep = kmem_cache_create("mm_struct",
1318 sizeof(struct mm_struct), 0,
1319 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);