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/ptrace.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>
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 /* SLAB cache for signal_struct structures (tsk->signal) */
86 static kmem_cache_t *signal_cachep;
88 /* SLAB cache for sighand_struct structures (tsk->sighand) */
89 kmem_cache_t *sighand_cachep;
91 /* SLAB cache for files_struct structures (tsk->files) */
92 kmem_cache_t *files_cachep;
94 /* SLAB cache for fs_struct structures (tsk->fs) */
95 kmem_cache_t *fs_cachep;
97 /* SLAB cache for vm_area_struct structures */
98 kmem_cache_t *vm_area_cachep;
100 /* SLAB cache for mm_struct structures (tsk->mm) */
101 static kmem_cache_t *mm_cachep;
103 void free_task(struct task_struct *tsk)
105 free_thread_info(tsk->thread_info);
106 rt_mutex_debug_task_free(tsk);
107 free_task_struct(tsk);
109 EXPORT_SYMBOL(free_task);
111 void __put_task_struct(struct task_struct *tsk)
113 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
114 WARN_ON(atomic_read(&tsk->usage));
115 WARN_ON(tsk == current);
117 security_task_free(tsk);
119 put_group_info(tsk->group_info);
121 if (!profile_handoff_task(tsk))
125 void __init fork_init(unsigned long mempages)
127 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
128 #ifndef ARCH_MIN_TASKALIGN
129 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
131 /* create a slab on which task_structs can be allocated */
133 kmem_cache_create("task_struct", sizeof(struct task_struct),
134 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
138 * The default maximum number of threads is set to a safe
139 * value: the thread structures can take up at most half
142 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
145 * we need to allow at least 20 threads to boot a system
150 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
151 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
152 init_task.signal->rlim[RLIMIT_SIGPENDING] =
153 init_task.signal->rlim[RLIMIT_NPROC];
156 static struct task_struct *dup_task_struct(struct task_struct *orig)
158 struct task_struct *tsk;
159 struct thread_info *ti;
161 prepare_to_copy(orig);
163 tsk = alloc_task_struct();
167 ti = alloc_thread_info(tsk);
169 free_task_struct(tsk);
174 tsk->thread_info = ti;
175 setup_thread_stack(tsk, orig);
177 /* One for us, one for whoever does the "release_task()" (usually parent) */
178 atomic_set(&tsk->usage,2);
179 atomic_set(&tsk->fs_excl, 0);
181 tsk->splice_pipe = NULL;
186 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
188 struct vm_area_struct *mpnt, *tmp, **pprev;
189 struct rb_node **rb_link, *rb_parent;
191 unsigned long charge;
192 struct mempolicy *pol;
194 down_write(&oldmm->mmap_sem);
195 flush_cache_mm(oldmm);
196 down_write(&mm->mmap_sem);
200 mm->mmap_cache = NULL;
201 mm->free_area_cache = oldmm->mmap_base;
202 mm->cached_hole_size = ~0UL;
204 cpus_clear(mm->cpu_vm_mask);
206 rb_link = &mm->mm_rb.rb_node;
210 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
213 if (mpnt->vm_flags & VM_DONTCOPY) {
214 long pages = vma_pages(mpnt);
215 mm->total_vm -= pages;
216 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
221 if (mpnt->vm_flags & VM_ACCOUNT) {
222 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
223 if (security_vm_enough_memory(len))
227 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
231 pol = mpol_copy(vma_policy(mpnt));
232 retval = PTR_ERR(pol);
234 goto fail_nomem_policy;
235 vma_set_policy(tmp, pol);
236 tmp->vm_flags &= ~VM_LOCKED;
242 struct inode *inode = file->f_dentry->d_inode;
244 if (tmp->vm_flags & VM_DENYWRITE)
245 atomic_dec(&inode->i_writecount);
247 /* insert tmp into the share list, just after mpnt */
248 spin_lock(&file->f_mapping->i_mmap_lock);
249 tmp->vm_truncate_count = mpnt->vm_truncate_count;
250 flush_dcache_mmap_lock(file->f_mapping);
251 vma_prio_tree_add(tmp, mpnt);
252 flush_dcache_mmap_unlock(file->f_mapping);
253 spin_unlock(&file->f_mapping->i_mmap_lock);
257 * Link in the new vma and copy the page table entries.
260 pprev = &tmp->vm_next;
262 __vma_link_rb(mm, tmp, rb_link, rb_parent);
263 rb_link = &tmp->vm_rb.rb_right;
264 rb_parent = &tmp->vm_rb;
267 retval = copy_page_range(mm, oldmm, mpnt);
269 if (tmp->vm_ops && tmp->vm_ops->open)
270 tmp->vm_ops->open(tmp);
277 up_write(&mm->mmap_sem);
279 up_write(&oldmm->mmap_sem);
282 kmem_cache_free(vm_area_cachep, tmp);
285 vm_unacct_memory(charge);
289 static inline int mm_alloc_pgd(struct mm_struct * mm)
291 mm->pgd = pgd_alloc(mm);
292 if (unlikely(!mm->pgd))
297 static inline void mm_free_pgd(struct mm_struct * mm)
302 #define dup_mmap(mm, oldmm) (0)
303 #define mm_alloc_pgd(mm) (0)
304 #define mm_free_pgd(mm)
305 #endif /* CONFIG_MMU */
307 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
309 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
310 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
312 #include <linux/init_task.h>
314 static struct mm_struct * mm_init(struct mm_struct * mm)
316 atomic_set(&mm->mm_users, 1);
317 atomic_set(&mm->mm_count, 1);
318 init_rwsem(&mm->mmap_sem);
319 INIT_LIST_HEAD(&mm->mmlist);
320 mm->core_waiters = 0;
322 set_mm_counter(mm, file_rss, 0);
323 set_mm_counter(mm, anon_rss, 0);
324 spin_lock_init(&mm->page_table_lock);
325 rwlock_init(&mm->ioctx_list_lock);
326 mm->ioctx_list = NULL;
327 mm->free_area_cache = TASK_UNMAPPED_BASE;
328 mm->cached_hole_size = ~0UL;
330 if (likely(!mm_alloc_pgd(mm))) {
339 * Allocate and initialize an mm_struct.
341 struct mm_struct * mm_alloc(void)
343 struct mm_struct * mm;
347 memset(mm, 0, sizeof(*mm));
354 * Called when the last reference to the mm
355 * is dropped: either by a lazy thread or by
356 * mmput. Free the page directory and the mm.
358 void fastcall __mmdrop(struct mm_struct *mm)
360 BUG_ON(mm == &init_mm);
367 * Decrement the use count and release all resources for an mm.
369 void mmput(struct mm_struct *mm)
373 if (atomic_dec_and_test(&mm->mm_users)) {
376 if (!list_empty(&mm->mmlist)) {
377 spin_lock(&mmlist_lock);
378 list_del(&mm->mmlist);
379 spin_unlock(&mmlist_lock);
385 EXPORT_SYMBOL_GPL(mmput);
388 * get_task_mm - acquire a reference to the task's mm
390 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
391 * this kernel workthread has transiently adopted a user mm with use_mm,
392 * to do its AIO) is not set and if so returns a reference to it, after
393 * bumping up the use count. User must release the mm via mmput()
394 * after use. Typically used by /proc and ptrace.
396 struct mm_struct *get_task_mm(struct task_struct *task)
398 struct mm_struct *mm;
403 if (task->flags & PF_BORROWED_MM)
406 atomic_inc(&mm->mm_users);
411 EXPORT_SYMBOL_GPL(get_task_mm);
413 /* Please note the differences between mmput and mm_release.
414 * mmput is called whenever we stop holding onto a mm_struct,
415 * error success whatever.
417 * mm_release is called after a mm_struct has been removed
418 * from the current process.
420 * This difference is important for error handling, when we
421 * only half set up a mm_struct for a new process and need to restore
422 * the old one. Because we mmput the new mm_struct before
423 * restoring the old one. . .
424 * Eric Biederman 10 January 1998
426 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
428 struct completion *vfork_done = tsk->vfork_done;
430 /* Get rid of any cached register state */
431 deactivate_mm(tsk, mm);
433 /* notify parent sleeping on vfork() */
435 tsk->vfork_done = NULL;
436 complete(vfork_done);
438 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
439 u32 __user * tidptr = tsk->clear_child_tid;
440 tsk->clear_child_tid = NULL;
443 * We don't check the error code - if userspace has
444 * not set up a proper pointer then tough luck.
447 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
452 * Allocate a new mm structure and copy contents from the
453 * mm structure of the passed in task structure.
455 static struct mm_struct *dup_mm(struct task_struct *tsk)
457 struct mm_struct *mm, *oldmm = current->mm;
467 memcpy(mm, oldmm, sizeof(*mm));
472 if (init_new_context(tsk, mm))
475 err = dup_mmap(mm, oldmm);
479 mm->hiwater_rss = get_mm_rss(mm);
480 mm->hiwater_vm = mm->total_vm;
492 * If init_new_context() failed, we cannot use mmput() to free the mm
493 * because it calls destroy_context()
500 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
502 struct mm_struct * mm, *oldmm;
505 tsk->min_flt = tsk->maj_flt = 0;
506 tsk->nvcsw = tsk->nivcsw = 0;
509 tsk->active_mm = NULL;
512 * Are we cloning a kernel thread?
514 * We need to steal a active VM for that..
520 if (clone_flags & CLONE_VM) {
521 atomic_inc(&oldmm->mm_users);
540 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
542 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
543 /* We don't need to lock fs - think why ;-) */
545 atomic_set(&fs->count, 1);
546 rwlock_init(&fs->lock);
547 fs->umask = old->umask;
548 read_lock(&old->lock);
549 fs->rootmnt = mntget(old->rootmnt);
550 fs->root = dget(old->root);
551 fs->pwdmnt = mntget(old->pwdmnt);
552 fs->pwd = dget(old->pwd);
554 fs->altrootmnt = mntget(old->altrootmnt);
555 fs->altroot = dget(old->altroot);
557 fs->altrootmnt = NULL;
560 read_unlock(&old->lock);
565 struct fs_struct *copy_fs_struct(struct fs_struct *old)
567 return __copy_fs_struct(old);
570 EXPORT_SYMBOL_GPL(copy_fs_struct);
572 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
574 if (clone_flags & CLONE_FS) {
575 atomic_inc(¤t->fs->count);
578 tsk->fs = __copy_fs_struct(current->fs);
584 static int count_open_files(struct fdtable *fdt)
586 int size = fdt->max_fdset;
589 /* Find the last open fd */
590 for (i = size/(8*sizeof(long)); i > 0; ) {
591 if (fdt->open_fds->fds_bits[--i])
594 i = (i+1) * 8 * sizeof(long);
598 static struct files_struct *alloc_files(void)
600 struct files_struct *newf;
603 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
607 atomic_set(&newf->count, 1);
609 spin_lock_init(&newf->file_lock);
612 fdt->max_fds = NR_OPEN_DEFAULT;
613 fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
614 fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
615 fdt->open_fds = (fd_set *)&newf->open_fds_init;
616 fdt->fd = &newf->fd_array[0];
617 INIT_RCU_HEAD(&fdt->rcu);
618 fdt->free_files = NULL;
620 rcu_assign_pointer(newf->fdt, fdt);
626 * Allocate a new files structure and copy contents from the
627 * passed in files structure.
628 * errorp will be valid only when the returned files_struct is NULL.
630 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
632 struct files_struct *newf;
633 struct file **old_fds, **new_fds;
634 int open_files, size, i, expand;
635 struct fdtable *old_fdt, *new_fdt;
638 newf = alloc_files();
642 spin_lock(&oldf->file_lock);
643 old_fdt = files_fdtable(oldf);
644 new_fdt = files_fdtable(newf);
645 size = old_fdt->max_fdset;
646 open_files = count_open_files(old_fdt);
650 * Check whether we need to allocate a larger fd array or fd set.
651 * Note: we're not a clone task, so the open count won't change.
653 if (open_files > new_fdt->max_fdset) {
654 new_fdt->max_fdset = 0;
657 if (open_files > new_fdt->max_fds) {
658 new_fdt->max_fds = 0;
662 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
664 spin_unlock(&oldf->file_lock);
665 spin_lock(&newf->file_lock);
666 *errorp = expand_files(newf, open_files-1);
667 spin_unlock(&newf->file_lock);
670 new_fdt = files_fdtable(newf);
672 * Reacquire the oldf lock and a pointer to its fd table
673 * who knows it may have a new bigger fd table. We need
674 * the latest pointer.
676 spin_lock(&oldf->file_lock);
677 old_fdt = files_fdtable(oldf);
680 old_fds = old_fdt->fd;
681 new_fds = new_fdt->fd;
683 memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
684 memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
686 for (i = open_files; i != 0; i--) {
687 struct file *f = *old_fds++;
692 * The fd may be claimed in the fd bitmap but not yet
693 * instantiated in the files array if a sibling thread
694 * is partway through open(). So make sure that this
695 * fd is available to the new process.
697 FD_CLR(open_files - i, new_fdt->open_fds);
699 rcu_assign_pointer(*new_fds++, f);
701 spin_unlock(&oldf->file_lock);
703 /* compute the remainder to be cleared */
704 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
706 /* This is long word aligned thus could use a optimized version */
707 memset(new_fds, 0, size);
709 if (new_fdt->max_fdset > open_files) {
710 int left = (new_fdt->max_fdset-open_files)/8;
711 int start = open_files / (8 * sizeof(unsigned long));
713 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
714 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
721 free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
722 free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
723 free_fd_array(new_fdt->fd, new_fdt->max_fds);
724 kmem_cache_free(files_cachep, newf);
728 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
730 struct files_struct *oldf, *newf;
734 * A background process may not have any files ...
736 oldf = current->files;
740 if (clone_flags & CLONE_FILES) {
741 atomic_inc(&oldf->count);
746 * Note: we may be using current for both targets (See exec.c)
747 * This works because we cache current->files (old) as oldf. Don't
751 newf = dup_fd(oldf, &error);
762 * Helper to unshare the files of the current task.
763 * We don't want to expose copy_files internals to
764 * the exec layer of the kernel.
767 int unshare_files(void)
769 struct files_struct *files = current->files;
774 /* This can race but the race causes us to copy when we don't
775 need to and drop the copy */
776 if(atomic_read(&files->count) == 1)
778 atomic_inc(&files->count);
781 rc = copy_files(0, current);
783 current->files = files;
787 EXPORT_SYMBOL(unshare_files);
789 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
791 struct sighand_struct *sig;
793 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
794 atomic_inc(¤t->sighand->count);
797 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
798 rcu_assign_pointer(tsk->sighand, sig);
801 atomic_set(&sig->count, 1);
802 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
806 void __cleanup_sighand(struct sighand_struct *sighand)
808 if (atomic_dec_and_test(&sighand->count))
809 kmem_cache_free(sighand_cachep, sighand);
812 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
814 struct signal_struct *sig;
817 if (clone_flags & CLONE_THREAD) {
818 atomic_inc(¤t->signal->count);
819 atomic_inc(¤t->signal->live);
822 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
827 ret = copy_thread_group_keys(tsk);
829 kmem_cache_free(signal_cachep, sig);
833 atomic_set(&sig->count, 1);
834 atomic_set(&sig->live, 1);
835 init_waitqueue_head(&sig->wait_chldexit);
837 sig->group_exit_code = 0;
838 sig->group_exit_task = NULL;
839 sig->group_stop_count = 0;
840 sig->curr_target = NULL;
841 init_sigpending(&sig->shared_pending);
842 INIT_LIST_HEAD(&sig->posix_timers);
844 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
845 sig->it_real_incr.tv64 = 0;
846 sig->real_timer.function = it_real_fn;
849 sig->it_virt_expires = cputime_zero;
850 sig->it_virt_incr = cputime_zero;
851 sig->it_prof_expires = cputime_zero;
852 sig->it_prof_incr = cputime_zero;
854 sig->leader = 0; /* session leadership doesn't inherit */
855 sig->tty_old_pgrp = 0;
857 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
858 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
859 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
861 INIT_LIST_HEAD(&sig->cpu_timers[0]);
862 INIT_LIST_HEAD(&sig->cpu_timers[1]);
863 INIT_LIST_HEAD(&sig->cpu_timers[2]);
865 task_lock(current->group_leader);
866 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
867 task_unlock(current->group_leader);
869 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
871 * New sole thread in the process gets an expiry time
872 * of the whole CPU time limit.
874 tsk->it_prof_expires =
875 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
877 acct_init_pacct(&sig->pacct);
882 void __cleanup_signal(struct signal_struct *sig)
884 exit_thread_group_keys(sig);
885 kmem_cache_free(signal_cachep, sig);
888 static inline void cleanup_signal(struct task_struct *tsk)
890 struct signal_struct *sig = tsk->signal;
892 atomic_dec(&sig->live);
894 if (atomic_dec_and_test(&sig->count))
895 __cleanup_signal(sig);
898 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
900 unsigned long new_flags = p->flags;
902 new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
903 new_flags |= PF_FORKNOEXEC;
904 if (!(clone_flags & CLONE_PTRACE))
906 p->flags = new_flags;
909 asmlinkage long sys_set_tid_address(int __user *tidptr)
911 current->clear_child_tid = tidptr;
916 static inline void rt_mutex_init_task(struct task_struct *p)
918 #ifdef CONFIG_RT_MUTEXES
919 spin_lock_init(&p->pi_lock);
920 plist_head_init(&p->pi_waiters, &p->pi_lock);
921 p->pi_blocked_on = NULL;
926 * This creates a new process as a copy of the old one,
927 * but does not actually start it yet.
929 * It copies the registers, and all the appropriate
930 * parts of the process environment (as per the clone
931 * flags). The actual kick-off is left to the caller.
933 static task_t *copy_process(unsigned long clone_flags,
934 unsigned long stack_start,
935 struct pt_regs *regs,
936 unsigned long stack_size,
937 int __user *parent_tidptr,
938 int __user *child_tidptr,
942 struct task_struct *p = NULL;
944 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
945 return ERR_PTR(-EINVAL);
948 * Thread groups must share signals as well, and detached threads
949 * can only be started up within the thread group.
951 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
952 return ERR_PTR(-EINVAL);
955 * Shared signal handlers imply shared VM. By way of the above,
956 * thread groups also imply shared VM. Blocking this case allows
957 * for various simplifications in other code.
959 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
960 return ERR_PTR(-EINVAL);
962 retval = security_task_create(clone_flags);
967 p = dup_task_struct(current);
971 #ifdef CONFIG_TRACE_IRQFLAGS
972 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
973 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
976 if (atomic_read(&p->user->processes) >=
977 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
978 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
979 p->user != &root_user)
983 atomic_inc(&p->user->__count);
984 atomic_inc(&p->user->processes);
985 get_group_info(p->group_info);
988 * If multiple threads are within copy_process(), then this check
989 * triggers too late. This doesn't hurt, the check is only there
990 * to stop root fork bombs.
992 if (nr_threads >= max_threads)
993 goto bad_fork_cleanup_count;
995 if (!try_module_get(task_thread_info(p)->exec_domain->module))
996 goto bad_fork_cleanup_count;
998 if (p->binfmt && !try_module_get(p->binfmt->module))
999 goto bad_fork_cleanup_put_domain;
1002 copy_flags(clone_flags, p);
1005 if (clone_flags & CLONE_PARENT_SETTID)
1006 if (put_user(p->pid, parent_tidptr))
1007 goto bad_fork_cleanup;
1009 INIT_LIST_HEAD(&p->children);
1010 INIT_LIST_HEAD(&p->sibling);
1011 p->vfork_done = NULL;
1012 spin_lock_init(&p->alloc_lock);
1014 clear_tsk_thread_flag(p, TIF_SIGPENDING);
1015 init_sigpending(&p->pending);
1017 p->utime = cputime_zero;
1018 p->stime = cputime_zero;
1020 p->rchar = 0; /* I/O counter: bytes read */
1021 p->wchar = 0; /* I/O counter: bytes written */
1022 p->syscr = 0; /* I/O counter: read syscalls */
1023 p->syscw = 0; /* I/O counter: write syscalls */
1024 acct_clear_integrals(p);
1026 p->it_virt_expires = cputime_zero;
1027 p->it_prof_expires = cputime_zero;
1028 p->it_sched_expires = 0;
1029 INIT_LIST_HEAD(&p->cpu_timers[0]);
1030 INIT_LIST_HEAD(&p->cpu_timers[1]);
1031 INIT_LIST_HEAD(&p->cpu_timers[2]);
1033 p->lock_depth = -1; /* -1 = no lock */
1034 do_posix_clock_monotonic_gettime(&p->start_time);
1036 p->io_context = NULL;
1038 p->audit_context = NULL;
1041 p->mempolicy = mpol_copy(p->mempolicy);
1042 if (IS_ERR(p->mempolicy)) {
1043 retval = PTR_ERR(p->mempolicy);
1044 p->mempolicy = NULL;
1045 goto bad_fork_cleanup_cpuset;
1047 mpol_fix_fork_child_flag(p);
1049 #ifdef CONFIG_TRACE_IRQFLAGS
1051 p->hardirqs_enabled = 0;
1052 p->hardirq_enable_ip = 0;
1053 p->hardirq_enable_event = 0;
1054 p->hardirq_disable_ip = _THIS_IP_;
1055 p->hardirq_disable_event = 0;
1056 p->softirqs_enabled = 1;
1057 p->softirq_enable_ip = _THIS_IP_;
1058 p->softirq_enable_event = 0;
1059 p->softirq_disable_ip = 0;
1060 p->softirq_disable_event = 0;
1061 p->hardirq_context = 0;
1062 p->softirq_context = 0;
1064 #ifdef CONFIG_LOCKDEP
1065 p->lockdep_depth = 0; /* no locks held yet */
1066 p->curr_chain_key = 0;
1067 p->lockdep_recursion = 0;
1070 rt_mutex_init_task(p);
1072 #ifdef CONFIG_DEBUG_MUTEXES
1073 p->blocked_on = NULL; /* not blocked yet */
1077 if (clone_flags & CLONE_THREAD)
1078 p->tgid = current->tgid;
1080 if ((retval = security_task_alloc(p)))
1081 goto bad_fork_cleanup_policy;
1082 if ((retval = audit_alloc(p)))
1083 goto bad_fork_cleanup_security;
1084 /* copy all the process information */
1085 if ((retval = copy_semundo(clone_flags, p)))
1086 goto bad_fork_cleanup_audit;
1087 if ((retval = copy_files(clone_flags, p)))
1088 goto bad_fork_cleanup_semundo;
1089 if ((retval = copy_fs(clone_flags, p)))
1090 goto bad_fork_cleanup_files;
1091 if ((retval = copy_sighand(clone_flags, p)))
1092 goto bad_fork_cleanup_fs;
1093 if ((retval = copy_signal(clone_flags, p)))
1094 goto bad_fork_cleanup_sighand;
1095 if ((retval = copy_mm(clone_flags, p)))
1096 goto bad_fork_cleanup_signal;
1097 if ((retval = copy_keys(clone_flags, p)))
1098 goto bad_fork_cleanup_mm;
1099 if ((retval = copy_namespace(clone_flags, p)))
1100 goto bad_fork_cleanup_keys;
1101 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1103 goto bad_fork_cleanup_namespace;
1105 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1107 * Clear TID on mm_release()?
1109 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1110 p->robust_list = NULL;
1111 #ifdef CONFIG_COMPAT
1112 p->compat_robust_list = NULL;
1114 INIT_LIST_HEAD(&p->pi_state_list);
1115 p->pi_state_cache = NULL;
1118 * sigaltstack should be cleared when sharing the same VM
1120 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1121 p->sas_ss_sp = p->sas_ss_size = 0;
1124 * Syscall tracing should be turned off in the child regardless
1127 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1128 #ifdef TIF_SYSCALL_EMU
1129 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1132 /* Our parent execution domain becomes current domain
1133 These must match for thread signalling to apply */
1135 p->parent_exec_id = p->self_exec_id;
1137 /* ok, now we should be set up.. */
1138 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1139 p->pdeath_signal = 0;
1143 * Ok, make it visible to the rest of the system.
1144 * We dont wake it up yet.
1146 p->group_leader = p;
1147 INIT_LIST_HEAD(&p->thread_group);
1148 INIT_LIST_HEAD(&p->ptrace_children);
1149 INIT_LIST_HEAD(&p->ptrace_list);
1151 /* Perform scheduler related setup. Assign this task to a CPU. */
1152 sched_fork(p, clone_flags);
1154 /* Need tasklist lock for parent etc handling! */
1155 write_lock_irq(&tasklist_lock);
1158 * The task hasn't been attached yet, so its cpus_allowed mask will
1159 * not be changed, nor will its assigned CPU.
1161 * The cpus_allowed mask of the parent may have changed after it was
1162 * copied first time - so re-copy it here, then check the child's CPU
1163 * to ensure it is on a valid CPU (and if not, just force it back to
1164 * parent's CPU). This avoids alot of nasty races.
1166 p->cpus_allowed = current->cpus_allowed;
1167 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1168 !cpu_online(task_cpu(p))))
1169 set_task_cpu(p, smp_processor_id());
1171 /* CLONE_PARENT re-uses the old parent */
1172 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1173 p->real_parent = current->real_parent;
1175 p->real_parent = current;
1176 p->parent = p->real_parent;
1178 spin_lock(¤t->sighand->siglock);
1181 * Process group and session signals need to be delivered to just the
1182 * parent before the fork or both the parent and the child after the
1183 * fork. Restart if a signal comes in before we add the new process to
1184 * it's process group.
1185 * A fatal signal pending means that current will exit, so the new
1186 * thread can't slip out of an OOM kill (or normal SIGKILL).
1188 recalc_sigpending();
1189 if (signal_pending(current)) {
1190 spin_unlock(¤t->sighand->siglock);
1191 write_unlock_irq(&tasklist_lock);
1192 retval = -ERESTARTNOINTR;
1193 goto bad_fork_cleanup_namespace;
1196 if (clone_flags & CLONE_THREAD) {
1197 p->group_leader = current->group_leader;
1198 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1200 if (!cputime_eq(current->signal->it_virt_expires,
1202 !cputime_eq(current->signal->it_prof_expires,
1204 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1205 !list_empty(¤t->signal->cpu_timers[0]) ||
1206 !list_empty(¤t->signal->cpu_timers[1]) ||
1207 !list_empty(¤t->signal->cpu_timers[2])) {
1209 * Have child wake up on its first tick to check
1210 * for process CPU timers.
1212 p->it_prof_expires = jiffies_to_cputime(1);
1219 p->ioprio = current->ioprio;
1221 if (likely(p->pid)) {
1223 if (unlikely(p->ptrace & PT_PTRACED))
1224 __ptrace_link(p, current->parent);
1226 if (thread_group_leader(p)) {
1227 p->signal->tty = current->signal->tty;
1228 p->signal->pgrp = process_group(current);
1229 p->signal->session = current->signal->session;
1230 attach_pid(p, PIDTYPE_PGID, process_group(p));
1231 attach_pid(p, PIDTYPE_SID, p->signal->session);
1233 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1234 __get_cpu_var(process_counts)++;
1236 attach_pid(p, PIDTYPE_PID, p->pid);
1241 spin_unlock(¤t->sighand->siglock);
1242 write_unlock_irq(&tasklist_lock);
1243 proc_fork_connector(p);
1246 bad_fork_cleanup_namespace:
1248 bad_fork_cleanup_keys:
1250 bad_fork_cleanup_mm:
1253 bad_fork_cleanup_signal:
1255 bad_fork_cleanup_sighand:
1256 __cleanup_sighand(p->sighand);
1257 bad_fork_cleanup_fs:
1258 exit_fs(p); /* blocking */
1259 bad_fork_cleanup_files:
1260 exit_files(p); /* blocking */
1261 bad_fork_cleanup_semundo:
1263 bad_fork_cleanup_audit:
1265 bad_fork_cleanup_security:
1266 security_task_free(p);
1267 bad_fork_cleanup_policy:
1269 mpol_free(p->mempolicy);
1270 bad_fork_cleanup_cpuset:
1275 module_put(p->binfmt->module);
1276 bad_fork_cleanup_put_domain:
1277 module_put(task_thread_info(p)->exec_domain->module);
1278 bad_fork_cleanup_count:
1279 put_group_info(p->group_info);
1280 atomic_dec(&p->user->processes);
1285 return ERR_PTR(retval);
1288 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1290 memset(regs, 0, sizeof(struct pt_regs));
1294 task_t * __devinit fork_idle(int cpu)
1297 struct pt_regs regs;
1299 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0);
1301 return ERR_PTR(-ENOMEM);
1302 init_idle(task, cpu);
1307 static inline int fork_traceflag (unsigned clone_flags)
1309 if (clone_flags & CLONE_UNTRACED)
1311 else if (clone_flags & CLONE_VFORK) {
1312 if (current->ptrace & PT_TRACE_VFORK)
1313 return PTRACE_EVENT_VFORK;
1314 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1315 if (current->ptrace & PT_TRACE_CLONE)
1316 return PTRACE_EVENT_CLONE;
1317 } else if (current->ptrace & PT_TRACE_FORK)
1318 return PTRACE_EVENT_FORK;
1324 * Ok, this is the main fork-routine.
1326 * It copies the process, and if successful kick-starts
1327 * it and waits for it to finish using the VM if required.
1329 long do_fork(unsigned long clone_flags,
1330 unsigned long stack_start,
1331 struct pt_regs *regs,
1332 unsigned long stack_size,
1333 int __user *parent_tidptr,
1334 int __user *child_tidptr)
1336 struct task_struct *p;
1338 struct pid *pid = alloc_pid();
1344 if (unlikely(current->ptrace)) {
1345 trace = fork_traceflag (clone_flags);
1347 clone_flags |= CLONE_PTRACE;
1350 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1352 * Do this prior waking up the new thread - the thread pointer
1353 * might get invalid after that point, if the thread exits quickly.
1356 struct completion vfork;
1358 if (clone_flags & CLONE_VFORK) {
1359 p->vfork_done = &vfork;
1360 init_completion(&vfork);
1363 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1365 * We'll start up with an immediate SIGSTOP.
1367 sigaddset(&p->pending.signal, SIGSTOP);
1368 set_tsk_thread_flag(p, TIF_SIGPENDING);
1371 if (!(clone_flags & CLONE_STOPPED))
1372 wake_up_new_task(p, clone_flags);
1374 p->state = TASK_STOPPED;
1376 if (unlikely (trace)) {
1377 current->ptrace_message = nr;
1378 ptrace_notify ((trace << 8) | SIGTRAP);
1381 if (clone_flags & CLONE_VFORK) {
1382 wait_for_completion(&vfork);
1383 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1384 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1393 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1394 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1397 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1399 struct sighand_struct *sighand = data;
1401 if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1402 SLAB_CTOR_CONSTRUCTOR)
1403 spin_lock_init(&sighand->siglock);
1406 void __init proc_caches_init(void)
1408 sighand_cachep = kmem_cache_create("sighand_cache",
1409 sizeof(struct sighand_struct), 0,
1410 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1411 sighand_ctor, NULL);
1412 signal_cachep = kmem_cache_create("signal_cache",
1413 sizeof(struct signal_struct), 0,
1414 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1415 files_cachep = kmem_cache_create("files_cache",
1416 sizeof(struct files_struct), 0,
1417 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1418 fs_cachep = kmem_cache_create("fs_cache",
1419 sizeof(struct fs_struct), 0,
1420 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1421 vm_area_cachep = kmem_cache_create("vm_area_struct",
1422 sizeof(struct vm_area_struct), 0,
1423 SLAB_PANIC, NULL, NULL);
1424 mm_cachep = kmem_cache_create("mm_struct",
1425 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1426 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1431 * Check constraints on flags passed to the unshare system call and
1432 * force unsharing of additional process context as appropriate.
1434 static inline void check_unshare_flags(unsigned long *flags_ptr)
1437 * If unsharing a thread from a thread group, must also
1440 if (*flags_ptr & CLONE_THREAD)
1441 *flags_ptr |= CLONE_VM;
1444 * If unsharing vm, must also unshare signal handlers.
1446 if (*flags_ptr & CLONE_VM)
1447 *flags_ptr |= CLONE_SIGHAND;
1450 * If unsharing signal handlers and the task was created
1451 * using CLONE_THREAD, then must unshare the thread
1453 if ((*flags_ptr & CLONE_SIGHAND) &&
1454 (atomic_read(¤t->signal->count) > 1))
1455 *flags_ptr |= CLONE_THREAD;
1458 * If unsharing namespace, must also unshare filesystem information.
1460 if (*flags_ptr & CLONE_NEWNS)
1461 *flags_ptr |= CLONE_FS;
1465 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1467 static int unshare_thread(unsigned long unshare_flags)
1469 if (unshare_flags & CLONE_THREAD)
1476 * Unshare the filesystem structure if it is being shared
1478 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1480 struct fs_struct *fs = current->fs;
1482 if ((unshare_flags & CLONE_FS) &&
1483 (fs && atomic_read(&fs->count) > 1)) {
1484 *new_fsp = __copy_fs_struct(current->fs);
1493 * Unshare the namespace structure if it is being shared
1495 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1497 struct namespace *ns = current->namespace;
1499 if ((unshare_flags & CLONE_NEWNS) &&
1500 (ns && atomic_read(&ns->count) > 1)) {
1501 if (!capable(CAP_SYS_ADMIN))
1504 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1513 * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1516 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1518 struct sighand_struct *sigh = current->sighand;
1520 if ((unshare_flags & CLONE_SIGHAND) &&
1521 (sigh && atomic_read(&sigh->count) > 1))
1528 * Unshare vm if it is being shared
1530 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1532 struct mm_struct *mm = current->mm;
1534 if ((unshare_flags & CLONE_VM) &&
1535 (mm && atomic_read(&mm->mm_users) > 1)) {
1543 * Unshare file descriptor table if it is being shared
1545 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1547 struct files_struct *fd = current->files;
1550 if ((unshare_flags & CLONE_FILES) &&
1551 (fd && atomic_read(&fd->count) > 1)) {
1552 *new_fdp = dup_fd(fd, &error);
1561 * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1564 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1566 if (unshare_flags & CLONE_SYSVSEM)
1573 * unshare allows a process to 'unshare' part of the process
1574 * context which was originally shared using clone. copy_*
1575 * functions used by do_fork() cannot be used here directly
1576 * because they modify an inactive task_struct that is being
1577 * constructed. Here we are modifying the current, active,
1580 asmlinkage long sys_unshare(unsigned long unshare_flags)
1583 struct fs_struct *fs, *new_fs = NULL;
1584 struct namespace *ns, *new_ns = NULL;
1585 struct sighand_struct *sigh, *new_sigh = NULL;
1586 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1587 struct files_struct *fd, *new_fd = NULL;
1588 struct sem_undo_list *new_ulist = NULL;
1590 check_unshare_flags(&unshare_flags);
1592 /* Return -EINVAL for all unsupported flags */
1594 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1595 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1596 goto bad_unshare_out;
1598 if ((err = unshare_thread(unshare_flags)))
1599 goto bad_unshare_out;
1600 if ((err = unshare_fs(unshare_flags, &new_fs)))
1601 goto bad_unshare_cleanup_thread;
1602 if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1603 goto bad_unshare_cleanup_fs;
1604 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1605 goto bad_unshare_cleanup_ns;
1606 if ((err = unshare_vm(unshare_flags, &new_mm)))
1607 goto bad_unshare_cleanup_sigh;
1608 if ((err = unshare_fd(unshare_flags, &new_fd)))
1609 goto bad_unshare_cleanup_vm;
1610 if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1611 goto bad_unshare_cleanup_fd;
1613 if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1619 current->fs = new_fs;
1624 ns = current->namespace;
1625 current->namespace = new_ns;
1630 sigh = current->sighand;
1631 rcu_assign_pointer(current->sighand, new_sigh);
1637 active_mm = current->active_mm;
1638 current->mm = new_mm;
1639 current->active_mm = new_mm;
1640 activate_mm(active_mm, new_mm);
1645 fd = current->files;
1646 current->files = new_fd;
1650 task_unlock(current);
1653 bad_unshare_cleanup_fd:
1655 put_files_struct(new_fd);
1657 bad_unshare_cleanup_vm:
1661 bad_unshare_cleanup_sigh:
1663 if (atomic_dec_and_test(&new_sigh->count))
1664 kmem_cache_free(sighand_cachep, new_sigh);
1666 bad_unshare_cleanup_ns:
1668 put_namespace(new_ns);
1670 bad_unshare_cleanup_fs:
1672 put_fs_struct(new_fs);
1674 bad_unshare_cleanup_thread: