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/acct.h>
43 #include <linux/vs_network.h>
44 #include <linux/vs_limit.h>
45 #include <linux/vs_memory.h>
47 #include <asm/pgtable.h>
48 #include <asm/pgalloc.h>
49 #include <asm/uaccess.h>
50 #include <asm/mmu_context.h>
51 #include <asm/cacheflush.h>
52 #include <asm/tlbflush.h>
55 * Protected counters by write_lock_irq(&tasklist_lock)
57 unsigned long total_forks; /* Handle normal Linux uptimes. */
58 int nr_threads; /* The idle threads do not count.. */
60 int max_threads; /* tunable limit on nr_threads */
62 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
64 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
66 EXPORT_SYMBOL(tasklist_lock);
68 int nr_processes(void)
73 for_each_online_cpu(cpu)
74 total += per_cpu(process_counts, cpu);
79 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
80 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
81 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
82 static kmem_cache_t *task_struct_cachep;
85 void free_task(struct task_struct *tsk)
87 free_thread_info(tsk->thread_info);
88 clr_vx_info(&tsk->vx_info);
89 clr_nx_info(&tsk->nx_info);
90 free_task_struct(tsk);
92 EXPORT_SYMBOL(free_task);
94 void __put_task_struct(struct task_struct *tsk)
96 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
97 WARN_ON(atomic_read(&tsk->usage));
98 WARN_ON(tsk == current);
100 if (unlikely(tsk->audit_context))
102 security_task_free(tsk);
104 put_group_info(tsk->group_info);
106 if (!profile_handoff_task(tsk))
110 void __init fork_init(unsigned long mempages)
112 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
113 #ifndef ARCH_MIN_TASKALIGN
114 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
116 /* create a slab on which task_structs can be allocated */
118 kmem_cache_create("task_struct", sizeof(struct task_struct),
119 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
123 * The default maximum number of threads is set to a safe
124 * value: the thread structures can take up at most half
127 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
130 * we need to allow at least 20 threads to boot a system
135 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
136 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
139 static struct task_struct *dup_task_struct(struct task_struct *orig)
141 struct task_struct *tsk;
142 struct thread_info *ti;
144 prepare_to_copy(orig);
146 tsk = alloc_task_struct();
150 ti = alloc_thread_info(tsk);
152 free_task_struct(tsk);
156 *ti = *orig->thread_info;
158 tsk->thread_info = ti;
161 /* One for us, one for whoever does the "release_task()" (usually parent) */
162 atomic_set(&tsk->usage,2);
167 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
169 struct vm_area_struct * mpnt, *tmp, **pprev;
170 struct rb_node **rb_link, *rb_parent;
172 unsigned long charge;
173 struct mempolicy *pol;
175 down_write(&oldmm->mmap_sem);
176 flush_cache_mm(current->mm);
179 mm->mmap_cache = NULL;
180 mm->free_area_cache = oldmm->mmap_base;
184 cpus_clear(mm->cpu_vm_mask);
186 rb_link = &mm->mm_rb.rb_node;
190 for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
193 if (mpnt->vm_flags & VM_DONTCOPY) {
194 __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
199 if (mpnt->vm_flags & VM_ACCOUNT) {
200 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
201 if (security_vm_enough_memory(len))
205 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
209 pol = mpol_copy(vma_policy(mpnt));
210 retval = PTR_ERR(pol);
212 goto fail_nomem_policy;
213 vma_set_policy(tmp, pol);
214 tmp->vm_flags &= ~VM_LOCKED;
220 struct inode *inode = file->f_dentry->d_inode;
222 if (tmp->vm_flags & VM_DENYWRITE)
223 atomic_dec(&inode->i_writecount);
225 /* insert tmp into the share list, just after mpnt */
226 spin_lock(&file->f_mapping->i_mmap_lock);
227 tmp->vm_truncate_count = mpnt->vm_truncate_count;
228 flush_dcache_mmap_lock(file->f_mapping);
229 vma_prio_tree_add(tmp, mpnt);
230 flush_dcache_mmap_unlock(file->f_mapping);
231 spin_unlock(&file->f_mapping->i_mmap_lock);
235 * Link in the new vma and copy the page table entries:
236 * link in first so that swapoff can see swap entries,
237 * and try_to_unmap_one's find_vma find the new vma.
239 spin_lock(&mm->page_table_lock);
241 pprev = &tmp->vm_next;
243 __vma_link_rb(mm, tmp, rb_link, rb_parent);
244 rb_link = &tmp->vm_rb.rb_right;
245 rb_parent = &tmp->vm_rb;
248 retval = copy_page_range(mm, current->mm, tmp);
249 spin_unlock(&mm->page_table_lock);
251 if (tmp->vm_ops && tmp->vm_ops->open)
252 tmp->vm_ops->open(tmp);
260 flush_tlb_mm(current->mm);
261 up_write(&oldmm->mmap_sem);
264 kmem_cache_free(vm_area_cachep, tmp);
267 vm_unacct_memory(charge);
271 static inline int mm_alloc_pgd(struct mm_struct * mm)
273 mm->pgd = pgd_alloc(mm);
274 if (unlikely(!mm->pgd))
279 static inline void mm_free_pgd(struct mm_struct * mm)
284 #define dup_mmap(mm, oldmm) (0)
285 #define mm_alloc_pgd(mm) (0)
286 #define mm_free_pgd(mm)
287 #endif /* CONFIG_MMU */
289 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
291 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
292 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
294 #include <linux/init_task.h>
296 static struct mm_struct * mm_init(struct mm_struct * mm)
298 atomic_set(&mm->mm_users, 1);
299 atomic_set(&mm->mm_count, 1);
300 init_rwsem(&mm->mmap_sem);
301 INIT_LIST_HEAD(&mm->mmlist);
302 mm->core_waiters = 0;
304 spin_lock_init(&mm->page_table_lock);
305 rwlock_init(&mm->ioctx_list_lock);
306 mm->ioctx_list = NULL;
307 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
308 mm->free_area_cache = TASK_UNMAPPED_BASE;
310 if (likely(!mm_alloc_pgd(mm))) {
312 set_vx_info(&mm->mm_vx_info, current->vx_info);
320 * Allocate and initialize an mm_struct.
322 struct mm_struct * mm_alloc(void)
324 struct mm_struct * mm;
328 memset(mm, 0, sizeof(*mm));
335 * Called when the last reference to the mm
336 * is dropped: either by a lazy thread or by
337 * mmput. Free the page directory and the mm.
339 void fastcall __mmdrop(struct mm_struct *mm)
341 BUG_ON(mm == &init_mm);
344 clr_vx_info(&mm->mm_vx_info);
349 * Decrement the use count and release all resources for an mm.
351 void mmput(struct mm_struct *mm)
353 if (atomic_dec_and_test(&mm->mm_users)) {
356 if (!list_empty(&mm->mmlist)) {
357 spin_lock(&mmlist_lock);
358 list_del(&mm->mmlist);
359 spin_unlock(&mmlist_lock);
365 EXPORT_SYMBOL_GPL(mmput);
368 * get_task_mm - acquire a reference to the task's mm
370 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
371 * this kernel workthread has transiently adopted a user mm with use_mm,
372 * to do its AIO) is not set and if so returns a reference to it, after
373 * bumping up the use count. User must release the mm via mmput()
374 * after use. Typically used by /proc and ptrace.
376 struct mm_struct *get_task_mm(struct task_struct *task)
378 struct mm_struct *mm;
383 if (task->flags & PF_BORROWED_MM)
386 atomic_inc(&mm->mm_users);
391 EXPORT_SYMBOL_GPL(get_task_mm);
393 /* Please note the differences between mmput and mm_release.
394 * mmput is called whenever we stop holding onto a mm_struct,
395 * error success whatever.
397 * mm_release is called after a mm_struct has been removed
398 * from the current process.
400 * This difference is important for error handling, when we
401 * only half set up a mm_struct for a new process and need to restore
402 * the old one. Because we mmput the new mm_struct before
403 * restoring the old one. . .
404 * Eric Biederman 10 January 1998
406 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
408 struct completion *vfork_done = tsk->vfork_done;
410 /* Get rid of any cached register state */
411 deactivate_mm(tsk, mm);
413 /* notify parent sleeping on vfork() */
415 tsk->vfork_done = NULL;
416 complete(vfork_done);
418 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
419 u32 __user * tidptr = tsk->clear_child_tid;
420 tsk->clear_child_tid = NULL;
423 * We don't check the error code - if userspace has
424 * not set up a proper pointer then tough luck.
427 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
431 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
433 struct mm_struct * mm, *oldmm;
436 tsk->min_flt = tsk->maj_flt = 0;
437 tsk->nvcsw = tsk->nivcsw = 0;
440 tsk->active_mm = NULL;
443 * Are we cloning a kernel thread?
445 * We need to steal a active VM for that..
451 if (clone_flags & CLONE_VM) {
452 atomic_inc(&oldmm->mm_users);
455 * There are cases where the PTL is held to ensure no
456 * new threads start up in user mode using an mm, which
457 * allows optimizing out ipis; the tlb_gather_mmu code
460 spin_unlock_wait(&oldmm->page_table_lock);
469 /* Copy the current MM stuff.. */
470 memcpy(mm, oldmm, sizeof(*mm));
471 mm->mm_vx_info = NULL;
475 if (init_new_context(tsk,mm))
478 retval = dup_mmap(mm, oldmm);
482 mm->hiwater_rss = mm->rss;
483 mm->hiwater_vm = mm->total_vm;
497 * If init_new_context() failed, we cannot use mmput() to free the mm
498 * because it calls destroy_context()
500 clr_vx_info(&mm->mm_vx_info);
506 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
508 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
509 /* We don't need to lock fs - think why ;-) */
511 atomic_set(&fs->count, 1);
512 rwlock_init(&fs->lock);
513 fs->umask = old->umask;
514 read_lock(&old->lock);
515 fs->rootmnt = mntget(old->rootmnt);
516 fs->root = dget(old->root);
517 fs->pwdmnt = mntget(old->pwdmnt);
518 fs->pwd = dget(old->pwd);
520 fs->altrootmnt = mntget(old->altrootmnt);
521 fs->altroot = dget(old->altroot);
523 fs->altrootmnt = NULL;
526 read_unlock(&old->lock);
531 struct fs_struct *copy_fs_struct(struct fs_struct *old)
533 return __copy_fs_struct(old);
536 EXPORT_SYMBOL_GPL(copy_fs_struct);
538 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
540 if (clone_flags & CLONE_FS) {
541 atomic_inc(¤t->fs->count);
544 tsk->fs = __copy_fs_struct(current->fs);
550 static int count_open_files(struct files_struct *files, int size)
554 /* Find the last open fd */
555 for (i = size/(8*sizeof(long)); i > 0; ) {
556 if (files->open_fds->fds_bits[--i])
559 i = (i+1) * 8 * sizeof(long);
563 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
565 struct files_struct *oldf, *newf;
566 struct file **old_fds, **new_fds;
567 int open_files, size, i, error = 0, expand;
570 * A background process may not have any files ...
572 oldf = current->files;
576 if (clone_flags & CLONE_FILES) {
577 atomic_inc(&oldf->count);
582 * Note: we may be using current for both targets (See exec.c)
583 * This works because we cache current->files (old) as oldf. Don't
588 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
592 atomic_set(&newf->count, 1);
594 spin_lock_init(&newf->file_lock);
596 newf->max_fds = NR_OPEN_DEFAULT;
597 newf->max_fdset = __FD_SETSIZE;
598 newf->close_on_exec = &newf->close_on_exec_init;
599 newf->open_fds = &newf->open_fds_init;
600 newf->fd = &newf->fd_array[0];
602 spin_lock(&oldf->file_lock);
604 open_files = count_open_files(oldf, oldf->max_fdset);
608 * Check whether we need to allocate a larger fd array or fd set.
609 * Note: we're not a clone task, so the open count won't change.
611 if (open_files > newf->max_fdset) {
615 if (open_files > newf->max_fds) {
620 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
622 spin_unlock(&oldf->file_lock);
623 spin_lock(&newf->file_lock);
624 error = expand_files(newf, open_files-1);
625 spin_unlock(&newf->file_lock);
628 spin_lock(&oldf->file_lock);
634 memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
635 memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
637 for (i = open_files; i != 0; i--) {
638 struct file *f = *old_fds++;
641 /* FIXME sum it first for avail check and performance */
642 vx_openfd_inc(open_files - i);
645 * The fd may be claimed in the fd bitmap but not yet
646 * instantiated in the files array if a sibling thread
647 * is partway through open(). So make sure that this
648 * fd is available to the new process.
650 FD_CLR(open_files - i, newf->open_fds);
654 spin_unlock(&oldf->file_lock);
656 /* compute the remainder to be cleared */
657 size = (newf->max_fds - open_files) * sizeof(struct file *);
659 /* This is long word aligned thus could use a optimized version */
660 memset(new_fds, 0, size);
662 if (newf->max_fdset > open_files) {
663 int left = (newf->max_fdset-open_files)/8;
664 int start = open_files / (8 * sizeof(unsigned long));
666 memset(&newf->open_fds->fds_bits[start], 0, left);
667 memset(&newf->close_on_exec->fds_bits[start], 0, left);
676 free_fdset (newf->close_on_exec, newf->max_fdset);
677 free_fdset (newf->open_fds, newf->max_fdset);
678 free_fd_array(newf->fd, newf->max_fds);
679 kmem_cache_free(files_cachep, newf);
684 * Helper to unshare the files of the current task.
685 * We don't want to expose copy_files internals to
686 * the exec layer of the kernel.
689 int unshare_files(void)
691 struct files_struct *files = current->files;
697 /* This can race but the race causes us to copy when we don't
698 need to and drop the copy */
699 if(atomic_read(&files->count) == 1)
701 atomic_inc(&files->count);
704 rc = copy_files(0, current);
706 current->files = files;
710 EXPORT_SYMBOL(unshare_files);
712 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
714 struct sighand_struct *sig;
716 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
717 atomic_inc(¤t->sighand->count);
720 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
724 spin_lock_init(&sig->siglock);
725 atomic_set(&sig->count, 1);
726 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
730 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
732 struct signal_struct *sig;
734 if (clone_flags & CLONE_THREAD) {
735 atomic_inc(¤t->signal->count);
736 atomic_inc(¤t->signal->live);
739 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
743 atomic_set(&sig->count, 1);
744 atomic_set(&sig->live, 1);
745 init_waitqueue_head(&sig->wait_chldexit);
747 sig->group_exit_code = 0;
748 sig->group_exit_task = NULL;
749 sig->group_stop_count = 0;
750 sig->curr_target = NULL;
751 init_sigpending(&sig->shared_pending);
752 INIT_LIST_HEAD(&sig->posix_timers);
754 sig->tty = current->signal->tty;
755 sig->pgrp = process_group(current);
756 sig->session = current->signal->session;
757 sig->leader = 0; /* session leadership doesn't inherit */
758 sig->tty_old_pgrp = 0;
760 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
761 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
762 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
764 task_lock(current->group_leader);
765 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
766 task_unlock(current->group_leader);
771 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
773 unsigned long new_flags = p->flags;
775 new_flags &= ~PF_SUPERPRIV;
776 new_flags |= PF_FORKNOEXEC;
777 if (!(clone_flags & CLONE_PTRACE))
779 p->flags = new_flags;
782 asmlinkage long sys_set_tid_address(int __user *tidptr)
784 current->clear_child_tid = tidptr;
790 * This creates a new process as a copy of the old one,
791 * but does not actually start it yet.
793 * It copies the registers, and all the appropriate
794 * parts of the process environment (as per the clone
795 * flags). The actual kick-off is left to the caller.
797 static task_t *copy_process(unsigned long clone_flags,
798 unsigned long stack_start,
799 struct pt_regs *regs,
800 unsigned long stack_size,
801 int __user *parent_tidptr,
802 int __user *child_tidptr,
806 struct task_struct *p = NULL;
809 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
810 return ERR_PTR(-EINVAL);
813 * Thread groups must share signals as well, and detached threads
814 * can only be started up within the thread group.
816 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
817 return ERR_PTR(-EINVAL);
820 * Shared signal handlers imply shared VM. By way of the above,
821 * thread groups also imply shared VM. Blocking this case allows
822 * for various simplifications in other code.
824 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
825 return ERR_PTR(-EINVAL);
827 retval = security_task_create(clone_flags);
832 p = dup_task_struct(current);
836 init_vx_info(&p->vx_info, current->vx_info);
838 set_nx_info(&p->nx_info, current->nx_info);
840 /* check vserver memory */
841 if (p->mm && !(clone_flags & CLONE_VM)) {
842 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
843 vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
847 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
848 if (!vx_rsspages_avail(p->mm, p->mm->rss))
849 goto bad_fork_cleanup_vm;
853 if (!vx_nproc_avail(1))
854 goto bad_fork_cleanup_vm;
856 if (atomic_read(&p->user->processes) >=
857 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
858 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
859 p->user != &root_user)
860 goto bad_fork_cleanup_vm;
863 atomic_inc(&p->user->__count);
864 atomic_inc(&p->user->processes);
865 get_group_info(p->group_info);
868 * If multiple threads are within copy_process(), then this check
869 * triggers too late. This doesn't hurt, the check is only there
870 * to stop root fork bombs.
872 if (nr_threads >= max_threads)
873 goto bad_fork_cleanup_count;
875 if (!try_module_get(p->thread_info->exec_domain->module))
876 goto bad_fork_cleanup_count;
878 if (p->binfmt && !try_module_get(p->binfmt->module))
879 goto bad_fork_cleanup_put_domain;
882 copy_flags(clone_flags, p);
885 if (clone_flags & CLONE_PARENT_SETTID)
886 if (put_user(p->pid, parent_tidptr))
887 goto bad_fork_cleanup;
889 p->proc_dentry = NULL;
891 INIT_LIST_HEAD(&p->children);
892 INIT_LIST_HEAD(&p->sibling);
893 p->vfork_done = NULL;
894 spin_lock_init(&p->alloc_lock);
895 spin_lock_init(&p->proc_lock);
897 clear_tsk_thread_flag(p, TIF_SIGPENDING);
898 init_sigpending(&p->pending);
900 p->it_real_value = 0;
902 p->it_virt_value = cputime_zero;
903 p->it_virt_incr = cputime_zero;
904 p->it_prof_value = cputime_zero;
905 p->it_prof_incr = cputime_zero;
906 init_timer(&p->real_timer);
907 p->real_timer.data = (unsigned long) p;
909 p->utime = cputime_zero;
910 p->stime = cputime_zero;
911 p->rchar = 0; /* I/O counter: bytes read */
912 p->wchar = 0; /* I/O counter: bytes written */
913 p->syscr = 0; /* I/O counter: read syscalls */
914 p->syscw = 0; /* I/O counter: write syscalls */
915 acct_clear_integrals(p);
917 p->lock_depth = -1; /* -1 = no lock */
918 do_posix_clock_monotonic_gettime(&p->start_time);
920 p->io_context = NULL;
922 p->audit_context = NULL;
924 p->mempolicy = mpol_copy(p->mempolicy);
925 if (IS_ERR(p->mempolicy)) {
926 retval = PTR_ERR(p->mempolicy);
928 goto bad_fork_cleanup;
933 if (clone_flags & CLONE_THREAD)
934 p->tgid = current->tgid;
936 if ((retval = security_task_alloc(p)))
937 goto bad_fork_cleanup_policy;
938 if ((retval = audit_alloc(p)))
939 goto bad_fork_cleanup_security;
940 /* copy all the process information */
941 if ((retval = copy_semundo(clone_flags, p)))
942 goto bad_fork_cleanup_audit;
943 if ((retval = copy_files(clone_flags, p)))
944 goto bad_fork_cleanup_semundo;
945 if ((retval = copy_fs(clone_flags, p)))
946 goto bad_fork_cleanup_files;
947 if ((retval = copy_sighand(clone_flags, p)))
948 goto bad_fork_cleanup_fs;
949 if ((retval = copy_signal(clone_flags, p)))
950 goto bad_fork_cleanup_sighand;
951 if ((retval = copy_mm(clone_flags, p)))
952 goto bad_fork_cleanup_signal;
953 if ((retval = copy_keys(clone_flags, p)))
954 goto bad_fork_cleanup_mm;
955 if ((retval = copy_namespace(clone_flags, p)))
956 goto bad_fork_cleanup_keys;
957 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
959 goto bad_fork_cleanup_namespace;
961 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
963 * Clear TID on mm_release()?
965 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
968 * Syscall tracing should be turned off in the child regardless
971 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
973 /* Our parent execution domain becomes current domain
974 These must match for thread signalling to apply */
976 p->parent_exec_id = p->self_exec_id;
978 /* ok, now we should be set up.. */
979 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
980 p->pdeath_signal = 0;
983 /* Perform scheduler related setup */
987 * Ok, make it visible to the rest of the system.
988 * We dont wake it up yet.
991 INIT_LIST_HEAD(&p->ptrace_children);
992 INIT_LIST_HEAD(&p->ptrace_list);
994 /* Need tasklist lock for parent etc handling! */
995 write_lock_irq(&tasklist_lock);
998 * The task hasn't been attached yet, so cpus_allowed mask cannot
999 * have changed. The cpus_allowed mask of the parent may have
1000 * changed after it was copied first time, and it may then move to
1001 * another CPU - so we re-copy it here and set the child's CPU to
1002 * the parent's CPU. This avoids alot of nasty races.
1004 p->cpus_allowed = current->cpus_allowed;
1005 set_task_cpu(p, smp_processor_id());
1008 * Check for pending SIGKILL! The new thread should not be allowed
1009 * to slip out of an OOM kill. (or normal SIGKILL.)
1011 if (sigismember(¤t->pending.signal, SIGKILL)) {
1012 write_unlock_irq(&tasklist_lock);
1014 goto bad_fork_cleanup_namespace;
1017 /* CLONE_PARENT re-uses the old parent */
1018 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1019 p->real_parent = current->real_parent;
1021 p->real_parent = current;
1022 p->parent = p->real_parent;
1024 if (clone_flags & CLONE_THREAD) {
1025 spin_lock(¤t->sighand->siglock);
1027 * Important: if an exit-all has been started then
1028 * do not create this new thread - the whole thread
1029 * group is supposed to exit anyway.
1031 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1032 spin_unlock(¤t->sighand->siglock);
1033 write_unlock_irq(&tasklist_lock);
1035 goto bad_fork_cleanup_namespace;
1037 p->group_leader = current->group_leader;
1039 if (current->signal->group_stop_count > 0) {
1041 * There is an all-stop in progress for the group.
1042 * We ourselves will stop as soon as we check signals.
1043 * Make the new thread part of that group stop too.
1045 current->signal->group_stop_count++;
1046 set_tsk_thread_flag(p, TIF_SIGPENDING);
1049 spin_unlock(¤t->sighand->siglock);
1053 if (unlikely(p->ptrace & PT_PTRACED))
1054 __ptrace_link(p, current->parent);
1056 attach_pid(p, PIDTYPE_PID, p->pid);
1057 attach_pid(p, PIDTYPE_TGID, p->tgid);
1058 if (thread_group_leader(p)) {
1059 attach_pid(p, PIDTYPE_PGID, process_group(p));
1060 attach_pid(p, PIDTYPE_SID, p->signal->session);
1062 __get_cpu_var(process_counts)++;
1068 /* p is copy of current */
1071 claim_vx_info(vxi, p);
1072 atomic_inc(&vxi->cvirt.nr_threads);
1073 atomic_inc(&vxi->cvirt.total_forks);
1076 write_unlock_irq(&tasklist_lock);
1081 return ERR_PTR(retval);
1084 bad_fork_cleanup_namespace:
1086 bad_fork_cleanup_keys:
1088 bad_fork_cleanup_mm:
1091 bad_fork_cleanup_signal:
1093 bad_fork_cleanup_sighand:
1095 bad_fork_cleanup_fs:
1096 exit_fs(p); /* blocking */
1097 bad_fork_cleanup_files:
1098 exit_files(p); /* blocking */
1099 bad_fork_cleanup_semundo:
1101 bad_fork_cleanup_audit:
1103 bad_fork_cleanup_security:
1104 security_task_free(p);
1105 bad_fork_cleanup_policy:
1107 mpol_free(p->mempolicy);
1111 module_put(p->binfmt->module);
1112 bad_fork_cleanup_put_domain:
1113 module_put(p->thread_info->exec_domain->module);
1114 bad_fork_cleanup_count:
1115 put_group_info(p->group_info);
1116 atomic_dec(&p->user->processes);
1118 bad_fork_cleanup_vm:
1119 if (p->mm && !(clone_flags & CLONE_VM))
1120 vx_pages_sub(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
1126 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1128 memset(regs, 0, sizeof(struct pt_regs));
1132 task_t * __devinit fork_idle(int cpu)
1135 struct pt_regs regs;
1137 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0);
1139 return ERR_PTR(-ENOMEM);
1140 init_idle(task, cpu);
1141 unhash_process(task);
1145 static inline int fork_traceflag (unsigned clone_flags)
1147 if (clone_flags & CLONE_UNTRACED)
1149 else if (clone_flags & CLONE_VFORK) {
1150 if (current->ptrace & PT_TRACE_VFORK)
1151 return PTRACE_EVENT_VFORK;
1152 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1153 if (current->ptrace & PT_TRACE_CLONE)
1154 return PTRACE_EVENT_CLONE;
1155 } else if (current->ptrace & PT_TRACE_FORK)
1156 return PTRACE_EVENT_FORK;
1162 * Ok, this is the main fork-routine.
1164 * It copies the process, and if successful kick-starts
1165 * it and waits for it to finish using the VM if required.
1167 long do_fork(unsigned long clone_flags,
1168 unsigned long stack_start,
1169 struct pt_regs *regs,
1170 unsigned long stack_size,
1171 int __user *parent_tidptr,
1172 int __user *child_tidptr)
1174 struct task_struct *p;
1176 long pid = alloc_pidmap();
1180 if (unlikely(current->ptrace)) {
1181 trace = fork_traceflag (clone_flags);
1183 clone_flags |= CLONE_PTRACE;
1186 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1188 * Do this prior waking up the new thread - the thread pointer
1189 * might get invalid after that point, if the thread exits quickly.
1192 struct completion vfork;
1194 if (clone_flags & CLONE_VFORK) {
1195 p->vfork_done = &vfork;
1196 init_completion(&vfork);
1199 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1201 * We'll start up with an immediate SIGSTOP.
1203 sigaddset(&p->pending.signal, SIGSTOP);
1204 set_tsk_thread_flag(p, TIF_SIGPENDING);
1207 if (!(clone_flags & CLONE_STOPPED))
1208 wake_up_new_task(p, clone_flags);
1210 p->state = TASK_STOPPED;
1212 if (unlikely (trace)) {
1213 current->ptrace_message = pid;
1214 ptrace_notify ((trace << 8) | SIGTRAP);
1217 if (clone_flags & CLONE_VFORK) {
1218 wait_for_completion(&vfork);
1219 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1220 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1229 /* SLAB cache for signal_struct structures (tsk->signal) */
1230 kmem_cache_t *signal_cachep;
1232 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1233 kmem_cache_t *sighand_cachep;
1235 /* SLAB cache for files_struct structures (tsk->files) */
1236 kmem_cache_t *files_cachep;
1238 /* SLAB cache for fs_struct structures (tsk->fs) */
1239 kmem_cache_t *fs_cachep;
1241 /* SLAB cache for vm_area_struct structures */
1242 kmem_cache_t *vm_area_cachep;
1244 /* SLAB cache for mm_struct structures (tsk->mm) */
1245 kmem_cache_t *mm_cachep;
1247 void __init proc_caches_init(void)
1249 sighand_cachep = kmem_cache_create("sighand_cache",
1250 sizeof(struct sighand_struct), 0,
1251 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1252 signal_cachep = kmem_cache_create("signal_cache",
1253 sizeof(struct signal_struct), 0,
1254 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1255 files_cachep = kmem_cache_create("files_cache",
1256 sizeof(struct files_struct), 0,
1257 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1258 fs_cachep = kmem_cache_create("fs_cache",
1259 sizeof(struct fs_struct), 0,
1260 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1261 vm_area_cachep = kmem_cache_create("vm_area_struct",
1262 sizeof(struct vm_area_struct), 0,
1263 SLAB_PANIC, NULL, NULL);
1264 mm_cachep = kmem_cache_create("mm_struct",
1265 sizeof(struct mm_struct), 0,
1266 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);