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/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/smp_lock.h>
18 #include <linux/module.h>
19 #include <linux/vmalloc.h>
20 #include <linux/completion.h>
21 #include <linux/namespace.h>
22 #include <linux/personality.h>
23 #include <linux/mempolicy.h>
24 #include <linux/sem.h>
25 #include <linux/file.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
30 #include <linux/capability.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/rcupdate.h>
39 #include <linux/tracehook.h>
40 #include <linux/mount.h>
41 #include <linux/audit.h>
42 #include <linux/profile.h>
43 #include <linux/rmap.h>
44 #include <linux/acct.h>
45 #include <linux/cn_proc.h>
46 #include <linux/delayacct.h>
47 #include <linux/taskstats_kern.h>
48 #include <linux/vs_context.h>
49 #include <linux/vs_network.h>
50 #include <linux/vs_limit.h>
51 #include <linux/vs_memory.h>
53 #include <asm/pgtable.h>
54 #include <asm/pgalloc.h>
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
57 #include <asm/cacheflush.h>
58 #include <asm/tlbflush.h>
61 * Protected counters by write_lock_irq(&tasklist_lock)
63 unsigned long total_forks; /* Handle normal Linux uptimes. */
64 int nr_threads; /* The idle threads do not count.. */
66 int max_threads; /* tunable limit on nr_threads */
68 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
70 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
72 int nr_processes(void)
77 for_each_online_cpu(cpu)
78 total += per_cpu(process_counts, cpu);
83 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
84 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
85 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
86 static kmem_cache_t *task_struct_cachep;
89 /* SLAB cache for signal_struct structures (tsk->signal) */
90 static kmem_cache_t *signal_cachep;
92 /* SLAB cache for sighand_struct structures (tsk->sighand) */
93 kmem_cache_t *sighand_cachep;
95 /* SLAB cache for files_struct structures (tsk->files) */
96 kmem_cache_t *files_cachep;
98 /* SLAB cache for fs_struct structures (tsk->fs) */
99 kmem_cache_t *fs_cachep;
101 /* SLAB cache for vm_area_struct structures */
102 kmem_cache_t *vm_area_cachep;
104 /* SLAB cache for mm_struct structures (tsk->mm) */
105 static kmem_cache_t *mm_cachep;
107 void free_task(struct task_struct *tsk)
109 free_thread_info(tsk->thread_info);
110 rt_mutex_debug_task_free(tsk);
111 clr_vx_info(&tsk->vx_info);
112 clr_nx_info(&tsk->nx_info);
113 free_task_struct(tsk);
115 EXPORT_SYMBOL(free_task);
117 void __put_task_struct(struct task_struct *tsk)
119 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
120 WARN_ON(atomic_read(&tsk->usage));
121 WARN_ON(tsk == current);
123 security_task_free(tsk);
125 put_group_info(tsk->group_info);
126 delayacct_tsk_free(tsk);
128 if (!profile_handoff_task(tsk))
132 void __init fork_init(unsigned long mempages)
134 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
135 #ifndef ARCH_MIN_TASKALIGN
136 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
138 /* create a slab on which task_structs can be allocated */
140 kmem_cache_create("task_struct", sizeof(struct task_struct),
141 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
145 * The default maximum number of threads is set to a safe
146 * value: the thread structures can take up at most half
149 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
152 * we need to allow at least 20 threads to boot a system
157 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
158 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
159 init_task.signal->rlim[RLIMIT_SIGPENDING] =
160 init_task.signal->rlim[RLIMIT_NPROC];
163 static struct task_struct *dup_task_struct(struct task_struct *orig)
165 struct task_struct *tsk;
166 struct thread_info *ti;
168 prepare_to_copy(orig);
170 tsk = alloc_task_struct();
174 ti = alloc_thread_info(tsk);
176 free_task_struct(tsk);
181 tsk->thread_info = ti;
182 setup_thread_stack(tsk, orig);
184 /* One for us, one for whoever does the "release_task()" (usually parent) */
185 atomic_set(&tsk->usage,2);
186 atomic_set(&tsk->fs_excl, 0);
188 tsk->splice_pipe = NULL;
193 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
195 struct vm_area_struct *mpnt, *tmp, **pprev;
196 struct rb_node **rb_link, *rb_parent;
198 unsigned long charge;
199 struct mempolicy *pol;
201 down_write(&oldmm->mmap_sem);
202 flush_cache_mm(oldmm);
204 * Not linked in yet - no deadlock potential:
206 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
210 mm->mmap_cache = NULL;
211 mm->free_area_cache = oldmm->mmap_base;
212 mm->cached_hole_size = ~0UL;
214 __set_mm_counter(mm, file_rss, 0);
215 __set_mm_counter(mm, anon_rss, 0);
216 cpus_clear(mm->cpu_vm_mask);
218 rb_link = &mm->mm_rb.rb_node;
222 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
225 if (mpnt->vm_flags & VM_DONTCOPY) {
226 long pages = vma_pages(mpnt);
227 vx_vmpages_sub(mm, pages);
228 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
233 if (mpnt->vm_flags & VM_ACCOUNT) {
234 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
235 if (security_vm_enough_memory(len))
239 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
243 pol = mpol_copy(vma_policy(mpnt));
244 retval = PTR_ERR(pol);
246 goto fail_nomem_policy;
247 vma_set_policy(tmp, pol);
248 tmp->vm_flags &= ~VM_LOCKED;
254 struct inode *inode = file->f_dentry->d_inode;
256 if (tmp->vm_flags & VM_DENYWRITE)
257 atomic_dec(&inode->i_writecount);
259 /* insert tmp into the share list, just after mpnt */
260 spin_lock(&file->f_mapping->i_mmap_lock);
261 tmp->vm_truncate_count = mpnt->vm_truncate_count;
262 flush_dcache_mmap_lock(file->f_mapping);
263 vma_prio_tree_add(tmp, mpnt);
264 flush_dcache_mmap_unlock(file->f_mapping);
265 spin_unlock(&file->f_mapping->i_mmap_lock);
269 * Link in the new vma and copy the page table entries.
272 pprev = &tmp->vm_next;
274 __vma_link_rb(mm, tmp, rb_link, rb_parent);
275 rb_link = &tmp->vm_rb.rb_right;
276 rb_parent = &tmp->vm_rb;
279 retval = copy_page_range(mm, oldmm, mpnt);
281 if (tmp->vm_ops && tmp->vm_ops->open)
282 tmp->vm_ops->open(tmp);
288 arch_dup_mmap(mm, oldmm);
292 up_write(&mm->mmap_sem);
294 up_write(&oldmm->mmap_sem);
297 kmem_cache_free(vm_area_cachep, tmp);
300 vm_unacct_memory(charge);
304 static inline int mm_alloc_pgd(struct mm_struct * mm)
306 mm->pgd = pgd_alloc(mm);
307 if (unlikely(!mm->pgd))
312 static inline void mm_free_pgd(struct mm_struct * mm)
317 #define dup_mmap(mm, oldmm) (0)
318 #define mm_alloc_pgd(mm) (0)
319 #define mm_free_pgd(mm)
320 #endif /* CONFIG_MMU */
322 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
324 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
325 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
327 #include <linux/init_task.h>
329 static struct mm_struct * mm_init(struct mm_struct * mm)
331 atomic_set(&mm->mm_users, 1);
332 atomic_set(&mm->mm_count, 1);
333 init_rwsem(&mm->mmap_sem);
334 INIT_LIST_HEAD(&mm->mmlist);
335 mm->core_waiters = 0;
337 spin_lock_init(&mm->page_table_lock);
338 rwlock_init(&mm->ioctx_list_lock);
339 mm->ioctx_list = NULL;
340 mm->free_area_cache = TASK_UNMAPPED_BASE;
341 mm->cached_hole_size = ~0UL;
343 if (likely(!mm_alloc_pgd(mm))) {
345 set_vx_info(&mm->mm_vx_info, current->vx_info);
353 * Allocate and initialize an mm_struct.
355 struct mm_struct * mm_alloc(void)
357 struct mm_struct * mm;
361 memset(mm, 0, sizeof(*mm));
368 * Called when the last reference to the mm
369 * is dropped: either by a lazy thread or by
370 * mmput. Free the page directory and the mm.
372 void fastcall __mmdrop(struct mm_struct *mm)
374 BUG_ON(mm == &init_mm);
377 clr_vx_info(&mm->mm_vx_info);
382 * Decrement the use count and release all resources for an mm.
384 void mmput(struct mm_struct *mm)
388 if (atomic_dec_and_test(&mm->mm_users)) {
391 if (!list_empty(&mm->mmlist)) {
392 spin_lock(&mmlist_lock);
393 list_del(&mm->mmlist);
394 spin_unlock(&mmlist_lock);
400 EXPORT_SYMBOL_GPL(mmput);
403 * get_task_mm - acquire a reference to the task's mm
405 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
406 * this kernel workthread has transiently adopted a user mm with use_mm,
407 * to do its AIO) is not set and if so returns a reference to it, after
408 * bumping up the use count. User must release the mm via mmput()
409 * after use. Typically used by /proc and ptrace.
411 struct mm_struct *get_task_mm(struct task_struct *task)
413 struct mm_struct *mm;
418 if (task->flags & PF_BORROWED_MM)
421 atomic_inc(&mm->mm_users);
426 EXPORT_SYMBOL_GPL(get_task_mm);
428 /* Please note the differences between mmput and mm_release.
429 * mmput is called whenever we stop holding onto a mm_struct,
430 * error success whatever.
432 * mm_release is called after a mm_struct has been removed
433 * from the current process.
435 * This difference is important for error handling, when we
436 * only half set up a mm_struct for a new process and need to restore
437 * the old one. Because we mmput the new mm_struct before
438 * restoring the old one. . .
439 * Eric Biederman 10 January 1998
441 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
443 struct completion *vfork_done = tsk->vfork_done;
445 /* Get rid of any cached register state */
446 deactivate_mm(tsk, mm);
448 /* notify parent sleeping on vfork() */
450 tsk->vfork_done = NULL;
451 complete(vfork_done);
453 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
454 u32 __user * tidptr = tsk->clear_child_tid;
455 tsk->clear_child_tid = NULL;
458 * We don't check the error code - if userspace has
459 * not set up a proper pointer then tough luck.
462 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
467 * Allocate a new mm structure and copy contents from the
468 * mm structure of the passed in task structure.
470 static struct mm_struct *dup_mm(struct task_struct *tsk)
472 struct mm_struct *mm, *oldmm = current->mm;
482 memcpy(mm, oldmm, sizeof(*mm));
483 mm->mm_vx_info = NULL;
488 if (init_new_context(tsk, mm))
491 err = dup_mmap(mm, oldmm);
495 mm->hiwater_rss = get_mm_rss(mm);
496 mm->hiwater_vm = mm->total_vm;
508 * If init_new_context() failed, we cannot use mmput() to free the mm
509 * because it calls destroy_context()
511 clr_vx_info(&mm->mm_vx_info);
517 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
519 struct mm_struct * mm, *oldmm;
522 tsk->min_flt = tsk->maj_flt = 0;
523 tsk->nvcsw = tsk->nivcsw = 0;
526 tsk->active_mm = NULL;
529 * Are we cloning a kernel thread?
531 * We need to steal a active VM for that..
537 if (clone_flags & CLONE_VM) {
538 atomic_inc(&oldmm->mm_users);
557 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
559 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
560 /* We don't need to lock fs - think why ;-) */
562 atomic_set(&fs->count, 1);
563 rwlock_init(&fs->lock);
564 fs->umask = old->umask;
565 read_lock(&old->lock);
566 fs->rootmnt = mntget(old->rootmnt);
567 fs->root = dget(old->root);
568 fs->pwdmnt = mntget(old->pwdmnt);
569 fs->pwd = dget(old->pwd);
571 fs->altrootmnt = mntget(old->altrootmnt);
572 fs->altroot = dget(old->altroot);
574 fs->altrootmnt = NULL;
577 read_unlock(&old->lock);
582 struct fs_struct *copy_fs_struct(struct fs_struct *old)
584 return __copy_fs_struct(old);
587 EXPORT_SYMBOL_GPL(copy_fs_struct);
589 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
591 if (clone_flags & CLONE_FS) {
592 atomic_inc(¤t->fs->count);
595 tsk->fs = __copy_fs_struct(current->fs);
601 static int count_open_files(struct fdtable *fdt)
603 int size = fdt->max_fdset;
606 /* Find the last open fd */
607 for (i = size/(8*sizeof(long)); i > 0; ) {
608 if (fdt->open_fds->fds_bits[--i])
611 i = (i+1) * 8 * sizeof(long);
615 static struct files_struct *alloc_files(void)
617 struct files_struct *newf;
620 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
624 atomic_set(&newf->count, 1);
626 spin_lock_init(&newf->file_lock);
629 fdt->max_fds = NR_OPEN_DEFAULT;
630 fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
631 fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
632 fdt->open_fds = (fd_set *)&newf->open_fds_init;
633 fdt->fd = &newf->fd_array[0];
634 INIT_RCU_HEAD(&fdt->rcu);
635 fdt->free_files = NULL;
637 rcu_assign_pointer(newf->fdt, fdt);
643 * Allocate a new files structure and copy contents from the
644 * passed in files structure.
645 * errorp will be valid only when the returned files_struct is NULL.
647 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
649 struct files_struct *newf;
650 struct file **old_fds, **new_fds;
651 int open_files, size, i, expand;
652 struct fdtable *old_fdt, *new_fdt;
655 newf = alloc_files();
659 spin_lock(&oldf->file_lock);
660 old_fdt = files_fdtable(oldf);
661 new_fdt = files_fdtable(newf);
662 size = old_fdt->max_fdset;
663 open_files = count_open_files(old_fdt);
667 * Check whether we need to allocate a larger fd array or fd set.
668 * Note: we're not a clone task, so the open count won't change.
670 if (open_files > new_fdt->max_fdset) {
671 new_fdt->max_fdset = 0;
674 if (open_files > new_fdt->max_fds) {
675 new_fdt->max_fds = 0;
679 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
681 spin_unlock(&oldf->file_lock);
682 spin_lock(&newf->file_lock);
683 *errorp = expand_files(newf, open_files-1);
684 spin_unlock(&newf->file_lock);
687 new_fdt = files_fdtable(newf);
689 * Reacquire the oldf lock and a pointer to its fd table
690 * who knows it may have a new bigger fd table. We need
691 * the latest pointer.
693 spin_lock(&oldf->file_lock);
694 old_fdt = files_fdtable(oldf);
697 old_fds = old_fdt->fd;
698 new_fds = new_fdt->fd;
700 memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
701 memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
703 for (i = open_files; i != 0; i--) {
704 struct file *f = *old_fds++;
707 /* FIXME: sum it first for check and performance */
708 vx_openfd_inc(open_files - i);
711 * The fd may be claimed in the fd bitmap but not yet
712 * instantiated in the files array if a sibling thread
713 * is partway through open(). So make sure that this
714 * fd is available to the new process.
716 FD_CLR(open_files - i, new_fdt->open_fds);
718 rcu_assign_pointer(*new_fds++, f);
720 spin_unlock(&oldf->file_lock);
722 /* compute the remainder to be cleared */
723 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
725 /* This is long word aligned thus could use a optimized version */
726 memset(new_fds, 0, size);
728 if (new_fdt->max_fdset > open_files) {
729 int left = (new_fdt->max_fdset-open_files)/8;
730 int start = open_files / (8 * sizeof(unsigned long));
732 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
733 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
740 free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
741 free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
742 free_fd_array(new_fdt->fd, new_fdt->max_fds);
743 kmem_cache_free(files_cachep, newf);
747 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
749 struct files_struct *oldf, *newf;
753 * A background process may not have any files ...
755 oldf = current->files;
759 if (clone_flags & CLONE_FILES) {
760 atomic_inc(&oldf->count);
765 * Note: we may be using current for both targets (See exec.c)
766 * This works because we cache current->files (old) as oldf. Don't
770 newf = dup_fd(oldf, &error);
781 * Helper to unshare the files of the current task.
782 * We don't want to expose copy_files internals to
783 * the exec layer of the kernel.
786 int unshare_files(void)
788 struct files_struct *files = current->files;
793 /* This can race but the race causes us to copy when we don't
794 need to and drop the copy */
795 if(atomic_read(&files->count) == 1)
797 atomic_inc(&files->count);
800 rc = copy_files(0, current);
802 current->files = files;
806 EXPORT_SYMBOL(unshare_files);
808 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
810 struct sighand_struct *sig;
812 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
813 atomic_inc(¤t->sighand->count);
816 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
817 rcu_assign_pointer(tsk->sighand, sig);
820 atomic_set(&sig->count, 1);
821 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
825 void __cleanup_sighand(struct sighand_struct *sighand)
827 if (atomic_dec_and_test(&sighand->count))
828 kmem_cache_free(sighand_cachep, sighand);
831 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
833 struct signal_struct *sig;
836 if (clone_flags & CLONE_THREAD) {
837 atomic_inc(¤t->signal->count);
838 atomic_inc(¤t->signal->live);
839 taskstats_tgid_alloc(current->signal);
842 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
847 ret = copy_thread_group_keys(tsk);
849 kmem_cache_free(signal_cachep, sig);
853 atomic_set(&sig->count, 1);
854 atomic_set(&sig->live, 1);
855 init_waitqueue_head(&sig->wait_chldexit);
857 sig->group_exit_code = 0;
858 sig->group_exit_task = NULL;
859 sig->group_stop_count = 0;
860 sig->curr_target = NULL;
861 init_sigpending(&sig->shared_pending);
862 INIT_LIST_HEAD(&sig->posix_timers);
864 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
865 sig->it_real_incr.tv64 = 0;
866 sig->real_timer.function = it_real_fn;
869 sig->it_virt_expires = cputime_zero;
870 sig->it_virt_incr = cputime_zero;
871 sig->it_prof_expires = cputime_zero;
872 sig->it_prof_incr = cputime_zero;
874 sig->leader = 0; /* session leadership doesn't inherit */
875 sig->tty_old_pgrp = 0;
877 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
878 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
879 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
881 INIT_LIST_HEAD(&sig->cpu_timers[0]);
882 INIT_LIST_HEAD(&sig->cpu_timers[1]);
883 INIT_LIST_HEAD(&sig->cpu_timers[2]);
884 taskstats_tgid_init(sig);
886 task_lock(current->group_leader);
887 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
888 task_unlock(current->group_leader);
890 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
892 * New sole thread in the process gets an expiry time
893 * of the whole CPU time limit.
895 tsk->it_prof_expires =
896 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
898 acct_init_pacct(&sig->pacct);
903 void __cleanup_signal(struct signal_struct *sig)
905 exit_thread_group_keys(sig);
906 taskstats_tgid_free(sig);
907 kmem_cache_free(signal_cachep, sig);
910 static inline void cleanup_signal(struct task_struct *tsk)
912 struct signal_struct *sig = tsk->signal;
914 atomic_dec(&sig->live);
916 if (atomic_dec_and_test(&sig->count))
917 __cleanup_signal(sig);
920 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
922 unsigned long new_flags = p->flags;
924 new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
925 new_flags |= PF_FORKNOEXEC;
926 new_flags |= PF_STARTING;
927 p->flags = new_flags;
930 asmlinkage long sys_set_tid_address(int __user *tidptr)
932 current->clear_child_tid = tidptr;
937 static inline void rt_mutex_init_task(struct task_struct *p)
939 #ifdef CONFIG_RT_MUTEXES
940 spin_lock_init(&p->pi_lock);
941 plist_head_init(&p->pi_waiters, &p->pi_lock);
942 p->pi_blocked_on = NULL;
947 * This creates a new process as a copy of the old one,
948 * but does not actually start it yet.
950 * It copies the registers, and all the appropriate
951 * parts of the process environment (as per the clone
952 * flags). The actual kick-off is left to the caller.
954 static struct task_struct *copy_process(unsigned long clone_flags,
955 unsigned long stack_start,
956 struct pt_regs *regs,
957 unsigned long stack_size,
958 int __user *parent_tidptr,
959 int __user *child_tidptr,
963 struct task_struct *p = NULL;
967 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
968 return ERR_PTR(-EINVAL);
971 * Thread groups must share signals as well, and detached threads
972 * can only be started up within the thread group.
974 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
975 return ERR_PTR(-EINVAL);
978 * Shared signal handlers imply shared VM. By way of the above,
979 * thread groups also imply shared VM. Blocking this case allows
980 * for various simplifications in other code.
982 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
983 return ERR_PTR(-EINVAL);
985 retval = security_task_create(clone_flags);
990 p = dup_task_struct(current);
994 rt_mutex_init_task(p);
998 #ifdef CONFIG_TRACE_IRQFLAGS
999 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1000 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1002 init_vx_info(&p->vx_info, current->vx_info);
1003 init_nx_info(&p->nx_info, current->nx_info);
1005 /* check vserver memory */
1006 if (p->mm && !(clone_flags & CLONE_VM)) {
1007 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
1008 vx_pages_add(p->vx_info, RLIMIT_AS, p->mm->total_vm);
1012 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
1013 if (!vx_rsspages_avail(p->mm, get_mm_counter(p->mm, file_rss)))
1014 goto bad_fork_cleanup_vm;
1018 if (!vx_nproc_avail(1))
1019 goto bad_fork_cleanup_vm;
1021 if (atomic_read(&p->user->processes) >=
1022 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1023 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1024 p->user != &root_user)
1025 goto bad_fork_cleanup_vm;
1028 atomic_inc(&p->user->__count);
1029 atomic_inc(&p->user->processes);
1030 get_group_info(p->group_info);
1033 * If multiple threads are within copy_process(), then this check
1034 * triggers too late. This doesn't hurt, the check is only there
1035 * to stop root fork bombs.
1037 if (nr_threads >= max_threads)
1038 goto bad_fork_cleanup_count;
1040 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1041 goto bad_fork_cleanup_count;
1043 if (p->binfmt && !try_module_get(p->binfmt->module))
1044 goto bad_fork_cleanup_put_domain;
1047 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1048 copy_flags(clone_flags, p);
1051 if (clone_flags & CLONE_PARENT_SETTID)
1052 if (put_user(p->pid, parent_tidptr))
1053 goto bad_fork_cleanup_delays_binfmt;
1055 INIT_LIST_HEAD(&p->children);
1056 INIT_LIST_HEAD(&p->sibling);
1057 p->vfork_done = NULL;
1058 spin_lock_init(&p->alloc_lock);
1059 #ifdef CONFIG_PTRACE
1060 INIT_LIST_HEAD(&p->ptracees);
1063 clear_tsk_thread_flag(p, TIF_SIGPENDING);
1064 init_sigpending(&p->pending);
1066 p->utime = cputime_zero;
1067 p->stime = cputime_zero;
1069 p->rchar = 0; /* I/O counter: bytes read */
1070 p->wchar = 0; /* I/O counter: bytes written */
1071 p->syscr = 0; /* I/O counter: read syscalls */
1072 p->syscw = 0; /* I/O counter: write syscalls */
1073 acct_clear_integrals(p);
1075 p->it_virt_expires = cputime_zero;
1076 p->it_prof_expires = cputime_zero;
1077 p->it_sched_expires = 0;
1078 INIT_LIST_HEAD(&p->cpu_timers[0]);
1079 INIT_LIST_HEAD(&p->cpu_timers[1]);
1080 INIT_LIST_HEAD(&p->cpu_timers[2]);
1082 p->lock_depth = -1; /* -1 = no lock */
1083 do_posix_clock_monotonic_gettime(&p->start_time);
1085 p->io_context = NULL;
1087 p->audit_context = NULL;
1090 p->mempolicy = mpol_copy(p->mempolicy);
1091 if (IS_ERR(p->mempolicy)) {
1092 retval = PTR_ERR(p->mempolicy);
1093 p->mempolicy = NULL;
1094 goto bad_fork_cleanup_cpuset;
1096 mpol_fix_fork_child_flag(p);
1098 #ifdef CONFIG_TRACE_IRQFLAGS
1100 p->hardirqs_enabled = 0;
1101 p->hardirq_enable_ip = 0;
1102 p->hardirq_enable_event = 0;
1103 p->hardirq_disable_ip = _THIS_IP_;
1104 p->hardirq_disable_event = 0;
1105 p->softirqs_enabled = 1;
1106 p->softirq_enable_ip = _THIS_IP_;
1107 p->softirq_enable_event = 0;
1108 p->softirq_disable_ip = 0;
1109 p->softirq_disable_event = 0;
1110 p->hardirq_context = 0;
1111 p->softirq_context = 0;
1113 #ifdef CONFIG_LOCKDEP
1114 p->lockdep_depth = 0; /* no locks held yet */
1115 p->curr_chain_key = 0;
1116 p->lockdep_recursion = 0;
1119 #ifdef CONFIG_DEBUG_MUTEXES
1120 p->blocked_on = NULL; /* not blocked yet */
1124 if (clone_flags & CLONE_THREAD)
1125 p->tgid = current->tgid;
1127 if ((retval = security_task_alloc(p)))
1128 goto bad_fork_cleanup_policy;
1129 if ((retval = audit_alloc(p)))
1130 goto bad_fork_cleanup_security;
1131 /* copy all the process information */
1132 if ((retval = copy_semundo(clone_flags, p)))
1133 goto bad_fork_cleanup_audit;
1134 if ((retval = copy_files(clone_flags, p)))
1135 goto bad_fork_cleanup_semundo;
1136 if ((retval = copy_fs(clone_flags, p)))
1137 goto bad_fork_cleanup_files;
1138 if ((retval = copy_sighand(clone_flags, p)))
1139 goto bad_fork_cleanup_fs;
1140 if ((retval = copy_signal(clone_flags, p)))
1141 goto bad_fork_cleanup_sighand;
1142 if ((retval = copy_mm(clone_flags, p)))
1143 goto bad_fork_cleanup_signal;
1144 if ((retval = copy_keys(clone_flags, p)))
1145 goto bad_fork_cleanup_mm;
1146 if ((retval = copy_namespace(clone_flags, p)))
1147 goto bad_fork_cleanup_keys;
1148 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1150 goto bad_fork_cleanup_namespace;
1152 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1154 * Clear TID on mm_release()?
1156 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1157 p->robust_list = NULL;
1158 #ifdef CONFIG_COMPAT
1159 p->compat_robust_list = NULL;
1161 INIT_LIST_HEAD(&p->pi_state_list);
1162 p->pi_state_cache = NULL;
1165 * sigaltstack should be cleared when sharing the same VM
1167 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1168 p->sas_ss_sp = p->sas_ss_size = 0;
1171 * Syscall tracing should be turned off in the child regardless
1174 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1175 #ifdef TIF_SYSCALL_EMU
1176 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1179 /* Our parent execution domain becomes current domain
1180 These must match for thread signalling to apply */
1182 p->parent_exec_id = p->self_exec_id;
1184 /* ok, now we should be set up.. */
1185 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1186 p->pdeath_signal = 0;
1190 * Ok, make it visible to the rest of the system.
1191 * We dont wake it up yet.
1193 p->group_leader = p;
1194 INIT_LIST_HEAD(&p->thread_group);
1196 /* Perform scheduler related setup. Assign this task to a CPU. */
1197 sched_fork(p, clone_flags);
1199 /* Need tasklist lock for parent etc handling! */
1200 write_lock_irq(&tasklist_lock);
1203 * The task hasn't been attached yet, so its cpus_allowed mask will
1204 * not be changed, nor will its assigned CPU.
1206 * The cpus_allowed mask of the parent may have changed after it was
1207 * copied first time - so re-copy it here, then check the child's CPU
1208 * to ensure it is on a valid CPU (and if not, just force it back to
1209 * parent's CPU). This avoids alot of nasty races.
1211 p->cpus_allowed = current->cpus_allowed;
1212 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1213 !cpu_online(task_cpu(p))))
1214 set_task_cpu(p, smp_processor_id());
1216 /* CLONE_PARENT re-uses the old parent */
1217 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1218 p->parent = current->parent;
1220 p->parent = current;
1222 spin_lock(¤t->sighand->siglock);
1225 * Process group and session signals need to be delivered to just the
1226 * parent before the fork or both the parent and the child after the
1227 * fork. Restart if a signal comes in before we add the new process to
1228 * it's process group.
1229 * A fatal signal pending means that current will exit, so the new
1230 * thread can't slip out of an OOM kill (or normal SIGKILL).
1232 recalc_sigpending();
1233 if (signal_pending(current)) {
1234 spin_unlock(¤t->sighand->siglock);
1235 write_unlock_irq(&tasklist_lock);
1236 retval = -ERESTARTNOINTR;
1237 goto bad_fork_cleanup_namespace;
1240 if (clone_flags & CLONE_THREAD) {
1241 p->group_leader = current->group_leader;
1242 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1244 if (!cputime_eq(current->signal->it_virt_expires,
1246 !cputime_eq(current->signal->it_prof_expires,
1248 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1249 !list_empty(¤t->signal->cpu_timers[0]) ||
1250 !list_empty(¤t->signal->cpu_timers[1]) ||
1251 !list_empty(¤t->signal->cpu_timers[2])) {
1253 * Have child wake up on its first tick to check
1254 * for process CPU timers.
1256 p->it_prof_expires = jiffies_to_cputime(1);
1263 p->ioprio = current->ioprio;
1265 if (likely(p->pid)) {
1267 tracehook_init_task(p);
1269 if (thread_group_leader(p)) {
1270 p->signal->tty = current->signal->tty;
1271 p->signal->pgrp = process_group(current);
1272 p->signal->session = current->signal->session;
1273 attach_pid(p, PIDTYPE_PGID, process_group(p));
1274 attach_pid(p, PIDTYPE_SID, p->signal->session);
1276 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1277 __get_cpu_var(process_counts)++;
1279 attach_pid(p, PIDTYPE_PID, p->pid);
1284 spin_unlock(¤t->sighand->siglock);
1286 /* p is copy of current */
1289 claim_vx_info(vxi, p);
1290 atomic_inc(&vxi->cvirt.nr_threads);
1291 atomic_inc(&vxi->cvirt.total_forks);
1296 claim_nx_info(nxi, p);
1297 write_unlock_irq(&tasklist_lock);
1298 proc_fork_connector(p);
1301 bad_fork_cleanup_namespace:
1303 bad_fork_cleanup_keys:
1305 bad_fork_cleanup_mm:
1308 bad_fork_cleanup_signal:
1310 bad_fork_cleanup_sighand:
1311 __cleanup_sighand(p->sighand);
1312 bad_fork_cleanup_fs:
1313 exit_fs(p); /* blocking */
1314 bad_fork_cleanup_files:
1315 exit_files(p); /* blocking */
1316 bad_fork_cleanup_semundo:
1318 bad_fork_cleanup_audit:
1320 bad_fork_cleanup_security:
1321 security_task_free(p);
1322 bad_fork_cleanup_policy:
1324 mpol_free(p->mempolicy);
1325 bad_fork_cleanup_cpuset:
1328 bad_fork_cleanup_delays_binfmt:
1329 delayacct_tsk_free(p);
1331 module_put(p->binfmt->module);
1332 bad_fork_cleanup_put_domain:
1333 module_put(task_thread_info(p)->exec_domain->module);
1334 bad_fork_cleanup_count:
1335 put_group_info(p->group_info);
1336 atomic_dec(&p->user->processes);
1338 bad_fork_cleanup_vm:
1339 if (p->mm && !(clone_flags & CLONE_VM))
1340 vx_pages_sub(p->vx_info, RLIMIT_AS, p->mm->total_vm);
1344 return ERR_PTR(retval);
1347 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1349 memset(regs, 0, sizeof(struct pt_regs));
1353 struct task_struct * __devinit fork_idle(int cpu)
1355 struct task_struct *task;
1356 struct pt_regs regs;
1358 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0);
1360 return ERR_PTR(-ENOMEM);
1361 init_idle(task, cpu);
1367 * Ok, this is the main fork-routine.
1369 * It copies the process, and if successful kick-starts
1370 * it and waits for it to finish using the VM if required.
1372 long do_fork(unsigned long clone_flags,
1373 unsigned long stack_start,
1374 struct pt_regs *regs,
1375 unsigned long stack_size,
1376 int __user *parent_tidptr,
1377 int __user *child_tidptr)
1379 struct task_struct *p;
1380 struct pid *pid = alloc_pid();
1387 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1389 * Do this prior waking up the new thread - the thread pointer
1390 * might get invalid after that point, if the thread exits quickly.
1393 struct completion vfork;
1395 if (clone_flags & CLONE_VFORK) {
1396 p->vfork_done = &vfork;
1397 init_completion(&vfork);
1400 tracehook_report_clone(clone_flags, p);
1402 p->flags &= ~PF_STARTING;
1404 if (clone_flags & CLONE_STOPPED) {
1406 * We'll start up with an immediate SIGSTOP.
1408 sigaddset(&p->pending.signal, SIGSTOP);
1409 set_tsk_thread_flag(p, TIF_SIGPENDING);
1410 p->state = TASK_STOPPED;
1413 wake_up_new_task(p, clone_flags);
1415 tracehook_report_clone_complete(clone_flags, nr, p);
1417 if (clone_flags & CLONE_VFORK) {
1418 wait_for_completion(&vfork);
1419 tracehook_report_vfork_done(p, nr);
1428 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1429 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1432 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1434 struct sighand_struct *sighand = data;
1436 if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1437 SLAB_CTOR_CONSTRUCTOR)
1438 spin_lock_init(&sighand->siglock);
1441 void __init proc_caches_init(void)
1443 sighand_cachep = kmem_cache_create("sighand_cache",
1444 sizeof(struct sighand_struct), 0,
1445 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1446 sighand_ctor, NULL);
1447 signal_cachep = kmem_cache_create("signal_cache",
1448 sizeof(struct signal_struct), 0,
1449 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1450 files_cachep = kmem_cache_create("files_cache",
1451 sizeof(struct files_struct), 0,
1452 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1453 fs_cachep = kmem_cache_create("fs_cache",
1454 sizeof(struct fs_struct), 0,
1455 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1456 vm_area_cachep = kmem_cache_create("vm_area_struct",
1457 sizeof(struct vm_area_struct), 0,
1458 SLAB_PANIC, NULL, NULL);
1459 mm_cachep = kmem_cache_create("mm_struct",
1460 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1461 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1466 * Check constraints on flags passed to the unshare system call and
1467 * force unsharing of additional process context as appropriate.
1469 static inline void check_unshare_flags(unsigned long *flags_ptr)
1472 * If unsharing a thread from a thread group, must also
1475 if (*flags_ptr & CLONE_THREAD)
1476 *flags_ptr |= CLONE_VM;
1479 * If unsharing vm, must also unshare signal handlers.
1481 if (*flags_ptr & CLONE_VM)
1482 *flags_ptr |= CLONE_SIGHAND;
1485 * If unsharing signal handlers and the task was created
1486 * using CLONE_THREAD, then must unshare the thread
1488 if ((*flags_ptr & CLONE_SIGHAND) &&
1489 (atomic_read(¤t->signal->count) > 1))
1490 *flags_ptr |= CLONE_THREAD;
1493 * If unsharing namespace, must also unshare filesystem information.
1495 if (*flags_ptr & CLONE_NEWNS)
1496 *flags_ptr |= CLONE_FS;
1500 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1502 static int unshare_thread(unsigned long unshare_flags)
1504 if (unshare_flags & CLONE_THREAD)
1511 * Unshare the filesystem structure if it is being shared
1513 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1515 struct fs_struct *fs = current->fs;
1517 if ((unshare_flags & CLONE_FS) &&
1518 (fs && atomic_read(&fs->count) > 1)) {
1519 *new_fsp = __copy_fs_struct(current->fs);
1528 * Unshare the namespace structure if it is being shared
1530 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1532 struct namespace *ns = current->namespace;
1534 if ((unshare_flags & CLONE_NEWNS) &&
1535 (ns && atomic_read(&ns->count) > 1)) {
1536 if (!capable(CAP_SYS_ADMIN))
1539 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1548 * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1551 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1553 struct sighand_struct *sigh = current->sighand;
1555 if ((unshare_flags & CLONE_SIGHAND) &&
1556 (sigh && atomic_read(&sigh->count) > 1))
1563 * Unshare vm if it is being shared
1565 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1567 struct mm_struct *mm = current->mm;
1569 if ((unshare_flags & CLONE_VM) &&
1570 (mm && atomic_read(&mm->mm_users) > 1)) {
1578 * Unshare file descriptor table if it is being shared
1580 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1582 struct files_struct *fd = current->files;
1585 if ((unshare_flags & CLONE_FILES) &&
1586 (fd && atomic_read(&fd->count) > 1)) {
1587 *new_fdp = dup_fd(fd, &error);
1596 * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1599 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1601 if (unshare_flags & CLONE_SYSVSEM)
1608 * unshare allows a process to 'unshare' part of the process
1609 * context which was originally shared using clone. copy_*
1610 * functions used by do_fork() cannot be used here directly
1611 * because they modify an inactive task_struct that is being
1612 * constructed. Here we are modifying the current, active,
1615 asmlinkage long sys_unshare(unsigned long unshare_flags)
1618 struct fs_struct *fs, *new_fs = NULL;
1619 struct namespace *ns, *new_ns = NULL;
1620 struct sighand_struct *sigh, *new_sigh = NULL;
1621 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1622 struct files_struct *fd, *new_fd = NULL;
1623 struct sem_undo_list *new_ulist = NULL;
1625 check_unshare_flags(&unshare_flags);
1627 /* Return -EINVAL for all unsupported flags */
1629 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1630 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1631 goto bad_unshare_out;
1633 if ((err = unshare_thread(unshare_flags)))
1634 goto bad_unshare_out;
1635 if ((err = unshare_fs(unshare_flags, &new_fs)))
1636 goto bad_unshare_cleanup_thread;
1637 if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1638 goto bad_unshare_cleanup_fs;
1639 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1640 goto bad_unshare_cleanup_ns;
1641 if ((err = unshare_vm(unshare_flags, &new_mm)))
1642 goto bad_unshare_cleanup_sigh;
1643 if ((err = unshare_fd(unshare_flags, &new_fd)))
1644 goto bad_unshare_cleanup_vm;
1645 if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1646 goto bad_unshare_cleanup_fd;
1648 if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1654 current->fs = new_fs;
1659 ns = current->namespace;
1660 current->namespace = new_ns;
1665 sigh = current->sighand;
1666 rcu_assign_pointer(current->sighand, new_sigh);
1672 active_mm = current->active_mm;
1673 current->mm = new_mm;
1674 current->active_mm = new_mm;
1675 activate_mm(active_mm, new_mm);
1680 fd = current->files;
1681 current->files = new_fd;
1685 task_unlock(current);
1688 bad_unshare_cleanup_fd:
1690 put_files_struct(new_fd);
1692 bad_unshare_cleanup_vm:
1696 bad_unshare_cleanup_sigh:
1698 if (atomic_dec_and_test(&new_sigh->count))
1699 kmem_cache_free(sighand_cachep, new_sigh);
1701 bad_unshare_cleanup_ns:
1703 put_namespace(new_ns);
1705 bad_unshare_cleanup_fs:
1707 put_fs_struct(new_fs);
1709 bad_unshare_cleanup_thread: