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/cpu.h>
32 #include <linux/cpuset.h>
33 #include <linux/security.h>
34 #include <linux/swap.h>
35 #include <linux/syscalls.h>
36 #include <linux/jiffies.h>
37 #include <linux/futex.h>
38 #include <linux/ptrace.h>
39 #include <linux/mount.h>
40 #include <linux/audit.h>
41 #include <linux/profile.h>
42 #include <linux/rmap.h>
43 #include <linux/acct.h>
45 #include <asm/pgtable.h>
46 #include <asm/pgalloc.h>
47 #include <asm/uaccess.h>
48 #include <asm/mmu_context.h>
49 #include <asm/cacheflush.h>
50 #include <asm/tlbflush.h>
53 * Protected counters by write_lock_irq(&tasklist_lock)
55 unsigned long total_forks
; /* Handle normal Linux uptimes. */
56 int nr_threads
; /* The idle threads do not count.. */
58 int max_threads
; /* tunable limit on nr_threads */
60 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
62 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
64 EXPORT_SYMBOL(tasklist_lock
);
66 int nr_processes(void)
71 for_each_online_cpu(cpu
)
72 total
+= per_cpu(process_counts
, cpu
);
77 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
78 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
79 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
80 static kmem_cache_t
*task_struct_cachep
;
83 /* SLAB cache for signal_struct structures (tsk->signal) */
84 kmem_cache_t
*signal_cachep
;
86 /* SLAB cache for sighand_struct structures (tsk->sighand) */
87 kmem_cache_t
*sighand_cachep
;
89 /* SLAB cache for files_struct structures (tsk->files) */
90 kmem_cache_t
*files_cachep
;
92 /* SLAB cache for fs_struct structures (tsk->fs) */
93 kmem_cache_t
*fs_cachep
;
95 /* SLAB cache for vm_area_struct structures */
96 kmem_cache_t
*vm_area_cachep
;
98 /* SLAB cache for mm_struct structures (tsk->mm) */
99 static kmem_cache_t
*mm_cachep
;
101 void free_task(struct task_struct
*tsk
)
103 free_thread_info(tsk
->thread_info
);
104 free_task_struct(tsk
);
106 EXPORT_SYMBOL(free_task
);
108 void __put_task_struct(struct task_struct
*tsk
)
110 WARN_ON(!(tsk
->exit_state
& (EXIT_DEAD
| EXIT_ZOMBIE
)));
111 WARN_ON(atomic_read(&tsk
->usage
));
112 WARN_ON(tsk
== current
);
114 if (unlikely(tsk
->audit_context
))
116 security_task_free(tsk
);
118 put_group_info(tsk
->group_info
);
120 if (!profile_handoff_task(tsk
))
124 void __init
fork_init(unsigned long mempages
)
126 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
127 #ifndef ARCH_MIN_TASKALIGN
128 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
130 /* create a slab on which task_structs can be allocated */
132 kmem_cache_create("task_struct", sizeof(struct task_struct
),
133 ARCH_MIN_TASKALIGN
, SLAB_PANIC
, NULL
, NULL
);
137 * The default maximum number of threads is set to a safe
138 * value: the thread structures can take up at most half
141 max_threads
= mempages
/ (8 * THREAD_SIZE
/ PAGE_SIZE
);
144 * we need to allow at least 20 threads to boot a system
149 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
150 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
151 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
152 init_task
.signal
->rlim
[RLIMIT_NPROC
];
155 static struct task_struct
*dup_task_struct(struct task_struct
*orig
)
157 struct task_struct
*tsk
;
158 struct thread_info
*ti
;
160 prepare_to_copy(orig
);
162 tsk
= alloc_task_struct();
166 ti
= alloc_thread_info(tsk
);
168 free_task_struct(tsk
);
172 *ti
= *orig
->thread_info
;
174 tsk
->thread_info
= ti
;
177 /* One for us, one for whoever does the "release_task()" (usually parent) */
178 atomic_set(&tsk
->usage
,2);
179 atomic_set(&tsk
->fs_excl
, 0);
184 static inline int dup_mmap(struct mm_struct
* mm
, struct mm_struct
* oldmm
)
186 struct vm_area_struct
* mpnt
, *tmp
, **pprev
;
187 struct rb_node
**rb_link
, *rb_parent
;
189 unsigned long charge
;
190 struct mempolicy
*pol
;
192 down_write(&oldmm
->mmap_sem
);
193 flush_cache_mm(current
->mm
);
196 mm
->mmap_cache
= NULL
;
197 mm
->free_area_cache
= oldmm
->mmap_base
;
198 mm
->cached_hole_size
= ~0UL;
200 set_mm_counter(mm
, rss
, 0);
201 set_mm_counter(mm
, anon_rss
, 0);
202 cpus_clear(mm
->cpu_vm_mask
);
204 rb_link
= &mm
->mm_rb
.rb_node
;
208 for (mpnt
= current
->mm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
211 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
212 long pages
= vma_pages(mpnt
);
213 mm
->total_vm
-= pages
;
214 __vm_stat_account(mm
, mpnt
->vm_flags
, mpnt
->vm_file
,
219 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
220 unsigned int len
= (mpnt
->vm_end
- mpnt
->vm_start
) >> PAGE_SHIFT
;
221 if (security_vm_enough_memory(len
))
225 tmp
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
229 pol
= mpol_copy(vma_policy(mpnt
));
230 retval
= PTR_ERR(pol
);
232 goto fail_nomem_policy
;
233 vma_set_policy(tmp
, pol
);
234 tmp
->vm_flags
&= ~VM_LOCKED
;
240 struct inode
*inode
= file
->f_dentry
->d_inode
;
242 if (tmp
->vm_flags
& VM_DENYWRITE
)
243 atomic_dec(&inode
->i_writecount
);
245 /* insert tmp into the share list, just after mpnt */
246 spin_lock(&file
->f_mapping
->i_mmap_lock
);
247 tmp
->vm_truncate_count
= mpnt
->vm_truncate_count
;
248 flush_dcache_mmap_lock(file
->f_mapping
);
249 vma_prio_tree_add(tmp
, mpnt
);
250 flush_dcache_mmap_unlock(file
->f_mapping
);
251 spin_unlock(&file
->f_mapping
->i_mmap_lock
);
255 * Link in the new vma and copy the page table entries:
256 * link in first so that swapoff can see swap entries.
257 * Note that, exceptionally, here the vma is inserted
258 * without holding mm->mmap_sem.
260 spin_lock(&mm
->page_table_lock
);
262 pprev
= &tmp
->vm_next
;
264 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
265 rb_link
= &tmp
->vm_rb
.rb_right
;
266 rb_parent
= &tmp
->vm_rb
;
269 retval
= copy_page_range(mm
, current
->mm
, tmp
);
270 spin_unlock(&mm
->page_table_lock
);
272 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
273 tmp
->vm_ops
->open(tmp
);
281 flush_tlb_mm(current
->mm
);
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 spin_lock_init(&mm
->page_table_lock
);
326 rwlock_init(&mm
->ioctx_list_lock
);
327 mm
->ioctx_list
= NULL
;
328 mm
->default_kioctx
= (struct kioctx
)INIT_KIOCTX(mm
->default_kioctx
, *mm
);
329 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
330 mm
->cached_hole_size
= ~0UL;
332 if (likely(!mm_alloc_pgd(mm
))) {
341 * Allocate and initialize an mm_struct.
343 struct mm_struct
* mm_alloc(void)
345 struct mm_struct
* mm
;
349 memset(mm
, 0, sizeof(*mm
));
356 * Called when the last reference to the mm
357 * is dropped: either by a lazy thread or by
358 * mmput. Free the page directory and the mm.
360 void fastcall
__mmdrop(struct mm_struct
*mm
)
362 BUG_ON(mm
== &init_mm
);
369 * Decrement the use count and release all resources for an mm.
371 void mmput(struct mm_struct
*mm
)
373 if (atomic_dec_and_test(&mm
->mm_users
)) {
376 if (!list_empty(&mm
->mmlist
)) {
377 spin_lock(&mmlist_lock
);
378 list_del(&mm
->mmlist
);
379 spin_unlock(&mmlist_lock
);
385 EXPORT_SYMBOL_GPL(mmput
);
388 * get_task_mm - acquire a reference to the task's mm
390 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
391 * this kernel workthread has transiently adopted a user mm with use_mm,
392 * to do its AIO) is not set and if so returns a reference to it, after
393 * bumping up the use count. User must release the mm via mmput()
394 * after use. Typically used by /proc and ptrace.
396 struct mm_struct
*get_task_mm(struct task_struct
*task
)
398 struct mm_struct
*mm
;
403 if (task
->flags
& PF_BORROWED_MM
)
406 atomic_inc(&mm
->mm_users
);
411 EXPORT_SYMBOL_GPL(get_task_mm
);
413 /* Please note the differences between mmput and mm_release.
414 * mmput is called whenever we stop holding onto a mm_struct,
415 * error success whatever.
417 * mm_release is called after a mm_struct has been removed
418 * from the current process.
420 * This difference is important for error handling, when we
421 * only half set up a mm_struct for a new process and need to restore
422 * the old one. Because we mmput the new mm_struct before
423 * restoring the old one. . .
424 * Eric Biederman 10 January 1998
426 void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
428 struct completion
*vfork_done
= tsk
->vfork_done
;
430 /* Get rid of any cached register state */
431 deactivate_mm(tsk
, mm
);
433 /* notify parent sleeping on vfork() */
435 tsk
->vfork_done
= NULL
;
436 complete(vfork_done
);
438 if (tsk
->clear_child_tid
&& atomic_read(&mm
->mm_users
) > 1) {
439 u32 __user
* tidptr
= tsk
->clear_child_tid
;
440 tsk
->clear_child_tid
= NULL
;
443 * We don't check the error code - if userspace has
444 * not set up a proper pointer then tough luck.
447 sys_futex(tidptr
, FUTEX_WAKE
, 1, NULL
, NULL
, 0);
451 static int copy_mm(unsigned long clone_flags
, struct task_struct
* tsk
)
453 struct mm_struct
* mm
, *oldmm
;
456 tsk
->min_flt
= tsk
->maj_flt
= 0;
457 tsk
->nvcsw
= tsk
->nivcsw
= 0;
460 tsk
->active_mm
= NULL
;
463 * Are we cloning a kernel thread?
465 * We need to steal a active VM for that..
471 if (clone_flags
& CLONE_VM
) {
472 atomic_inc(&oldmm
->mm_users
);
475 * There are cases where the PTL is held to ensure no
476 * new threads start up in user mode using an mm, which
477 * allows optimizing out ipis; the tlb_gather_mmu code
480 spin_unlock_wait(&oldmm
->page_table_lock
);
489 /* Copy the current MM stuff.. */
490 memcpy(mm
, oldmm
, sizeof(*mm
));
494 if (init_new_context(tsk
,mm
))
497 retval
= dup_mmap(mm
, oldmm
);
501 mm
->hiwater_rss
= get_mm_counter(mm
,rss
);
502 mm
->hiwater_vm
= mm
->total_vm
;
516 * If init_new_context() failed, we cannot use mmput() to free the mm
517 * because it calls destroy_context()
524 static inline struct fs_struct
*__copy_fs_struct(struct fs_struct
*old
)
526 struct fs_struct
*fs
= kmem_cache_alloc(fs_cachep
, GFP_KERNEL
);
527 /* We don't need to lock fs - think why ;-) */
529 atomic_set(&fs
->count
, 1);
530 rwlock_init(&fs
->lock
);
531 fs
->umask
= old
->umask
;
532 read_lock(&old
->lock
);
533 fs
->rootmnt
= mntget(old
->rootmnt
);
534 fs
->root
= dget(old
->root
);
535 fs
->pwdmnt
= mntget(old
->pwdmnt
);
536 fs
->pwd
= dget(old
->pwd
);
538 fs
->altrootmnt
= mntget(old
->altrootmnt
);
539 fs
->altroot
= dget(old
->altroot
);
541 fs
->altrootmnt
= NULL
;
544 read_unlock(&old
->lock
);
549 struct fs_struct
*copy_fs_struct(struct fs_struct
*old
)
551 return __copy_fs_struct(old
);
554 EXPORT_SYMBOL_GPL(copy_fs_struct
);
556 static inline int copy_fs(unsigned long clone_flags
, struct task_struct
* tsk
)
558 if (clone_flags
& CLONE_FS
) {
559 atomic_inc(¤t
->fs
->count
);
562 tsk
->fs
= __copy_fs_struct(current
->fs
);
568 static int count_open_files(struct files_struct
*files
, int size
)
572 /* Find the last open fd */
573 for (i
= size
/(8*sizeof(long)); i
> 0; ) {
574 if (files
->open_fds
->fds_bits
[--i
])
577 i
= (i
+1) * 8 * sizeof(long);
581 static int copy_files(unsigned long clone_flags
, struct task_struct
* tsk
)
583 struct files_struct
*oldf
, *newf
;
584 struct file
**old_fds
, **new_fds
;
585 int open_files
, size
, i
, error
= 0, expand
;
588 * A background process may not have any files ...
590 oldf
= current
->files
;
594 if (clone_flags
& CLONE_FILES
) {
595 atomic_inc(&oldf
->count
);
600 * Note: we may be using current for both targets (See exec.c)
601 * This works because we cache current->files (old) as oldf. Don't
606 newf
= kmem_cache_alloc(files_cachep
, SLAB_KERNEL
);
610 atomic_set(&newf
->count
, 1);
612 spin_lock_init(&newf
->file_lock
);
614 newf
->max_fds
= NR_OPEN_DEFAULT
;
615 newf
->max_fdset
= __FD_SETSIZE
;
616 newf
->close_on_exec
= &newf
->close_on_exec_init
;
617 newf
->open_fds
= &newf
->open_fds_init
;
618 newf
->fd
= &newf
->fd_array
[0];
620 spin_lock(&oldf
->file_lock
);
622 open_files
= count_open_files(oldf
, oldf
->max_fdset
);
626 * Check whether we need to allocate a larger fd array or fd set.
627 * Note: we're not a clone task, so the open count won't change.
629 if (open_files
> newf
->max_fdset
) {
633 if (open_files
> newf
->max_fds
) {
638 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
640 spin_unlock(&oldf
->file_lock
);
641 spin_lock(&newf
->file_lock
);
642 error
= expand_files(newf
, open_files
-1);
643 spin_unlock(&newf
->file_lock
);
646 spin_lock(&oldf
->file_lock
);
652 memcpy(newf
->open_fds
->fds_bits
, oldf
->open_fds
->fds_bits
, open_files
/8);
653 memcpy(newf
->close_on_exec
->fds_bits
, oldf
->close_on_exec
->fds_bits
, open_files
/8);
655 for (i
= open_files
; i
!= 0; i
--) {
656 struct file
*f
= *old_fds
++;
661 * The fd may be claimed in the fd bitmap but not yet
662 * instantiated in the files array if a sibling thread
663 * is partway through open(). So make sure that this
664 * fd is available to the new process.
666 FD_CLR(open_files
- i
, newf
->open_fds
);
670 spin_unlock(&oldf
->file_lock
);
672 /* compute the remainder to be cleared */
673 size
= (newf
->max_fds
- open_files
) * sizeof(struct file
*);
675 /* This is long word aligned thus could use a optimized version */
676 memset(new_fds
, 0, size
);
678 if (newf
->max_fdset
> open_files
) {
679 int left
= (newf
->max_fdset
-open_files
)/8;
680 int start
= open_files
/ (8 * sizeof(unsigned long));
682 memset(&newf
->open_fds
->fds_bits
[start
], 0, left
);
683 memset(&newf
->close_on_exec
->fds_bits
[start
], 0, left
);
692 free_fdset (newf
->close_on_exec
, newf
->max_fdset
);
693 free_fdset (newf
->open_fds
, newf
->max_fdset
);
694 free_fd_array(newf
->fd
, newf
->max_fds
);
695 kmem_cache_free(files_cachep
, newf
);
700 * Helper to unshare the files of the current task.
701 * We don't want to expose copy_files internals to
702 * the exec layer of the kernel.
705 int unshare_files(void)
707 struct files_struct
*files
= current
->files
;
713 /* This can race but the race causes us to copy when we don't
714 need to and drop the copy */
715 if(atomic_read(&files
->count
) == 1)
717 atomic_inc(&files
->count
);
720 rc
= copy_files(0, current
);
722 current
->files
= files
;
726 EXPORT_SYMBOL(unshare_files
);
728 static inline int copy_sighand(unsigned long clone_flags
, struct task_struct
* tsk
)
730 struct sighand_struct
*sig
;
732 if (clone_flags
& (CLONE_SIGHAND
| CLONE_THREAD
)) {
733 atomic_inc(¤t
->sighand
->count
);
736 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
740 spin_lock_init(&sig
->siglock
);
741 atomic_set(&sig
->count
, 1);
742 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
746 static inline int copy_signal(unsigned long clone_flags
, struct task_struct
* tsk
)
748 struct signal_struct
*sig
;
751 if (clone_flags
& CLONE_THREAD
) {
752 atomic_inc(¤t
->signal
->count
);
753 atomic_inc(¤t
->signal
->live
);
756 sig
= kmem_cache_alloc(signal_cachep
, GFP_KERNEL
);
761 ret
= copy_thread_group_keys(tsk
);
763 kmem_cache_free(signal_cachep
, sig
);
767 atomic_set(&sig
->count
, 1);
768 atomic_set(&sig
->live
, 1);
769 init_waitqueue_head(&sig
->wait_chldexit
);
771 sig
->group_exit_code
= 0;
772 sig
->group_exit_task
= NULL
;
773 sig
->group_stop_count
= 0;
774 sig
->curr_target
= NULL
;
775 init_sigpending(&sig
->shared_pending
);
776 INIT_LIST_HEAD(&sig
->posix_timers
);
778 sig
->it_real_value
= sig
->it_real_incr
= 0;
779 sig
->real_timer
.function
= it_real_fn
;
780 sig
->real_timer
.data
= (unsigned long) tsk
;
781 init_timer(&sig
->real_timer
);
783 sig
->it_virt_expires
= cputime_zero
;
784 sig
->it_virt_incr
= cputime_zero
;
785 sig
->it_prof_expires
= cputime_zero
;
786 sig
->it_prof_incr
= cputime_zero
;
788 sig
->tty
= current
->signal
->tty
;
789 sig
->pgrp
= process_group(current
);
790 sig
->session
= current
->signal
->session
;
791 sig
->leader
= 0; /* session leadership doesn't inherit */
792 sig
->tty_old_pgrp
= 0;
794 sig
->utime
= sig
->stime
= sig
->cutime
= sig
->cstime
= cputime_zero
;
795 sig
->nvcsw
= sig
->nivcsw
= sig
->cnvcsw
= sig
->cnivcsw
= 0;
796 sig
->min_flt
= sig
->maj_flt
= sig
->cmin_flt
= sig
->cmaj_flt
= 0;
798 INIT_LIST_HEAD(&sig
->cpu_timers
[0]);
799 INIT_LIST_HEAD(&sig
->cpu_timers
[1]);
800 INIT_LIST_HEAD(&sig
->cpu_timers
[2]);
802 task_lock(current
->group_leader
);
803 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
804 task_unlock(current
->group_leader
);
806 if (sig
->rlim
[RLIMIT_CPU
].rlim_cur
!= RLIM_INFINITY
) {
808 * New sole thread in the process gets an expiry time
809 * of the whole CPU time limit.
811 tsk
->it_prof_expires
=
812 secs_to_cputime(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
818 static inline void copy_flags(unsigned long clone_flags
, struct task_struct
*p
)
820 unsigned long new_flags
= p
->flags
;
822 new_flags
&= ~PF_SUPERPRIV
;
823 new_flags
|= PF_FORKNOEXEC
;
824 if (!(clone_flags
& CLONE_PTRACE
))
826 p
->flags
= new_flags
;
829 asmlinkage
long sys_set_tid_address(int __user
*tidptr
)
831 current
->clear_child_tid
= tidptr
;
837 * This creates a new process as a copy of the old one,
838 * but does not actually start it yet.
840 * It copies the registers, and all the appropriate
841 * parts of the process environment (as per the clone
842 * flags). The actual kick-off is left to the caller.
844 static task_t
*copy_process(unsigned long clone_flags
,
845 unsigned long stack_start
,
846 struct pt_regs
*regs
,
847 unsigned long stack_size
,
848 int __user
*parent_tidptr
,
849 int __user
*child_tidptr
,
853 struct task_struct
*p
= NULL
;
855 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
856 return ERR_PTR(-EINVAL
);
859 * Thread groups must share signals as well, and detached threads
860 * can only be started up within the thread group.
862 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
863 return ERR_PTR(-EINVAL
);
866 * Shared signal handlers imply shared VM. By way of the above,
867 * thread groups also imply shared VM. Blocking this case allows
868 * for various simplifications in other code.
870 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
871 return ERR_PTR(-EINVAL
);
873 retval
= security_task_create(clone_flags
);
878 p
= dup_task_struct(current
);
883 if (atomic_read(&p
->user
->processes
) >=
884 p
->signal
->rlim
[RLIMIT_NPROC
].rlim_cur
) {
885 if (!capable(CAP_SYS_ADMIN
) && !capable(CAP_SYS_RESOURCE
) &&
886 p
->user
!= &root_user
)
890 atomic_inc(&p
->user
->__count
);
891 atomic_inc(&p
->user
->processes
);
892 get_group_info(p
->group_info
);
895 * If multiple threads are within copy_process(), then this check
896 * triggers too late. This doesn't hurt, the check is only there
897 * to stop root fork bombs.
899 if (nr_threads
>= max_threads
)
900 goto bad_fork_cleanup_count
;
902 if (!try_module_get(p
->thread_info
->exec_domain
->module
))
903 goto bad_fork_cleanup_count
;
905 if (p
->binfmt
&& !try_module_get(p
->binfmt
->module
))
906 goto bad_fork_cleanup_put_domain
;
909 copy_flags(clone_flags
, p
);
912 if (clone_flags
& CLONE_PARENT_SETTID
)
913 if (put_user(p
->pid
, parent_tidptr
))
914 goto bad_fork_cleanup
;
916 p
->proc_dentry
= NULL
;
918 INIT_LIST_HEAD(&p
->children
);
919 INIT_LIST_HEAD(&p
->sibling
);
920 p
->vfork_done
= NULL
;
921 spin_lock_init(&p
->alloc_lock
);
922 spin_lock_init(&p
->proc_lock
);
924 clear_tsk_thread_flag(p
, TIF_SIGPENDING
);
925 init_sigpending(&p
->pending
);
927 p
->utime
= cputime_zero
;
928 p
->stime
= cputime_zero
;
930 p
->rchar
= 0; /* I/O counter: bytes read */
931 p
->wchar
= 0; /* I/O counter: bytes written */
932 p
->syscr
= 0; /* I/O counter: read syscalls */
933 p
->syscw
= 0; /* I/O counter: write syscalls */
934 acct_clear_integrals(p
);
936 p
->it_virt_expires
= cputime_zero
;
937 p
->it_prof_expires
= cputime_zero
;
938 p
->it_sched_expires
= 0;
939 INIT_LIST_HEAD(&p
->cpu_timers
[0]);
940 INIT_LIST_HEAD(&p
->cpu_timers
[1]);
941 INIT_LIST_HEAD(&p
->cpu_timers
[2]);
943 p
->lock_depth
= -1; /* -1 = no lock */
944 do_posix_clock_monotonic_gettime(&p
->start_time
);
946 p
->io_context
= NULL
;
948 p
->audit_context
= NULL
;
950 p
->mempolicy
= mpol_copy(p
->mempolicy
);
951 if (IS_ERR(p
->mempolicy
)) {
952 retval
= PTR_ERR(p
->mempolicy
);
954 goto bad_fork_cleanup
;
959 if (clone_flags
& CLONE_THREAD
)
960 p
->tgid
= current
->tgid
;
962 if ((retval
= security_task_alloc(p
)))
963 goto bad_fork_cleanup_policy
;
964 if ((retval
= audit_alloc(p
)))
965 goto bad_fork_cleanup_security
;
966 /* copy all the process information */
967 if ((retval
= copy_semundo(clone_flags
, p
)))
968 goto bad_fork_cleanup_audit
;
969 if ((retval
= copy_files(clone_flags
, p
)))
970 goto bad_fork_cleanup_semundo
;
971 if ((retval
= copy_fs(clone_flags
, p
)))
972 goto bad_fork_cleanup_files
;
973 if ((retval
= copy_sighand(clone_flags
, p
)))
974 goto bad_fork_cleanup_fs
;
975 if ((retval
= copy_signal(clone_flags
, p
)))
976 goto bad_fork_cleanup_sighand
;
977 if ((retval
= copy_mm(clone_flags
, p
)))
978 goto bad_fork_cleanup_signal
;
979 if ((retval
= copy_keys(clone_flags
, p
)))
980 goto bad_fork_cleanup_mm
;
981 if ((retval
= copy_namespace(clone_flags
, p
)))
982 goto bad_fork_cleanup_keys
;
983 retval
= copy_thread(0, clone_flags
, stack_start
, stack_size
, p
, regs
);
985 goto bad_fork_cleanup_namespace
;
987 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? child_tidptr
: NULL
;
989 * Clear TID on mm_release()?
991 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? child_tidptr
: NULL
;
994 * Syscall tracing should be turned off in the child regardless
997 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
998 #ifdef TIF_SYSCALL_EMU
999 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
1002 /* Our parent execution domain becomes current domain
1003 These must match for thread signalling to apply */
1005 p
->parent_exec_id
= p
->self_exec_id
;
1007 /* ok, now we should be set up.. */
1008 p
->exit_signal
= (clone_flags
& CLONE_THREAD
) ? -1 : (clone_flags
& CSIGNAL
);
1009 p
->pdeath_signal
= 0;
1013 * Ok, make it visible to the rest of the system.
1014 * We dont wake it up yet.
1016 p
->group_leader
= p
;
1017 INIT_LIST_HEAD(&p
->ptrace_children
);
1018 INIT_LIST_HEAD(&p
->ptrace_list
);
1020 /* Perform scheduler related setup. Assign this task to a CPU. */
1021 sched_fork(p
, clone_flags
);
1023 /* Need tasklist lock for parent etc handling! */
1024 write_lock_irq(&tasklist_lock
);
1027 * The task hasn't been attached yet, so its cpus_allowed mask will
1028 * not be changed, nor will its assigned CPU.
1030 * The cpus_allowed mask of the parent may have changed after it was
1031 * copied first time - so re-copy it here, then check the child's CPU
1032 * to ensure it is on a valid CPU (and if not, just force it back to
1033 * parent's CPU). This avoids alot of nasty races.
1035 p
->cpus_allowed
= current
->cpus_allowed
;
1036 if (unlikely(!cpu_isset(task_cpu(p
), p
->cpus_allowed
)))
1037 set_task_cpu(p
, smp_processor_id());
1040 * Check for pending SIGKILL! The new thread should not be allowed
1041 * to slip out of an OOM kill. (or normal SIGKILL.)
1043 if (sigismember(¤t
->pending
.signal
, SIGKILL
)) {
1044 write_unlock_irq(&tasklist_lock
);
1046 goto bad_fork_cleanup_namespace
;
1049 /* CLONE_PARENT re-uses the old parent */
1050 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
))
1051 p
->real_parent
= current
->real_parent
;
1053 p
->real_parent
= current
;
1054 p
->parent
= p
->real_parent
;
1056 if (clone_flags
& CLONE_THREAD
) {
1057 spin_lock(¤t
->sighand
->siglock
);
1059 * Important: if an exit-all has been started then
1060 * do not create this new thread - the whole thread
1061 * group is supposed to exit anyway.
1063 if (current
->signal
->flags
& SIGNAL_GROUP_EXIT
) {
1064 spin_unlock(¤t
->sighand
->siglock
);
1065 write_unlock_irq(&tasklist_lock
);
1067 goto bad_fork_cleanup_namespace
;
1069 p
->group_leader
= current
->group_leader
;
1071 if (current
->signal
->group_stop_count
> 0) {
1073 * There is an all-stop in progress for the group.
1074 * We ourselves will stop as soon as we check signals.
1075 * Make the new thread part of that group stop too.
1077 current
->signal
->group_stop_count
++;
1078 set_tsk_thread_flag(p
, TIF_SIGPENDING
);
1081 if (!cputime_eq(current
->signal
->it_virt_expires
,
1083 !cputime_eq(current
->signal
->it_prof_expires
,
1085 current
->signal
->rlim
[RLIMIT_CPU
].rlim_cur
!= RLIM_INFINITY
||
1086 !list_empty(¤t
->signal
->cpu_timers
[0]) ||
1087 !list_empty(¤t
->signal
->cpu_timers
[1]) ||
1088 !list_empty(¤t
->signal
->cpu_timers
[2])) {
1090 * Have child wake up on its first tick to check
1091 * for process CPU timers.
1093 p
->it_prof_expires
= jiffies_to_cputime(1);
1096 spin_unlock(¤t
->sighand
->siglock
);
1102 p
->ioprio
= current
->ioprio
;
1105 if (unlikely(p
->ptrace
& PT_PTRACED
))
1106 __ptrace_link(p
, current
->parent
);
1110 attach_pid(p
, PIDTYPE_PID
, p
->pid
);
1111 attach_pid(p
, PIDTYPE_TGID
, p
->tgid
);
1112 if (thread_group_leader(p
)) {
1113 attach_pid(p
, PIDTYPE_PGID
, process_group(p
));
1114 attach_pid(p
, PIDTYPE_SID
, p
->signal
->session
);
1116 __get_cpu_var(process_counts
)++;
1119 if (!current
->signal
->tty
&& p
->signal
->tty
)
1120 p
->signal
->tty
= NULL
;
1124 write_unlock_irq(&tasklist_lock
);
1129 return ERR_PTR(retval
);
1132 bad_fork_cleanup_namespace
:
1134 bad_fork_cleanup_keys
:
1136 bad_fork_cleanup_mm
:
1139 bad_fork_cleanup_signal
:
1141 bad_fork_cleanup_sighand
:
1143 bad_fork_cleanup_fs
:
1144 exit_fs(p
); /* blocking */
1145 bad_fork_cleanup_files
:
1146 exit_files(p
); /* blocking */
1147 bad_fork_cleanup_semundo
:
1149 bad_fork_cleanup_audit
:
1151 bad_fork_cleanup_security
:
1152 security_task_free(p
);
1153 bad_fork_cleanup_policy
:
1155 mpol_free(p
->mempolicy
);
1159 module_put(p
->binfmt
->module
);
1160 bad_fork_cleanup_put_domain
:
1161 module_put(p
->thread_info
->exec_domain
->module
);
1162 bad_fork_cleanup_count
:
1163 put_group_info(p
->group_info
);
1164 atomic_dec(&p
->user
->processes
);
1171 struct pt_regs
* __devinit
__attribute__((weak
)) idle_regs(struct pt_regs
*regs
)
1173 memset(regs
, 0, sizeof(struct pt_regs
));
1177 task_t
* __devinit
fork_idle(int cpu
)
1180 struct pt_regs regs
;
1182 task
= copy_process(CLONE_VM
, 0, idle_regs(®s
), 0, NULL
, NULL
, 0);
1184 return ERR_PTR(-ENOMEM
);
1185 init_idle(task
, cpu
);
1186 unhash_process(task
);
1190 static inline int fork_traceflag (unsigned clone_flags
)
1192 if (clone_flags
& CLONE_UNTRACED
)
1194 else if (clone_flags
& CLONE_VFORK
) {
1195 if (current
->ptrace
& PT_TRACE_VFORK
)
1196 return PTRACE_EVENT_VFORK
;
1197 } else if ((clone_flags
& CSIGNAL
) != SIGCHLD
) {
1198 if (current
->ptrace
& PT_TRACE_CLONE
)
1199 return PTRACE_EVENT_CLONE
;
1200 } else if (current
->ptrace
& PT_TRACE_FORK
)
1201 return PTRACE_EVENT_FORK
;
1207 * Ok, this is the main fork-routine.
1209 * It copies the process, and if successful kick-starts
1210 * it and waits for it to finish using the VM if required.
1212 long do_fork(unsigned long clone_flags
,
1213 unsigned long stack_start
,
1214 struct pt_regs
*regs
,
1215 unsigned long stack_size
,
1216 int __user
*parent_tidptr
,
1217 int __user
*child_tidptr
)
1219 struct task_struct
*p
;
1221 long pid
= alloc_pidmap();
1225 if (unlikely(current
->ptrace
)) {
1226 trace
= fork_traceflag (clone_flags
);
1228 clone_flags
|= CLONE_PTRACE
;
1231 p
= copy_process(clone_flags
, stack_start
, regs
, stack_size
, parent_tidptr
, child_tidptr
, pid
);
1233 * Do this prior waking up the new thread - the thread pointer
1234 * might get invalid after that point, if the thread exits quickly.
1237 struct completion vfork
;
1239 if (clone_flags
& CLONE_VFORK
) {
1240 p
->vfork_done
= &vfork
;
1241 init_completion(&vfork
);
1244 if ((p
->ptrace
& PT_PTRACED
) || (clone_flags
& CLONE_STOPPED
)) {
1246 * We'll start up with an immediate SIGSTOP.
1248 sigaddset(&p
->pending
.signal
, SIGSTOP
);
1249 set_tsk_thread_flag(p
, TIF_SIGPENDING
);
1252 if (!(clone_flags
& CLONE_STOPPED
))
1253 wake_up_new_task(p
, clone_flags
);
1255 p
->state
= TASK_STOPPED
;
1257 if (unlikely (trace
)) {
1258 current
->ptrace_message
= pid
;
1259 ptrace_notify ((trace
<< 8) | SIGTRAP
);
1262 if (clone_flags
& CLONE_VFORK
) {
1263 wait_for_completion(&vfork
);
1264 if (unlikely (current
->ptrace
& PT_TRACE_VFORK_DONE
))
1265 ptrace_notify ((PTRACE_EVENT_VFORK_DONE
<< 8) | SIGTRAP
);
1274 void __init
proc_caches_init(void)
1276 sighand_cachep
= kmem_cache_create("sighand_cache",
1277 sizeof(struct sighand_struct
), 0,
1278 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
1279 signal_cachep
= kmem_cache_create("signal_cache",
1280 sizeof(struct signal_struct
), 0,
1281 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
1282 files_cachep
= kmem_cache_create("files_cache",
1283 sizeof(struct files_struct
), 0,
1284 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
1285 fs_cachep
= kmem_cache_create("fs_cache",
1286 sizeof(struct fs_struct
), 0,
1287 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
1288 vm_area_cachep
= kmem_cache_create("vm_area_struct",
1289 sizeof(struct vm_area_struct
), 0,
1290 SLAB_PANIC
, NULL
, NULL
);
1291 mm_cachep
= kmem_cache_create("mm_struct",
1292 sizeof(struct mm_struct
), 0,
1293 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);