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/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/mnt_namespace.h>
21 #include <linux/personality.h>
22 #include <linux/mempolicy.h>
23 #include <linux/sem.h>
24 #include <linux/file.h>
25 #include <linux/fdtable.h>
26 #include <linux/iocontext.h>
27 #include <linux/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.h>
30 #include <linux/mmu_notifier.h>
32 #include <linux/nsproxy.h>
33 #include <linux/capability.h>
34 #include <linux/cpu.h>
35 #include <linux/cgroup.h>
36 #include <linux/security.h>
37 #include <linux/hugetlb.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/tracehook.h>
42 #include <linux/futex.h>
43 #include <linux/task_io_accounting_ops.h>
44 #include <linux/rcupdate.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/audit.h>
48 #include <linux/memcontrol.h>
49 #include <linux/profile.h>
50 #include <linux/rmap.h>
51 #include <linux/acct.h>
52 #include <linux/tsacct_kern.h>
53 #include <linux/cn_proc.h>
54 #include <linux/freezer.h>
55 #include <linux/delayacct.h>
56 #include <linux/taskstats_kern.h>
57 #include <linux/random.h>
58 #include <linux/tty.h>
59 #include <linux/proc_fs.h>
60 #include <linux/blkdev.h>
62 #include <asm/pgtable.h>
63 #include <asm/pgalloc.h>
64 #include <asm/uaccess.h>
65 #include <asm/mmu_context.h>
66 #include <asm/cacheflush.h>
67 #include <asm/tlbflush.h>
70 * Protected counters by write_lock_irq(&tasklist_lock)
72 unsigned long total_forks; /* Handle normal Linux uptimes. */
73 int nr_threads; /* The idle threads do not count.. */
75 int max_threads; /* tunable limit on nr_threads */
77 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
79 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
81 int nr_processes(void)
86 for_each_online_cpu(cpu)
87 total += per_cpu(process_counts, cpu);
92 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
93 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
94 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
95 static struct kmem_cache *task_struct_cachep;
98 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
99 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
101 #ifdef CONFIG_DEBUG_STACK_USAGE
102 gfp_t mask = GFP_KERNEL | __GFP_ZERO;
104 gfp_t mask = GFP_KERNEL;
106 return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
109 static inline void free_thread_info(struct thread_info *ti)
111 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
115 /* SLAB cache for signal_struct structures (tsk->signal) */
116 static struct kmem_cache *signal_cachep;
118 /* SLAB cache for sighand_struct structures (tsk->sighand) */
119 struct kmem_cache *sighand_cachep;
121 /* SLAB cache for files_struct structures (tsk->files) */
122 struct kmem_cache *files_cachep;
124 /* SLAB cache for fs_struct structures (tsk->fs) */
125 struct kmem_cache *fs_cachep;
127 /* SLAB cache for vm_area_struct structures */
128 struct kmem_cache *vm_area_cachep;
130 /* SLAB cache for mm_struct structures (tsk->mm) */
131 static struct kmem_cache *mm_cachep;
133 void free_task(struct task_struct *tsk)
135 prop_local_destroy_single(&tsk->dirties);
136 free_thread_info(tsk->stack);
137 rt_mutex_debug_task_free(tsk);
138 free_task_struct(tsk);
140 EXPORT_SYMBOL(free_task);
142 void __put_task_struct(struct task_struct *tsk)
144 WARN_ON(!tsk->exit_state);
145 WARN_ON(atomic_read(&tsk->usage));
146 WARN_ON(tsk == current);
148 security_task_free(tsk);
150 put_group_info(tsk->group_info);
151 delayacct_tsk_free(tsk);
153 if (!profile_handoff_task(tsk))
158 * macro override instead of weak attribute alias, to workaround
159 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
161 #ifndef arch_task_cache_init
162 #define arch_task_cache_init()
165 void __init fork_init(unsigned long mempages)
167 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
168 #ifndef ARCH_MIN_TASKALIGN
169 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
171 /* create a slab on which task_structs can be allocated */
173 kmem_cache_create("task_struct", sizeof(struct task_struct),
174 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
177 /* do the arch specific task caches init */
178 arch_task_cache_init();
181 * The default maximum number of threads is set to a safe
182 * value: the thread structures can take up at most half
185 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
188 * we need to allow at least 20 threads to boot a system
193 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
194 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
195 init_task.signal->rlim[RLIMIT_SIGPENDING] =
196 init_task.signal->rlim[RLIMIT_NPROC];
199 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
200 struct task_struct *src)
206 static struct task_struct *dup_task_struct(struct task_struct *orig)
208 struct task_struct *tsk;
209 struct thread_info *ti;
212 prepare_to_copy(orig);
214 tsk = alloc_task_struct();
218 ti = alloc_thread_info(tsk);
220 free_task_struct(tsk);
224 err = arch_dup_task_struct(tsk, orig);
230 err = prop_local_init_single(&tsk->dirties);
234 setup_thread_stack(tsk, orig);
236 #ifdef CONFIG_CC_STACKPROTECTOR
237 tsk->stack_canary = get_random_int();
240 /* One for us, one for whoever does the "release_task()" (usually parent) */
241 atomic_set(&tsk->usage,2);
242 atomic_set(&tsk->fs_excl, 0);
243 #ifdef CONFIG_BLK_DEV_IO_TRACE
246 tsk->splice_pipe = NULL;
250 free_thread_info(ti);
251 free_task_struct(tsk);
256 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
258 struct vm_area_struct *mpnt, *tmp, **pprev;
259 struct rb_node **rb_link, *rb_parent;
261 unsigned long charge;
262 struct mempolicy *pol;
264 down_write(&oldmm->mmap_sem);
265 flush_cache_dup_mm(oldmm);
267 * Not linked in yet - no deadlock potential:
269 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
273 mm->mmap_cache = NULL;
274 mm->free_area_cache = oldmm->mmap_base;
275 mm->cached_hole_size = ~0UL;
277 cpus_clear(mm->cpu_vm_mask);
279 rb_link = &mm->mm_rb.rb_node;
283 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
286 if (mpnt->vm_flags & VM_DONTCOPY) {
287 long pages = vma_pages(mpnt);
288 mm->total_vm -= pages;
289 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
294 if (mpnt->vm_flags & VM_ACCOUNT) {
295 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
296 if (security_vm_enough_memory(len))
300 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
304 pol = mpol_dup(vma_policy(mpnt));
305 retval = PTR_ERR(pol);
307 goto fail_nomem_policy;
308 vma_set_policy(tmp, pol);
309 tmp->vm_flags &= ~VM_LOCKED;
315 struct inode *inode = file->f_path.dentry->d_inode;
317 if (tmp->vm_flags & VM_DENYWRITE)
318 atomic_dec(&inode->i_writecount);
320 /* insert tmp into the share list, just after mpnt */
321 spin_lock(&file->f_mapping->i_mmap_lock);
322 tmp->vm_truncate_count = mpnt->vm_truncate_count;
323 flush_dcache_mmap_lock(file->f_mapping);
324 vma_prio_tree_add(tmp, mpnt);
325 flush_dcache_mmap_unlock(file->f_mapping);
326 spin_unlock(&file->f_mapping->i_mmap_lock);
330 * Clear hugetlb-related page reserves for children. This only
331 * affects MAP_PRIVATE mappings. Faults generated by the child
332 * are not guaranteed to succeed, even if read-only
334 if (is_vm_hugetlb_page(tmp))
335 reset_vma_resv_huge_pages(tmp);
338 * Link in the new vma and copy the page table entries.
341 pprev = &tmp->vm_next;
343 __vma_link_rb(mm, tmp, rb_link, rb_parent);
344 rb_link = &tmp->vm_rb.rb_right;
345 rb_parent = &tmp->vm_rb;
348 retval = copy_page_range(mm, oldmm, mpnt);
350 if (tmp->vm_ops && tmp->vm_ops->open)
351 tmp->vm_ops->open(tmp);
356 /* a new mm has just been created */
357 arch_dup_mmap(oldmm, mm);
360 up_write(&mm->mmap_sem);
362 up_write(&oldmm->mmap_sem);
365 kmem_cache_free(vm_area_cachep, tmp);
368 vm_unacct_memory(charge);
372 static inline int mm_alloc_pgd(struct mm_struct * mm)
374 mm->pgd = pgd_alloc(mm);
375 if (unlikely(!mm->pgd))
380 static inline void mm_free_pgd(struct mm_struct * mm)
382 pgd_free(mm, mm->pgd);
385 #define dup_mmap(mm, oldmm) (0)
386 #define mm_alloc_pgd(mm) (0)
387 #define mm_free_pgd(mm)
388 #endif /* CONFIG_MMU */
390 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
392 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
393 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
395 #include <linux/init_task.h>
397 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
399 atomic_set(&mm->mm_users, 1);
400 atomic_set(&mm->mm_count, 1);
401 init_rwsem(&mm->mmap_sem);
402 INIT_LIST_HEAD(&mm->mmlist);
403 mm->flags = (current->mm) ? current->mm->flags
404 : MMF_DUMP_FILTER_DEFAULT;
405 mm->core_state = NULL;
407 set_mm_counter(mm, file_rss, 0);
408 set_mm_counter(mm, anon_rss, 0);
409 spin_lock_init(&mm->page_table_lock);
410 rwlock_init(&mm->ioctx_list_lock);
411 mm->ioctx_list = NULL;
412 mm->free_area_cache = TASK_UNMAPPED_BASE;
413 mm->cached_hole_size = ~0UL;
414 mm_init_owner(mm, p);
416 if (likely(!mm_alloc_pgd(mm))) {
418 mmu_notifier_mm_init(mm);
427 * Allocate and initialize an mm_struct.
429 struct mm_struct * mm_alloc(void)
431 struct mm_struct * mm;
435 memset(mm, 0, sizeof(*mm));
436 mm = mm_init(mm, current);
442 * Called when the last reference to the mm
443 * is dropped: either by a lazy thread or by
444 * mmput. Free the page directory and the mm.
446 void __mmdrop(struct mm_struct *mm)
448 BUG_ON(mm == &init_mm);
451 mmu_notifier_mm_destroy(mm);
454 EXPORT_SYMBOL_GPL(__mmdrop);
457 * Decrement the use count and release all resources for an mm.
459 void mmput(struct mm_struct *mm)
463 if (atomic_dec_and_test(&mm->mm_users)) {
466 set_mm_exe_file(mm, NULL);
467 if (!list_empty(&mm->mmlist)) {
468 spin_lock(&mmlist_lock);
469 list_del(&mm->mmlist);
470 spin_unlock(&mmlist_lock);
476 EXPORT_SYMBOL_GPL(mmput);
479 * get_task_mm - acquire a reference to the task's mm
481 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
482 * this kernel workthread has transiently adopted a user mm with use_mm,
483 * to do its AIO) is not set and if so returns a reference to it, after
484 * bumping up the use count. User must release the mm via mmput()
485 * after use. Typically used by /proc and ptrace.
487 struct mm_struct *get_task_mm(struct task_struct *task)
489 struct mm_struct *mm;
494 if (task->flags & PF_KTHREAD)
497 atomic_inc(&mm->mm_users);
502 EXPORT_SYMBOL_GPL(get_task_mm);
504 /* Please note the differences between mmput and mm_release.
505 * mmput is called whenever we stop holding onto a mm_struct,
506 * error success whatever.
508 * mm_release is called after a mm_struct has been removed
509 * from the current process.
511 * This difference is important for error handling, when we
512 * only half set up a mm_struct for a new process and need to restore
513 * the old one. Because we mmput the new mm_struct before
514 * restoring the old one. . .
515 * Eric Biederman 10 January 1998
517 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
519 struct completion *vfork_done = tsk->vfork_done;
521 /* Get rid of any cached register state */
522 deactivate_mm(tsk, mm);
524 /* notify parent sleeping on vfork() */
526 tsk->vfork_done = NULL;
527 complete(vfork_done);
531 * If we're exiting normally, clear a user-space tid field if
532 * requested. We leave this alone when dying by signal, to leave
533 * the value intact in a core dump, and to save the unnecessary
534 * trouble otherwise. Userland only wants this done for a sys_exit.
536 if (tsk->clear_child_tid
537 && !(tsk->flags & PF_SIGNALED)
538 && atomic_read(&mm->mm_users) > 1) {
539 u32 __user * tidptr = tsk->clear_child_tid;
540 tsk->clear_child_tid = NULL;
543 * We don't check the error code - if userspace has
544 * not set up a proper pointer then tough luck.
547 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
552 * Allocate a new mm structure and copy contents from the
553 * mm structure of the passed in task structure.
555 struct mm_struct *dup_mm(struct task_struct *tsk)
557 struct mm_struct *mm, *oldmm = current->mm;
567 memcpy(mm, oldmm, sizeof(*mm));
569 /* Initializing for Swap token stuff */
570 mm->token_priority = 0;
571 mm->last_interval = 0;
573 if (!mm_init(mm, tsk))
576 if (init_new_context(tsk, mm))
579 dup_mm_exe_file(oldmm, mm);
581 err = dup_mmap(mm, oldmm);
585 mm->hiwater_rss = get_mm_rss(mm);
586 mm->hiwater_vm = mm->total_vm;
598 * If init_new_context() failed, we cannot use mmput() to free the mm
599 * because it calls destroy_context()
606 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
608 struct mm_struct * mm, *oldmm;
611 tsk->min_flt = tsk->maj_flt = 0;
612 tsk->nvcsw = tsk->nivcsw = 0;
615 tsk->active_mm = NULL;
618 * Are we cloning a kernel thread?
620 * We need to steal a active VM for that..
626 if (clone_flags & CLONE_VM) {
627 atomic_inc(&oldmm->mm_users);
638 /* Initializing for Swap token stuff */
639 mm->token_priority = 0;
640 mm->last_interval = 0;
650 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
652 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
653 /* We don't need to lock fs - think why ;-) */
655 atomic_set(&fs->count, 1);
656 rwlock_init(&fs->lock);
657 fs->umask = old->umask;
658 read_lock(&old->lock);
659 fs->root = old->root;
660 path_get(&old->root);
663 read_unlock(&old->lock);
668 struct fs_struct *copy_fs_struct(struct fs_struct *old)
670 return __copy_fs_struct(old);
673 EXPORT_SYMBOL_GPL(copy_fs_struct);
675 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
677 if (clone_flags & CLONE_FS) {
678 atomic_inc(¤t->fs->count);
681 tsk->fs = __copy_fs_struct(current->fs);
687 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
689 struct files_struct *oldf, *newf;
693 * A background process may not have any files ...
695 oldf = current->files;
699 if (clone_flags & CLONE_FILES) {
700 atomic_inc(&oldf->count);
704 newf = dup_fd(oldf, &error);
714 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
717 struct io_context *ioc = current->io_context;
722 * Share io context with parent, if CLONE_IO is set
724 if (clone_flags & CLONE_IO) {
725 tsk->io_context = ioc_task_link(ioc);
726 if (unlikely(!tsk->io_context))
728 } else if (ioprio_valid(ioc->ioprio)) {
729 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
730 if (unlikely(!tsk->io_context))
733 tsk->io_context->ioprio = ioc->ioprio;
739 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
741 struct sighand_struct *sig;
743 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
744 atomic_inc(¤t->sighand->count);
747 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
748 rcu_assign_pointer(tsk->sighand, sig);
751 atomic_set(&sig->count, 1);
752 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
756 void __cleanup_sighand(struct sighand_struct *sighand)
758 if (atomic_dec_and_test(&sighand->count))
759 kmem_cache_free(sighand_cachep, sighand);
764 * Initialize POSIX timer handling for a thread group.
766 static void posix_cpu_timers_init_group(struct signal_struct *sig)
768 /* Thread group counters. */
769 thread_group_cputime_init(sig);
771 /* Expiration times and increments. */
772 sig->it_virt_expires = cputime_zero;
773 sig->it_virt_incr = cputime_zero;
774 sig->it_prof_expires = cputime_zero;
775 sig->it_prof_incr = cputime_zero;
777 /* Cached expiration times. */
778 sig->cputime_expires.prof_exp = cputime_zero;
779 sig->cputime_expires.virt_exp = cputime_zero;
780 sig->cputime_expires.sched_exp = 0;
782 /* The timer lists. */
783 INIT_LIST_HEAD(&sig->cpu_timers[0]);
784 INIT_LIST_HEAD(&sig->cpu_timers[1]);
785 INIT_LIST_HEAD(&sig->cpu_timers[2]);
788 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
790 struct signal_struct *sig;
793 if (clone_flags & CLONE_THREAD) {
794 ret = thread_group_cputime_clone_thread(current);
796 atomic_inc(¤t->signal->count);
797 atomic_inc(¤t->signal->live);
801 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
806 ret = copy_thread_group_keys(tsk);
808 kmem_cache_free(signal_cachep, sig);
812 atomic_set(&sig->count, 1);
813 atomic_set(&sig->live, 1);
814 init_waitqueue_head(&sig->wait_chldexit);
816 sig->group_exit_code = 0;
817 sig->group_exit_task = NULL;
818 sig->group_stop_count = 0;
819 sig->curr_target = tsk;
820 init_sigpending(&sig->shared_pending);
821 INIT_LIST_HEAD(&sig->posix_timers);
823 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
824 sig->it_real_incr.tv64 = 0;
825 sig->real_timer.function = it_real_fn;
827 sig->leader = 0; /* session leadership doesn't inherit */
828 sig->tty_old_pgrp = NULL;
831 sig->cutime = sig->cstime = cputime_zero;
832 sig->gtime = cputime_zero;
833 sig->cgtime = cputime_zero;
834 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
835 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
836 sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
837 task_io_accounting_init(&sig->ioac);
838 taskstats_tgid_init(sig);
840 task_lock(current->group_leader);
841 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
842 task_unlock(current->group_leader);
844 posix_cpu_timers_init_group(sig);
846 acct_init_pacct(&sig->pacct);
853 void __cleanup_signal(struct signal_struct *sig)
855 thread_group_cputime_free(sig);
856 exit_thread_group_keys(sig);
857 tty_kref_put(sig->tty);
858 kmem_cache_free(signal_cachep, sig);
861 static void cleanup_signal(struct task_struct *tsk)
863 struct signal_struct *sig = tsk->signal;
865 atomic_dec(&sig->live);
867 if (atomic_dec_and_test(&sig->count))
868 __cleanup_signal(sig);
871 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
873 unsigned long new_flags = p->flags;
875 new_flags &= ~PF_SUPERPRIV;
876 new_flags |= PF_FORKNOEXEC;
877 new_flags |= PF_STARTING;
878 p->flags = new_flags;
879 clear_freeze_flag(p);
882 asmlinkage long sys_set_tid_address(int __user *tidptr)
884 current->clear_child_tid = tidptr;
886 return task_pid_vnr(current);
889 static void rt_mutex_init_task(struct task_struct *p)
891 spin_lock_init(&p->pi_lock);
892 #ifdef CONFIG_RT_MUTEXES
893 plist_head_init(&p->pi_waiters, &p->pi_lock);
894 p->pi_blocked_on = NULL;
898 #ifdef CONFIG_MM_OWNER
899 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
903 #endif /* CONFIG_MM_OWNER */
906 * Initialize POSIX timer handling for a single task.
908 static void posix_cpu_timers_init(struct task_struct *tsk)
910 tsk->cputime_expires.prof_exp = cputime_zero;
911 tsk->cputime_expires.virt_exp = cputime_zero;
912 tsk->cputime_expires.sched_exp = 0;
913 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
914 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
915 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
919 * This creates a new process as a copy of the old one,
920 * but does not actually start it yet.
922 * It copies the registers, and all the appropriate
923 * parts of the process environment (as per the clone
924 * flags). The actual kick-off is left to the caller.
926 static struct task_struct *copy_process(unsigned long clone_flags,
927 unsigned long stack_start,
928 struct pt_regs *regs,
929 unsigned long stack_size,
930 int __user *child_tidptr,
935 struct task_struct *p;
936 int cgroup_callbacks_done = 0;
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);
965 rt_mutex_init_task(p);
967 #ifdef CONFIG_PROVE_LOCKING
968 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
969 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
972 if (atomic_read(&p->user->processes) >=
973 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
974 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
975 p->user != current->nsproxy->user_ns->root_user)
979 atomic_inc(&p->user->__count);
980 atomic_inc(&p->user->processes);
981 get_group_info(p->group_info);
984 * If multiple threads are within copy_process(), then this check
985 * triggers too late. This doesn't hurt, the check is only there
986 * to stop root fork bombs.
988 if (nr_threads >= max_threads)
989 goto bad_fork_cleanup_count;
991 if (!try_module_get(task_thread_info(p)->exec_domain->module))
992 goto bad_fork_cleanup_count;
994 if (p->binfmt && !try_module_get(p->binfmt->module))
995 goto bad_fork_cleanup_put_domain;
998 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
999 copy_flags(clone_flags, p);
1000 INIT_LIST_HEAD(&p->children);
1001 INIT_LIST_HEAD(&p->sibling);
1002 #ifdef CONFIG_PREEMPT_RCU
1003 p->rcu_read_lock_nesting = 0;
1004 p->rcu_flipctr_idx = 0;
1005 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1006 p->vfork_done = NULL;
1007 spin_lock_init(&p->alloc_lock);
1009 clear_tsk_thread_flag(p, TIF_SIGPENDING);
1010 init_sigpending(&p->pending);
1012 p->utime = cputime_zero;
1013 p->stime = cputime_zero;
1014 p->gtime = cputime_zero;
1015 p->utimescaled = cputime_zero;
1016 p->stimescaled = cputime_zero;
1017 p->prev_utime = cputime_zero;
1018 p->prev_stime = cputime_zero;
1020 #ifdef CONFIG_DETECT_SOFTLOCKUP
1021 p->last_switch_count = 0;
1022 p->last_switch_timestamp = 0;
1025 task_io_accounting_init(&p->ioac);
1026 acct_clear_integrals(p);
1028 posix_cpu_timers_init(p);
1030 p->lock_depth = -1; /* -1 = no lock */
1031 do_posix_clock_monotonic_gettime(&p->start_time);
1032 p->real_start_time = p->start_time;
1033 monotonic_to_bootbased(&p->real_start_time);
1034 #ifdef CONFIG_SECURITY
1037 p->cap_bset = current->cap_bset;
1038 p->io_context = NULL;
1039 p->audit_context = NULL;
1042 p->mempolicy = mpol_dup(p->mempolicy);
1043 if (IS_ERR(p->mempolicy)) {
1044 retval = PTR_ERR(p->mempolicy);
1045 p->mempolicy = NULL;
1046 goto bad_fork_cleanup_cgroup;
1048 mpol_fix_fork_child_flag(p);
1050 #ifdef CONFIG_TRACE_IRQFLAGS
1052 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1053 p->hardirqs_enabled = 1;
1055 p->hardirqs_enabled = 0;
1057 p->hardirq_enable_ip = 0;
1058 p->hardirq_enable_event = 0;
1059 p->hardirq_disable_ip = _THIS_IP_;
1060 p->hardirq_disable_event = 0;
1061 p->softirqs_enabled = 1;
1062 p->softirq_enable_ip = _THIS_IP_;
1063 p->softirq_enable_event = 0;
1064 p->softirq_disable_ip = 0;
1065 p->softirq_disable_event = 0;
1066 p->hardirq_context = 0;
1067 p->softirq_context = 0;
1069 #ifdef CONFIG_LOCKDEP
1070 p->lockdep_depth = 0; /* no locks held yet */
1071 p->curr_chain_key = 0;
1072 p->lockdep_recursion = 0;
1075 #ifdef CONFIG_DEBUG_MUTEXES
1076 p->blocked_on = NULL; /* not blocked yet */
1079 /* Perform scheduler related setup. Assign this task to a CPU. */
1080 sched_fork(p, clone_flags);
1082 if ((retval = security_task_alloc(p)))
1083 goto bad_fork_cleanup_policy;
1084 if ((retval = audit_alloc(p)))
1085 goto bad_fork_cleanup_security;
1086 /* copy all the process information */
1087 if ((retval = copy_semundo(clone_flags, p)))
1088 goto bad_fork_cleanup_audit;
1089 if ((retval = copy_files(clone_flags, p)))
1090 goto bad_fork_cleanup_semundo;
1091 if ((retval = copy_fs(clone_flags, p)))
1092 goto bad_fork_cleanup_files;
1093 if ((retval = copy_sighand(clone_flags, p)))
1094 goto bad_fork_cleanup_fs;
1095 if ((retval = copy_signal(clone_flags, p)))
1096 goto bad_fork_cleanup_sighand;
1097 if ((retval = copy_mm(clone_flags, p)))
1098 goto bad_fork_cleanup_signal;
1099 if ((retval = copy_keys(clone_flags, p)))
1100 goto bad_fork_cleanup_mm;
1101 if ((retval = copy_namespaces(clone_flags, p)))
1102 goto bad_fork_cleanup_keys;
1103 if ((retval = copy_io(clone_flags, p)))
1104 goto bad_fork_cleanup_namespaces;
1105 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1107 goto bad_fork_cleanup_io;
1109 if (pid != &init_struct_pid) {
1111 pid = alloc_pid(task_active_pid_ns(p));
1113 goto bad_fork_cleanup_io;
1115 if (clone_flags & CLONE_NEWPID) {
1116 retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1118 goto bad_fork_free_pid;
1122 p->pid = pid_nr(pid);
1124 if (clone_flags & CLONE_THREAD)
1125 p->tgid = current->tgid;
1127 if (current->nsproxy != p->nsproxy) {
1128 retval = ns_cgroup_clone(p, pid);
1130 goto bad_fork_free_pid;
1133 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1135 * Clear TID on mm_release()?
1137 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1139 p->robust_list = NULL;
1140 #ifdef CONFIG_COMPAT
1141 p->compat_robust_list = NULL;
1143 INIT_LIST_HEAD(&p->pi_state_list);
1144 p->pi_state_cache = NULL;
1147 * sigaltstack should be cleared when sharing the same VM
1149 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1150 p->sas_ss_sp = p->sas_ss_size = 0;
1153 * Syscall tracing should be turned off in the child regardless
1156 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1157 #ifdef TIF_SYSCALL_EMU
1158 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1160 clear_all_latency_tracing(p);
1162 /* Our parent execution domain becomes current domain
1163 These must match for thread signalling to apply */
1164 p->parent_exec_id = p->self_exec_id;
1166 /* ok, now we should be set up.. */
1167 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1168 p->pdeath_signal = 0;
1172 * Ok, make it visible to the rest of the system.
1173 * We dont wake it up yet.
1175 p->group_leader = p;
1176 INIT_LIST_HEAD(&p->thread_group);
1178 /* Now that the task is set up, run cgroup callbacks if
1179 * necessary. We need to run them before the task is visible
1180 * on the tasklist. */
1181 cgroup_fork_callbacks(p);
1182 cgroup_callbacks_done = 1;
1184 /* Need tasklist lock for parent etc handling! */
1185 write_lock_irq(&tasklist_lock);
1188 * The task hasn't been attached yet, so its cpus_allowed mask will
1189 * not be changed, nor will its assigned CPU.
1191 * The cpus_allowed mask of the parent may have changed after it was
1192 * copied first time - so re-copy it here, then check the child's CPU
1193 * to ensure it is on a valid CPU (and if not, just force it back to
1194 * parent's CPU). This avoids alot of nasty races.
1196 p->cpus_allowed = current->cpus_allowed;
1197 p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1198 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1199 !cpu_online(task_cpu(p))))
1200 set_task_cpu(p, smp_processor_id());
1202 /* CLONE_PARENT re-uses the old parent */
1203 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1204 p->real_parent = current->real_parent;
1206 p->real_parent = current;
1208 spin_lock(¤t->sighand->siglock);
1211 * Process group and session signals need to be delivered to just the
1212 * parent before the fork or both the parent and the child after the
1213 * fork. Restart if a signal comes in before we add the new process to
1214 * it's process group.
1215 * A fatal signal pending means that current will exit, so the new
1216 * thread can't slip out of an OOM kill (or normal SIGKILL).
1218 recalc_sigpending();
1219 if (signal_pending(current)) {
1220 spin_unlock(¤t->sighand->siglock);
1221 write_unlock_irq(&tasklist_lock);
1222 retval = -ERESTARTNOINTR;
1223 goto bad_fork_free_pid;
1226 if (clone_flags & CLONE_THREAD) {
1227 p->group_leader = current->group_leader;
1228 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1231 if (likely(p->pid)) {
1232 list_add_tail(&p->sibling, &p->real_parent->children);
1233 tracehook_finish_clone(p, clone_flags, trace);
1235 if (thread_group_leader(p)) {
1236 if (clone_flags & CLONE_NEWPID)
1237 p->nsproxy->pid_ns->child_reaper = p;
1239 p->signal->leader_pid = pid;
1240 tty_kref_put(p->signal->tty);
1241 p->signal->tty = tty_kref_get(current->signal->tty);
1242 set_task_pgrp(p, task_pgrp_nr(current));
1243 set_task_session(p, task_session_nr(current));
1244 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1245 attach_pid(p, PIDTYPE_SID, task_session(current));
1246 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1247 __get_cpu_var(process_counts)++;
1249 attach_pid(p, PIDTYPE_PID, pid);
1254 spin_unlock(¤t->sighand->siglock);
1255 write_unlock_irq(&tasklist_lock);
1256 proc_fork_connector(p);
1257 cgroup_post_fork(p);
1261 if (pid != &init_struct_pid)
1263 bad_fork_cleanup_io:
1264 put_io_context(p->io_context);
1265 bad_fork_cleanup_namespaces:
1266 exit_task_namespaces(p);
1267 bad_fork_cleanup_keys:
1269 bad_fork_cleanup_mm:
1272 bad_fork_cleanup_signal:
1274 bad_fork_cleanup_sighand:
1275 __cleanup_sighand(p->sighand);
1276 bad_fork_cleanup_fs:
1277 exit_fs(p); /* blocking */
1278 bad_fork_cleanup_files:
1279 exit_files(p); /* blocking */
1280 bad_fork_cleanup_semundo:
1282 bad_fork_cleanup_audit:
1284 bad_fork_cleanup_security:
1285 security_task_free(p);
1286 bad_fork_cleanup_policy:
1288 mpol_put(p->mempolicy);
1289 bad_fork_cleanup_cgroup:
1291 cgroup_exit(p, cgroup_callbacks_done);
1292 delayacct_tsk_free(p);
1294 module_put(p->binfmt->module);
1295 bad_fork_cleanup_put_domain:
1296 module_put(task_thread_info(p)->exec_domain->module);
1297 bad_fork_cleanup_count:
1298 put_group_info(p->group_info);
1299 atomic_dec(&p->user->processes);
1304 return ERR_PTR(retval);
1307 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1309 memset(regs, 0, sizeof(struct pt_regs));
1313 struct task_struct * __cpuinit fork_idle(int cpu)
1315 struct task_struct *task;
1316 struct pt_regs regs;
1318 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL,
1319 &init_struct_pid, 0);
1321 init_idle(task, cpu);
1327 * Ok, this is the main fork-routine.
1329 * It copies the process, and if successful kick-starts
1330 * it and waits for it to finish using the VM if required.
1332 long do_fork(unsigned long clone_flags,
1333 unsigned long stack_start,
1334 struct pt_regs *regs,
1335 unsigned long stack_size,
1336 int __user *parent_tidptr,
1337 int __user *child_tidptr)
1339 struct task_struct *p;
1344 * We hope to recycle these flags after 2.6.26
1346 if (unlikely(clone_flags & CLONE_STOPPED)) {
1347 static int __read_mostly count = 100;
1349 if (count > 0 && printk_ratelimit()) {
1350 char comm[TASK_COMM_LEN];
1353 printk(KERN_INFO "fork(): process `%s' used deprecated "
1354 "clone flags 0x%lx\n",
1355 get_task_comm(comm, current),
1356 clone_flags & CLONE_STOPPED);
1361 * When called from kernel_thread, don't do user tracing stuff.
1363 if (likely(user_mode(regs)))
1364 trace = tracehook_prepare_clone(clone_flags);
1366 p = copy_process(clone_flags, stack_start, regs, stack_size,
1367 child_tidptr, NULL, trace);
1369 * Do this prior waking up the new thread - the thread pointer
1370 * might get invalid after that point, if the thread exits quickly.
1373 struct completion vfork;
1375 nr = task_pid_vnr(p);
1377 if (clone_flags & CLONE_PARENT_SETTID)
1378 put_user(nr, parent_tidptr);
1380 if (clone_flags & CLONE_VFORK) {
1381 p->vfork_done = &vfork;
1382 init_completion(&vfork);
1385 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1388 * We set PF_STARTING at creation in case tracing wants to
1389 * use this to distinguish a fully live task from one that
1390 * hasn't gotten to tracehook_report_clone() yet. Now we
1391 * clear it and set the child going.
1393 p->flags &= ~PF_STARTING;
1395 if (unlikely(clone_flags & CLONE_STOPPED)) {
1397 * We'll start up with an immediate SIGSTOP.
1399 sigaddset(&p->pending.signal, SIGSTOP);
1400 set_tsk_thread_flag(p, TIF_SIGPENDING);
1401 __set_task_state(p, TASK_STOPPED);
1403 wake_up_new_task(p, clone_flags);
1406 tracehook_report_clone_complete(trace, regs,
1407 clone_flags, nr, p);
1409 if (clone_flags & CLONE_VFORK) {
1410 freezer_do_not_count();
1411 wait_for_completion(&vfork);
1413 tracehook_report_vfork_done(p, nr);
1421 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1422 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1425 static void sighand_ctor(void *data)
1427 struct sighand_struct *sighand = data;
1429 spin_lock_init(&sighand->siglock);
1430 init_waitqueue_head(&sighand->signalfd_wqh);
1433 void __init proc_caches_init(void)
1435 sighand_cachep = kmem_cache_create("sighand_cache",
1436 sizeof(struct sighand_struct), 0,
1437 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1439 signal_cachep = kmem_cache_create("signal_cache",
1440 sizeof(struct signal_struct), 0,
1441 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1442 files_cachep = kmem_cache_create("files_cache",
1443 sizeof(struct files_struct), 0,
1444 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1445 fs_cachep = kmem_cache_create("fs_cache",
1446 sizeof(struct fs_struct), 0,
1447 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1448 vm_area_cachep = kmem_cache_create("vm_area_struct",
1449 sizeof(struct vm_area_struct), 0,
1451 mm_cachep = kmem_cache_create("mm_struct",
1452 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1453 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1457 * Check constraints on flags passed to the unshare system call and
1458 * force unsharing of additional process context as appropriate.
1460 static void check_unshare_flags(unsigned long *flags_ptr)
1463 * If unsharing a thread from a thread group, must also
1466 if (*flags_ptr & CLONE_THREAD)
1467 *flags_ptr |= CLONE_VM;
1470 * If unsharing vm, must also unshare signal handlers.
1472 if (*flags_ptr & CLONE_VM)
1473 *flags_ptr |= CLONE_SIGHAND;
1476 * If unsharing signal handlers and the task was created
1477 * using CLONE_THREAD, then must unshare the thread
1479 if ((*flags_ptr & CLONE_SIGHAND) &&
1480 (atomic_read(¤t->signal->count) > 1))
1481 *flags_ptr |= CLONE_THREAD;
1484 * If unsharing namespace, must also unshare filesystem information.
1486 if (*flags_ptr & CLONE_NEWNS)
1487 *flags_ptr |= CLONE_FS;
1491 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1493 static int unshare_thread(unsigned long unshare_flags)
1495 if (unshare_flags & CLONE_THREAD)
1502 * Unshare the filesystem structure if it is being shared
1504 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1506 struct fs_struct *fs = current->fs;
1508 if ((unshare_flags & CLONE_FS) &&
1509 (fs && atomic_read(&fs->count) > 1)) {
1510 *new_fsp = __copy_fs_struct(current->fs);
1519 * Unsharing of sighand is not supported yet
1521 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1523 struct sighand_struct *sigh = current->sighand;
1525 if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1532 * Unshare vm if it is being shared
1534 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1536 struct mm_struct *mm = current->mm;
1538 if ((unshare_flags & CLONE_VM) &&
1539 (mm && atomic_read(&mm->mm_users) > 1)) {
1547 * Unshare file descriptor table if it is being shared
1549 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1551 struct files_struct *fd = current->files;
1554 if ((unshare_flags & CLONE_FILES) &&
1555 (fd && atomic_read(&fd->count) > 1)) {
1556 *new_fdp = dup_fd(fd, &error);
1565 * unshare allows a process to 'unshare' part of the process
1566 * context which was originally shared using clone. copy_*
1567 * functions used by do_fork() cannot be used here directly
1568 * because they modify an inactive task_struct that is being
1569 * constructed. Here we are modifying the current, active,
1572 asmlinkage long sys_unshare(unsigned long unshare_flags)
1575 struct fs_struct *fs, *new_fs = NULL;
1576 struct sighand_struct *new_sigh = NULL;
1577 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1578 struct files_struct *fd, *new_fd = NULL;
1579 struct nsproxy *new_nsproxy = NULL;
1582 check_unshare_flags(&unshare_flags);
1584 /* Return -EINVAL for all unsupported flags */
1586 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1587 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1588 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1590 goto bad_unshare_out;
1593 * CLONE_NEWIPC must also detach from the undolist: after switching
1594 * to a new ipc namespace, the semaphore arrays from the old
1595 * namespace are unreachable.
1597 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1599 if ((err = unshare_thread(unshare_flags)))
1600 goto bad_unshare_out;
1601 if ((err = unshare_fs(unshare_flags, &new_fs)))
1602 goto bad_unshare_cleanup_thread;
1603 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1604 goto bad_unshare_cleanup_fs;
1605 if ((err = unshare_vm(unshare_flags, &new_mm)))
1606 goto bad_unshare_cleanup_sigh;
1607 if ((err = unshare_fd(unshare_flags, &new_fd)))
1608 goto bad_unshare_cleanup_vm;
1609 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1611 goto bad_unshare_cleanup_fd;
1613 if (new_fs || new_mm || new_fd || do_sysvsem || new_nsproxy) {
1616 * CLONE_SYSVSEM is equivalent to sys_exit().
1622 switch_task_namespaces(current, new_nsproxy);
1630 current->fs = new_fs;
1636 active_mm = current->active_mm;
1637 current->mm = new_mm;
1638 current->active_mm = new_mm;
1639 activate_mm(active_mm, new_mm);
1644 fd = current->files;
1645 current->files = new_fd;
1649 task_unlock(current);
1653 put_nsproxy(new_nsproxy);
1655 bad_unshare_cleanup_fd:
1657 put_files_struct(new_fd);
1659 bad_unshare_cleanup_vm:
1663 bad_unshare_cleanup_sigh:
1665 if (atomic_dec_and_test(&new_sigh->count))
1666 kmem_cache_free(sighand_cachep, new_sigh);
1668 bad_unshare_cleanup_fs:
1670 put_fs_struct(new_fs);
1672 bad_unshare_cleanup_thread:
1678 * Helper to unshare the files of the current task.
1679 * We don't want to expose copy_files internals to
1680 * the exec layer of the kernel.
1683 int unshare_files(struct files_struct **displaced)
1685 struct task_struct *task = current;
1686 struct files_struct *copy = NULL;
1689 error = unshare_fd(CLONE_FILES, ©);
1690 if (error || !copy) {
1694 *displaced = task->files;