include: replace linux/module.h with "struct module" wherever possible
[linux-2.6/next.git] / kernel / fork.c
blobd0a56f041d4aa4b8cea6410c0e0e6ba67fdef144
1 /*
2 * linux/kernel/fork.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 /*
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/export.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/module.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>
31 #include <linux/fs.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/futex.h>
42 #include <linux/compat.h>
43 #include <linux/kthread.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/profile.h>
52 #include <linux/rmap.h>
53 #include <linux/ksm.h>
54 #include <linux/acct.h>
55 #include <linux/tsacct_kern.h>
56 #include <linux/cn_proc.h>
57 #include <linux/freezer.h>
58 #include <linux/delayacct.h>
59 #include <linux/taskstats_kern.h>
60 #include <linux/random.h>
61 #include <linux/tty.h>
62 #include <linux/blkdev.h>
63 #include <linux/fs_struct.h>
64 #include <linux/magic.h>
65 #include <linux/perf_event.h>
66 #include <linux/posix-timers.h>
67 #include <linux/user-return-notifier.h>
68 #include <linux/oom.h>
69 #include <linux/khugepaged.h>
71 #include <asm/pgtable.h>
72 #include <asm/pgalloc.h>
73 #include <asm/uaccess.h>
74 #include <asm/mmu_context.h>
75 #include <asm/cacheflush.h>
76 #include <asm/tlbflush.h>
78 #include <trace/events/sched.h>
81 * Protected counters by write_lock_irq(&tasklist_lock)
83 unsigned long total_forks; /* Handle normal Linux uptimes. */
84 int nr_threads; /* The idle threads do not count.. */
86 int max_threads; /* tunable limit on nr_threads */
88 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
90 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
92 #ifdef CONFIG_PROVE_RCU
93 int lockdep_tasklist_lock_is_held(void)
95 return lockdep_is_held(&tasklist_lock);
97 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
98 #endif /* #ifdef CONFIG_PROVE_RCU */
100 int nr_processes(void)
102 int cpu;
103 int total = 0;
105 for_each_possible_cpu(cpu)
106 total += per_cpu(process_counts, cpu);
108 return total;
111 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
112 # define alloc_task_struct_node(node) \
113 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
114 # define free_task_struct(tsk) \
115 kmem_cache_free(task_struct_cachep, (tsk))
116 static struct kmem_cache *task_struct_cachep;
117 #endif
119 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
120 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
121 int node)
123 #ifdef CONFIG_DEBUG_STACK_USAGE
124 gfp_t mask = GFP_KERNEL | __GFP_ZERO;
125 #else
126 gfp_t mask = GFP_KERNEL;
127 #endif
128 struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
130 return page ? page_address(page) : NULL;
133 static inline void free_thread_info(struct thread_info *ti)
135 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
137 #endif
139 /* SLAB cache for signal_struct structures (tsk->signal) */
140 static struct kmem_cache *signal_cachep;
142 /* SLAB cache for sighand_struct structures (tsk->sighand) */
143 struct kmem_cache *sighand_cachep;
145 /* SLAB cache for files_struct structures (tsk->files) */
146 struct kmem_cache *files_cachep;
148 /* SLAB cache for fs_struct structures (tsk->fs) */
149 struct kmem_cache *fs_cachep;
151 /* SLAB cache for vm_area_struct structures */
152 struct kmem_cache *vm_area_cachep;
154 /* SLAB cache for mm_struct structures (tsk->mm) */
155 static struct kmem_cache *mm_cachep;
157 static void account_kernel_stack(struct thread_info *ti, int account)
159 struct zone *zone = page_zone(virt_to_page(ti));
161 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
164 void free_task(struct task_struct *tsk)
166 prop_local_destroy_single(&tsk->dirties);
167 account_kernel_stack(tsk->stack, -1);
168 free_thread_info(tsk->stack);
169 rt_mutex_debug_task_free(tsk);
170 ftrace_graph_exit_task(tsk);
171 free_task_struct(tsk);
173 EXPORT_SYMBOL(free_task);
175 static inline void free_signal_struct(struct signal_struct *sig)
177 taskstats_tgid_free(sig);
178 sched_autogroup_exit(sig);
179 kmem_cache_free(signal_cachep, sig);
182 static inline void put_signal_struct(struct signal_struct *sig)
184 if (atomic_dec_and_test(&sig->sigcnt))
185 free_signal_struct(sig);
188 void __put_task_struct(struct task_struct *tsk)
190 WARN_ON(!tsk->exit_state);
191 WARN_ON(atomic_read(&tsk->usage));
192 WARN_ON(tsk == current);
194 exit_creds(tsk);
195 delayacct_tsk_free(tsk);
196 put_signal_struct(tsk->signal);
198 if (!profile_handoff_task(tsk))
199 free_task(tsk);
201 EXPORT_SYMBOL_GPL(__put_task_struct);
204 * macro override instead of weak attribute alias, to workaround
205 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
207 #ifndef arch_task_cache_init
208 #define arch_task_cache_init()
209 #endif
211 void __init fork_init(unsigned long mempages)
213 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
214 #ifndef ARCH_MIN_TASKALIGN
215 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
216 #endif
217 /* create a slab on which task_structs can be allocated */
218 task_struct_cachep =
219 kmem_cache_create("task_struct", sizeof(struct task_struct),
220 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
221 #endif
223 /* do the arch specific task caches init */
224 arch_task_cache_init();
227 * The default maximum number of threads is set to a safe
228 * value: the thread structures can take up at most half
229 * of memory.
231 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
234 * we need to allow at least 20 threads to boot a system
236 if (max_threads < 20)
237 max_threads = 20;
239 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
240 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
241 init_task.signal->rlim[RLIMIT_SIGPENDING] =
242 init_task.signal->rlim[RLIMIT_NPROC];
245 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
246 struct task_struct *src)
248 *dst = *src;
249 return 0;
252 static struct task_struct *dup_task_struct(struct task_struct *orig)
254 struct task_struct *tsk;
255 struct thread_info *ti;
256 unsigned long *stackend;
257 int node = tsk_fork_get_node(orig);
258 int err;
260 prepare_to_copy(orig);
262 tsk = alloc_task_struct_node(node);
263 if (!tsk)
264 return NULL;
266 ti = alloc_thread_info_node(tsk, node);
267 if (!ti) {
268 free_task_struct(tsk);
269 return NULL;
272 err = arch_dup_task_struct(tsk, orig);
273 if (err)
274 goto out;
276 tsk->stack = ti;
278 err = prop_local_init_single(&tsk->dirties);
279 if (err)
280 goto out;
282 setup_thread_stack(tsk, orig);
283 clear_user_return_notifier(tsk);
284 clear_tsk_need_resched(tsk);
285 stackend = end_of_stack(tsk);
286 *stackend = STACK_END_MAGIC; /* for overflow detection */
288 #ifdef CONFIG_CC_STACKPROTECTOR
289 tsk->stack_canary = get_random_int();
290 #endif
293 * One for us, one for whoever does the "release_task()" (usually
294 * parent)
296 atomic_set(&tsk->usage, 2);
297 #ifdef CONFIG_BLK_DEV_IO_TRACE
298 tsk->btrace_seq = 0;
299 #endif
300 tsk->splice_pipe = NULL;
302 account_kernel_stack(ti, 1);
304 return tsk;
306 out:
307 free_thread_info(ti);
308 free_task_struct(tsk);
309 return NULL;
312 #ifdef CONFIG_MMU
313 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
315 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
316 struct rb_node **rb_link, *rb_parent;
317 int retval;
318 unsigned long charge;
319 struct mempolicy *pol;
321 down_write(&oldmm->mmap_sem);
322 flush_cache_dup_mm(oldmm);
324 * Not linked in yet - no deadlock potential:
326 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
328 mm->locked_vm = 0;
329 mm->mmap = NULL;
330 mm->mmap_cache = NULL;
331 mm->free_area_cache = oldmm->mmap_base;
332 mm->cached_hole_size = ~0UL;
333 mm->map_count = 0;
334 cpumask_clear(mm_cpumask(mm));
335 mm->mm_rb = RB_ROOT;
336 rb_link = &mm->mm_rb.rb_node;
337 rb_parent = NULL;
338 pprev = &mm->mmap;
339 retval = ksm_fork(mm, oldmm);
340 if (retval)
341 goto out;
342 retval = khugepaged_fork(mm, oldmm);
343 if (retval)
344 goto out;
346 prev = NULL;
347 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
348 struct file *file;
350 if (mpnt->vm_flags & VM_DONTCOPY) {
351 long pages = vma_pages(mpnt);
352 mm->total_vm -= pages;
353 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
354 -pages);
355 continue;
357 charge = 0;
358 if (mpnt->vm_flags & VM_ACCOUNT) {
359 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
360 if (security_vm_enough_memory(len))
361 goto fail_nomem;
362 charge = len;
364 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
365 if (!tmp)
366 goto fail_nomem;
367 *tmp = *mpnt;
368 INIT_LIST_HEAD(&tmp->anon_vma_chain);
369 pol = mpol_dup(vma_policy(mpnt));
370 retval = PTR_ERR(pol);
371 if (IS_ERR(pol))
372 goto fail_nomem_policy;
373 vma_set_policy(tmp, pol);
374 tmp->vm_mm = mm;
375 if (anon_vma_fork(tmp, mpnt))
376 goto fail_nomem_anon_vma_fork;
377 tmp->vm_flags &= ~VM_LOCKED;
378 tmp->vm_next = tmp->vm_prev = NULL;
379 file = tmp->vm_file;
380 if (file) {
381 struct inode *inode = file->f_path.dentry->d_inode;
382 struct address_space *mapping = file->f_mapping;
384 get_file(file);
385 if (tmp->vm_flags & VM_DENYWRITE)
386 atomic_dec(&inode->i_writecount);
387 mutex_lock(&mapping->i_mmap_mutex);
388 if (tmp->vm_flags & VM_SHARED)
389 mapping->i_mmap_writable++;
390 flush_dcache_mmap_lock(mapping);
391 /* insert tmp into the share list, just after mpnt */
392 vma_prio_tree_add(tmp, mpnt);
393 flush_dcache_mmap_unlock(mapping);
394 mutex_unlock(&mapping->i_mmap_mutex);
398 * Clear hugetlb-related page reserves for children. This only
399 * affects MAP_PRIVATE mappings. Faults generated by the child
400 * are not guaranteed to succeed, even if read-only
402 if (is_vm_hugetlb_page(tmp))
403 reset_vma_resv_huge_pages(tmp);
406 * Link in the new vma and copy the page table entries.
408 *pprev = tmp;
409 pprev = &tmp->vm_next;
410 tmp->vm_prev = prev;
411 prev = tmp;
413 __vma_link_rb(mm, tmp, rb_link, rb_parent);
414 rb_link = &tmp->vm_rb.rb_right;
415 rb_parent = &tmp->vm_rb;
417 mm->map_count++;
418 retval = copy_page_range(mm, oldmm, mpnt);
420 if (tmp->vm_ops && tmp->vm_ops->open)
421 tmp->vm_ops->open(tmp);
423 if (retval)
424 goto out;
426 /* a new mm has just been created */
427 arch_dup_mmap(oldmm, mm);
428 retval = 0;
429 out:
430 up_write(&mm->mmap_sem);
431 flush_tlb_mm(oldmm);
432 up_write(&oldmm->mmap_sem);
433 return retval;
434 fail_nomem_anon_vma_fork:
435 mpol_put(pol);
436 fail_nomem_policy:
437 kmem_cache_free(vm_area_cachep, tmp);
438 fail_nomem:
439 retval = -ENOMEM;
440 vm_unacct_memory(charge);
441 goto out;
444 static inline int mm_alloc_pgd(struct mm_struct *mm)
446 mm->pgd = pgd_alloc(mm);
447 if (unlikely(!mm->pgd))
448 return -ENOMEM;
449 return 0;
452 static inline void mm_free_pgd(struct mm_struct *mm)
454 pgd_free(mm, mm->pgd);
456 #else
457 #define dup_mmap(mm, oldmm) (0)
458 #define mm_alloc_pgd(mm) (0)
459 #define mm_free_pgd(mm)
460 #endif /* CONFIG_MMU */
462 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
464 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
465 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
467 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
469 static int __init coredump_filter_setup(char *s)
471 default_dump_filter =
472 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
473 MMF_DUMP_FILTER_MASK;
474 return 1;
477 __setup("coredump_filter=", coredump_filter_setup);
479 #include <linux/init_task.h>
481 static void mm_init_aio(struct mm_struct *mm)
483 #ifdef CONFIG_AIO
484 spin_lock_init(&mm->ioctx_lock);
485 INIT_HLIST_HEAD(&mm->ioctx_list);
486 #endif
489 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
491 atomic_set(&mm->mm_users, 1);
492 atomic_set(&mm->mm_count, 1);
493 init_rwsem(&mm->mmap_sem);
494 INIT_LIST_HEAD(&mm->mmlist);
495 mm->flags = (current->mm) ?
496 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
497 mm->core_state = NULL;
498 mm->nr_ptes = 0;
499 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
500 spin_lock_init(&mm->page_table_lock);
501 mm->free_area_cache = TASK_UNMAPPED_BASE;
502 mm->cached_hole_size = ~0UL;
503 mm_init_aio(mm);
504 mm_init_owner(mm, p);
505 atomic_set(&mm->oom_disable_count, 0);
507 if (likely(!mm_alloc_pgd(mm))) {
508 mm->def_flags = 0;
509 mmu_notifier_mm_init(mm);
510 return mm;
513 free_mm(mm);
514 return NULL;
518 * Allocate and initialize an mm_struct.
520 struct mm_struct *mm_alloc(void)
522 struct mm_struct *mm;
524 mm = allocate_mm();
525 if (!mm)
526 return NULL;
528 memset(mm, 0, sizeof(*mm));
529 mm_init_cpumask(mm);
530 return mm_init(mm, current);
534 * Called when the last reference to the mm
535 * is dropped: either by a lazy thread or by
536 * mmput. Free the page directory and the mm.
538 void __mmdrop(struct mm_struct *mm)
540 BUG_ON(mm == &init_mm);
541 mm_free_pgd(mm);
542 destroy_context(mm);
543 mmu_notifier_mm_destroy(mm);
544 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
545 VM_BUG_ON(mm->pmd_huge_pte);
546 #endif
547 free_mm(mm);
549 EXPORT_SYMBOL_GPL(__mmdrop);
552 * Decrement the use count and release all resources for an mm.
554 void mmput(struct mm_struct *mm)
556 might_sleep();
558 if (atomic_dec_and_test(&mm->mm_users)) {
559 exit_aio(mm);
560 ksm_exit(mm);
561 khugepaged_exit(mm); /* must run before exit_mmap */
562 exit_mmap(mm);
563 set_mm_exe_file(mm, NULL);
564 if (!list_empty(&mm->mmlist)) {
565 spin_lock(&mmlist_lock);
566 list_del(&mm->mmlist);
567 spin_unlock(&mmlist_lock);
569 put_swap_token(mm);
570 if (mm->binfmt)
571 module_put(mm->binfmt->module);
572 mmdrop(mm);
575 EXPORT_SYMBOL_GPL(mmput);
578 * We added or removed a vma mapping the executable. The vmas are only mapped
579 * during exec and are not mapped with the mmap system call.
580 * Callers must hold down_write() on the mm's mmap_sem for these
582 void added_exe_file_vma(struct mm_struct *mm)
584 mm->num_exe_file_vmas++;
587 void removed_exe_file_vma(struct mm_struct *mm)
589 mm->num_exe_file_vmas--;
590 if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
591 fput(mm->exe_file);
592 mm->exe_file = NULL;
597 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
599 if (new_exe_file)
600 get_file(new_exe_file);
601 if (mm->exe_file)
602 fput(mm->exe_file);
603 mm->exe_file = new_exe_file;
604 mm->num_exe_file_vmas = 0;
607 struct file *get_mm_exe_file(struct mm_struct *mm)
609 struct file *exe_file;
611 /* We need mmap_sem to protect against races with removal of
612 * VM_EXECUTABLE vmas */
613 down_read(&mm->mmap_sem);
614 exe_file = mm->exe_file;
615 if (exe_file)
616 get_file(exe_file);
617 up_read(&mm->mmap_sem);
618 return exe_file;
621 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
623 /* It's safe to write the exe_file pointer without exe_file_lock because
624 * this is called during fork when the task is not yet in /proc */
625 newmm->exe_file = get_mm_exe_file(oldmm);
629 * get_task_mm - acquire a reference to the task's mm
631 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
632 * this kernel workthread has transiently adopted a user mm with use_mm,
633 * to do its AIO) is not set and if so returns a reference to it, after
634 * bumping up the use count. User must release the mm via mmput()
635 * after use. Typically used by /proc and ptrace.
637 struct mm_struct *get_task_mm(struct task_struct *task)
639 struct mm_struct *mm;
641 task_lock(task);
642 mm = task->mm;
643 if (mm) {
644 if (task->flags & PF_KTHREAD)
645 mm = NULL;
646 else
647 atomic_inc(&mm->mm_users);
649 task_unlock(task);
650 return mm;
652 EXPORT_SYMBOL_GPL(get_task_mm);
654 /* Please note the differences between mmput and mm_release.
655 * mmput is called whenever we stop holding onto a mm_struct,
656 * error success whatever.
658 * mm_release is called after a mm_struct has been removed
659 * from the current process.
661 * This difference is important for error handling, when we
662 * only half set up a mm_struct for a new process and need to restore
663 * the old one. Because we mmput the new mm_struct before
664 * restoring the old one. . .
665 * Eric Biederman 10 January 1998
667 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
669 struct completion *vfork_done = tsk->vfork_done;
671 /* Get rid of any futexes when releasing the mm */
672 #ifdef CONFIG_FUTEX
673 if (unlikely(tsk->robust_list)) {
674 exit_robust_list(tsk);
675 tsk->robust_list = NULL;
677 #ifdef CONFIG_COMPAT
678 if (unlikely(tsk->compat_robust_list)) {
679 compat_exit_robust_list(tsk);
680 tsk->compat_robust_list = NULL;
682 #endif
683 if (unlikely(!list_empty(&tsk->pi_state_list)))
684 exit_pi_state_list(tsk);
685 #endif
687 /* Get rid of any cached register state */
688 deactivate_mm(tsk, mm);
690 /* notify parent sleeping on vfork() */
691 if (vfork_done) {
692 tsk->vfork_done = NULL;
693 complete(vfork_done);
697 * If we're exiting normally, clear a user-space tid field if
698 * requested. We leave this alone when dying by signal, to leave
699 * the value intact in a core dump, and to save the unnecessary
700 * trouble otherwise. Userland only wants this done for a sys_exit.
702 if (tsk->clear_child_tid) {
703 if (!(tsk->flags & PF_SIGNALED) &&
704 atomic_read(&mm->mm_users) > 1) {
706 * We don't check the error code - if userspace has
707 * not set up a proper pointer then tough luck.
709 put_user(0, tsk->clear_child_tid);
710 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
711 1, NULL, NULL, 0);
713 tsk->clear_child_tid = NULL;
718 * Allocate a new mm structure and copy contents from the
719 * mm structure of the passed in task structure.
721 struct mm_struct *dup_mm(struct task_struct *tsk)
723 struct mm_struct *mm, *oldmm = current->mm;
724 int err;
726 if (!oldmm)
727 return NULL;
729 mm = allocate_mm();
730 if (!mm)
731 goto fail_nomem;
733 memcpy(mm, oldmm, sizeof(*mm));
734 mm_init_cpumask(mm);
736 /* Initializing for Swap token stuff */
737 mm->token_priority = 0;
738 mm->last_interval = 0;
740 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
741 mm->pmd_huge_pte = NULL;
742 #endif
744 if (!mm_init(mm, tsk))
745 goto fail_nomem;
747 if (init_new_context(tsk, mm))
748 goto fail_nocontext;
750 dup_mm_exe_file(oldmm, mm);
752 err = dup_mmap(mm, oldmm);
753 if (err)
754 goto free_pt;
756 mm->hiwater_rss = get_mm_rss(mm);
757 mm->hiwater_vm = mm->total_vm;
759 if (mm->binfmt && !try_module_get(mm->binfmt->module))
760 goto free_pt;
762 return mm;
764 free_pt:
765 /* don't put binfmt in mmput, we haven't got module yet */
766 mm->binfmt = NULL;
767 mmput(mm);
769 fail_nomem:
770 return NULL;
772 fail_nocontext:
774 * If init_new_context() failed, we cannot use mmput() to free the mm
775 * because it calls destroy_context()
777 mm_free_pgd(mm);
778 free_mm(mm);
779 return NULL;
782 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
784 struct mm_struct *mm, *oldmm;
785 int retval;
787 tsk->min_flt = tsk->maj_flt = 0;
788 tsk->nvcsw = tsk->nivcsw = 0;
789 #ifdef CONFIG_DETECT_HUNG_TASK
790 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
791 #endif
793 tsk->mm = NULL;
794 tsk->active_mm = NULL;
797 * Are we cloning a kernel thread?
799 * We need to steal a active VM for that..
801 oldmm = current->mm;
802 if (!oldmm)
803 return 0;
805 if (clone_flags & CLONE_VM) {
806 atomic_inc(&oldmm->mm_users);
807 mm = oldmm;
808 goto good_mm;
811 retval = -ENOMEM;
812 mm = dup_mm(tsk);
813 if (!mm)
814 goto fail_nomem;
816 good_mm:
817 /* Initializing for Swap token stuff */
818 mm->token_priority = 0;
819 mm->last_interval = 0;
820 if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
821 atomic_inc(&mm->oom_disable_count);
823 tsk->mm = mm;
824 tsk->active_mm = mm;
825 return 0;
827 fail_nomem:
828 return retval;
831 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
833 struct fs_struct *fs = current->fs;
834 if (clone_flags & CLONE_FS) {
835 /* tsk->fs is already what we want */
836 spin_lock(&fs->lock);
837 if (fs->in_exec) {
838 spin_unlock(&fs->lock);
839 return -EAGAIN;
841 fs->users++;
842 spin_unlock(&fs->lock);
843 return 0;
845 tsk->fs = copy_fs_struct(fs);
846 if (!tsk->fs)
847 return -ENOMEM;
848 return 0;
851 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
853 struct files_struct *oldf, *newf;
854 int error = 0;
857 * A background process may not have any files ...
859 oldf = current->files;
860 if (!oldf)
861 goto out;
863 if (clone_flags & CLONE_FILES) {
864 atomic_inc(&oldf->count);
865 goto out;
868 newf = dup_fd(oldf, &error);
869 if (!newf)
870 goto out;
872 tsk->files = newf;
873 error = 0;
874 out:
875 return error;
878 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
880 #ifdef CONFIG_BLOCK
881 struct io_context *ioc = current->io_context;
883 if (!ioc)
884 return 0;
886 * Share io context with parent, if CLONE_IO is set
888 if (clone_flags & CLONE_IO) {
889 tsk->io_context = ioc_task_link(ioc);
890 if (unlikely(!tsk->io_context))
891 return -ENOMEM;
892 } else if (ioprio_valid(ioc->ioprio)) {
893 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
894 if (unlikely(!tsk->io_context))
895 return -ENOMEM;
897 tsk->io_context->ioprio = ioc->ioprio;
899 #endif
900 return 0;
903 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
905 struct sighand_struct *sig;
907 if (clone_flags & CLONE_SIGHAND) {
908 atomic_inc(&current->sighand->count);
909 return 0;
911 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
912 rcu_assign_pointer(tsk->sighand, sig);
913 if (!sig)
914 return -ENOMEM;
915 atomic_set(&sig->count, 1);
916 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
917 return 0;
920 void __cleanup_sighand(struct sighand_struct *sighand)
922 if (atomic_dec_and_test(&sighand->count))
923 kmem_cache_free(sighand_cachep, sighand);
928 * Initialize POSIX timer handling for a thread group.
930 static void posix_cpu_timers_init_group(struct signal_struct *sig)
932 unsigned long cpu_limit;
934 /* Thread group counters. */
935 thread_group_cputime_init(sig);
937 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
938 if (cpu_limit != RLIM_INFINITY) {
939 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
940 sig->cputimer.running = 1;
943 /* The timer lists. */
944 INIT_LIST_HEAD(&sig->cpu_timers[0]);
945 INIT_LIST_HEAD(&sig->cpu_timers[1]);
946 INIT_LIST_HEAD(&sig->cpu_timers[2]);
949 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
951 struct signal_struct *sig;
953 if (clone_flags & CLONE_THREAD)
954 return 0;
956 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
957 tsk->signal = sig;
958 if (!sig)
959 return -ENOMEM;
961 sig->nr_threads = 1;
962 atomic_set(&sig->live, 1);
963 atomic_set(&sig->sigcnt, 1);
964 init_waitqueue_head(&sig->wait_chldexit);
965 if (clone_flags & CLONE_NEWPID)
966 sig->flags |= SIGNAL_UNKILLABLE;
967 sig->curr_target = tsk;
968 init_sigpending(&sig->shared_pending);
969 INIT_LIST_HEAD(&sig->posix_timers);
971 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
972 sig->real_timer.function = it_real_fn;
974 task_lock(current->group_leader);
975 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
976 task_unlock(current->group_leader);
978 posix_cpu_timers_init_group(sig);
980 tty_audit_fork(sig);
981 sched_autogroup_fork(sig);
983 #ifdef CONFIG_CGROUPS
984 init_rwsem(&sig->threadgroup_fork_lock);
985 #endif
987 sig->oom_adj = current->signal->oom_adj;
988 sig->oom_score_adj = current->signal->oom_score_adj;
989 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
991 mutex_init(&sig->cred_guard_mutex);
993 return 0;
996 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
998 unsigned long new_flags = p->flags;
1000 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1001 new_flags |= PF_FORKNOEXEC;
1002 new_flags |= PF_STARTING;
1003 p->flags = new_flags;
1004 clear_freeze_flag(p);
1007 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1009 current->clear_child_tid = tidptr;
1011 return task_pid_vnr(current);
1014 static void rt_mutex_init_task(struct task_struct *p)
1016 raw_spin_lock_init(&p->pi_lock);
1017 #ifdef CONFIG_RT_MUTEXES
1018 plist_head_init(&p->pi_waiters);
1019 p->pi_blocked_on = NULL;
1020 #endif
1023 #ifdef CONFIG_MM_OWNER
1024 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1026 mm->owner = p;
1028 #endif /* CONFIG_MM_OWNER */
1031 * Initialize POSIX timer handling for a single task.
1033 static void posix_cpu_timers_init(struct task_struct *tsk)
1035 tsk->cputime_expires.prof_exp = cputime_zero;
1036 tsk->cputime_expires.virt_exp = cputime_zero;
1037 tsk->cputime_expires.sched_exp = 0;
1038 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1039 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1040 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1044 * This creates a new process as a copy of the old one,
1045 * but does not actually start it yet.
1047 * It copies the registers, and all the appropriate
1048 * parts of the process environment (as per the clone
1049 * flags). The actual kick-off is left to the caller.
1051 static struct task_struct *copy_process(unsigned long clone_flags,
1052 unsigned long stack_start,
1053 struct pt_regs *regs,
1054 unsigned long stack_size,
1055 int __user *child_tidptr,
1056 struct pid *pid,
1057 int trace)
1059 int retval;
1060 struct task_struct *p;
1061 int cgroup_callbacks_done = 0;
1063 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1064 return ERR_PTR(-EINVAL);
1067 * Thread groups must share signals as well, and detached threads
1068 * can only be started up within the thread group.
1070 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1071 return ERR_PTR(-EINVAL);
1074 * Shared signal handlers imply shared VM. By way of the above,
1075 * thread groups also imply shared VM. Blocking this case allows
1076 * for various simplifications in other code.
1078 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1079 return ERR_PTR(-EINVAL);
1082 * Siblings of global init remain as zombies on exit since they are
1083 * not reaped by their parent (swapper). To solve this and to avoid
1084 * multi-rooted process trees, prevent global and container-inits
1085 * from creating siblings.
1087 if ((clone_flags & CLONE_PARENT) &&
1088 current->signal->flags & SIGNAL_UNKILLABLE)
1089 return ERR_PTR(-EINVAL);
1091 retval = security_task_create(clone_flags);
1092 if (retval)
1093 goto fork_out;
1095 retval = -ENOMEM;
1096 p = dup_task_struct(current);
1097 if (!p)
1098 goto fork_out;
1100 ftrace_graph_init_task(p);
1102 rt_mutex_init_task(p);
1104 #ifdef CONFIG_PROVE_LOCKING
1105 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1106 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1107 #endif
1108 retval = -EAGAIN;
1109 if (atomic_read(&p->real_cred->user->processes) >=
1110 task_rlimit(p, RLIMIT_NPROC)) {
1111 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1112 p->real_cred->user != INIT_USER)
1113 goto bad_fork_free;
1115 current->flags &= ~PF_NPROC_EXCEEDED;
1117 retval = copy_creds(p, clone_flags);
1118 if (retval < 0)
1119 goto bad_fork_free;
1122 * If multiple threads are within copy_process(), then this check
1123 * triggers too late. This doesn't hurt, the check is only there
1124 * to stop root fork bombs.
1126 retval = -EAGAIN;
1127 if (nr_threads >= max_threads)
1128 goto bad_fork_cleanup_count;
1130 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1131 goto bad_fork_cleanup_count;
1133 p->did_exec = 0;
1134 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1135 copy_flags(clone_flags, p);
1136 INIT_LIST_HEAD(&p->children);
1137 INIT_LIST_HEAD(&p->sibling);
1138 rcu_copy_process(p);
1139 p->vfork_done = NULL;
1140 spin_lock_init(&p->alloc_lock);
1142 init_sigpending(&p->pending);
1144 p->utime = cputime_zero;
1145 p->stime = cputime_zero;
1146 p->gtime = cputime_zero;
1147 p->utimescaled = cputime_zero;
1148 p->stimescaled = cputime_zero;
1149 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1150 p->prev_utime = cputime_zero;
1151 p->prev_stime = cputime_zero;
1152 #endif
1153 #if defined(SPLIT_RSS_COUNTING)
1154 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1155 #endif
1157 p->default_timer_slack_ns = current->timer_slack_ns;
1159 task_io_accounting_init(&p->ioac);
1160 acct_clear_integrals(p);
1162 posix_cpu_timers_init(p);
1164 do_posix_clock_monotonic_gettime(&p->start_time);
1165 p->real_start_time = p->start_time;
1166 monotonic_to_bootbased(&p->real_start_time);
1167 p->io_context = NULL;
1168 p->audit_context = NULL;
1169 if (clone_flags & CLONE_THREAD)
1170 threadgroup_fork_read_lock(current);
1171 cgroup_fork(p);
1172 #ifdef CONFIG_NUMA
1173 p->mempolicy = mpol_dup(p->mempolicy);
1174 if (IS_ERR(p->mempolicy)) {
1175 retval = PTR_ERR(p->mempolicy);
1176 p->mempolicy = NULL;
1177 goto bad_fork_cleanup_cgroup;
1179 mpol_fix_fork_child_flag(p);
1180 #endif
1181 #ifdef CONFIG_CPUSETS
1182 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1183 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1184 #endif
1185 #ifdef CONFIG_TRACE_IRQFLAGS
1186 p->irq_events = 0;
1187 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1188 p->hardirqs_enabled = 1;
1189 #else
1190 p->hardirqs_enabled = 0;
1191 #endif
1192 p->hardirq_enable_ip = 0;
1193 p->hardirq_enable_event = 0;
1194 p->hardirq_disable_ip = _THIS_IP_;
1195 p->hardirq_disable_event = 0;
1196 p->softirqs_enabled = 1;
1197 p->softirq_enable_ip = _THIS_IP_;
1198 p->softirq_enable_event = 0;
1199 p->softirq_disable_ip = 0;
1200 p->softirq_disable_event = 0;
1201 p->hardirq_context = 0;
1202 p->softirq_context = 0;
1203 #endif
1204 #ifdef CONFIG_LOCKDEP
1205 p->lockdep_depth = 0; /* no locks held yet */
1206 p->curr_chain_key = 0;
1207 p->lockdep_recursion = 0;
1208 #endif
1210 #ifdef CONFIG_DEBUG_MUTEXES
1211 p->blocked_on = NULL; /* not blocked yet */
1212 #endif
1213 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1214 p->memcg_batch.do_batch = 0;
1215 p->memcg_batch.memcg = NULL;
1216 #endif
1218 /* Perform scheduler related setup. Assign this task to a CPU. */
1219 sched_fork(p);
1221 retval = perf_event_init_task(p);
1222 if (retval)
1223 goto bad_fork_cleanup_policy;
1224 retval = audit_alloc(p);
1225 if (retval)
1226 goto bad_fork_cleanup_policy;
1227 /* copy all the process information */
1228 retval = copy_semundo(clone_flags, p);
1229 if (retval)
1230 goto bad_fork_cleanup_audit;
1231 retval = copy_files(clone_flags, p);
1232 if (retval)
1233 goto bad_fork_cleanup_semundo;
1234 retval = copy_fs(clone_flags, p);
1235 if (retval)
1236 goto bad_fork_cleanup_files;
1237 retval = copy_sighand(clone_flags, p);
1238 if (retval)
1239 goto bad_fork_cleanup_fs;
1240 retval = copy_signal(clone_flags, p);
1241 if (retval)
1242 goto bad_fork_cleanup_sighand;
1243 retval = copy_mm(clone_flags, p);
1244 if (retval)
1245 goto bad_fork_cleanup_signal;
1246 retval = copy_namespaces(clone_flags, p);
1247 if (retval)
1248 goto bad_fork_cleanup_mm;
1249 retval = copy_io(clone_flags, p);
1250 if (retval)
1251 goto bad_fork_cleanup_namespaces;
1252 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1253 if (retval)
1254 goto bad_fork_cleanup_io;
1256 if (pid != &init_struct_pid) {
1257 retval = -ENOMEM;
1258 pid = alloc_pid(p->nsproxy->pid_ns);
1259 if (!pid)
1260 goto bad_fork_cleanup_io;
1263 p->pid = pid_nr(pid);
1264 p->tgid = p->pid;
1265 if (clone_flags & CLONE_THREAD)
1266 p->tgid = current->tgid;
1268 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1270 * Clear TID on mm_release()?
1272 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1273 #ifdef CONFIG_BLOCK
1274 p->plug = NULL;
1275 #endif
1276 #ifdef CONFIG_FUTEX
1277 p->robust_list = NULL;
1278 #ifdef CONFIG_COMPAT
1279 p->compat_robust_list = NULL;
1280 #endif
1281 INIT_LIST_HEAD(&p->pi_state_list);
1282 p->pi_state_cache = NULL;
1283 #endif
1285 * sigaltstack should be cleared when sharing the same VM
1287 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1288 p->sas_ss_sp = p->sas_ss_size = 0;
1291 * Syscall tracing and stepping should be turned off in the
1292 * child regardless of CLONE_PTRACE.
1294 user_disable_single_step(p);
1295 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1296 #ifdef TIF_SYSCALL_EMU
1297 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1298 #endif
1299 clear_all_latency_tracing(p);
1301 /* ok, now we should be set up.. */
1302 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1303 p->pdeath_signal = 0;
1304 p->exit_state = 0;
1307 * Ok, make it visible to the rest of the system.
1308 * We dont wake it up yet.
1310 p->group_leader = p;
1311 INIT_LIST_HEAD(&p->thread_group);
1313 /* Now that the task is set up, run cgroup callbacks if
1314 * necessary. We need to run them before the task is visible
1315 * on the tasklist. */
1316 cgroup_fork_callbacks(p);
1317 cgroup_callbacks_done = 1;
1319 /* Need tasklist lock for parent etc handling! */
1320 write_lock_irq(&tasklist_lock);
1322 /* CLONE_PARENT re-uses the old parent */
1323 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1324 p->real_parent = current->real_parent;
1325 p->parent_exec_id = current->parent_exec_id;
1326 } else {
1327 p->real_parent = current;
1328 p->parent_exec_id = current->self_exec_id;
1331 spin_lock(&current->sighand->siglock);
1334 * Process group and session signals need to be delivered to just the
1335 * parent before the fork or both the parent and the child after the
1336 * fork. Restart if a signal comes in before we add the new process to
1337 * it's process group.
1338 * A fatal signal pending means that current will exit, so the new
1339 * thread can't slip out of an OOM kill (or normal SIGKILL).
1341 recalc_sigpending();
1342 if (signal_pending(current)) {
1343 spin_unlock(&current->sighand->siglock);
1344 write_unlock_irq(&tasklist_lock);
1345 retval = -ERESTARTNOINTR;
1346 goto bad_fork_free_pid;
1349 if (clone_flags & CLONE_THREAD) {
1350 current->signal->nr_threads++;
1351 atomic_inc(&current->signal->live);
1352 atomic_inc(&current->signal->sigcnt);
1353 p->group_leader = current->group_leader;
1354 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1357 if (likely(p->pid)) {
1358 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1360 if (thread_group_leader(p)) {
1361 if (is_child_reaper(pid))
1362 p->nsproxy->pid_ns->child_reaper = p;
1364 p->signal->leader_pid = pid;
1365 p->signal->tty = tty_kref_get(current->signal->tty);
1366 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1367 attach_pid(p, PIDTYPE_SID, task_session(current));
1368 list_add_tail(&p->sibling, &p->real_parent->children);
1369 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1370 __this_cpu_inc(process_counts);
1372 attach_pid(p, PIDTYPE_PID, pid);
1373 nr_threads++;
1376 total_forks++;
1377 spin_unlock(&current->sighand->siglock);
1378 write_unlock_irq(&tasklist_lock);
1379 proc_fork_connector(p);
1380 cgroup_post_fork(p);
1381 if (clone_flags & CLONE_THREAD)
1382 threadgroup_fork_read_unlock(current);
1383 perf_event_fork(p);
1384 return p;
1386 bad_fork_free_pid:
1387 if (pid != &init_struct_pid)
1388 free_pid(pid);
1389 bad_fork_cleanup_io:
1390 if (p->io_context)
1391 exit_io_context(p);
1392 bad_fork_cleanup_namespaces:
1393 exit_task_namespaces(p);
1394 bad_fork_cleanup_mm:
1395 if (p->mm) {
1396 task_lock(p);
1397 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1398 atomic_dec(&p->mm->oom_disable_count);
1399 task_unlock(p);
1400 mmput(p->mm);
1402 bad_fork_cleanup_signal:
1403 if (!(clone_flags & CLONE_THREAD))
1404 free_signal_struct(p->signal);
1405 bad_fork_cleanup_sighand:
1406 __cleanup_sighand(p->sighand);
1407 bad_fork_cleanup_fs:
1408 exit_fs(p); /* blocking */
1409 bad_fork_cleanup_files:
1410 exit_files(p); /* blocking */
1411 bad_fork_cleanup_semundo:
1412 exit_sem(p);
1413 bad_fork_cleanup_audit:
1414 audit_free(p);
1415 bad_fork_cleanup_policy:
1416 perf_event_free_task(p);
1417 #ifdef CONFIG_NUMA
1418 mpol_put(p->mempolicy);
1419 bad_fork_cleanup_cgroup:
1420 #endif
1421 if (clone_flags & CLONE_THREAD)
1422 threadgroup_fork_read_unlock(current);
1423 cgroup_exit(p, cgroup_callbacks_done);
1424 delayacct_tsk_free(p);
1425 module_put(task_thread_info(p)->exec_domain->module);
1426 bad_fork_cleanup_count:
1427 atomic_dec(&p->cred->user->processes);
1428 exit_creds(p);
1429 bad_fork_free:
1430 free_task(p);
1431 fork_out:
1432 return ERR_PTR(retval);
1435 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1437 memset(regs, 0, sizeof(struct pt_regs));
1438 return regs;
1441 static inline void init_idle_pids(struct pid_link *links)
1443 enum pid_type type;
1445 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1446 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1447 links[type].pid = &init_struct_pid;
1451 struct task_struct * __cpuinit fork_idle(int cpu)
1453 struct task_struct *task;
1454 struct pt_regs regs;
1456 task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1457 &init_struct_pid, 0);
1458 if (!IS_ERR(task)) {
1459 init_idle_pids(task->pids);
1460 init_idle(task, cpu);
1463 return task;
1467 * Ok, this is the main fork-routine.
1469 * It copies the process, and if successful kick-starts
1470 * it and waits for it to finish using the VM if required.
1472 long do_fork(unsigned long clone_flags,
1473 unsigned long stack_start,
1474 struct pt_regs *regs,
1475 unsigned long stack_size,
1476 int __user *parent_tidptr,
1477 int __user *child_tidptr)
1479 struct task_struct *p;
1480 int trace = 0;
1481 long nr;
1484 * Do some preliminary argument and permissions checking before we
1485 * actually start allocating stuff
1487 if (clone_flags & CLONE_NEWUSER) {
1488 if (clone_flags & CLONE_THREAD)
1489 return -EINVAL;
1490 /* hopefully this check will go away when userns support is
1491 * complete
1493 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1494 !capable(CAP_SETGID))
1495 return -EPERM;
1499 * Determine whether and which event to report to ptracer. When
1500 * called from kernel_thread or CLONE_UNTRACED is explicitly
1501 * requested, no event is reported; otherwise, report if the event
1502 * for the type of forking is enabled.
1504 if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1505 if (clone_flags & CLONE_VFORK)
1506 trace = PTRACE_EVENT_VFORK;
1507 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1508 trace = PTRACE_EVENT_CLONE;
1509 else
1510 trace = PTRACE_EVENT_FORK;
1512 if (likely(!ptrace_event_enabled(current, trace)))
1513 trace = 0;
1516 p = copy_process(clone_flags, stack_start, regs, stack_size,
1517 child_tidptr, NULL, trace);
1519 * Do this prior waking up the new thread - the thread pointer
1520 * might get invalid after that point, if the thread exits quickly.
1522 if (!IS_ERR(p)) {
1523 struct completion vfork;
1525 trace_sched_process_fork(current, p);
1527 nr = task_pid_vnr(p);
1529 if (clone_flags & CLONE_PARENT_SETTID)
1530 put_user(nr, parent_tidptr);
1532 if (clone_flags & CLONE_VFORK) {
1533 p->vfork_done = &vfork;
1534 init_completion(&vfork);
1537 audit_finish_fork(p);
1540 * We set PF_STARTING at creation in case tracing wants to
1541 * use this to distinguish a fully live task from one that
1542 * hasn't finished SIGSTOP raising yet. Now we clear it
1543 * and set the child going.
1545 p->flags &= ~PF_STARTING;
1547 wake_up_new_task(p);
1549 /* forking complete and child started to run, tell ptracer */
1550 if (unlikely(trace))
1551 ptrace_event(trace, nr);
1553 if (clone_flags & CLONE_VFORK) {
1554 freezer_do_not_count();
1555 wait_for_completion(&vfork);
1556 freezer_count();
1557 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1559 } else {
1560 nr = PTR_ERR(p);
1562 return nr;
1565 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1566 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1567 #endif
1569 static void sighand_ctor(void *data)
1571 struct sighand_struct *sighand = data;
1573 spin_lock_init(&sighand->siglock);
1574 init_waitqueue_head(&sighand->signalfd_wqh);
1577 void __init proc_caches_init(void)
1579 sighand_cachep = kmem_cache_create("sighand_cache",
1580 sizeof(struct sighand_struct), 0,
1581 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1582 SLAB_NOTRACK, sighand_ctor);
1583 signal_cachep = kmem_cache_create("signal_cache",
1584 sizeof(struct signal_struct), 0,
1585 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1586 files_cachep = kmem_cache_create("files_cache",
1587 sizeof(struct files_struct), 0,
1588 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1589 fs_cachep = kmem_cache_create("fs_cache",
1590 sizeof(struct fs_struct), 0,
1591 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1593 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1594 * whole struct cpumask for the OFFSTACK case. We could change
1595 * this to *only* allocate as much of it as required by the
1596 * maximum number of CPU's we can ever have. The cpumask_allocation
1597 * is at the end of the structure, exactly for that reason.
1599 mm_cachep = kmem_cache_create("mm_struct",
1600 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1601 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1602 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1603 mmap_init();
1604 nsproxy_cache_init();
1608 * Check constraints on flags passed to the unshare system call.
1610 static int check_unshare_flags(unsigned long unshare_flags)
1612 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1613 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1614 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1615 return -EINVAL;
1617 * Not implemented, but pretend it works if there is nothing to
1618 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1619 * needs to unshare vm.
1621 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1622 /* FIXME: get_task_mm() increments ->mm_users */
1623 if (atomic_read(&current->mm->mm_users) > 1)
1624 return -EINVAL;
1627 return 0;
1631 * Unshare the filesystem structure if it is being shared
1633 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1635 struct fs_struct *fs = current->fs;
1637 if (!(unshare_flags & CLONE_FS) || !fs)
1638 return 0;
1640 /* don't need lock here; in the worst case we'll do useless copy */
1641 if (fs->users == 1)
1642 return 0;
1644 *new_fsp = copy_fs_struct(fs);
1645 if (!*new_fsp)
1646 return -ENOMEM;
1648 return 0;
1652 * Unshare file descriptor table if it is being shared
1654 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1656 struct files_struct *fd = current->files;
1657 int error = 0;
1659 if ((unshare_flags & CLONE_FILES) &&
1660 (fd && atomic_read(&fd->count) > 1)) {
1661 *new_fdp = dup_fd(fd, &error);
1662 if (!*new_fdp)
1663 return error;
1666 return 0;
1670 * unshare allows a process to 'unshare' part of the process
1671 * context which was originally shared using clone. copy_*
1672 * functions used by do_fork() cannot be used here directly
1673 * because they modify an inactive task_struct that is being
1674 * constructed. Here we are modifying the current, active,
1675 * task_struct.
1677 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1679 struct fs_struct *fs, *new_fs = NULL;
1680 struct files_struct *fd, *new_fd = NULL;
1681 struct nsproxy *new_nsproxy = NULL;
1682 int do_sysvsem = 0;
1683 int err;
1685 err = check_unshare_flags(unshare_flags);
1686 if (err)
1687 goto bad_unshare_out;
1690 * If unsharing namespace, must also unshare filesystem information.
1692 if (unshare_flags & CLONE_NEWNS)
1693 unshare_flags |= CLONE_FS;
1695 * CLONE_NEWIPC must also detach from the undolist: after switching
1696 * to a new ipc namespace, the semaphore arrays from the old
1697 * namespace are unreachable.
1699 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1700 do_sysvsem = 1;
1701 err = unshare_fs(unshare_flags, &new_fs);
1702 if (err)
1703 goto bad_unshare_out;
1704 err = unshare_fd(unshare_flags, &new_fd);
1705 if (err)
1706 goto bad_unshare_cleanup_fs;
1707 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1708 if (err)
1709 goto bad_unshare_cleanup_fd;
1711 if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1712 if (do_sysvsem) {
1714 * CLONE_SYSVSEM is equivalent to sys_exit().
1716 exit_sem(current);
1719 if (new_nsproxy) {
1720 switch_task_namespaces(current, new_nsproxy);
1721 new_nsproxy = NULL;
1724 task_lock(current);
1726 if (new_fs) {
1727 fs = current->fs;
1728 spin_lock(&fs->lock);
1729 current->fs = new_fs;
1730 if (--fs->users)
1731 new_fs = NULL;
1732 else
1733 new_fs = fs;
1734 spin_unlock(&fs->lock);
1737 if (new_fd) {
1738 fd = current->files;
1739 current->files = new_fd;
1740 new_fd = fd;
1743 task_unlock(current);
1746 if (new_nsproxy)
1747 put_nsproxy(new_nsproxy);
1749 bad_unshare_cleanup_fd:
1750 if (new_fd)
1751 put_files_struct(new_fd);
1753 bad_unshare_cleanup_fs:
1754 if (new_fs)
1755 free_fs_struct(new_fs);
1757 bad_unshare_out:
1758 return err;
1762 * Helper to unshare the files of the current task.
1763 * We don't want to expose copy_files internals to
1764 * the exec layer of the kernel.
1767 int unshare_files(struct files_struct **displaced)
1769 struct task_struct *task = current;
1770 struct files_struct *copy = NULL;
1771 int error;
1773 error = unshare_fd(CLONE_FILES, &copy);
1774 if (error || !copy) {
1775 *displaced = NULL;
1776 return error;
1778 *displaced = task->files;
1779 task_lock(task);
1780 task->files = copy;
1781 task_unlock(task);
1782 return 0;