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>
47 #include <linux/vs_context.h>
48 #include <linux/vs_network.h>
49 #include <linux/vs_limit.h>
50 #include <linux/vs_memory.h>
52 #include <asm/pgtable.h>
53 #include <asm/pgalloc.h>
54 #include <asm/uaccess.h>
55 #include <asm/mmu_context.h>
56 #include <asm/cacheflush.h>
57 #include <asm/tlbflush.h>
60 * Protected counters by write_lock_irq(&tasklist_lock)
62 unsigned long total_forks; /* Handle normal Linux uptimes. */
63 int nr_threads; /* The idle threads do not count.. */
65 int max_threads; /* tunable limit on nr_threads */
67 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
69 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
71 EXPORT_SYMBOL(tasklist_lock);
73 int nr_processes(void)
78 for_each_online_cpu(cpu)
79 total += per_cpu(process_counts, cpu);
84 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
85 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
86 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
87 static kmem_cache_t *task_struct_cachep;
90 /* SLAB cache for signal_struct structures (tsk->signal) */
91 static kmem_cache_t *signal_cachep;
93 /* SLAB cache for sighand_struct structures (tsk->sighand) */
94 kmem_cache_t *sighand_cachep;
96 /* SLAB cache for files_struct structures (tsk->files) */
97 kmem_cache_t *files_cachep;
99 /* SLAB cache for fs_struct structures (tsk->fs) */
100 kmem_cache_t *fs_cachep;
102 /* SLAB cache for vm_area_struct structures */
103 kmem_cache_t *vm_area_cachep;
105 /* SLAB cache for mm_struct structures (tsk->mm) */
106 static kmem_cache_t *mm_cachep;
108 void free_task(struct task_struct *tsk)
110 free_thread_info(tsk->thread_info);
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);
127 if (!profile_handoff_task(tsk))
131 void __init fork_init(unsigned long mempages)
133 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
134 #ifndef ARCH_MIN_TASKALIGN
135 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
137 /* create a slab on which task_structs can be allocated */
139 kmem_cache_create("task_struct", sizeof(struct task_struct),
140 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
144 * The default maximum number of threads is set to a safe
145 * value: the thread structures can take up at most half
148 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
151 * we need to allow at least 20 threads to boot a system
156 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
157 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
158 init_task.signal->rlim[RLIMIT_SIGPENDING] =
159 init_task.signal->rlim[RLIMIT_NPROC];
162 static struct task_struct *dup_task_struct(struct task_struct *orig)
164 struct task_struct *tsk;
165 struct thread_info *ti;
167 prepare_to_copy(orig);
169 tsk = alloc_task_struct();
173 ti = alloc_thread_info(tsk);
175 free_task_struct(tsk);
180 tsk->thread_info = ti;
181 setup_thread_stack(tsk, orig);
183 /* One for us, one for whoever does the "release_task()" (usually parent) */
184 atomic_set(&tsk->usage,2);
185 atomic_set(&tsk->fs_excl, 0);
187 tsk->splice_pipe = NULL;
192 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
194 struct vm_area_struct *mpnt, *tmp, **pprev;
195 struct rb_node **rb_link, *rb_parent;
197 unsigned long charge;
198 struct mempolicy *pol;
200 down_write(&oldmm->mmap_sem);
201 flush_cache_mm(oldmm);
202 down_write(&mm->mmap_sem);
206 mm->mmap_cache = NULL;
207 mm->free_area_cache = oldmm->mmap_base;
208 mm->cached_hole_size = ~0UL;
210 __set_mm_counter(mm, file_rss, 0);
211 __set_mm_counter(mm, anon_rss, 0);
212 cpus_clear(mm->cpu_vm_mask);
214 rb_link = &mm->mm_rb.rb_node;
218 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
221 if (mpnt->vm_flags & VM_DONTCOPY) {
222 long pages = vma_pages(mpnt);
223 vx_vmpages_sub(mm, pages);
224 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
229 if (mpnt->vm_flags & VM_ACCOUNT) {
230 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
231 if (security_vm_enough_memory(len))
235 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
239 pol = mpol_copy(vma_policy(mpnt));
240 retval = PTR_ERR(pol);
242 goto fail_nomem_policy;
243 vma_set_policy(tmp, pol);
244 tmp->vm_flags &= ~VM_LOCKED;
250 struct inode *inode = file->f_dentry->d_inode;
252 if (tmp->vm_flags & VM_DENYWRITE)
253 atomic_dec(&inode->i_writecount);
255 /* insert tmp into the share list, just after mpnt */
256 spin_lock(&file->f_mapping->i_mmap_lock);
257 tmp->vm_truncate_count = mpnt->vm_truncate_count;
258 flush_dcache_mmap_lock(file->f_mapping);
259 vma_prio_tree_add(tmp, mpnt);
260 flush_dcache_mmap_unlock(file->f_mapping);
261 spin_unlock(&file->f_mapping->i_mmap_lock);
265 * Link in the new vma and copy the page table entries.
268 pprev = &tmp->vm_next;
270 __vma_link_rb(mm, tmp, rb_link, rb_parent);
271 rb_link = &tmp->vm_rb.rb_right;
272 rb_parent = &tmp->vm_rb;
275 retval = copy_page_range(mm, oldmm, mpnt);
277 if (tmp->vm_ops && tmp->vm_ops->open)
278 tmp->vm_ops->open(tmp);
284 arch_dup_mmap(mm, oldmm);
288 up_write(&mm->mmap_sem);
290 up_write(&oldmm->mmap_sem);
293 kmem_cache_free(vm_area_cachep, tmp);
296 vm_unacct_memory(charge);
300 static inline int mm_alloc_pgd(struct mm_struct * mm)
302 mm->pgd = pgd_alloc(mm);
303 if (unlikely(!mm->pgd))
308 static inline void mm_free_pgd(struct mm_struct * mm)
313 #define dup_mmap(mm, oldmm) (0)
314 #define mm_alloc_pgd(mm) (0)
315 #define mm_free_pgd(mm)
316 #endif /* CONFIG_MMU */
318 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
320 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
321 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
323 #include <linux/init_task.h>
325 static struct mm_struct * mm_init(struct mm_struct * mm)
327 atomic_set(&mm->mm_users, 1);
328 atomic_set(&mm->mm_count, 1);
329 init_rwsem(&mm->mmap_sem);
330 INIT_LIST_HEAD(&mm->mmlist);
331 mm->core_waiters = 0;
333 spin_lock_init(&mm->page_table_lock);
334 rwlock_init(&mm->ioctx_list_lock);
335 mm->ioctx_list = NULL;
336 mm->free_area_cache = TASK_UNMAPPED_BASE;
337 mm->cached_hole_size = ~0UL;
339 if (likely(!mm_alloc_pgd(mm))) {
341 set_vx_info(&mm->mm_vx_info, current->vx_info);
349 * Allocate and initialize an mm_struct.
351 struct mm_struct * mm_alloc(void)
353 struct mm_struct * mm;
357 memset(mm, 0, sizeof(*mm));
364 * Called when the last reference to the mm
365 * is dropped: either by a lazy thread or by
366 * mmput. Free the page directory and the mm.
368 void fastcall __mmdrop(struct mm_struct *mm)
370 BUG_ON(mm == &init_mm);
373 clr_vx_info(&mm->mm_vx_info);
378 * Decrement the use count and release all resources for an mm.
380 void mmput(struct mm_struct *mm)
382 if (atomic_dec_and_test(&mm->mm_users)) {
385 if (!list_empty(&mm->mmlist)) {
386 spin_lock(&mmlist_lock);
387 list_del(&mm->mmlist);
388 spin_unlock(&mmlist_lock);
394 EXPORT_SYMBOL_GPL(mmput);
397 * get_task_mm - acquire a reference to the task's mm
399 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
400 * this kernel workthread has transiently adopted a user mm with use_mm,
401 * to do its AIO) is not set and if so returns a reference to it, after
402 * bumping up the use count. User must release the mm via mmput()
403 * after use. Typically used by /proc and ptrace.
405 struct mm_struct *get_task_mm(struct task_struct *task)
407 struct mm_struct *mm;
412 if (task->flags & PF_BORROWED_MM)
415 atomic_inc(&mm->mm_users);
420 EXPORT_SYMBOL_GPL(get_task_mm);
422 /* Please note the differences between mmput and mm_release.
423 * mmput is called whenever we stop holding onto a mm_struct,
424 * error success whatever.
426 * mm_release is called after a mm_struct has been removed
427 * from the current process.
429 * This difference is important for error handling, when we
430 * only half set up a mm_struct for a new process and need to restore
431 * the old one. Because we mmput the new mm_struct before
432 * restoring the old one. . .
433 * Eric Biederman 10 January 1998
435 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
437 struct completion *vfork_done = tsk->vfork_done;
439 /* Get rid of any cached register state */
440 deactivate_mm(tsk, mm);
442 /* notify parent sleeping on vfork() */
444 tsk->vfork_done = NULL;
445 complete(vfork_done);
447 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
448 u32 __user * tidptr = tsk->clear_child_tid;
449 tsk->clear_child_tid = NULL;
452 * We don't check the error code - if userspace has
453 * not set up a proper pointer then tough luck.
456 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
461 * Allocate a new mm structure and copy contents from the
462 * mm structure of the passed in task structure.
464 static struct mm_struct *dup_mm(struct task_struct *tsk)
466 struct mm_struct *mm, *oldmm = current->mm;
476 memcpy(mm, oldmm, sizeof(*mm));
477 mm->mm_vx_info = NULL;
482 if (init_new_context(tsk, mm))
485 err = dup_mmap(mm, oldmm);
489 mm->hiwater_rss = get_mm_rss(mm);
490 mm->hiwater_vm = mm->total_vm;
502 * If init_new_context() failed, we cannot use mmput() to free the mm
503 * because it calls destroy_context()
505 clr_vx_info(&mm->mm_vx_info);
511 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
513 struct mm_struct * mm, *oldmm;
516 tsk->min_flt = tsk->maj_flt = 0;
517 tsk->nvcsw = tsk->nivcsw = 0;
520 tsk->active_mm = NULL;
523 * Are we cloning a kernel thread?
525 * We need to steal a active VM for that..
531 if (clone_flags & CLONE_VM) {
532 atomic_inc(&oldmm->mm_users);
551 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
553 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
554 /* We don't need to lock fs - think why ;-) */
556 atomic_set(&fs->count, 1);
557 rwlock_init(&fs->lock);
558 fs->umask = old->umask;
559 read_lock(&old->lock);
560 fs->rootmnt = mntget(old->rootmnt);
561 fs->root = dget(old->root);
562 fs->pwdmnt = mntget(old->pwdmnt);
563 fs->pwd = dget(old->pwd);
565 fs->altrootmnt = mntget(old->altrootmnt);
566 fs->altroot = dget(old->altroot);
568 fs->altrootmnt = NULL;
571 read_unlock(&old->lock);
576 struct fs_struct *copy_fs_struct(struct fs_struct *old)
578 return __copy_fs_struct(old);
581 EXPORT_SYMBOL_GPL(copy_fs_struct);
583 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
585 if (clone_flags & CLONE_FS) {
586 atomic_inc(¤t->fs->count);
589 tsk->fs = __copy_fs_struct(current->fs);
595 static int count_open_files(struct fdtable *fdt)
597 int size = fdt->max_fdset;
600 /* Find the last open fd */
601 for (i = size/(8*sizeof(long)); i > 0; ) {
602 if (fdt->open_fds->fds_bits[--i])
605 i = (i+1) * 8 * sizeof(long);
609 static struct files_struct *alloc_files(void)
611 struct files_struct *newf;
614 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
618 atomic_set(&newf->count, 1);
620 spin_lock_init(&newf->file_lock);
623 fdt->max_fds = NR_OPEN_DEFAULT;
624 fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
625 fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
626 fdt->open_fds = (fd_set *)&newf->open_fds_init;
627 fdt->fd = &newf->fd_array[0];
628 INIT_RCU_HEAD(&fdt->rcu);
629 fdt->free_files = NULL;
631 rcu_assign_pointer(newf->fdt, fdt);
637 * Allocate a new files structure and copy contents from the
638 * passed in files structure.
640 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
642 struct files_struct *newf;
643 struct file **old_fds, **new_fds;
644 int open_files, size, i, expand;
645 struct fdtable *old_fdt, *new_fdt;
647 newf = alloc_files();
651 spin_lock(&oldf->file_lock);
652 old_fdt = files_fdtable(oldf);
653 new_fdt = files_fdtable(newf);
654 size = old_fdt->max_fdset;
655 open_files = count_open_files(old_fdt);
659 * Check whether we need to allocate a larger fd array or fd set.
660 * Note: we're not a clone task, so the open count won't change.
662 if (open_files > new_fdt->max_fdset) {
663 new_fdt->max_fdset = 0;
666 if (open_files > new_fdt->max_fds) {
667 new_fdt->max_fds = 0;
671 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
673 spin_unlock(&oldf->file_lock);
674 spin_lock(&newf->file_lock);
675 *errorp = expand_files(newf, open_files-1);
676 spin_unlock(&newf->file_lock);
679 new_fdt = files_fdtable(newf);
681 * Reacquire the oldf lock and a pointer to its fd table
682 * who knows it may have a new bigger fd table. We need
683 * the latest pointer.
685 spin_lock(&oldf->file_lock);
686 old_fdt = files_fdtable(oldf);
689 old_fds = old_fdt->fd;
690 new_fds = new_fdt->fd;
692 memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
693 memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
695 for (i = open_files; i != 0; i--) {
696 struct file *f = *old_fds++;
699 /* FIXME: sum it first for check and performance */
700 vx_openfd_inc(open_files - i);
703 * The fd may be claimed in the fd bitmap but not yet
704 * instantiated in the files array if a sibling thread
705 * is partway through open(). So make sure that this
706 * fd is available to the new process.
708 FD_CLR(open_files - i, new_fdt->open_fds);
710 rcu_assign_pointer(*new_fds++, f);
712 spin_unlock(&oldf->file_lock);
714 /* compute the remainder to be cleared */
715 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
717 /* This is long word aligned thus could use a optimized version */
718 memset(new_fds, 0, size);
720 if (new_fdt->max_fdset > open_files) {
721 int left = (new_fdt->max_fdset-open_files)/8;
722 int start = open_files / (8 * sizeof(unsigned long));
724 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
725 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
732 free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
733 free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
734 free_fd_array(new_fdt->fd, new_fdt->max_fds);
735 kmem_cache_free(files_cachep, newf);
739 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
741 struct files_struct *oldf, *newf;
745 * A background process may not have any files ...
747 oldf = current->files;
751 if (clone_flags & CLONE_FILES) {
752 atomic_inc(&oldf->count);
757 * Note: we may be using current for both targets (See exec.c)
758 * This works because we cache current->files (old) as oldf. Don't
763 newf = dup_fd(oldf, &error);
774 * Helper to unshare the files of the current task.
775 * We don't want to expose copy_files internals to
776 * the exec layer of the kernel.
779 int unshare_files(void)
781 struct files_struct *files = current->files;
786 /* This can race but the race causes us to copy when we don't
787 need to and drop the copy */
788 if(atomic_read(&files->count) == 1)
790 atomic_inc(&files->count);
793 rc = copy_files(0, current);
795 current->files = files;
799 EXPORT_SYMBOL(unshare_files);
801 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
803 struct sighand_struct *sig;
805 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
806 atomic_inc(¤t->sighand->count);
809 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
810 rcu_assign_pointer(tsk->sighand, sig);
813 atomic_set(&sig->count, 1);
814 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
818 void __cleanup_sighand(struct sighand_struct *sighand)
820 if (atomic_dec_and_test(&sighand->count))
821 kmem_cache_free(sighand_cachep, sighand);
824 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
826 struct signal_struct *sig;
829 if (clone_flags & CLONE_THREAD) {
830 atomic_inc(¤t->signal->count);
831 atomic_inc(¤t->signal->live);
834 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
839 ret = copy_thread_group_keys(tsk);
841 kmem_cache_free(signal_cachep, sig);
845 atomic_set(&sig->count, 1);
846 atomic_set(&sig->live, 1);
847 init_waitqueue_head(&sig->wait_chldexit);
849 sig->group_exit_code = 0;
850 sig->group_exit_task = NULL;
851 sig->group_stop_count = 0;
852 sig->curr_target = NULL;
853 init_sigpending(&sig->shared_pending);
854 INIT_LIST_HEAD(&sig->posix_timers);
856 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
857 sig->it_real_incr.tv64 = 0;
858 sig->real_timer.function = it_real_fn;
861 sig->it_virt_expires = cputime_zero;
862 sig->it_virt_incr = cputime_zero;
863 sig->it_prof_expires = cputime_zero;
864 sig->it_prof_incr = cputime_zero;
866 sig->leader = 0; /* session leadership doesn't inherit */
867 sig->tty_old_pgrp = 0;
869 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
870 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
871 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
873 INIT_LIST_HEAD(&sig->cpu_timers[0]);
874 INIT_LIST_HEAD(&sig->cpu_timers[1]);
875 INIT_LIST_HEAD(&sig->cpu_timers[2]);
877 task_lock(current->group_leader);
878 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
879 task_unlock(current->group_leader);
881 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
883 * New sole thread in the process gets an expiry time
884 * of the whole CPU time limit.
886 tsk->it_prof_expires =
887 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
893 void __cleanup_signal(struct signal_struct *sig)
895 exit_thread_group_keys(sig);
896 kmem_cache_free(signal_cachep, sig);
899 static inline void cleanup_signal(struct task_struct *tsk)
901 struct signal_struct *sig = tsk->signal;
903 atomic_dec(&sig->live);
905 if (atomic_dec_and_test(&sig->count))
906 __cleanup_signal(sig);
909 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
911 unsigned long new_flags = p->flags;
913 new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
914 new_flags |= PF_FORKNOEXEC;
915 if (!(clone_flags & CLONE_PTRACE))
917 p->flags = new_flags;
920 asmlinkage long sys_set_tid_address(int __user *tidptr)
922 current->clear_child_tid = tidptr;
928 * This creates a new process as a copy of the old one,
929 * but does not actually start it yet.
931 * It copies the registers, and all the appropriate
932 * parts of the process environment (as per the clone
933 * flags). The actual kick-off is left to the caller.
935 static task_t *copy_process(unsigned long clone_flags,
936 unsigned long stack_start,
937 struct pt_regs *regs,
938 unsigned long stack_size,
939 int __user *parent_tidptr,
940 int __user *child_tidptr,
944 struct task_struct *p = NULL;
948 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
949 return ERR_PTR(-EINVAL);
952 * Thread groups must share signals as well, and detached threads
953 * can only be started up within the thread group.
955 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
956 return ERR_PTR(-EINVAL);
959 * Shared signal handlers imply shared VM. By way of the above,
960 * thread groups also imply shared VM. Blocking this case allows
961 * for various simplifications in other code.
963 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
964 return ERR_PTR(-EINVAL);
966 retval = security_task_create(clone_flags);
971 p = dup_task_struct(current);
976 init_vx_info(&p->vx_info, current->vx_info);
977 init_nx_info(&p->nx_info, current->nx_info);
979 /* check vserver memory */
980 if (p->mm && !(clone_flags & CLONE_VM)) {
981 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
982 vx_pages_add(p->vx_info, RLIMIT_AS, p->mm->total_vm);
986 if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
987 if (!vx_rsspages_avail(p->mm, get_mm_counter(p->mm, file_rss)))
988 goto bad_fork_cleanup_vm;
992 if (!vx_nproc_avail(1))
993 goto bad_fork_cleanup_vm;
995 if (atomic_read(&p->user->processes) >=
996 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
997 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
998 p->user != &root_user)
999 goto bad_fork_cleanup_vm;
1002 atomic_inc(&p->user->__count);
1003 atomic_inc(&p->user->processes);
1004 get_group_info(p->group_info);
1007 * If multiple threads are within copy_process(), then this check
1008 * triggers too late. This doesn't hurt, the check is only there
1009 * to stop root fork bombs.
1011 if (nr_threads >= max_threads)
1012 goto bad_fork_cleanup_count;
1014 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1015 goto bad_fork_cleanup_count;
1017 if (p->binfmt && !try_module_get(p->binfmt->module))
1018 goto bad_fork_cleanup_put_domain;
1021 copy_flags(clone_flags, p);
1024 if (clone_flags & CLONE_PARENT_SETTID)
1025 if (put_user(p->pid, parent_tidptr))
1026 goto bad_fork_cleanup;
1028 p->proc_dentry = NULL;
1030 INIT_LIST_HEAD(&p->children);
1031 INIT_LIST_HEAD(&p->sibling);
1032 p->vfork_done = NULL;
1033 spin_lock_init(&p->alloc_lock);
1034 spin_lock_init(&p->proc_lock);
1036 clear_tsk_thread_flag(p, TIF_SIGPENDING);
1037 init_sigpending(&p->pending);
1039 p->utime = cputime_zero;
1040 p->stime = cputime_zero;
1042 p->rchar = 0; /* I/O counter: bytes read */
1043 p->wchar = 0; /* I/O counter: bytes written */
1044 p->syscr = 0; /* I/O counter: read syscalls */
1045 p->syscw = 0; /* I/O counter: write syscalls */
1046 acct_clear_integrals(p);
1048 p->it_virt_expires = cputime_zero;
1049 p->it_prof_expires = cputime_zero;
1050 p->it_sched_expires = 0;
1051 INIT_LIST_HEAD(&p->cpu_timers[0]);
1052 INIT_LIST_HEAD(&p->cpu_timers[1]);
1053 INIT_LIST_HEAD(&p->cpu_timers[2]);
1055 p->lock_depth = -1; /* -1 = no lock */
1056 do_posix_clock_monotonic_gettime(&p->start_time);
1058 p->io_context = NULL;
1060 p->audit_context = NULL;
1063 p->mempolicy = mpol_copy(p->mempolicy);
1064 if (IS_ERR(p->mempolicy)) {
1065 retval = PTR_ERR(p->mempolicy);
1066 p->mempolicy = NULL;
1067 goto bad_fork_cleanup_cpuset;
1069 mpol_fix_fork_child_flag(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;
1115 * sigaltstack should be cleared when sharing the same VM
1117 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1118 p->sas_ss_sp = p->sas_ss_size = 0;
1121 * Syscall tracing should be turned off in the child regardless
1124 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1125 #ifdef TIF_SYSCALL_EMU
1126 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1129 /* Our parent execution domain becomes current domain
1130 These must match for thread signalling to apply */
1132 p->parent_exec_id = p->self_exec_id;
1134 /* ok, now we should be set up.. */
1135 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1136 p->pdeath_signal = 0;
1140 * Ok, make it visible to the rest of the system.
1141 * We dont wake it up yet.
1143 p->group_leader = p;
1144 INIT_LIST_HEAD(&p->thread_group);
1145 INIT_LIST_HEAD(&p->ptrace_children);
1146 INIT_LIST_HEAD(&p->ptrace_list);
1148 /* Perform scheduler related setup. Assign this task to a CPU. */
1149 sched_fork(p, clone_flags);
1151 /* Need tasklist lock for parent etc handling! */
1152 write_lock_irq(&tasklist_lock);
1155 * The task hasn't been attached yet, so its cpus_allowed mask will
1156 * not be changed, nor will its assigned CPU.
1158 * The cpus_allowed mask of the parent may have changed after it was
1159 * copied first time - so re-copy it here, then check the child's CPU
1160 * to ensure it is on a valid CPU (and if not, just force it back to
1161 * parent's CPU). This avoids alot of nasty races.
1163 p->cpus_allowed = current->cpus_allowed;
1164 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1165 !cpu_online(task_cpu(p))))
1166 set_task_cpu(p, smp_processor_id());
1168 /* CLONE_PARENT re-uses the old parent */
1169 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1170 p->real_parent = current->real_parent;
1172 p->real_parent = current;
1173 p->parent = p->real_parent;
1175 spin_lock(¤t->sighand->siglock);
1178 * Process group and session signals need to be delivered to just the
1179 * parent before the fork or both the parent and the child after the
1180 * fork. Restart if a signal comes in before we add the new process to
1181 * it's process group.
1182 * A fatal signal pending means that current will exit, so the new
1183 * thread can't slip out of an OOM kill (or normal SIGKILL).
1185 recalc_sigpending();
1186 if (signal_pending(current)) {
1187 spin_unlock(¤t->sighand->siglock);
1188 write_unlock_irq(&tasklist_lock);
1189 retval = -ERESTARTNOINTR;
1190 goto bad_fork_cleanup_namespace;
1193 if (clone_flags & CLONE_THREAD) {
1195 * Important: if an exit-all has been started then
1196 * do not create this new thread - the whole thread
1197 * group is supposed to exit anyway.
1199 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1200 spin_unlock(¤t->sighand->siglock);
1201 write_unlock_irq(&tasklist_lock);
1203 goto bad_fork_cleanup_namespace;
1206 p->group_leader = current->group_leader;
1207 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1209 if (!cputime_eq(current->signal->it_virt_expires,
1211 !cputime_eq(current->signal->it_prof_expires,
1213 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1214 !list_empty(¤t->signal->cpu_timers[0]) ||
1215 !list_empty(¤t->signal->cpu_timers[1]) ||
1216 !list_empty(¤t->signal->cpu_timers[2])) {
1218 * Have child wake up on its first tick to check
1219 * for process CPU timers.
1221 p->it_prof_expires = jiffies_to_cputime(1);
1228 p->ioprio = current->ioprio;
1230 if (likely(p->pid)) {
1232 if (unlikely(p->ptrace & PT_PTRACED))
1233 __ptrace_link(p, current->parent);
1235 if (thread_group_leader(p)) {
1236 p->signal->tty = current->signal->tty;
1237 p->signal->pgrp = process_group(current);
1238 p->signal->session = current->signal->session;
1239 attach_pid(p, PIDTYPE_PGID, process_group(p));
1240 attach_pid(p, PIDTYPE_SID, p->signal->session);
1242 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1243 __get_cpu_var(process_counts)++;
1245 attach_pid(p, PIDTYPE_PID, p->pid);
1250 spin_unlock(¤t->sighand->siglock);
1252 /* p is copy of current */
1255 claim_vx_info(vxi, p);
1256 atomic_inc(&vxi->cvirt.nr_threads);
1257 atomic_inc(&vxi->cvirt.total_forks);
1262 claim_nx_info(nxi, p);
1263 write_unlock_irq(&tasklist_lock);
1264 proc_fork_connector(p);
1267 bad_fork_cleanup_namespace:
1269 bad_fork_cleanup_keys:
1271 bad_fork_cleanup_mm:
1274 bad_fork_cleanup_signal:
1276 bad_fork_cleanup_sighand:
1277 __cleanup_sighand(p->sighand);
1278 bad_fork_cleanup_fs:
1279 exit_fs(p); /* blocking */
1280 bad_fork_cleanup_files:
1281 exit_files(p); /* blocking */
1282 bad_fork_cleanup_semundo:
1284 bad_fork_cleanup_audit:
1286 bad_fork_cleanup_security:
1287 security_task_free(p);
1288 bad_fork_cleanup_policy:
1290 mpol_free(p->mempolicy);
1291 bad_fork_cleanup_cpuset:
1296 module_put(p->binfmt->module);
1297 bad_fork_cleanup_put_domain:
1298 module_put(task_thread_info(p)->exec_domain->module);
1299 bad_fork_cleanup_count:
1300 put_group_info(p->group_info);
1301 atomic_dec(&p->user->processes);
1303 bad_fork_cleanup_vm:
1304 if (p->mm && !(clone_flags & CLONE_VM))
1305 vx_pages_sub(p->vx_info, RLIMIT_AS, p->mm->total_vm);
1309 return ERR_PTR(retval);
1312 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1314 memset(regs, 0, sizeof(struct pt_regs));
1318 task_t * __devinit fork_idle(int cpu)
1321 struct pt_regs regs;
1323 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0);
1325 return ERR_PTR(-ENOMEM);
1326 init_idle(task, cpu);
1331 static inline int fork_traceflag (unsigned clone_flags)
1333 if (clone_flags & CLONE_UNTRACED)
1335 else if (clone_flags & CLONE_VFORK) {
1336 if (current->ptrace & PT_TRACE_VFORK)
1337 return PTRACE_EVENT_VFORK;
1338 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1339 if (current->ptrace & PT_TRACE_CLONE)
1340 return PTRACE_EVENT_CLONE;
1341 } else if (current->ptrace & PT_TRACE_FORK)
1342 return PTRACE_EVENT_FORK;
1348 * Ok, this is the main fork-routine.
1350 * It copies the process, and if successful kick-starts
1351 * it and waits for it to finish using the VM if required.
1353 long do_fork(unsigned long clone_flags,
1354 unsigned long stack_start,
1355 struct pt_regs *regs,
1356 unsigned long stack_size,
1357 int __user *parent_tidptr,
1358 int __user *child_tidptr)
1360 struct task_struct *p;
1362 struct pid *pid = alloc_pid();
1368 if (unlikely(current->ptrace)) {
1369 trace = fork_traceflag (clone_flags);
1371 clone_flags |= CLONE_PTRACE;
1374 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1376 * Do this prior waking up the new thread - the thread pointer
1377 * might get invalid after that point, if the thread exits quickly.
1380 struct completion vfork;
1382 if (clone_flags & CLONE_VFORK) {
1383 p->vfork_done = &vfork;
1384 init_completion(&vfork);
1387 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1389 * We'll start up with an immediate SIGSTOP.
1391 sigaddset(&p->pending.signal, SIGSTOP);
1392 set_tsk_thread_flag(p, TIF_SIGPENDING);
1395 if (!(clone_flags & CLONE_STOPPED))
1396 wake_up_new_task(p, clone_flags);
1398 p->state = TASK_STOPPED;
1400 if (unlikely (trace)) {
1401 current->ptrace_message = nr;
1402 ptrace_notify ((trace << 8) | SIGTRAP);
1405 if (clone_flags & CLONE_VFORK) {
1406 wait_for_completion(&vfork);
1407 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1408 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1417 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1418 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1421 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1423 struct sighand_struct *sighand = data;
1425 if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1426 SLAB_CTOR_CONSTRUCTOR)
1427 spin_lock_init(&sighand->siglock);
1430 void __init proc_caches_init(void)
1432 sighand_cachep = kmem_cache_create("sighand_cache",
1433 sizeof(struct sighand_struct), 0,
1434 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1435 sighand_ctor, NULL);
1436 signal_cachep = kmem_cache_create("signal_cache",
1437 sizeof(struct signal_struct), 0,
1438 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1439 files_cachep = kmem_cache_create("files_cache",
1440 sizeof(struct files_struct), 0,
1441 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1442 fs_cachep = kmem_cache_create("fs_cache",
1443 sizeof(struct fs_struct), 0,
1444 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1445 vm_area_cachep = kmem_cache_create("vm_area_struct",
1446 sizeof(struct vm_area_struct), 0,
1447 SLAB_PANIC, NULL, NULL);
1448 mm_cachep = kmem_cache_create("mm_struct",
1449 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1450 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1455 * Check constraints on flags passed to the unshare system call and
1456 * force unsharing of additional process context as appropriate.
1458 static inline void check_unshare_flags(unsigned long *flags_ptr)
1461 * If unsharing a thread from a thread group, must also
1464 if (*flags_ptr & CLONE_THREAD)
1465 *flags_ptr |= CLONE_VM;
1468 * If unsharing vm, must also unshare signal handlers.
1470 if (*flags_ptr & CLONE_VM)
1471 *flags_ptr |= CLONE_SIGHAND;
1474 * If unsharing signal handlers and the task was created
1475 * using CLONE_THREAD, then must unshare the thread
1477 if ((*flags_ptr & CLONE_SIGHAND) &&
1478 (atomic_read(¤t->signal->count) > 1))
1479 *flags_ptr |= CLONE_THREAD;
1482 * If unsharing namespace, must also unshare filesystem information.
1484 if (*flags_ptr & CLONE_NEWNS)
1485 *flags_ptr |= CLONE_FS;
1489 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1491 static int unshare_thread(unsigned long unshare_flags)
1493 if (unshare_flags & CLONE_THREAD)
1500 * Unshare the filesystem structure if it is being shared
1502 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1504 struct fs_struct *fs = current->fs;
1506 if ((unshare_flags & CLONE_FS) &&
1507 (fs && atomic_read(&fs->count) > 1)) {
1508 *new_fsp = __copy_fs_struct(current->fs);
1517 * Unshare the namespace structure if it is being shared
1519 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1521 struct namespace *ns = current->namespace;
1523 if ((unshare_flags & CLONE_NEWNS) &&
1524 (ns && atomic_read(&ns->count) > 1)) {
1525 if (!capable(CAP_SYS_ADMIN))
1528 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1537 * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1540 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1542 struct sighand_struct *sigh = current->sighand;
1544 if ((unshare_flags & CLONE_SIGHAND) &&
1545 (sigh && atomic_read(&sigh->count) > 1))
1552 * Unshare vm if it is being shared
1554 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1556 struct mm_struct *mm = current->mm;
1558 if ((unshare_flags & CLONE_VM) &&
1559 (mm && atomic_read(&mm->mm_users) > 1)) {
1567 * Unshare file descriptor table if it is being shared
1569 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1571 struct files_struct *fd = current->files;
1574 if ((unshare_flags & CLONE_FILES) &&
1575 (fd && atomic_read(&fd->count) > 1)) {
1576 *new_fdp = dup_fd(fd, &error);
1585 * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1588 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1590 if (unshare_flags & CLONE_SYSVSEM)
1597 * unshare allows a process to 'unshare' part of the process
1598 * context which was originally shared using clone. copy_*
1599 * functions used by do_fork() cannot be used here directly
1600 * because they modify an inactive task_struct that is being
1601 * constructed. Here we are modifying the current, active,
1604 asmlinkage long sys_unshare(unsigned long unshare_flags)
1607 struct fs_struct *fs, *new_fs = NULL;
1608 struct namespace *ns, *new_ns = NULL;
1609 struct sighand_struct *sigh, *new_sigh = NULL;
1610 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1611 struct files_struct *fd, *new_fd = NULL;
1612 struct sem_undo_list *new_ulist = NULL;
1614 check_unshare_flags(&unshare_flags);
1616 /* Return -EINVAL for all unsupported flags */
1618 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1619 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1620 goto bad_unshare_out;
1622 if ((err = unshare_thread(unshare_flags)))
1623 goto bad_unshare_out;
1624 if ((err = unshare_fs(unshare_flags, &new_fs)))
1625 goto bad_unshare_cleanup_thread;
1626 if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1627 goto bad_unshare_cleanup_fs;
1628 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1629 goto bad_unshare_cleanup_ns;
1630 if ((err = unshare_vm(unshare_flags, &new_mm)))
1631 goto bad_unshare_cleanup_sigh;
1632 if ((err = unshare_fd(unshare_flags, &new_fd)))
1633 goto bad_unshare_cleanup_vm;
1634 if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1635 goto bad_unshare_cleanup_fd;
1637 if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1643 current->fs = new_fs;
1648 ns = current->namespace;
1649 current->namespace = new_ns;
1654 sigh = current->sighand;
1655 rcu_assign_pointer(current->sighand, new_sigh);
1661 active_mm = current->active_mm;
1662 current->mm = new_mm;
1663 current->active_mm = new_mm;
1664 activate_mm(active_mm, new_mm);
1669 fd = current->files;
1670 current->files = new_fd;
1674 task_unlock(current);
1677 bad_unshare_cleanup_fd:
1679 put_files_struct(new_fd);
1681 bad_unshare_cleanup_vm:
1685 bad_unshare_cleanup_sigh:
1687 if (atomic_dec_and_test(&new_sigh->count))
1688 kmem_cache_free(sighand_cachep, new_sigh);
1690 bad_unshare_cleanup_ns:
1692 put_namespace(new_ns);
1694 bad_unshare_cleanup_fs:
1696 put_fs_struct(new_fs);
1698 bad_unshare_cleanup_thread: