IPv4: Send gratuitous ARP for secondary IP addresses also
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / fork.c
blob17bf7c8d65114497d5cd076a850d9a013e4d4df5
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/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/futex.h>
41 #include <linux/compat.h>
42 #include <linux/kthread.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/ftrace.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/ksm.h>
53 #include <linux/acct.h>
54 #include <linux/tsacct_kern.h>
55 #include <linux/cn_proc.h>
56 #include <linux/freezer.h>
57 #include <linux/delayacct.h>
58 #include <linux/taskstats_kern.h>
59 #include <linux/random.h>
60 #include <linux/tty.h>
61 #include <linux/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_event.h>
65 #include <linux/posix-timers.h>
66 #include <linux/user-return-notifier.h>
67 #include <linux/oom.h>
68 #include <linux/khugepaged.h>
70 #include <asm/pgtable.h>
71 #include <asm/pgalloc.h>
72 #include <asm/uaccess.h>
73 #include <asm/mmu_context.h>
74 #include <asm/cacheflush.h>
75 #include <asm/tlbflush.h>
77 #include <trace/events/sched.h>
80 * Protected counters by write_lock_irq(&tasklist_lock)
82 unsigned long total_forks; /* Handle normal Linux uptimes. */
83 int nr_threads; /* The idle threads do not count.. */
85 int max_threads; /* tunable limit on nr_threads */
87 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
89 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
91 #ifdef CONFIG_PROVE_RCU
92 int lockdep_tasklist_lock_is_held(void)
94 return lockdep_is_held(&tasklist_lock);
96 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
97 #endif /* #ifdef CONFIG_PROVE_RCU */
99 int nr_processes(void)
101 int cpu;
102 int total = 0;
104 for_each_possible_cpu(cpu)
105 total += per_cpu(process_counts, cpu);
107 return total;
110 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
111 # define alloc_task_struct_node(node) \
112 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
113 # define free_task_struct(tsk) \
114 kmem_cache_free(task_struct_cachep, (tsk))
115 static struct kmem_cache *task_struct_cachep;
116 #endif
118 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
119 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
120 int node)
122 #ifdef CONFIG_DEBUG_STACK_USAGE
123 gfp_t mask = GFP_KERNEL | __GFP_ZERO;
124 #else
125 gfp_t mask = GFP_KERNEL;
126 #endif
127 struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
129 return page ? page_address(page) : NULL;
132 static inline void free_thread_info(struct thread_info *ti)
134 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
136 #endif
138 /* SLAB cache for signal_struct structures (tsk->signal) */
139 static struct kmem_cache *signal_cachep;
141 /* SLAB cache for sighand_struct structures (tsk->sighand) */
142 struct kmem_cache *sighand_cachep;
144 /* SLAB cache for files_struct structures (tsk->files) */
145 struct kmem_cache *files_cachep;
147 /* SLAB cache for fs_struct structures (tsk->fs) */
148 struct kmem_cache *fs_cachep;
150 /* SLAB cache for vm_area_struct structures */
151 struct kmem_cache *vm_area_cachep;
153 /* SLAB cache for mm_struct structures (tsk->mm) */
154 static struct kmem_cache *mm_cachep;
156 static void account_kernel_stack(struct thread_info *ti, int account)
158 struct zone *zone = page_zone(virt_to_page(ti));
160 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
163 void free_task(struct task_struct *tsk)
165 prop_local_destroy_single(&tsk->dirties);
166 account_kernel_stack(tsk->stack, -1);
167 free_thread_info(tsk->stack);
168 rt_mutex_debug_task_free(tsk);
169 ftrace_graph_exit_task(tsk);
170 free_task_struct(tsk);
172 EXPORT_SYMBOL(free_task);
174 static inline void free_signal_struct(struct signal_struct *sig)
176 taskstats_tgid_free(sig);
177 sched_autogroup_exit(sig);
178 kmem_cache_free(signal_cachep, sig);
181 static inline void put_signal_struct(struct signal_struct *sig)
183 if (atomic_dec_and_test(&sig->sigcnt))
184 free_signal_struct(sig);
187 void __put_task_struct(struct task_struct *tsk)
189 WARN_ON(!tsk->exit_state);
190 WARN_ON(atomic_read(&tsk->usage));
191 WARN_ON(tsk == current);
193 exit_creds(tsk);
194 delayacct_tsk_free(tsk);
195 put_signal_struct(tsk->signal);
197 if (!profile_handoff_task(tsk))
198 free_task(tsk);
200 EXPORT_SYMBOL_GPL(__put_task_struct);
203 * macro override instead of weak attribute alias, to workaround
204 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
206 #ifndef arch_task_cache_init
207 #define arch_task_cache_init()
208 #endif
210 void __init fork_init(unsigned long mempages)
212 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
213 #ifndef ARCH_MIN_TASKALIGN
214 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
215 #endif
216 /* create a slab on which task_structs can be allocated */
217 task_struct_cachep =
218 kmem_cache_create("task_struct", sizeof(struct task_struct),
219 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
220 #endif
222 /* do the arch specific task caches init */
223 arch_task_cache_init();
226 * The default maximum number of threads is set to a safe
227 * value: the thread structures can take up at most half
228 * of memory.
230 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
233 * we need to allow at least 20 threads to boot a system
235 if(max_threads < 20)
236 max_threads = 20;
238 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
239 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
240 init_task.signal->rlim[RLIMIT_SIGPENDING] =
241 init_task.signal->rlim[RLIMIT_NPROC];
244 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
245 struct task_struct *src)
247 *dst = *src;
248 return 0;
251 static struct task_struct *dup_task_struct(struct task_struct *orig)
253 struct task_struct *tsk;
254 struct thread_info *ti;
255 unsigned long *stackend;
256 int node = tsk_fork_get_node(orig);
257 int err;
259 prepare_to_copy(orig);
261 tsk = alloc_task_struct_node(node);
262 if (!tsk)
263 return NULL;
265 ti = alloc_thread_info_node(tsk, node);
266 if (!ti) {
267 free_task_struct(tsk);
268 return NULL;
271 err = arch_dup_task_struct(tsk, orig);
272 if (err)
273 goto out;
275 tsk->stack = ti;
277 err = prop_local_init_single(&tsk->dirties);
278 if (err)
279 goto out;
281 setup_thread_stack(tsk, orig);
282 clear_user_return_notifier(tsk);
283 clear_tsk_need_resched(tsk);
284 stackend = end_of_stack(tsk);
285 *stackend = STACK_END_MAGIC; /* for overflow detection */
287 #ifdef CONFIG_CC_STACKPROTECTOR
288 tsk->stack_canary = get_random_int();
289 #endif
291 /* One for us, one for whoever does the "release_task()" (usually parent) */
292 atomic_set(&tsk->usage,2);
293 #ifdef CONFIG_BLK_DEV_IO_TRACE
294 tsk->btrace_seq = 0;
295 #endif
296 tsk->splice_pipe = NULL;
298 account_kernel_stack(ti, 1);
300 return tsk;
302 out:
303 free_thread_info(ti);
304 free_task_struct(tsk);
305 return NULL;
308 #ifdef CONFIG_MMU
309 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
311 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
312 struct rb_node **rb_link, *rb_parent;
313 int retval;
314 unsigned long charge;
315 struct mempolicy *pol;
317 down_write(&oldmm->mmap_sem);
318 flush_cache_dup_mm(oldmm);
320 * Not linked in yet - no deadlock potential:
322 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
324 mm->locked_vm = 0;
325 mm->mmap = NULL;
326 mm->mmap_cache = NULL;
327 mm->free_area_cache = oldmm->mmap_base;
328 mm->cached_hole_size = ~0UL;
329 mm->map_count = 0;
330 cpumask_clear(mm_cpumask(mm));
331 mm->mm_rb = RB_ROOT;
332 rb_link = &mm->mm_rb.rb_node;
333 rb_parent = NULL;
334 pprev = &mm->mmap;
335 retval = ksm_fork(mm, oldmm);
336 if (retval)
337 goto out;
338 retval = khugepaged_fork(mm, oldmm);
339 if (retval)
340 goto out;
342 prev = NULL;
343 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
344 struct file *file;
346 if (mpnt->vm_flags & VM_DONTCOPY) {
347 long pages = vma_pages(mpnt);
348 mm->total_vm -= pages;
349 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
350 -pages);
351 continue;
353 charge = 0;
354 if (mpnt->vm_flags & VM_ACCOUNT) {
355 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
356 if (security_vm_enough_memory(len))
357 goto fail_nomem;
358 charge = len;
360 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
361 if (!tmp)
362 goto fail_nomem;
363 *tmp = *mpnt;
364 INIT_LIST_HEAD(&tmp->anon_vma_chain);
365 pol = mpol_dup(vma_policy(mpnt));
366 retval = PTR_ERR(pol);
367 if (IS_ERR(pol))
368 goto fail_nomem_policy;
369 vma_set_policy(tmp, pol);
370 tmp->vm_mm = mm;
371 if (anon_vma_fork(tmp, mpnt))
372 goto fail_nomem_anon_vma_fork;
373 tmp->vm_flags &= ~VM_LOCKED;
374 tmp->vm_next = tmp->vm_prev = NULL;
375 file = tmp->vm_file;
376 if (file) {
377 struct inode *inode = file->f_path.dentry->d_inode;
378 struct address_space *mapping = file->f_mapping;
380 get_file(file);
381 if (tmp->vm_flags & VM_DENYWRITE)
382 atomic_dec(&inode->i_writecount);
383 mutex_lock(&mapping->i_mmap_mutex);
384 if (tmp->vm_flags & VM_SHARED)
385 mapping->i_mmap_writable++;
386 flush_dcache_mmap_lock(mapping);
387 /* insert tmp into the share list, just after mpnt */
388 vma_prio_tree_add(tmp, mpnt);
389 flush_dcache_mmap_unlock(mapping);
390 mutex_unlock(&mapping->i_mmap_mutex);
394 * Clear hugetlb-related page reserves for children. This only
395 * affects MAP_PRIVATE mappings. Faults generated by the child
396 * are not guaranteed to succeed, even if read-only
398 if (is_vm_hugetlb_page(tmp))
399 reset_vma_resv_huge_pages(tmp);
402 * Link in the new vma and copy the page table entries.
404 *pprev = tmp;
405 pprev = &tmp->vm_next;
406 tmp->vm_prev = prev;
407 prev = tmp;
409 __vma_link_rb(mm, tmp, rb_link, rb_parent);
410 rb_link = &tmp->vm_rb.rb_right;
411 rb_parent = &tmp->vm_rb;
413 mm->map_count++;
414 retval = copy_page_range(mm, oldmm, mpnt);
416 if (tmp->vm_ops && tmp->vm_ops->open)
417 tmp->vm_ops->open(tmp);
419 if (retval)
420 goto out;
422 /* a new mm has just been created */
423 arch_dup_mmap(oldmm, mm);
424 retval = 0;
425 out:
426 up_write(&mm->mmap_sem);
427 flush_tlb_mm(oldmm);
428 up_write(&oldmm->mmap_sem);
429 return retval;
430 fail_nomem_anon_vma_fork:
431 mpol_put(pol);
432 fail_nomem_policy:
433 kmem_cache_free(vm_area_cachep, tmp);
434 fail_nomem:
435 retval = -ENOMEM;
436 vm_unacct_memory(charge);
437 goto out;
440 static inline int mm_alloc_pgd(struct mm_struct * mm)
442 mm->pgd = pgd_alloc(mm);
443 if (unlikely(!mm->pgd))
444 return -ENOMEM;
445 return 0;
448 static inline void mm_free_pgd(struct mm_struct * mm)
450 pgd_free(mm, mm->pgd);
452 #else
453 #define dup_mmap(mm, oldmm) (0)
454 #define mm_alloc_pgd(mm) (0)
455 #define mm_free_pgd(mm)
456 #endif /* CONFIG_MMU */
458 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
460 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
461 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
463 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
465 static int __init coredump_filter_setup(char *s)
467 default_dump_filter =
468 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
469 MMF_DUMP_FILTER_MASK;
470 return 1;
473 __setup("coredump_filter=", coredump_filter_setup);
475 #include <linux/init_task.h>
477 static void mm_init_aio(struct mm_struct *mm)
479 #ifdef CONFIG_AIO
480 spin_lock_init(&mm->ioctx_lock);
481 INIT_HLIST_HEAD(&mm->ioctx_list);
482 #endif
485 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
487 atomic_set(&mm->mm_users, 1);
488 atomic_set(&mm->mm_count, 1);
489 init_rwsem(&mm->mmap_sem);
490 INIT_LIST_HEAD(&mm->mmlist);
491 mm->flags = (current->mm) ?
492 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
493 mm->core_state = NULL;
494 mm->nr_ptes = 0;
495 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
496 spin_lock_init(&mm->page_table_lock);
497 mm->free_area_cache = TASK_UNMAPPED_BASE;
498 mm->cached_hole_size = ~0UL;
499 mm_init_aio(mm);
500 mm_init_owner(mm, p);
501 atomic_set(&mm->oom_disable_count, 0);
503 if (likely(!mm_alloc_pgd(mm))) {
504 mm->def_flags = 0;
505 mmu_notifier_mm_init(mm);
506 return mm;
509 free_mm(mm);
510 return NULL;
514 * Allocate and initialize an mm_struct.
516 struct mm_struct * mm_alloc(void)
518 struct mm_struct * mm;
520 mm = allocate_mm();
521 if (!mm)
522 return NULL;
524 memset(mm, 0, sizeof(*mm));
525 mm_init_cpumask(mm);
526 return mm_init(mm, current);
530 * Called when the last reference to the mm
531 * is dropped: either by a lazy thread or by
532 * mmput. Free the page directory and the mm.
534 void __mmdrop(struct mm_struct *mm)
536 BUG_ON(mm == &init_mm);
537 mm_free_pgd(mm);
538 destroy_context(mm);
539 mmu_notifier_mm_destroy(mm);
540 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
541 VM_BUG_ON(mm->pmd_huge_pte);
542 #endif
543 free_mm(mm);
545 EXPORT_SYMBOL_GPL(__mmdrop);
548 * Decrement the use count and release all resources for an mm.
550 void mmput(struct mm_struct *mm)
552 might_sleep();
554 if (atomic_dec_and_test(&mm->mm_users)) {
555 exit_aio(mm);
556 ksm_exit(mm);
557 khugepaged_exit(mm); /* must run before exit_mmap */
558 exit_mmap(mm);
559 set_mm_exe_file(mm, NULL);
560 if (!list_empty(&mm->mmlist)) {
561 spin_lock(&mmlist_lock);
562 list_del(&mm->mmlist);
563 spin_unlock(&mmlist_lock);
565 put_swap_token(mm);
566 if (mm->binfmt)
567 module_put(mm->binfmt->module);
568 mmdrop(mm);
571 EXPORT_SYMBOL_GPL(mmput);
574 * We added or removed a vma mapping the executable. The vmas are only mapped
575 * during exec and are not mapped with the mmap system call.
576 * Callers must hold down_write() on the mm's mmap_sem for these
578 void added_exe_file_vma(struct mm_struct *mm)
580 mm->num_exe_file_vmas++;
583 void removed_exe_file_vma(struct mm_struct *mm)
585 mm->num_exe_file_vmas--;
586 if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
587 fput(mm->exe_file);
588 mm->exe_file = NULL;
593 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
595 if (new_exe_file)
596 get_file(new_exe_file);
597 if (mm->exe_file)
598 fput(mm->exe_file);
599 mm->exe_file = new_exe_file;
600 mm->num_exe_file_vmas = 0;
603 struct file *get_mm_exe_file(struct mm_struct *mm)
605 struct file *exe_file;
607 /* We need mmap_sem to protect against races with removal of
608 * VM_EXECUTABLE vmas */
609 down_read(&mm->mmap_sem);
610 exe_file = mm->exe_file;
611 if (exe_file)
612 get_file(exe_file);
613 up_read(&mm->mmap_sem);
614 return exe_file;
617 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
619 /* It's safe to write the exe_file pointer without exe_file_lock because
620 * this is called during fork when the task is not yet in /proc */
621 newmm->exe_file = get_mm_exe_file(oldmm);
625 * get_task_mm - acquire a reference to the task's mm
627 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
628 * this kernel workthread has transiently adopted a user mm with use_mm,
629 * to do its AIO) is not set and if so returns a reference to it, after
630 * bumping up the use count. User must release the mm via mmput()
631 * after use. Typically used by /proc and ptrace.
633 struct mm_struct *get_task_mm(struct task_struct *task)
635 struct mm_struct *mm;
637 task_lock(task);
638 mm = task->mm;
639 if (mm) {
640 if (task->flags & PF_KTHREAD)
641 mm = NULL;
642 else
643 atomic_inc(&mm->mm_users);
645 task_unlock(task);
646 return mm;
648 EXPORT_SYMBOL_GPL(get_task_mm);
650 /* Please note the differences between mmput and mm_release.
651 * mmput is called whenever we stop holding onto a mm_struct,
652 * error success whatever.
654 * mm_release is called after a mm_struct has been removed
655 * from the current process.
657 * This difference is important for error handling, when we
658 * only half set up a mm_struct for a new process and need to restore
659 * the old one. Because we mmput the new mm_struct before
660 * restoring the old one. . .
661 * Eric Biederman 10 January 1998
663 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
665 struct completion *vfork_done = tsk->vfork_done;
667 /* Get rid of any futexes when releasing the mm */
668 #ifdef CONFIG_FUTEX
669 if (unlikely(tsk->robust_list)) {
670 exit_robust_list(tsk);
671 tsk->robust_list = NULL;
673 #ifdef CONFIG_COMPAT
674 if (unlikely(tsk->compat_robust_list)) {
675 compat_exit_robust_list(tsk);
676 tsk->compat_robust_list = NULL;
678 #endif
679 if (unlikely(!list_empty(&tsk->pi_state_list)))
680 exit_pi_state_list(tsk);
681 #endif
683 /* Get rid of any cached register state */
684 deactivate_mm(tsk, mm);
686 /* notify parent sleeping on vfork() */
687 if (vfork_done) {
688 tsk->vfork_done = NULL;
689 complete(vfork_done);
693 * If we're exiting normally, clear a user-space tid field if
694 * requested. We leave this alone when dying by signal, to leave
695 * the value intact in a core dump, and to save the unnecessary
696 * trouble otherwise. Userland only wants this done for a sys_exit.
698 if (tsk->clear_child_tid) {
699 if (!(tsk->flags & PF_SIGNALED) &&
700 atomic_read(&mm->mm_users) > 1) {
702 * We don't check the error code - if userspace has
703 * not set up a proper pointer then tough luck.
705 put_user(0, tsk->clear_child_tid);
706 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
707 1, NULL, NULL, 0);
709 tsk->clear_child_tid = NULL;
714 * Allocate a new mm structure and copy contents from the
715 * mm structure of the passed in task structure.
717 struct mm_struct *dup_mm(struct task_struct *tsk)
719 struct mm_struct *mm, *oldmm = current->mm;
720 int err;
722 if (!oldmm)
723 return NULL;
725 mm = allocate_mm();
726 if (!mm)
727 goto fail_nomem;
729 memcpy(mm, oldmm, sizeof(*mm));
730 mm_init_cpumask(mm);
732 /* Initializing for Swap token stuff */
733 mm->token_priority = 0;
734 mm->last_interval = 0;
736 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
737 mm->pmd_huge_pte = NULL;
738 #endif
740 if (!mm_init(mm, tsk))
741 goto fail_nomem;
743 if (init_new_context(tsk, mm))
744 goto fail_nocontext;
746 dup_mm_exe_file(oldmm, mm);
748 err = dup_mmap(mm, oldmm);
749 if (err)
750 goto free_pt;
752 mm->hiwater_rss = get_mm_rss(mm);
753 mm->hiwater_vm = mm->total_vm;
755 if (mm->binfmt && !try_module_get(mm->binfmt->module))
756 goto free_pt;
758 return mm;
760 free_pt:
761 /* don't put binfmt in mmput, we haven't got module yet */
762 mm->binfmt = NULL;
763 mmput(mm);
765 fail_nomem:
766 return NULL;
768 fail_nocontext:
770 * If init_new_context() failed, we cannot use mmput() to free the mm
771 * because it calls destroy_context()
773 mm_free_pgd(mm);
774 free_mm(mm);
775 return NULL;
778 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
780 struct mm_struct * mm, *oldmm;
781 int retval;
783 tsk->min_flt = tsk->maj_flt = 0;
784 tsk->nvcsw = tsk->nivcsw = 0;
785 #ifdef CONFIG_DETECT_HUNG_TASK
786 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
787 #endif
789 tsk->mm = NULL;
790 tsk->active_mm = NULL;
793 * Are we cloning a kernel thread?
795 * We need to steal a active VM for that..
797 oldmm = current->mm;
798 if (!oldmm)
799 return 0;
801 if (clone_flags & CLONE_VM) {
802 atomic_inc(&oldmm->mm_users);
803 mm = oldmm;
804 goto good_mm;
807 retval = -ENOMEM;
808 mm = dup_mm(tsk);
809 if (!mm)
810 goto fail_nomem;
812 good_mm:
813 /* Initializing for Swap token stuff */
814 mm->token_priority = 0;
815 mm->last_interval = 0;
816 if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
817 atomic_inc(&mm->oom_disable_count);
819 tsk->mm = mm;
820 tsk->active_mm = mm;
821 return 0;
823 fail_nomem:
824 return retval;
827 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
829 struct fs_struct *fs = current->fs;
830 if (clone_flags & CLONE_FS) {
831 /* tsk->fs is already what we want */
832 spin_lock(&fs->lock);
833 if (fs->in_exec) {
834 spin_unlock(&fs->lock);
835 return -EAGAIN;
837 fs->users++;
838 spin_unlock(&fs->lock);
839 return 0;
841 tsk->fs = copy_fs_struct(fs);
842 if (!tsk->fs)
843 return -ENOMEM;
844 return 0;
847 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
849 struct files_struct *oldf, *newf;
850 int error = 0;
853 * A background process may not have any files ...
855 oldf = current->files;
856 if (!oldf)
857 goto out;
859 if (clone_flags & CLONE_FILES) {
860 atomic_inc(&oldf->count);
861 goto out;
864 newf = dup_fd(oldf, &error);
865 if (!newf)
866 goto out;
868 tsk->files = newf;
869 error = 0;
870 out:
871 return error;
874 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
876 #ifdef CONFIG_BLOCK
877 struct io_context *ioc = current->io_context;
879 if (!ioc)
880 return 0;
882 * Share io context with parent, if CLONE_IO is set
884 if (clone_flags & CLONE_IO) {
885 tsk->io_context = ioc_task_link(ioc);
886 if (unlikely(!tsk->io_context))
887 return -ENOMEM;
888 } else if (ioprio_valid(ioc->ioprio)) {
889 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
890 if (unlikely(!tsk->io_context))
891 return -ENOMEM;
893 tsk->io_context->ioprio = ioc->ioprio;
895 #endif
896 return 0;
899 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
901 struct sighand_struct *sig;
903 if (clone_flags & CLONE_SIGHAND) {
904 atomic_inc(&current->sighand->count);
905 return 0;
907 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
908 rcu_assign_pointer(tsk->sighand, sig);
909 if (!sig)
910 return -ENOMEM;
911 atomic_set(&sig->count, 1);
912 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
913 return 0;
916 void __cleanup_sighand(struct sighand_struct *sighand)
918 if (atomic_dec_and_test(&sighand->count))
919 kmem_cache_free(sighand_cachep, sighand);
924 * Initialize POSIX timer handling for a thread group.
926 static void posix_cpu_timers_init_group(struct signal_struct *sig)
928 unsigned long cpu_limit;
930 /* Thread group counters. */
931 thread_group_cputime_init(sig);
933 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
934 if (cpu_limit != RLIM_INFINITY) {
935 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
936 sig->cputimer.running = 1;
939 /* The timer lists. */
940 INIT_LIST_HEAD(&sig->cpu_timers[0]);
941 INIT_LIST_HEAD(&sig->cpu_timers[1]);
942 INIT_LIST_HEAD(&sig->cpu_timers[2]);
945 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
947 struct signal_struct *sig;
949 if (clone_flags & CLONE_THREAD)
950 return 0;
952 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
953 tsk->signal = sig;
954 if (!sig)
955 return -ENOMEM;
957 sig->nr_threads = 1;
958 atomic_set(&sig->live, 1);
959 atomic_set(&sig->sigcnt, 1);
960 init_waitqueue_head(&sig->wait_chldexit);
961 if (clone_flags & CLONE_NEWPID)
962 sig->flags |= SIGNAL_UNKILLABLE;
963 sig->curr_target = tsk;
964 init_sigpending(&sig->shared_pending);
965 INIT_LIST_HEAD(&sig->posix_timers);
967 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
968 sig->real_timer.function = it_real_fn;
970 task_lock(current->group_leader);
971 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
972 task_unlock(current->group_leader);
974 posix_cpu_timers_init_group(sig);
976 tty_audit_fork(sig);
977 sched_autogroup_fork(sig);
979 #ifdef CONFIG_CGROUPS
980 init_rwsem(&sig->threadgroup_fork_lock);
981 #endif
983 sig->oom_adj = current->signal->oom_adj;
984 sig->oom_score_adj = current->signal->oom_score_adj;
985 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
987 mutex_init(&sig->cred_guard_mutex);
989 return 0;
992 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
994 unsigned long new_flags = p->flags;
996 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
997 new_flags |= PF_FORKNOEXEC;
998 new_flags |= PF_STARTING;
999 p->flags = new_flags;
1000 clear_freeze_flag(p);
1003 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1005 current->clear_child_tid = tidptr;
1007 return task_pid_vnr(current);
1010 static void rt_mutex_init_task(struct task_struct *p)
1012 raw_spin_lock_init(&p->pi_lock);
1013 #ifdef CONFIG_RT_MUTEXES
1014 plist_head_init(&p->pi_waiters);
1015 p->pi_blocked_on = NULL;
1016 #endif
1019 #ifdef CONFIG_MM_OWNER
1020 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1022 mm->owner = p;
1024 #endif /* CONFIG_MM_OWNER */
1027 * Initialize POSIX timer handling for a single task.
1029 static void posix_cpu_timers_init(struct task_struct *tsk)
1031 tsk->cputime_expires.prof_exp = cputime_zero;
1032 tsk->cputime_expires.virt_exp = cputime_zero;
1033 tsk->cputime_expires.sched_exp = 0;
1034 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1035 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1036 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1040 * This creates a new process as a copy of the old one,
1041 * but does not actually start it yet.
1043 * It copies the registers, and all the appropriate
1044 * parts of the process environment (as per the clone
1045 * flags). The actual kick-off is left to the caller.
1047 static struct task_struct *copy_process(unsigned long clone_flags,
1048 unsigned long stack_start,
1049 struct pt_regs *regs,
1050 unsigned long stack_size,
1051 int __user *child_tidptr,
1052 struct pid *pid,
1053 int trace)
1055 int retval;
1056 struct task_struct *p;
1057 int cgroup_callbacks_done = 0;
1059 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1060 return ERR_PTR(-EINVAL);
1063 * Thread groups must share signals as well, and detached threads
1064 * can only be started up within the thread group.
1066 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1067 return ERR_PTR(-EINVAL);
1070 * Shared signal handlers imply shared VM. By way of the above,
1071 * thread groups also imply shared VM. Blocking this case allows
1072 * for various simplifications in other code.
1074 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1075 return ERR_PTR(-EINVAL);
1078 * Siblings of global init remain as zombies on exit since they are
1079 * not reaped by their parent (swapper). To solve this and to avoid
1080 * multi-rooted process trees, prevent global and container-inits
1081 * from creating siblings.
1083 if ((clone_flags & CLONE_PARENT) &&
1084 current->signal->flags & SIGNAL_UNKILLABLE)
1085 return ERR_PTR(-EINVAL);
1087 retval = security_task_create(clone_flags);
1088 if (retval)
1089 goto fork_out;
1091 retval = -ENOMEM;
1092 p = dup_task_struct(current);
1093 if (!p)
1094 goto fork_out;
1096 ftrace_graph_init_task(p);
1098 rt_mutex_init_task(p);
1100 #ifdef CONFIG_PROVE_LOCKING
1101 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1102 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1103 #endif
1104 retval = -EAGAIN;
1105 if (atomic_read(&p->real_cred->user->processes) >=
1106 task_rlimit(p, RLIMIT_NPROC)) {
1107 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1108 p->real_cred->user != INIT_USER)
1109 goto bad_fork_free;
1112 retval = copy_creds(p, clone_flags);
1113 if (retval < 0)
1114 goto bad_fork_free;
1117 * If multiple threads are within copy_process(), then this check
1118 * triggers too late. This doesn't hurt, the check is only there
1119 * to stop root fork bombs.
1121 retval = -EAGAIN;
1122 if (nr_threads >= max_threads)
1123 goto bad_fork_cleanup_count;
1125 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1126 goto bad_fork_cleanup_count;
1128 p->did_exec = 0;
1129 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1130 copy_flags(clone_flags, p);
1131 INIT_LIST_HEAD(&p->children);
1132 INIT_LIST_HEAD(&p->sibling);
1133 rcu_copy_process(p);
1134 p->vfork_done = NULL;
1135 spin_lock_init(&p->alloc_lock);
1137 init_sigpending(&p->pending);
1139 p->utime = cputime_zero;
1140 p->stime = cputime_zero;
1141 p->gtime = cputime_zero;
1142 p->utimescaled = cputime_zero;
1143 p->stimescaled = cputime_zero;
1144 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1145 p->prev_utime = cputime_zero;
1146 p->prev_stime = cputime_zero;
1147 #endif
1148 #if defined(SPLIT_RSS_COUNTING)
1149 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1150 #endif
1152 p->default_timer_slack_ns = current->timer_slack_ns;
1154 task_io_accounting_init(&p->ioac);
1155 acct_clear_integrals(p);
1157 posix_cpu_timers_init(p);
1159 do_posix_clock_monotonic_gettime(&p->start_time);
1160 p->real_start_time = p->start_time;
1161 monotonic_to_bootbased(&p->real_start_time);
1162 p->io_context = NULL;
1163 p->audit_context = NULL;
1164 if (clone_flags & CLONE_THREAD)
1165 threadgroup_fork_read_lock(current);
1166 cgroup_fork(p);
1167 #ifdef CONFIG_NUMA
1168 p->mempolicy = mpol_dup(p->mempolicy);
1169 if (IS_ERR(p->mempolicy)) {
1170 retval = PTR_ERR(p->mempolicy);
1171 p->mempolicy = NULL;
1172 goto bad_fork_cleanup_cgroup;
1174 mpol_fix_fork_child_flag(p);
1175 #endif
1176 #ifdef CONFIG_TRACE_IRQFLAGS
1177 p->irq_events = 0;
1178 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1179 p->hardirqs_enabled = 1;
1180 #else
1181 p->hardirqs_enabled = 0;
1182 #endif
1183 p->hardirq_enable_ip = 0;
1184 p->hardirq_enable_event = 0;
1185 p->hardirq_disable_ip = _THIS_IP_;
1186 p->hardirq_disable_event = 0;
1187 p->softirqs_enabled = 1;
1188 p->softirq_enable_ip = _THIS_IP_;
1189 p->softirq_enable_event = 0;
1190 p->softirq_disable_ip = 0;
1191 p->softirq_disable_event = 0;
1192 p->hardirq_context = 0;
1193 p->softirq_context = 0;
1194 #endif
1195 #ifdef CONFIG_LOCKDEP
1196 p->lockdep_depth = 0; /* no locks held yet */
1197 p->curr_chain_key = 0;
1198 p->lockdep_recursion = 0;
1199 #endif
1201 #ifdef CONFIG_DEBUG_MUTEXES
1202 p->blocked_on = NULL; /* not blocked yet */
1203 #endif
1204 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1205 p->memcg_batch.do_batch = 0;
1206 p->memcg_batch.memcg = NULL;
1207 #endif
1209 /* Perform scheduler related setup. Assign this task to a CPU. */
1210 sched_fork(p);
1212 retval = perf_event_init_task(p);
1213 if (retval)
1214 goto bad_fork_cleanup_policy;
1216 if ((retval = audit_alloc(p)))
1217 goto bad_fork_cleanup_policy;
1218 /* copy all the process information */
1219 if ((retval = copy_semundo(clone_flags, p)))
1220 goto bad_fork_cleanup_audit;
1221 if ((retval = copy_files(clone_flags, p)))
1222 goto bad_fork_cleanup_semundo;
1223 if ((retval = copy_fs(clone_flags, p)))
1224 goto bad_fork_cleanup_files;
1225 if ((retval = copy_sighand(clone_flags, p)))
1226 goto bad_fork_cleanup_fs;
1227 if ((retval = copy_signal(clone_flags, p)))
1228 goto bad_fork_cleanup_sighand;
1229 if ((retval = copy_mm(clone_flags, p)))
1230 goto bad_fork_cleanup_signal;
1231 if ((retval = copy_namespaces(clone_flags, p)))
1232 goto bad_fork_cleanup_mm;
1233 if ((retval = copy_io(clone_flags, p)))
1234 goto bad_fork_cleanup_namespaces;
1235 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1236 if (retval)
1237 goto bad_fork_cleanup_io;
1239 if (pid != &init_struct_pid) {
1240 retval = -ENOMEM;
1241 pid = alloc_pid(p->nsproxy->pid_ns);
1242 if (!pid)
1243 goto bad_fork_cleanup_io;
1246 p->pid = pid_nr(pid);
1247 p->tgid = p->pid;
1248 if (clone_flags & CLONE_THREAD)
1249 p->tgid = current->tgid;
1251 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1253 * Clear TID on mm_release()?
1255 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1256 #ifdef CONFIG_BLOCK
1257 p->plug = NULL;
1258 #endif
1259 #ifdef CONFIG_FUTEX
1260 p->robust_list = NULL;
1261 #ifdef CONFIG_COMPAT
1262 p->compat_robust_list = NULL;
1263 #endif
1264 INIT_LIST_HEAD(&p->pi_state_list);
1265 p->pi_state_cache = NULL;
1266 #endif
1268 * sigaltstack should be cleared when sharing the same VM
1270 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1271 p->sas_ss_sp = p->sas_ss_size = 0;
1274 * Syscall tracing and stepping should be turned off in the
1275 * child regardless of CLONE_PTRACE.
1277 user_disable_single_step(p);
1278 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1279 #ifdef TIF_SYSCALL_EMU
1280 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1281 #endif
1282 clear_all_latency_tracing(p);
1284 /* ok, now we should be set up.. */
1285 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1286 p->pdeath_signal = 0;
1287 p->exit_state = 0;
1290 * Ok, make it visible to the rest of the system.
1291 * We dont wake it up yet.
1293 p->group_leader = p;
1294 INIT_LIST_HEAD(&p->thread_group);
1296 /* Now that the task is set up, run cgroup callbacks if
1297 * necessary. We need to run them before the task is visible
1298 * on the tasklist. */
1299 cgroup_fork_callbacks(p);
1300 cgroup_callbacks_done = 1;
1302 /* Need tasklist lock for parent etc handling! */
1303 write_lock_irq(&tasklist_lock);
1305 /* CLONE_PARENT re-uses the old parent */
1306 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1307 p->real_parent = current->real_parent;
1308 p->parent_exec_id = current->parent_exec_id;
1309 } else {
1310 p->real_parent = current;
1311 p->parent_exec_id = current->self_exec_id;
1314 spin_lock(&current->sighand->siglock);
1317 * Process group and session signals need to be delivered to just the
1318 * parent before the fork or both the parent and the child after the
1319 * fork. Restart if a signal comes in before we add the new process to
1320 * it's process group.
1321 * A fatal signal pending means that current will exit, so the new
1322 * thread can't slip out of an OOM kill (or normal SIGKILL).
1324 recalc_sigpending();
1325 if (signal_pending(current)) {
1326 spin_unlock(&current->sighand->siglock);
1327 write_unlock_irq(&tasklist_lock);
1328 retval = -ERESTARTNOINTR;
1329 goto bad_fork_free_pid;
1332 if (clone_flags & CLONE_THREAD) {
1333 current->signal->nr_threads++;
1334 atomic_inc(&current->signal->live);
1335 atomic_inc(&current->signal->sigcnt);
1336 p->group_leader = current->group_leader;
1337 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1340 if (likely(p->pid)) {
1341 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1343 if (thread_group_leader(p)) {
1344 if (is_child_reaper(pid))
1345 p->nsproxy->pid_ns->child_reaper = p;
1347 p->signal->leader_pid = pid;
1348 p->signal->tty = tty_kref_get(current->signal->tty);
1349 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1350 attach_pid(p, PIDTYPE_SID, task_session(current));
1351 list_add_tail(&p->sibling, &p->real_parent->children);
1352 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1353 __this_cpu_inc(process_counts);
1355 attach_pid(p, PIDTYPE_PID, pid);
1356 nr_threads++;
1359 total_forks++;
1360 spin_unlock(&current->sighand->siglock);
1361 write_unlock_irq(&tasklist_lock);
1362 proc_fork_connector(p);
1363 cgroup_post_fork(p);
1364 if (clone_flags & CLONE_THREAD)
1365 threadgroup_fork_read_unlock(current);
1366 perf_event_fork(p);
1367 return p;
1369 bad_fork_free_pid:
1370 if (pid != &init_struct_pid)
1371 free_pid(pid);
1372 bad_fork_cleanup_io:
1373 if (p->io_context)
1374 exit_io_context(p);
1375 bad_fork_cleanup_namespaces:
1376 exit_task_namespaces(p);
1377 bad_fork_cleanup_mm:
1378 if (p->mm) {
1379 task_lock(p);
1380 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1381 atomic_dec(&p->mm->oom_disable_count);
1382 task_unlock(p);
1383 mmput(p->mm);
1385 bad_fork_cleanup_signal:
1386 if (!(clone_flags & CLONE_THREAD))
1387 free_signal_struct(p->signal);
1388 bad_fork_cleanup_sighand:
1389 __cleanup_sighand(p->sighand);
1390 bad_fork_cleanup_fs:
1391 exit_fs(p); /* blocking */
1392 bad_fork_cleanup_files:
1393 exit_files(p); /* blocking */
1394 bad_fork_cleanup_semundo:
1395 exit_sem(p);
1396 bad_fork_cleanup_audit:
1397 audit_free(p);
1398 bad_fork_cleanup_policy:
1399 perf_event_free_task(p);
1400 #ifdef CONFIG_NUMA
1401 mpol_put(p->mempolicy);
1402 bad_fork_cleanup_cgroup:
1403 #endif
1404 if (clone_flags & CLONE_THREAD)
1405 threadgroup_fork_read_unlock(current);
1406 cgroup_exit(p, cgroup_callbacks_done);
1407 delayacct_tsk_free(p);
1408 module_put(task_thread_info(p)->exec_domain->module);
1409 bad_fork_cleanup_count:
1410 atomic_dec(&p->cred->user->processes);
1411 exit_creds(p);
1412 bad_fork_free:
1413 free_task(p);
1414 fork_out:
1415 return ERR_PTR(retval);
1418 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1420 memset(regs, 0, sizeof(struct pt_regs));
1421 return regs;
1424 static inline void init_idle_pids(struct pid_link *links)
1426 enum pid_type type;
1428 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1429 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1430 links[type].pid = &init_struct_pid;
1434 struct task_struct * __cpuinit fork_idle(int cpu)
1436 struct task_struct *task;
1437 struct pt_regs regs;
1439 task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1440 &init_struct_pid, 0);
1441 if (!IS_ERR(task)) {
1442 init_idle_pids(task->pids);
1443 init_idle(task, cpu);
1446 return task;
1450 * Ok, this is the main fork-routine.
1452 * It copies the process, and if successful kick-starts
1453 * it and waits for it to finish using the VM if required.
1455 long do_fork(unsigned long clone_flags,
1456 unsigned long stack_start,
1457 struct pt_regs *regs,
1458 unsigned long stack_size,
1459 int __user *parent_tidptr,
1460 int __user *child_tidptr)
1462 struct task_struct *p;
1463 int trace = 0;
1464 long nr;
1467 * Do some preliminary argument and permissions checking before we
1468 * actually start allocating stuff
1470 if (clone_flags & CLONE_NEWUSER) {
1471 if (clone_flags & CLONE_THREAD)
1472 return -EINVAL;
1473 /* hopefully this check will go away when userns support is
1474 * complete
1476 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1477 !capable(CAP_SETGID))
1478 return -EPERM;
1482 * Determine whether and which event to report to ptracer. When
1483 * called from kernel_thread or CLONE_UNTRACED is explicitly
1484 * requested, no event is reported; otherwise, report if the event
1485 * for the type of forking is enabled.
1487 if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1488 if (clone_flags & CLONE_VFORK)
1489 trace = PTRACE_EVENT_VFORK;
1490 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1491 trace = PTRACE_EVENT_CLONE;
1492 else
1493 trace = PTRACE_EVENT_FORK;
1495 if (likely(!ptrace_event_enabled(current, trace)))
1496 trace = 0;
1499 p = copy_process(clone_flags, stack_start, regs, stack_size,
1500 child_tidptr, NULL, trace);
1502 * Do this prior waking up the new thread - the thread pointer
1503 * might get invalid after that point, if the thread exits quickly.
1505 if (!IS_ERR(p)) {
1506 struct completion vfork;
1508 trace_sched_process_fork(current, p);
1510 nr = task_pid_vnr(p);
1512 if (clone_flags & CLONE_PARENT_SETTID)
1513 put_user(nr, parent_tidptr);
1515 if (clone_flags & CLONE_VFORK) {
1516 p->vfork_done = &vfork;
1517 init_completion(&vfork);
1520 audit_finish_fork(p);
1523 * We set PF_STARTING at creation in case tracing wants to
1524 * use this to distinguish a fully live task from one that
1525 * hasn't finished SIGSTOP raising yet. Now we clear it
1526 * and set the child going.
1528 p->flags &= ~PF_STARTING;
1530 wake_up_new_task(p);
1532 /* forking complete and child started to run, tell ptracer */
1533 if (unlikely(trace))
1534 ptrace_event(trace, nr);
1536 if (clone_flags & CLONE_VFORK) {
1537 freezer_do_not_count();
1538 wait_for_completion(&vfork);
1539 freezer_count();
1540 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1542 } else {
1543 nr = PTR_ERR(p);
1545 return nr;
1548 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1549 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1550 #endif
1552 static void sighand_ctor(void *data)
1554 struct sighand_struct *sighand = data;
1556 spin_lock_init(&sighand->siglock);
1557 init_waitqueue_head(&sighand->signalfd_wqh);
1560 void __init proc_caches_init(void)
1562 sighand_cachep = kmem_cache_create("sighand_cache",
1563 sizeof(struct sighand_struct), 0,
1564 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1565 SLAB_NOTRACK, sighand_ctor);
1566 signal_cachep = kmem_cache_create("signal_cache",
1567 sizeof(struct signal_struct), 0,
1568 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1569 files_cachep = kmem_cache_create("files_cache",
1570 sizeof(struct files_struct), 0,
1571 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1572 fs_cachep = kmem_cache_create("fs_cache",
1573 sizeof(struct fs_struct), 0,
1574 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1576 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1577 * whole struct cpumask for the OFFSTACK case. We could change
1578 * this to *only* allocate as much of it as required by the
1579 * maximum number of CPU's we can ever have. The cpumask_allocation
1580 * is at the end of the structure, exactly for that reason.
1582 mm_cachep = kmem_cache_create("mm_struct",
1583 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1584 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1585 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1586 mmap_init();
1587 nsproxy_cache_init();
1591 * Check constraints on flags passed to the unshare system call.
1593 static int check_unshare_flags(unsigned long unshare_flags)
1595 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1596 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1597 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1598 return -EINVAL;
1600 * Not implemented, but pretend it works if there is nothing to
1601 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1602 * needs to unshare vm.
1604 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1605 /* FIXME: get_task_mm() increments ->mm_users */
1606 if (atomic_read(&current->mm->mm_users) > 1)
1607 return -EINVAL;
1610 return 0;
1614 * Unshare the filesystem structure if it is being shared
1616 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1618 struct fs_struct *fs = current->fs;
1620 if (!(unshare_flags & CLONE_FS) || !fs)
1621 return 0;
1623 /* don't need lock here; in the worst case we'll do useless copy */
1624 if (fs->users == 1)
1625 return 0;
1627 *new_fsp = copy_fs_struct(fs);
1628 if (!*new_fsp)
1629 return -ENOMEM;
1631 return 0;
1635 * Unshare file descriptor table if it is being shared
1637 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1639 struct files_struct *fd = current->files;
1640 int error = 0;
1642 if ((unshare_flags & CLONE_FILES) &&
1643 (fd && atomic_read(&fd->count) > 1)) {
1644 *new_fdp = dup_fd(fd, &error);
1645 if (!*new_fdp)
1646 return error;
1649 return 0;
1653 * unshare allows a process to 'unshare' part of the process
1654 * context which was originally shared using clone. copy_*
1655 * functions used by do_fork() cannot be used here directly
1656 * because they modify an inactive task_struct that is being
1657 * constructed. Here we are modifying the current, active,
1658 * task_struct.
1660 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1662 struct fs_struct *fs, *new_fs = NULL;
1663 struct files_struct *fd, *new_fd = NULL;
1664 struct nsproxy *new_nsproxy = NULL;
1665 int do_sysvsem = 0;
1666 int err;
1668 err = check_unshare_flags(unshare_flags);
1669 if (err)
1670 goto bad_unshare_out;
1673 * If unsharing namespace, must also unshare filesystem information.
1675 if (unshare_flags & CLONE_NEWNS)
1676 unshare_flags |= CLONE_FS;
1678 * CLONE_NEWIPC must also detach from the undolist: after switching
1679 * to a new ipc namespace, the semaphore arrays from the old
1680 * namespace are unreachable.
1682 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1683 do_sysvsem = 1;
1684 if ((err = unshare_fs(unshare_flags, &new_fs)))
1685 goto bad_unshare_out;
1686 if ((err = unshare_fd(unshare_flags, &new_fd)))
1687 goto bad_unshare_cleanup_fs;
1688 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1689 new_fs)))
1690 goto bad_unshare_cleanup_fd;
1692 if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1693 if (do_sysvsem) {
1695 * CLONE_SYSVSEM is equivalent to sys_exit().
1697 exit_sem(current);
1700 if (new_nsproxy) {
1701 switch_task_namespaces(current, new_nsproxy);
1702 new_nsproxy = NULL;
1705 task_lock(current);
1707 if (new_fs) {
1708 fs = current->fs;
1709 spin_lock(&fs->lock);
1710 current->fs = new_fs;
1711 if (--fs->users)
1712 new_fs = NULL;
1713 else
1714 new_fs = fs;
1715 spin_unlock(&fs->lock);
1718 if (new_fd) {
1719 fd = current->files;
1720 current->files = new_fd;
1721 new_fd = fd;
1724 task_unlock(current);
1727 if (new_nsproxy)
1728 put_nsproxy(new_nsproxy);
1730 bad_unshare_cleanup_fd:
1731 if (new_fd)
1732 put_files_struct(new_fd);
1734 bad_unshare_cleanup_fs:
1735 if (new_fs)
1736 free_fs_struct(new_fs);
1738 bad_unshare_out:
1739 return err;
1743 * Helper to unshare the files of the current task.
1744 * We don't want to expose copy_files internals to
1745 * the exec layer of the kernel.
1748 int unshare_files(struct files_struct **displaced)
1750 struct task_struct *task = current;
1751 struct files_struct *copy = NULL;
1752 int error;
1754 error = unshare_fd(CLONE_FILES, &copy);
1755 if (error || !copy) {
1756 *displaced = NULL;
1757 return error;
1759 *displaced = task->files;
1760 task_lock(task);
1761 task->files = copy;
1762 task_unlock(task);
1763 return 0;