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/capability.h>
32 #include <linux/cpu.h>
33 #include <linux/cpuset.h>
34 #include <linux/security.h>
35 #include <linux/swap.h>
36 #include <linux/syscalls.h>
37 #include <linux/jiffies.h>
38 #include <linux/futex.h>
39 #include <linux/rcupdate.h>
40 #include <linux/ptrace.h>
41 #include <linux/mount.h>
42 #include <linux/audit.h>
43 #include <linux/profile.h>
44 #include <linux/rmap.h>
45 #include <linux/acct.h>
46 #include <linux/cn_proc.h>
48 #include <asm/pgtable.h>
49 #include <asm/pgalloc.h>
50 #include <asm/uaccess.h>
51 #include <asm/mmu_context.h>
52 #include <asm/cacheflush.h>
53 #include <asm/tlbflush.h>
56 * Protected counters by write_lock_irq(&tasklist_lock)
58 unsigned long total_forks; /* Handle normal Linux uptimes. */
59 int nr_threads; /* The idle threads do not count.. */
61 int max_threads; /* tunable limit on nr_threads */
63 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
65 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
67 EXPORT_SYMBOL(tasklist_lock);
69 int nr_processes(void)
74 for_each_online_cpu(cpu)
75 total += per_cpu(process_counts, cpu);
80 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
81 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
82 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
83 static kmem_cache_t *task_struct_cachep;
86 /* SLAB cache for signal_struct structures (tsk->signal) */
87 static kmem_cache_t *signal_cachep;
89 /* SLAB cache for sighand_struct structures (tsk->sighand) */
90 kmem_cache_t *sighand_cachep;
92 /* SLAB cache for files_struct structures (tsk->files) */
93 kmem_cache_t *files_cachep;
95 /* SLAB cache for fs_struct structures (tsk->fs) */
96 kmem_cache_t *fs_cachep;
98 /* SLAB cache for vm_area_struct structures */
99 kmem_cache_t *vm_area_cachep;
101 /* SLAB cache for mm_struct structures (tsk->mm) */
102 static kmem_cache_t *mm_cachep;
104 void free_task(struct task_struct *tsk)
106 free_thread_info(tsk->thread_info);
107 free_task_struct(tsk);
109 EXPORT_SYMBOL(free_task);
111 void __put_task_struct_cb(struct rcu_head *rhp)
113 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
115 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
116 WARN_ON(atomic_read(&tsk->usage));
117 WARN_ON(tsk == current);
119 if (unlikely(tsk->audit_context))
121 security_task_free(tsk);
123 put_group_info(tsk->group_info);
125 if (!profile_handoff_task(tsk))
129 void __init fork_init(unsigned long mempages)
131 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
132 #ifndef ARCH_MIN_TASKALIGN
133 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
135 /* create a slab on which task_structs can be allocated */
137 kmem_cache_create("task_struct", sizeof(struct task_struct),
138 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
142 * The default maximum number of threads is set to a safe
143 * value: the thread structures can take up at most half
146 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
149 * we need to allow at least 20 threads to boot a system
154 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
155 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
156 init_task.signal->rlim[RLIMIT_SIGPENDING] =
157 init_task.signal->rlim[RLIMIT_NPROC];
160 static struct task_struct *dup_task_struct(struct task_struct *orig)
162 struct task_struct *tsk;
163 struct thread_info *ti;
165 prepare_to_copy(orig);
167 tsk = alloc_task_struct();
171 ti = alloc_thread_info(tsk);
173 free_task_struct(tsk);
178 tsk->thread_info = ti;
179 setup_thread_stack(tsk, orig);
181 /* One for us, one for whoever does the "release_task()" (usually parent) */
182 atomic_set(&tsk->usage,2);
183 atomic_set(&tsk->fs_excl, 0);
189 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
191 struct vm_area_struct *mpnt, *tmp, **pprev;
192 struct rb_node **rb_link, *rb_parent;
194 unsigned long charge;
195 struct mempolicy *pol;
197 down_write(&oldmm->mmap_sem);
198 flush_cache_mm(oldmm);
199 down_write(&mm->mmap_sem);
203 mm->mmap_cache = NULL;
204 mm->free_area_cache = oldmm->mmap_base;
205 mm->cached_hole_size = ~0UL;
207 cpus_clear(mm->cpu_vm_mask);
209 rb_link = &mm->mm_rb.rb_node;
213 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
216 if (mpnt->vm_flags & VM_DONTCOPY) {
217 long pages = vma_pages(mpnt);
218 mm->total_vm -= pages;
219 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
224 if (mpnt->vm_flags & VM_ACCOUNT) {
225 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
226 if (security_vm_enough_memory(len))
230 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
234 pol = mpol_copy(vma_policy(mpnt));
235 retval = PTR_ERR(pol);
237 goto fail_nomem_policy;
238 vma_set_policy(tmp, pol);
239 tmp->vm_flags &= ~VM_LOCKED;
245 struct inode *inode = file->f_dentry->d_inode;
247 if (tmp->vm_flags & VM_DENYWRITE)
248 atomic_dec(&inode->i_writecount);
250 /* insert tmp into the share list, just after mpnt */
251 spin_lock(&file->f_mapping->i_mmap_lock);
252 tmp->vm_truncate_count = mpnt->vm_truncate_count;
253 flush_dcache_mmap_lock(file->f_mapping);
254 vma_prio_tree_add(tmp, mpnt);
255 flush_dcache_mmap_unlock(file->f_mapping);
256 spin_unlock(&file->f_mapping->i_mmap_lock);
260 * Link in the new vma and copy the page table entries.
263 pprev = &tmp->vm_next;
265 __vma_link_rb(mm, tmp, rb_link, rb_parent);
266 rb_link = &tmp->vm_rb.rb_right;
267 rb_parent = &tmp->vm_rb;
270 retval = copy_page_range(mm, oldmm, mpnt);
272 if (tmp->vm_ops && tmp->vm_ops->open)
273 tmp->vm_ops->open(tmp);
280 up_write(&mm->mmap_sem);
282 up_write(&oldmm->mmap_sem);
285 kmem_cache_free(vm_area_cachep, tmp);
288 vm_unacct_memory(charge);
292 static inline int mm_alloc_pgd(struct mm_struct * mm)
294 mm->pgd = pgd_alloc(mm);
295 if (unlikely(!mm->pgd))
300 static inline void mm_free_pgd(struct mm_struct * mm)
305 #define dup_mmap(mm, oldmm) (0)
306 #define mm_alloc_pgd(mm) (0)
307 #define mm_free_pgd(mm)
308 #endif /* CONFIG_MMU */
310 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
312 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
313 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
315 #include <linux/init_task.h>
317 static struct mm_struct * mm_init(struct mm_struct * mm)
319 atomic_set(&mm->mm_users, 1);
320 atomic_set(&mm->mm_count, 1);
321 init_rwsem(&mm->mmap_sem);
322 INIT_LIST_HEAD(&mm->mmlist);
323 mm->core_waiters = 0;
325 set_mm_counter(mm, file_rss, 0);
326 set_mm_counter(mm, anon_rss, 0);
327 spin_lock_init(&mm->page_table_lock);
328 rwlock_init(&mm->ioctx_list_lock);
329 mm->ioctx_list = NULL;
330 mm->free_area_cache = TASK_UNMAPPED_BASE;
331 mm->cached_hole_size = ~0UL;
333 if (likely(!mm_alloc_pgd(mm))) {
342 * Allocate and initialize an mm_struct.
344 struct mm_struct * mm_alloc(void)
346 struct mm_struct * mm;
350 memset(mm, 0, sizeof(*mm));
357 * Called when the last reference to the mm
358 * is dropped: either by a lazy thread or by
359 * mmput. Free the page directory and the mm.
361 void fastcall __mmdrop(struct mm_struct *mm)
363 BUG_ON(mm == &init_mm);
370 * Decrement the use count and release all resources for an mm.
372 void mmput(struct mm_struct *mm)
374 if (atomic_dec_and_test(&mm->mm_users)) {
377 if (!list_empty(&mm->mmlist)) {
378 spin_lock(&mmlist_lock);
379 list_del(&mm->mmlist);
380 spin_unlock(&mmlist_lock);
386 EXPORT_SYMBOL_GPL(mmput);
389 * get_task_mm - acquire a reference to the task's mm
391 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
392 * this kernel workthread has transiently adopted a user mm with use_mm,
393 * to do its AIO) is not set and if so returns a reference to it, after
394 * bumping up the use count. User must release the mm via mmput()
395 * after use. Typically used by /proc and ptrace.
397 struct mm_struct *get_task_mm(struct task_struct *task)
399 struct mm_struct *mm;
404 if (task->flags & PF_BORROWED_MM)
407 atomic_inc(&mm->mm_users);
412 EXPORT_SYMBOL_GPL(get_task_mm);
414 /* Please note the differences between mmput and mm_release.
415 * mmput is called whenever we stop holding onto a mm_struct,
416 * error success whatever.
418 * mm_release is called after a mm_struct has been removed
419 * from the current process.
421 * This difference is important for error handling, when we
422 * only half set up a mm_struct for a new process and need to restore
423 * the old one. Because we mmput the new mm_struct before
424 * restoring the old one. . .
425 * Eric Biederman 10 January 1998
427 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
429 struct completion *vfork_done = tsk->vfork_done;
431 /* Get rid of any cached register state */
432 deactivate_mm(tsk, mm);
434 /* notify parent sleeping on vfork() */
436 tsk->vfork_done = NULL;
437 complete(vfork_done);
439 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
440 u32 __user * tidptr = tsk->clear_child_tid;
441 tsk->clear_child_tid = NULL;
444 * We don't check the error code - if userspace has
445 * not set up a proper pointer then tough luck.
448 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
453 * Allocate a new mm structure and copy contents from the
454 * mm structure of the passed in task structure.
456 static struct mm_struct *dup_mm(struct task_struct *tsk)
458 struct mm_struct *mm, *oldmm = current->mm;
468 memcpy(mm, oldmm, sizeof(*mm));
473 if (init_new_context(tsk, mm))
476 err = dup_mmap(mm, oldmm);
480 mm->hiwater_rss = get_mm_rss(mm);
481 mm->hiwater_vm = mm->total_vm;
493 * If init_new_context() failed, we cannot use mmput() to free the mm
494 * because it calls destroy_context()
501 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
503 struct mm_struct * mm, *oldmm;
506 tsk->min_flt = tsk->maj_flt = 0;
507 tsk->nvcsw = tsk->nivcsw = 0;
510 tsk->active_mm = NULL;
513 * Are we cloning a kernel thread?
515 * We need to steal a active VM for that..
521 if (clone_flags & CLONE_VM) {
522 atomic_inc(&oldmm->mm_users);
541 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
543 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
544 /* We don't need to lock fs - think why ;-) */
546 atomic_set(&fs->count, 1);
547 rwlock_init(&fs->lock);
548 fs->umask = old->umask;
549 read_lock(&old->lock);
550 fs->rootmnt = mntget(old->rootmnt);
551 fs->root = dget(old->root);
552 fs->pwdmnt = mntget(old->pwdmnt);
553 fs->pwd = dget(old->pwd);
555 fs->altrootmnt = mntget(old->altrootmnt);
556 fs->altroot = dget(old->altroot);
558 fs->altrootmnt = NULL;
561 read_unlock(&old->lock);
566 struct fs_struct *copy_fs_struct(struct fs_struct *old)
568 return __copy_fs_struct(old);
571 EXPORT_SYMBOL_GPL(copy_fs_struct);
573 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
575 if (clone_flags & CLONE_FS) {
576 atomic_inc(¤t->fs->count);
579 tsk->fs = __copy_fs_struct(current->fs);
585 static int count_open_files(struct fdtable *fdt)
587 int size = fdt->max_fdset;
590 /* Find the last open fd */
591 for (i = size/(8*sizeof(long)); i > 0; ) {
592 if (fdt->open_fds->fds_bits[--i])
595 i = (i+1) * 8 * sizeof(long);
599 static struct files_struct *alloc_files(void)
601 struct files_struct *newf;
604 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
608 atomic_set(&newf->count, 1);
610 spin_lock_init(&newf->file_lock);
613 fdt->max_fds = NR_OPEN_DEFAULT;
614 fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
615 fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
616 fdt->open_fds = (fd_set *)&newf->open_fds_init;
617 fdt->fd = &newf->fd_array[0];
618 INIT_RCU_HEAD(&fdt->rcu);
619 fdt->free_files = NULL;
621 rcu_assign_pointer(newf->fdt, fdt);
627 * Allocate a new files structure and copy contents from the
628 * passed in files structure.
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;
637 newf = alloc_files();
641 spin_lock(&oldf->file_lock);
642 old_fdt = files_fdtable(oldf);
643 new_fdt = files_fdtable(newf);
644 size = old_fdt->max_fdset;
645 open_files = count_open_files(old_fdt);
649 * Check whether we need to allocate a larger fd array or fd set.
650 * Note: we're not a clone task, so the open count won't change.
652 if (open_files > new_fdt->max_fdset) {
653 new_fdt->max_fdset = 0;
656 if (open_files > new_fdt->max_fds) {
657 new_fdt->max_fds = 0;
661 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
663 spin_unlock(&oldf->file_lock);
664 spin_lock(&newf->file_lock);
665 *errorp = expand_files(newf, open_files-1);
666 spin_unlock(&newf->file_lock);
669 new_fdt = files_fdtable(newf);
671 * Reacquire the oldf lock and a pointer to its fd table
672 * who knows it may have a new bigger fd table. We need
673 * the latest pointer.
675 spin_lock(&oldf->file_lock);
676 old_fdt = files_fdtable(oldf);
679 old_fds = old_fdt->fd;
680 new_fds = new_fdt->fd;
682 memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
683 memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
685 for (i = open_files; i != 0; i--) {
686 struct file *f = *old_fds++;
691 * The fd may be claimed in the fd bitmap but not yet
692 * instantiated in the files array if a sibling thread
693 * is partway through open(). So make sure that this
694 * fd is available to the new process.
696 FD_CLR(open_files - i, new_fdt->open_fds);
698 rcu_assign_pointer(*new_fds++, f);
700 spin_unlock(&oldf->file_lock);
702 /* compute the remainder to be cleared */
703 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
705 /* This is long word aligned thus could use a optimized version */
706 memset(new_fds, 0, size);
708 if (new_fdt->max_fdset > open_files) {
709 int left = (new_fdt->max_fdset-open_files)/8;
710 int start = open_files / (8 * sizeof(unsigned long));
712 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
713 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
720 free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
721 free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
722 free_fd_array(new_fdt->fd, new_fdt->max_fds);
723 kmem_cache_free(files_cachep, newf);
727 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
729 struct files_struct *oldf, *newf;
733 * A background process may not have any files ...
735 oldf = current->files;
739 if (clone_flags & CLONE_FILES) {
740 atomic_inc(&oldf->count);
745 * Note: we may be using current for both targets (See exec.c)
746 * 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 task_struct *tsk)
808 struct sighand_struct * sighand = tsk->sighand;
810 /* Ok, we're done with the signal handlers */
812 if (atomic_dec_and_test(&sighand->count))
813 kmem_cache_free(sighand_cachep, sighand);
816 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
818 struct signal_struct *sig;
821 if (clone_flags & CLONE_THREAD) {
822 atomic_inc(¤t->signal->count);
823 atomic_inc(¤t->signal->live);
826 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
831 ret = copy_thread_group_keys(tsk);
833 kmem_cache_free(signal_cachep, sig);
837 atomic_set(&sig->count, 1);
838 atomic_set(&sig->live, 1);
839 init_waitqueue_head(&sig->wait_chldexit);
841 sig->group_exit_code = 0;
842 sig->group_exit_task = NULL;
843 sig->group_stop_count = 0;
844 sig->curr_target = NULL;
845 init_sigpending(&sig->shared_pending);
846 INIT_LIST_HEAD(&sig->posix_timers);
848 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
849 sig->it_real_incr.tv64 = 0;
850 sig->real_timer.function = it_real_fn;
853 sig->it_virt_expires = cputime_zero;
854 sig->it_virt_incr = cputime_zero;
855 sig->it_prof_expires = cputime_zero;
856 sig->it_prof_incr = cputime_zero;
858 sig->leader = 0; /* session leadership doesn't inherit */
859 sig->tty_old_pgrp = 0;
861 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
862 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
863 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
865 INIT_LIST_HEAD(&sig->cpu_timers[0]);
866 INIT_LIST_HEAD(&sig->cpu_timers[1]);
867 INIT_LIST_HEAD(&sig->cpu_timers[2]);
869 task_lock(current->group_leader);
870 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
871 task_unlock(current->group_leader);
873 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
875 * New sole thread in the process gets an expiry time
876 * of the whole CPU time limit.
878 tsk->it_prof_expires =
879 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
885 void __cleanup_signal(struct signal_struct *sig)
887 exit_thread_group_keys(sig);
888 kmem_cache_free(signal_cachep, sig);
891 static inline void cleanup_signal(struct task_struct *tsk)
893 struct signal_struct *sig = tsk->signal;
895 atomic_dec(&sig->live);
897 if (atomic_dec_and_test(&sig->count))
898 __cleanup_signal(sig);
901 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
903 unsigned long new_flags = p->flags;
905 new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
906 new_flags |= PF_FORKNOEXEC;
907 if (!(clone_flags & CLONE_PTRACE))
909 p->flags = new_flags;
912 asmlinkage long sys_set_tid_address(int __user *tidptr)
914 current->clear_child_tid = tidptr;
920 * This creates a new process as a copy of the old one,
921 * but does not actually start it yet.
923 * It copies the registers, and all the appropriate
924 * parts of the process environment (as per the clone
925 * flags). The actual kick-off is left to the caller.
927 static task_t *copy_process(unsigned long clone_flags,
928 unsigned long stack_start,
929 struct pt_regs *regs,
930 unsigned long stack_size,
931 int __user *parent_tidptr,
932 int __user *child_tidptr,
936 struct task_struct *p = NULL;
938 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
939 return ERR_PTR(-EINVAL);
942 * Thread groups must share signals as well, and detached threads
943 * can only be started up within the thread group.
945 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
946 return ERR_PTR(-EINVAL);
949 * Shared signal handlers imply shared VM. By way of the above,
950 * thread groups also imply shared VM. Blocking this case allows
951 * for various simplifications in other code.
953 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
954 return ERR_PTR(-EINVAL);
956 retval = security_task_create(clone_flags);
961 p = dup_task_struct(current);
966 if (atomic_read(&p->user->processes) >=
967 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
968 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
969 p->user != &root_user)
973 atomic_inc(&p->user->__count);
974 atomic_inc(&p->user->processes);
975 get_group_info(p->group_info);
978 * If multiple threads are within copy_process(), then this check
979 * triggers too late. This doesn't hurt, the check is only there
980 * to stop root fork bombs.
982 if (nr_threads >= max_threads)
983 goto bad_fork_cleanup_count;
985 if (!try_module_get(task_thread_info(p)->exec_domain->module))
986 goto bad_fork_cleanup_count;
988 if (p->binfmt && !try_module_get(p->binfmt->module))
989 goto bad_fork_cleanup_put_domain;
992 copy_flags(clone_flags, p);
995 if (clone_flags & CLONE_PARENT_SETTID)
996 if (put_user(p->pid, parent_tidptr))
997 goto bad_fork_cleanup;
999 p->proc_dentry = NULL;
1001 INIT_LIST_HEAD(&p->children);
1002 INIT_LIST_HEAD(&p->sibling);
1003 p->vfork_done = NULL;
1004 spin_lock_init(&p->alloc_lock);
1005 spin_lock_init(&p->proc_lock);
1007 clear_tsk_thread_flag(p, TIF_SIGPENDING);
1008 init_sigpending(&p->pending);
1010 p->utime = cputime_zero;
1011 p->stime = cputime_zero;
1013 p->rchar = 0; /* I/O counter: bytes read */
1014 p->wchar = 0; /* I/O counter: bytes written */
1015 p->syscr = 0; /* I/O counter: read syscalls */
1016 p->syscw = 0; /* I/O counter: write syscalls */
1017 acct_clear_integrals(p);
1019 p->it_virt_expires = cputime_zero;
1020 p->it_prof_expires = cputime_zero;
1021 p->it_sched_expires = 0;
1022 INIT_LIST_HEAD(&p->cpu_timers[0]);
1023 INIT_LIST_HEAD(&p->cpu_timers[1]);
1024 INIT_LIST_HEAD(&p->cpu_timers[2]);
1026 p->lock_depth = -1; /* -1 = no lock */
1027 do_posix_clock_monotonic_gettime(&p->start_time);
1029 p->io_context = NULL;
1031 p->audit_context = NULL;
1034 p->mempolicy = mpol_copy(p->mempolicy);
1035 if (IS_ERR(p->mempolicy)) {
1036 retval = PTR_ERR(p->mempolicy);
1037 p->mempolicy = NULL;
1038 goto bad_fork_cleanup_cpuset;
1040 mpol_fix_fork_child_flag(p);
1043 #ifdef CONFIG_DEBUG_MUTEXES
1044 p->blocked_on = NULL; /* not blocked yet */
1048 if (clone_flags & CLONE_THREAD)
1049 p->tgid = current->tgid;
1051 if ((retval = security_task_alloc(p)))
1052 goto bad_fork_cleanup_policy;
1053 if ((retval = audit_alloc(p)))
1054 goto bad_fork_cleanup_security;
1055 /* copy all the process information */
1056 if ((retval = copy_semundo(clone_flags, p)))
1057 goto bad_fork_cleanup_audit;
1058 if ((retval = copy_files(clone_flags, p)))
1059 goto bad_fork_cleanup_semundo;
1060 if ((retval = copy_fs(clone_flags, p)))
1061 goto bad_fork_cleanup_files;
1062 if ((retval = copy_sighand(clone_flags, p)))
1063 goto bad_fork_cleanup_fs;
1064 if ((retval = copy_signal(clone_flags, p)))
1065 goto bad_fork_cleanup_sighand;
1066 if ((retval = copy_mm(clone_flags, p)))
1067 goto bad_fork_cleanup_signal;
1068 if ((retval = copy_keys(clone_flags, p)))
1069 goto bad_fork_cleanup_mm;
1070 if ((retval = copy_namespace(clone_flags, p)))
1071 goto bad_fork_cleanup_keys;
1072 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1074 goto bad_fork_cleanup_namespace;
1076 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1078 * Clear TID on mm_release()?
1080 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1081 p->robust_list = NULL;
1082 #ifdef CONFIG_COMPAT
1083 p->compat_robust_list = NULL;
1086 * sigaltstack should be cleared when sharing the same VM
1088 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1089 p->sas_ss_sp = p->sas_ss_size = 0;
1092 * Syscall tracing should be turned off in the child regardless
1095 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1096 #ifdef TIF_SYSCALL_EMU
1097 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1100 /* Our parent execution domain becomes current domain
1101 These must match for thread signalling to apply */
1103 p->parent_exec_id = p->self_exec_id;
1105 /* ok, now we should be set up.. */
1106 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1107 p->pdeath_signal = 0;
1111 * Ok, make it visible to the rest of the system.
1112 * We dont wake it up yet.
1114 p->group_leader = p;
1115 INIT_LIST_HEAD(&p->ptrace_children);
1116 INIT_LIST_HEAD(&p->ptrace_list);
1118 /* Perform scheduler related setup. Assign this task to a CPU. */
1119 sched_fork(p, clone_flags);
1121 /* Need tasklist lock for parent etc handling! */
1122 write_lock_irq(&tasklist_lock);
1125 * The task hasn't been attached yet, so its cpus_allowed mask will
1126 * not be changed, nor will its assigned CPU.
1128 * The cpus_allowed mask of the parent may have changed after it was
1129 * copied first time - so re-copy it here, then check the child's CPU
1130 * to ensure it is on a valid CPU (and if not, just force it back to
1131 * parent's CPU). This avoids alot of nasty races.
1133 p->cpus_allowed = current->cpus_allowed;
1134 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1135 !cpu_online(task_cpu(p))))
1136 set_task_cpu(p, smp_processor_id());
1139 * Check for pending SIGKILL! The new thread should not be allowed
1140 * to slip out of an OOM kill. (or normal SIGKILL.)
1142 if (sigismember(¤t->pending.signal, SIGKILL)) {
1143 write_unlock_irq(&tasklist_lock);
1145 goto bad_fork_cleanup_namespace;
1148 /* CLONE_PARENT re-uses the old parent */
1149 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1150 p->real_parent = current->real_parent;
1152 p->real_parent = current;
1153 p->parent = p->real_parent;
1155 spin_lock(¤t->sighand->siglock);
1156 if (clone_flags & CLONE_THREAD) {
1158 * Important: if an exit-all has been started then
1159 * do not create this new thread - the whole thread
1160 * group is supposed to exit anyway.
1162 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1163 spin_unlock(¤t->sighand->siglock);
1164 write_unlock_irq(&tasklist_lock);
1166 goto bad_fork_cleanup_namespace;
1168 p->group_leader = current->group_leader;
1170 if (current->signal->group_stop_count > 0) {
1172 * There is an all-stop in progress for the group.
1173 * We ourselves will stop as soon as we check signals.
1174 * Make the new thread part of that group stop too.
1176 current->signal->group_stop_count++;
1177 set_tsk_thread_flag(p, TIF_SIGPENDING);
1180 if (!cputime_eq(current->signal->it_virt_expires,
1182 !cputime_eq(current->signal->it_prof_expires,
1184 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1185 !list_empty(¤t->signal->cpu_timers[0]) ||
1186 !list_empty(¤t->signal->cpu_timers[1]) ||
1187 !list_empty(¤t->signal->cpu_timers[2])) {
1189 * Have child wake up on its first tick to check
1190 * for process CPU timers.
1192 p->it_prof_expires = jiffies_to_cputime(1);
1199 p->ioprio = current->ioprio;
1201 if (likely(p->pid)) {
1203 if (unlikely(p->ptrace & PT_PTRACED))
1204 __ptrace_link(p, current->parent);
1206 if (thread_group_leader(p)) {
1207 p->signal->tty = current->signal->tty;
1208 p->signal->pgrp = process_group(current);
1209 p->signal->session = current->signal->session;
1210 attach_pid(p, PIDTYPE_PGID, process_group(p));
1211 attach_pid(p, PIDTYPE_SID, p->signal->session);
1213 list_add_tail(&p->tasks, &init_task.tasks);
1214 __get_cpu_var(process_counts)++;
1216 attach_pid(p, PIDTYPE_TGID, p->tgid);
1217 attach_pid(p, PIDTYPE_PID, p->pid);
1222 spin_unlock(¤t->sighand->siglock);
1223 write_unlock_irq(&tasklist_lock);
1224 proc_fork_connector(p);
1227 bad_fork_cleanup_namespace:
1229 bad_fork_cleanup_keys:
1231 bad_fork_cleanup_mm:
1234 bad_fork_cleanup_signal:
1236 bad_fork_cleanup_sighand:
1238 bad_fork_cleanup_fs:
1239 exit_fs(p); /* blocking */
1240 bad_fork_cleanup_files:
1241 exit_files(p); /* blocking */
1242 bad_fork_cleanup_semundo:
1244 bad_fork_cleanup_audit:
1246 bad_fork_cleanup_security:
1247 security_task_free(p);
1248 bad_fork_cleanup_policy:
1250 mpol_free(p->mempolicy);
1251 bad_fork_cleanup_cpuset:
1256 module_put(p->binfmt->module);
1257 bad_fork_cleanup_put_domain:
1258 module_put(task_thread_info(p)->exec_domain->module);
1259 bad_fork_cleanup_count:
1260 put_group_info(p->group_info);
1261 atomic_dec(&p->user->processes);
1266 return ERR_PTR(retval);
1269 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1271 memset(regs, 0, sizeof(struct pt_regs));
1275 task_t * __devinit fork_idle(int cpu)
1278 struct pt_regs regs;
1280 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0);
1282 return ERR_PTR(-ENOMEM);
1283 init_idle(task, cpu);
1288 static inline int fork_traceflag (unsigned clone_flags)
1290 if (clone_flags & CLONE_UNTRACED)
1292 else if (clone_flags & CLONE_VFORK) {
1293 if (current->ptrace & PT_TRACE_VFORK)
1294 return PTRACE_EVENT_VFORK;
1295 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1296 if (current->ptrace & PT_TRACE_CLONE)
1297 return PTRACE_EVENT_CLONE;
1298 } else if (current->ptrace & PT_TRACE_FORK)
1299 return PTRACE_EVENT_FORK;
1305 * Ok, this is the main fork-routine.
1307 * It copies the process, and if successful kick-starts
1308 * it and waits for it to finish using the VM if required.
1310 long do_fork(unsigned long clone_flags,
1311 unsigned long stack_start,
1312 struct pt_regs *regs,
1313 unsigned long stack_size,
1314 int __user *parent_tidptr,
1315 int __user *child_tidptr)
1317 struct task_struct *p;
1319 long pid = alloc_pidmap();
1323 if (unlikely(current->ptrace)) {
1324 trace = fork_traceflag (clone_flags);
1326 clone_flags |= CLONE_PTRACE;
1329 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1331 * Do this prior waking up the new thread - the thread pointer
1332 * might get invalid after that point, if the thread exits quickly.
1335 struct completion vfork;
1337 if (clone_flags & CLONE_VFORK) {
1338 p->vfork_done = &vfork;
1339 init_completion(&vfork);
1342 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1344 * We'll start up with an immediate SIGSTOP.
1346 sigaddset(&p->pending.signal, SIGSTOP);
1347 set_tsk_thread_flag(p, TIF_SIGPENDING);
1350 if (!(clone_flags & CLONE_STOPPED))
1351 wake_up_new_task(p, clone_flags);
1353 p->state = TASK_STOPPED;
1355 if (unlikely (trace)) {
1356 current->ptrace_message = pid;
1357 ptrace_notify ((trace << 8) | SIGTRAP);
1360 if (clone_flags & CLONE_VFORK) {
1361 wait_for_completion(&vfork);
1362 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1363 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1372 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1373 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1376 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1378 struct sighand_struct *sighand = data;
1380 if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1381 SLAB_CTOR_CONSTRUCTOR)
1382 spin_lock_init(&sighand->siglock);
1385 void __init proc_caches_init(void)
1387 sighand_cachep = kmem_cache_create("sighand_cache",
1388 sizeof(struct sighand_struct), 0,
1389 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1390 sighand_ctor, NULL);
1391 signal_cachep = kmem_cache_create("signal_cache",
1392 sizeof(struct signal_struct), 0,
1393 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1394 files_cachep = kmem_cache_create("files_cache",
1395 sizeof(struct files_struct), 0,
1396 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1397 fs_cachep = kmem_cache_create("fs_cache",
1398 sizeof(struct fs_struct), 0,
1399 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1400 vm_area_cachep = kmem_cache_create("vm_area_struct",
1401 sizeof(struct vm_area_struct), 0,
1402 SLAB_PANIC, NULL, NULL);
1403 mm_cachep = kmem_cache_create("mm_struct",
1404 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1405 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1410 * Check constraints on flags passed to the unshare system call and
1411 * force unsharing of additional process context as appropriate.
1413 static inline void check_unshare_flags(unsigned long *flags_ptr)
1416 * If unsharing a thread from a thread group, must also
1419 if (*flags_ptr & CLONE_THREAD)
1420 *flags_ptr |= CLONE_VM;
1423 * If unsharing vm, must also unshare signal handlers.
1425 if (*flags_ptr & CLONE_VM)
1426 *flags_ptr |= CLONE_SIGHAND;
1429 * If unsharing signal handlers and the task was created
1430 * using CLONE_THREAD, then must unshare the thread
1432 if ((*flags_ptr & CLONE_SIGHAND) &&
1433 (atomic_read(¤t->signal->count) > 1))
1434 *flags_ptr |= CLONE_THREAD;
1437 * If unsharing namespace, must also unshare filesystem information.
1439 if (*flags_ptr & CLONE_NEWNS)
1440 *flags_ptr |= CLONE_FS;
1444 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1446 static int unshare_thread(unsigned long unshare_flags)
1448 if (unshare_flags & CLONE_THREAD)
1455 * Unshare the filesystem structure if it is being shared
1457 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1459 struct fs_struct *fs = current->fs;
1461 if ((unshare_flags & CLONE_FS) &&
1462 (fs && atomic_read(&fs->count) > 1)) {
1463 *new_fsp = __copy_fs_struct(current->fs);
1472 * Unshare the namespace structure if it is being shared
1474 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1476 struct namespace *ns = current->namespace;
1478 if ((unshare_flags & CLONE_NEWNS) &&
1479 (ns && atomic_read(&ns->count) > 1)) {
1480 if (!capable(CAP_SYS_ADMIN))
1483 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1492 * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1495 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1497 struct sighand_struct *sigh = current->sighand;
1499 if ((unshare_flags & CLONE_SIGHAND) &&
1500 (sigh && atomic_read(&sigh->count) > 1))
1507 * Unshare vm if it is being shared
1509 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1511 struct mm_struct *mm = current->mm;
1513 if ((unshare_flags & CLONE_VM) &&
1514 (mm && atomic_read(&mm->mm_users) > 1)) {
1522 * Unshare file descriptor table if it is being shared
1524 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1526 struct files_struct *fd = current->files;
1529 if ((unshare_flags & CLONE_FILES) &&
1530 (fd && atomic_read(&fd->count) > 1)) {
1531 *new_fdp = dup_fd(fd, &error);
1540 * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1543 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1545 if (unshare_flags & CLONE_SYSVSEM)
1552 * unshare allows a process to 'unshare' part of the process
1553 * context which was originally shared using clone. copy_*
1554 * functions used by do_fork() cannot be used here directly
1555 * because they modify an inactive task_struct that is being
1556 * constructed. Here we are modifying the current, active,
1559 asmlinkage long sys_unshare(unsigned long unshare_flags)
1562 struct fs_struct *fs, *new_fs = NULL;
1563 struct namespace *ns, *new_ns = NULL;
1564 struct sighand_struct *sigh, *new_sigh = NULL;
1565 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1566 struct files_struct *fd, *new_fd = NULL;
1567 struct sem_undo_list *new_ulist = NULL;
1569 check_unshare_flags(&unshare_flags);
1571 /* Return -EINVAL for all unsupported flags */
1573 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1574 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1575 goto bad_unshare_out;
1577 if ((err = unshare_thread(unshare_flags)))
1578 goto bad_unshare_out;
1579 if ((err = unshare_fs(unshare_flags, &new_fs)))
1580 goto bad_unshare_cleanup_thread;
1581 if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1582 goto bad_unshare_cleanup_fs;
1583 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1584 goto bad_unshare_cleanup_ns;
1585 if ((err = unshare_vm(unshare_flags, &new_mm)))
1586 goto bad_unshare_cleanup_sigh;
1587 if ((err = unshare_fd(unshare_flags, &new_fd)))
1588 goto bad_unshare_cleanup_vm;
1589 if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1590 goto bad_unshare_cleanup_fd;
1592 if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1598 current->fs = new_fs;
1603 ns = current->namespace;
1604 current->namespace = new_ns;
1609 sigh = current->sighand;
1610 rcu_assign_pointer(current->sighand, new_sigh);
1616 active_mm = current->active_mm;
1617 current->mm = new_mm;
1618 current->active_mm = new_mm;
1619 activate_mm(active_mm, new_mm);
1624 fd = current->files;
1625 current->files = new_fd;
1629 task_unlock(current);
1632 bad_unshare_cleanup_fd:
1634 put_files_struct(new_fd);
1636 bad_unshare_cleanup_vm:
1640 bad_unshare_cleanup_sigh:
1642 if (atomic_dec_and_test(&new_sigh->count))
1643 kmem_cache_free(sighand_cachep, new_sigh);
1645 bad_unshare_cleanup_ns:
1647 put_namespace(new_ns);
1649 bad_unshare_cleanup_fs:
1651 put_fs_struct(new_fs);
1653 bad_unshare_cleanup_thread: