cpusets: randomize node rotor used in cpuset_mem_spread_node()
[linux-2.6.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
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()'
12  */
13
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>
69
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>
76
77 #include <trace/events/sched.h>
78
79 /*
80  * Protected counters by write_lock_irq(&tasklist_lock)
81  */
82 unsigned long total_forks;      /* Handle normal Linux uptimes. */
83 int nr_threads;                 /* The idle threads do not count.. */
84
85 int max_threads;                /* tunable limit on nr_threads */
86
87 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
88
89 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
90
91 #ifdef CONFIG_PROVE_RCU
92 int lockdep_tasklist_lock_is_held(void)
93 {
94         return lockdep_is_held(&tasklist_lock);
95 }
96 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
97 #endif /* #ifdef CONFIG_PROVE_RCU */
98
99 int nr_processes(void)
100 {
101         int cpu;
102         int total = 0;
103
104         for_each_possible_cpu(cpu)
105                 total += per_cpu(process_counts, cpu);
106
107         return total;
108 }
109
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
117
118 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
119 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
120                                                   int node)
121 {
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);
128
129         return page ? page_address(page) : NULL;
130 }
131
132 static inline void free_thread_info(struct thread_info *ti)
133 {
134         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
135 }
136 #endif
137
138 /* SLAB cache for signal_struct structures (tsk->signal) */
139 static struct kmem_cache *signal_cachep;
140
141 /* SLAB cache for sighand_struct structures (tsk->sighand) */
142 struct kmem_cache *sighand_cachep;
143
144 /* SLAB cache for files_struct structures (tsk->files) */
145 struct kmem_cache *files_cachep;
146
147 /* SLAB cache for fs_struct structures (tsk->fs) */
148 struct kmem_cache *fs_cachep;
149
150 /* SLAB cache for vm_area_struct structures */
151 struct kmem_cache *vm_area_cachep;
152
153 /* SLAB cache for mm_struct structures (tsk->mm) */
154 static struct kmem_cache *mm_cachep;
155
156 static void account_kernel_stack(struct thread_info *ti, int account)
157 {
158         struct zone *zone = page_zone(virt_to_page(ti));
159
160         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
161 }
162
163 void free_task(struct task_struct *tsk)
164 {
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);
171 }
172 EXPORT_SYMBOL(free_task);
173
174 static inline void free_signal_struct(struct signal_struct *sig)
175 {
176         taskstats_tgid_free(sig);
177         sched_autogroup_exit(sig);
178         kmem_cache_free(signal_cachep, sig);
179 }
180
181 static inline void put_signal_struct(struct signal_struct *sig)
182 {
183         if (atomic_dec_and_test(&sig->sigcnt))
184                 free_signal_struct(sig);
185 }
186
187 void __put_task_struct(struct task_struct *tsk)
188 {
189         WARN_ON(!tsk->exit_state);
190         WARN_ON(atomic_read(&tsk->usage));
191         WARN_ON(tsk == current);
192
193         exit_creds(tsk);
194         delayacct_tsk_free(tsk);
195         put_signal_struct(tsk->signal);
196
197         if (!profile_handoff_task(tsk))
198                 free_task(tsk);
199 }
200 EXPORT_SYMBOL_GPL(__put_task_struct);
201
202 /*
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.
205  */
206 #ifndef arch_task_cache_init
207 #define arch_task_cache_init()
208 #endif
209
210 void __init fork_init(unsigned long mempages)
211 {
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
221
222         /* do the arch specific task caches init */
223         arch_task_cache_init();
224
225         /*
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.
229          */
230         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
231
232         /*
233          * we need to allow at least 20 threads to boot a system
234          */
235         if(max_threads < 20)
236                 max_threads = 20;
237
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];
242 }
243
244 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
245                                                struct task_struct *src)
246 {
247         *dst = *src;
248         return 0;
249 }
250
251 static struct task_struct *dup_task_struct(struct task_struct *orig)
252 {
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;
258
259         prepare_to_copy(orig);
260
261         tsk = alloc_task_struct_node(node);
262         if (!tsk)
263                 return NULL;
264
265         ti = alloc_thread_info_node(tsk, node);
266         if (!ti) {
267                 free_task_struct(tsk);
268                 return NULL;
269         }
270
271         err = arch_dup_task_struct(tsk, orig);
272         if (err)
273                 goto out;
274
275         tsk->stack = ti;
276
277         err = prop_local_init_single(&tsk->dirties);
278         if (err)
279                 goto out;
280
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 */
286
287 #ifdef CONFIG_CC_STACKPROTECTOR
288         tsk->stack_canary = get_random_int();
289 #endif
290
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;
297
298         account_kernel_stack(ti, 1);
299
300         return tsk;
301
302 out:
303         free_thread_info(ti);
304         free_task_struct(tsk);
305         return NULL;
306 }
307
308 #ifdef CONFIG_MMU
309 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
310 {
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;
316
317         down_write(&oldmm->mmap_sem);
318         flush_cache_dup_mm(oldmm);
319         /*
320          * Not linked in yet - no deadlock potential:
321          */
322         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
323
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;
341
342         prev = NULL;
343         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
344                 struct file *file;
345
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;
352                 }
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;
359                 }
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;
379
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);
391                 }
392
393                 /*
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
397                  */
398                 if (is_vm_hugetlb_page(tmp))
399                         reset_vma_resv_huge_pages(tmp);
400
401                 /*
402                  * Link in the new vma and copy the page table entries.
403                  */
404                 *pprev = tmp;
405                 pprev = &tmp->vm_next;
406                 tmp->vm_prev = prev;
407                 prev = tmp;
408
409                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
410                 rb_link = &tmp->vm_rb.rb_right;
411                 rb_parent = &tmp->vm_rb;
412
413                 mm->map_count++;
414                 retval = copy_page_range(mm, oldmm, mpnt);
415
416                 if (tmp->vm_ops && tmp->vm_ops->open)
417                         tmp->vm_ops->open(tmp);
418
419                 if (retval)
420                         goto out;
421         }
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;
438 }
439
440 static inline int mm_alloc_pgd(struct mm_struct * mm)
441 {
442         mm->pgd = pgd_alloc(mm);
443         if (unlikely(!mm->pgd))
444                 return -ENOMEM;
445         return 0;
446 }
447
448 static inline void mm_free_pgd(struct mm_struct * mm)
449 {
450         pgd_free(mm, mm->pgd);
451 }
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 */
457
458 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
459
460 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
461 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
462
463 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
464
465 static int __init coredump_filter_setup(char *s)
466 {
467         default_dump_filter =
468                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
469                 MMF_DUMP_FILTER_MASK;
470         return 1;
471 }
472
473 __setup("coredump_filter=", coredump_filter_setup);
474
475 #include <linux/init_task.h>
476
477 static void mm_init_aio(struct mm_struct *mm)
478 {
479 #ifdef CONFIG_AIO
480         spin_lock_init(&mm->ioctx_lock);
481         INIT_HLIST_HEAD(&mm->ioctx_list);
482 #endif
483 }
484
485 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
486 {
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);
502
503         if (likely(!mm_alloc_pgd(mm))) {
504                 mm->def_flags = 0;
505                 mmu_notifier_mm_init(mm);
506                 return mm;
507         }
508
509         free_mm(mm);
510         return NULL;
511 }
512
513 /*
514  * Allocate and initialize an mm_struct.
515  */
516 struct mm_struct * mm_alloc(void)
517 {
518         struct mm_struct * mm;
519
520         mm = allocate_mm();
521         if (!mm)
522                 return NULL;
523
524         memset(mm, 0, sizeof(*mm));
525         mm_init_cpumask(mm);
526         return mm_init(mm, current);
527 }
528
529 /*
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.
533  */
534 void __mmdrop(struct mm_struct *mm)
535 {
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);
544 }
545 EXPORT_SYMBOL_GPL(__mmdrop);
546
547 /*
548  * Decrement the use count and release all resources for an mm.
549  */
550 void mmput(struct mm_struct *mm)
551 {
552         might_sleep();
553
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);
564                 }
565                 put_swap_token(mm);
566                 if (mm->binfmt)
567                         module_put(mm->binfmt->module);
568                 mmdrop(mm);
569         }
570 }
571 EXPORT_SYMBOL_GPL(mmput);
572
573 /*
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
577  */
578 void added_exe_file_vma(struct mm_struct *mm)
579 {
580         mm->num_exe_file_vmas++;
581 }
582
583 void removed_exe_file_vma(struct mm_struct *mm)
584 {
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;
589         }
590
591 }
592
593 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
594 {
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;
601 }
602
603 struct file *get_mm_exe_file(struct mm_struct *mm)
604 {
605         struct file *exe_file;
606
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;
615 }
616
617 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
618 {
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);
622 }
623
624 /**
625  * get_task_mm - acquire a reference to the task's mm
626  *
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.
632  */
633 struct mm_struct *get_task_mm(struct task_struct *task)
634 {
635         struct mm_struct *mm;
636
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);
644         }
645         task_unlock(task);
646         return mm;
647 }
648 EXPORT_SYMBOL_GPL(get_task_mm);
649
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.
653  *
654  * mm_release is called after a mm_struct has been removed
655  * from the current process.
656  *
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
662  */
663 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
664 {
665         struct completion *vfork_done = tsk->vfork_done;
666
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;
672         }
673 #ifdef CONFIG_COMPAT
674         if (unlikely(tsk->compat_robust_list)) {
675                 compat_exit_robust_list(tsk);
676                 tsk->compat_robust_list = NULL;
677         }
678 #endif
679         if (unlikely(!list_empty(&tsk->pi_state_list)))
680                 exit_pi_state_list(tsk);
681 #endif
682
683         /* Get rid of any cached register state */
684         deactivate_mm(tsk, mm);
685
686         /* notify parent sleeping on vfork() */
687         if (vfork_done) {
688                 tsk->vfork_done = NULL;
689                 complete(vfork_done);
690         }
691
692         /*
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.
697          */
698         if (tsk->clear_child_tid) {
699                 if (!(tsk->flags & PF_SIGNALED) &&
700                     atomic_read(&mm->mm_users) > 1) {
701                         /*
702                          * We don't check the error code - if userspace has
703                          * not set up a proper pointer then tough luck.
704                          */
705                         put_user(0, tsk->clear_child_tid);
706                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
707                                         1, NULL, NULL, 0);
708                 }
709                 tsk->clear_child_tid = NULL;
710         }
711 }
712
713 /*
714  * Allocate a new mm structure and copy contents from the
715  * mm structure of the passed in task structure.
716  */
717 struct mm_struct *dup_mm(struct task_struct *tsk)
718 {
719         struct mm_struct *mm, *oldmm = current->mm;
720         int err;
721
722         if (!oldmm)
723                 return NULL;
724
725         mm = allocate_mm();
726         if (!mm)
727                 goto fail_nomem;
728
729         memcpy(mm, oldmm, sizeof(*mm));
730         mm_init_cpumask(mm);
731
732         /* Initializing for Swap token stuff */
733         mm->token_priority = 0;
734         mm->last_interval = 0;
735
736 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
737         mm->pmd_huge_pte = NULL;
738 #endif
739
740         if (!mm_init(mm, tsk))
741                 goto fail_nomem;
742
743         if (init_new_context(tsk, mm))
744                 goto fail_nocontext;
745
746         dup_mm_exe_file(oldmm, mm);
747
748         err = dup_mmap(mm, oldmm);
749         if (err)
750                 goto free_pt;
751
752         mm->hiwater_rss = get_mm_rss(mm);
753         mm->hiwater_vm = mm->total_vm;
754
755         if (mm->binfmt && !try_module_get(mm->binfmt->module))
756                 goto free_pt;
757
758         return mm;
759
760 free_pt:
761         /* don't put binfmt in mmput, we haven't got module yet */
762         mm->binfmt = NULL;
763         mmput(mm);
764
765 fail_nomem:
766         return NULL;
767
768 fail_nocontext:
769         /*
770          * If init_new_context() failed, we cannot use mmput() to free the mm
771          * because it calls destroy_context()
772          */
773         mm_free_pgd(mm);
774         free_mm(mm);
775         return NULL;
776 }
777
778 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
779 {
780         struct mm_struct * mm, *oldmm;
781         int retval;
782
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
788
789         tsk->mm = NULL;
790         tsk->active_mm = NULL;
791
792         /*
793          * Are we cloning a kernel thread?
794          *
795          * We need to steal a active VM for that..
796          */
797         oldmm = current->mm;
798         if (!oldmm)
799                 return 0;
800
801         if (clone_flags & CLONE_VM) {
802                 atomic_inc(&oldmm->mm_users);
803                 mm = oldmm;
804                 goto good_mm;
805         }
806
807         retval = -ENOMEM;
808         mm = dup_mm(tsk);
809         if (!mm)
810                 goto fail_nomem;
811
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);
818
819         tsk->mm = mm;
820         tsk->active_mm = mm;
821         return 0;
822
823 fail_nomem:
824         return retval;
825 }
826
827 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
828 {
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;
836                 }
837                 fs->users++;
838                 spin_unlock(&fs->lock);
839                 return 0;
840         }
841         tsk->fs = copy_fs_struct(fs);
842         if (!tsk->fs)
843                 return -ENOMEM;
844         return 0;
845 }
846
847 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
848 {
849         struct files_struct *oldf, *newf;
850         int error = 0;
851
852         /*
853          * A background process may not have any files ...
854          */
855         oldf = current->files;
856         if (!oldf)
857                 goto out;
858
859         if (clone_flags & CLONE_FILES) {
860                 atomic_inc(&oldf->count);
861                 goto out;
862         }
863
864         newf = dup_fd(oldf, &error);
865         if (!newf)
866                 goto out;
867
868         tsk->files = newf;
869         error = 0;
870 out:
871         return error;
872 }
873
874 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
875 {
876 #ifdef CONFIG_BLOCK
877         struct io_context *ioc = current->io_context;
878
879         if (!ioc)
880                 return 0;
881         /*
882          * Share io context with parent, if CLONE_IO is set
883          */
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;
892
893                 tsk->io_context->ioprio = ioc->ioprio;
894         }
895 #endif
896         return 0;
897 }
898
899 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
900 {
901         struct sighand_struct *sig;
902
903         if (clone_flags & CLONE_SIGHAND) {
904                 atomic_inc(&current->sighand->count);
905                 return 0;
906         }
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;
914 }
915
916 void __cleanup_sighand(struct sighand_struct *sighand)
917 {
918         if (atomic_dec_and_test(&sighand->count))
919                 kmem_cache_free(sighand_cachep, sighand);
920 }
921
922
923 /*
924  * Initialize POSIX timer handling for a thread group.
925  */
926 static void posix_cpu_timers_init_group(struct signal_struct *sig)
927 {
928         unsigned long cpu_limit;
929
930         /* Thread group counters. */
931         thread_group_cputime_init(sig);
932
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;
937         }
938
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]);
943 }
944
945 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
946 {
947         struct signal_struct *sig;
948
949         if (clone_flags & CLONE_THREAD)
950                 return 0;
951
952         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
953         tsk->signal = sig;
954         if (!sig)
955                 return -ENOMEM;
956
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);
966
967         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
968         sig->real_timer.function = it_real_fn;
969
970         task_lock(current->group_leader);
971         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
972         task_unlock(current->group_leader);
973
974         posix_cpu_timers_init_group(sig);
975
976         tty_audit_fork(sig);
977         sched_autogroup_fork(sig);
978
979 #ifdef CONFIG_CGROUPS
980         init_rwsem(&sig->threadgroup_fork_lock);
981 #endif
982
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;
986
987         mutex_init(&sig->cred_guard_mutex);
988
989         return 0;
990 }
991
992 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
993 {
994         unsigned long new_flags = p->flags;
995
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);
1001 }
1002
1003 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1004 {
1005         current->clear_child_tid = tidptr;
1006
1007         return task_pid_vnr(current);
1008 }
1009
1010 static void rt_mutex_init_task(struct task_struct *p)
1011 {
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
1017 }
1018
1019 #ifdef CONFIG_MM_OWNER
1020 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1021 {
1022         mm->owner = p;
1023 }
1024 #endif /* CONFIG_MM_OWNER */
1025
1026 /*
1027  * Initialize POSIX timer handling for a single task.
1028  */
1029 static void posix_cpu_timers_init(struct task_struct *tsk)
1030 {
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]);
1037 }
1038
1039 /*
1040  * This creates a new process as a copy of the old one,
1041  * but does not actually start it yet.
1042  *
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.
1046  */
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)
1054 {
1055         int retval;
1056         struct task_struct *p;
1057         int cgroup_callbacks_done = 0;
1058
1059         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1060                 return ERR_PTR(-EINVAL);
1061
1062         /*
1063          * Thread groups must share signals as well, and detached threads
1064          * can only be started up within the thread group.
1065          */
1066         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1067                 return ERR_PTR(-EINVAL);
1068
1069         /*
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.
1073          */
1074         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1075                 return ERR_PTR(-EINVAL);
1076
1077         /*
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.
1082          */
1083         if ((clone_flags & CLONE_PARENT) &&
1084                                 current->signal->flags & SIGNAL_UNKILLABLE)
1085                 return ERR_PTR(-EINVAL);
1086
1087         retval = security_task_create(clone_flags);
1088         if (retval)
1089                 goto fork_out;
1090
1091         retval = -ENOMEM;
1092         p = dup_task_struct(current);
1093         if (!p)
1094                 goto fork_out;
1095
1096         ftrace_graph_init_task(p);
1097
1098         rt_mutex_init_task(p);
1099
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;
1110         }
1111
1112         retval = copy_creds(p, clone_flags);
1113         if (retval < 0)
1114                 goto bad_fork_free;
1115
1116         /*
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.
1120          */
1121         retval = -EAGAIN;
1122         if (nr_threads >= max_threads)
1123                 goto bad_fork_cleanup_count;
1124
1125         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1126                 goto bad_fork_cleanup_count;
1127
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);
1136
1137         init_sigpending(&p->pending);
1138
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
1151
1152         p->default_timer_slack_ns = current->timer_slack_ns;
1153
1154         task_io_accounting_init(&p->ioac);
1155         acct_clear_integrals(p);
1156
1157         posix_cpu_timers_init(p);
1158
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;
1173         }
1174         mpol_fix_fork_child_flag(p);
1175 #endif
1176 #ifdef CONFIG_CPUSETS
1177         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1178         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1179 #endif
1180 #ifdef CONFIG_TRACE_IRQFLAGS
1181         p->irq_events = 0;
1182 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1183         p->hardirqs_enabled = 1;
1184 #else
1185         p->hardirqs_enabled = 0;
1186 #endif
1187         p->hardirq_enable_ip = 0;
1188         p->hardirq_enable_event = 0;
1189         p->hardirq_disable_ip = _THIS_IP_;
1190         p->hardirq_disable_event = 0;
1191         p->softirqs_enabled = 1;
1192         p->softirq_enable_ip = _THIS_IP_;
1193         p->softirq_enable_event = 0;
1194         p->softirq_disable_ip = 0;
1195         p->softirq_disable_event = 0;
1196         p->hardirq_context = 0;
1197         p->softirq_context = 0;
1198 #endif
1199 #ifdef CONFIG_LOCKDEP
1200         p->lockdep_depth = 0; /* no locks held yet */
1201         p->curr_chain_key = 0;
1202         p->lockdep_recursion = 0;
1203 #endif
1204
1205 #ifdef CONFIG_DEBUG_MUTEXES
1206         p->blocked_on = NULL; /* not blocked yet */
1207 #endif
1208 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1209         p->memcg_batch.do_batch = 0;
1210         p->memcg_batch.memcg = NULL;
1211 #endif
1212
1213         /* Perform scheduler related setup. Assign this task to a CPU. */
1214         sched_fork(p);
1215
1216         retval = perf_event_init_task(p);
1217         if (retval)
1218                 goto bad_fork_cleanup_policy;
1219
1220         if ((retval = audit_alloc(p)))
1221                 goto bad_fork_cleanup_policy;
1222         /* copy all the process information */
1223         if ((retval = copy_semundo(clone_flags, p)))
1224                 goto bad_fork_cleanup_audit;
1225         if ((retval = copy_files(clone_flags, p)))
1226                 goto bad_fork_cleanup_semundo;
1227         if ((retval = copy_fs(clone_flags, p)))
1228                 goto bad_fork_cleanup_files;
1229         if ((retval = copy_sighand(clone_flags, p)))
1230                 goto bad_fork_cleanup_fs;
1231         if ((retval = copy_signal(clone_flags, p)))
1232                 goto bad_fork_cleanup_sighand;
1233         if ((retval = copy_mm(clone_flags, p)))
1234                 goto bad_fork_cleanup_signal;
1235         if ((retval = copy_namespaces(clone_flags, p)))
1236                 goto bad_fork_cleanup_mm;
1237         if ((retval = copy_io(clone_flags, p)))
1238                 goto bad_fork_cleanup_namespaces;
1239         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1240         if (retval)
1241                 goto bad_fork_cleanup_io;
1242
1243         if (pid != &init_struct_pid) {
1244                 retval = -ENOMEM;
1245                 pid = alloc_pid(p->nsproxy->pid_ns);
1246                 if (!pid)
1247                         goto bad_fork_cleanup_io;
1248         }
1249
1250         p->pid = pid_nr(pid);
1251         p->tgid = p->pid;
1252         if (clone_flags & CLONE_THREAD)
1253                 p->tgid = current->tgid;
1254
1255         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1256         /*
1257          * Clear TID on mm_release()?
1258          */
1259         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1260 #ifdef CONFIG_BLOCK
1261         p->plug = NULL;
1262 #endif
1263 #ifdef CONFIG_FUTEX
1264         p->robust_list = NULL;
1265 #ifdef CONFIG_COMPAT
1266         p->compat_robust_list = NULL;
1267 #endif
1268         INIT_LIST_HEAD(&p->pi_state_list);
1269         p->pi_state_cache = NULL;
1270 #endif
1271         /*
1272          * sigaltstack should be cleared when sharing the same VM
1273          */
1274         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1275                 p->sas_ss_sp = p->sas_ss_size = 0;
1276
1277         /*
1278          * Syscall tracing and stepping should be turned off in the
1279          * child regardless of CLONE_PTRACE.
1280          */
1281         user_disable_single_step(p);
1282         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1283 #ifdef TIF_SYSCALL_EMU
1284         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1285 #endif
1286         clear_all_latency_tracing(p);
1287
1288         /* ok, now we should be set up.. */
1289         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1290         p->pdeath_signal = 0;
1291         p->exit_state = 0;
1292
1293         /*
1294          * Ok, make it visible to the rest of the system.
1295          * We dont wake it up yet.
1296          */
1297         p->group_leader = p;
1298         INIT_LIST_HEAD(&p->thread_group);
1299
1300         /* Now that the task is set up, run cgroup callbacks if
1301          * necessary. We need to run them before the task is visible
1302          * on the tasklist. */
1303         cgroup_fork_callbacks(p);
1304         cgroup_callbacks_done = 1;
1305
1306         /* Need tasklist lock for parent etc handling! */
1307         write_lock_irq(&tasklist_lock);
1308
1309         /* CLONE_PARENT re-uses the old parent */
1310         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1311                 p->real_parent = current->real_parent;
1312                 p->parent_exec_id = current->parent_exec_id;
1313         } else {
1314                 p->real_parent = current;
1315                 p->parent_exec_id = current->self_exec_id;
1316         }
1317
1318         spin_lock(&current->sighand->siglock);
1319
1320         /*
1321          * Process group and session signals need to be delivered to just the
1322          * parent before the fork or both the parent and the child after the
1323          * fork. Restart if a signal comes in before we add the new process to
1324          * it's process group.
1325          * A fatal signal pending means that current will exit, so the new
1326          * thread can't slip out of an OOM kill (or normal SIGKILL).
1327          */
1328         recalc_sigpending();
1329         if (signal_pending(current)) {
1330                 spin_unlock(&current->sighand->siglock);
1331                 write_unlock_irq(&tasklist_lock);
1332                 retval = -ERESTARTNOINTR;
1333                 goto bad_fork_free_pid;
1334         }
1335
1336         if (clone_flags & CLONE_THREAD) {
1337                 current->signal->nr_threads++;
1338                 atomic_inc(&current->signal->live);
1339                 atomic_inc(&current->signal->sigcnt);
1340                 p->group_leader = current->group_leader;
1341                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1342         }
1343
1344         if (likely(p->pid)) {
1345                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1346
1347                 if (thread_group_leader(p)) {
1348                         if (is_child_reaper(pid))
1349                                 p->nsproxy->pid_ns->child_reaper = p;
1350
1351                         p->signal->leader_pid = pid;
1352                         p->signal->tty = tty_kref_get(current->signal->tty);
1353                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1354                         attach_pid(p, PIDTYPE_SID, task_session(current));
1355                         list_add_tail(&p->sibling, &p->real_parent->children);
1356                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1357                         __this_cpu_inc(process_counts);
1358                 }
1359                 attach_pid(p, PIDTYPE_PID, pid);
1360                 nr_threads++;
1361         }
1362
1363         total_forks++;
1364         spin_unlock(&current->sighand->siglock);
1365         write_unlock_irq(&tasklist_lock);
1366         proc_fork_connector(p);
1367         cgroup_post_fork(p);
1368         if (clone_flags & CLONE_THREAD)
1369                 threadgroup_fork_read_unlock(current);
1370         perf_event_fork(p);
1371         return p;
1372
1373 bad_fork_free_pid:
1374         if (pid != &init_struct_pid)
1375                 free_pid(pid);
1376 bad_fork_cleanup_io:
1377         if (p->io_context)
1378                 exit_io_context(p);
1379 bad_fork_cleanup_namespaces:
1380         exit_task_namespaces(p);
1381 bad_fork_cleanup_mm:
1382         if (p->mm) {
1383                 task_lock(p);
1384                 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1385                         atomic_dec(&p->mm->oom_disable_count);
1386                 task_unlock(p);
1387                 mmput(p->mm);
1388         }
1389 bad_fork_cleanup_signal:
1390         if (!(clone_flags & CLONE_THREAD))
1391                 free_signal_struct(p->signal);
1392 bad_fork_cleanup_sighand:
1393         __cleanup_sighand(p->sighand);
1394 bad_fork_cleanup_fs:
1395         exit_fs(p); /* blocking */
1396 bad_fork_cleanup_files:
1397         exit_files(p); /* blocking */
1398 bad_fork_cleanup_semundo:
1399         exit_sem(p);
1400 bad_fork_cleanup_audit:
1401         audit_free(p);
1402 bad_fork_cleanup_policy:
1403         perf_event_free_task(p);
1404 #ifdef CONFIG_NUMA
1405         mpol_put(p->mempolicy);
1406 bad_fork_cleanup_cgroup:
1407 #endif
1408         if (clone_flags & CLONE_THREAD)
1409                 threadgroup_fork_read_unlock(current);
1410         cgroup_exit(p, cgroup_callbacks_done);
1411         delayacct_tsk_free(p);
1412         module_put(task_thread_info(p)->exec_domain->module);
1413 bad_fork_cleanup_count:
1414         atomic_dec(&p->cred->user->processes);
1415         exit_creds(p);
1416 bad_fork_free:
1417         free_task(p);
1418 fork_out:
1419         return ERR_PTR(retval);
1420 }
1421
1422 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1423 {
1424         memset(regs, 0, sizeof(struct pt_regs));
1425         return regs;
1426 }
1427
1428 static inline void init_idle_pids(struct pid_link *links)
1429 {
1430         enum pid_type type;
1431
1432         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1433                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1434                 links[type].pid = &init_struct_pid;
1435         }
1436 }
1437
1438 struct task_struct * __cpuinit fork_idle(int cpu)
1439 {
1440         struct task_struct *task;
1441         struct pt_regs regs;
1442
1443         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1444                             &init_struct_pid, 0);
1445         if (!IS_ERR(task)) {
1446                 init_idle_pids(task->pids);
1447                 init_idle(task, cpu);
1448         }
1449
1450         return task;
1451 }
1452
1453 /*
1454  *  Ok, this is the main fork-routine.
1455  *
1456  * It copies the process, and if successful kick-starts
1457  * it and waits for it to finish using the VM if required.
1458  */
1459 long do_fork(unsigned long clone_flags,
1460               unsigned long stack_start,
1461               struct pt_regs *regs,
1462               unsigned long stack_size,
1463               int __user *parent_tidptr,
1464               int __user *child_tidptr)
1465 {
1466         struct task_struct *p;
1467         int trace = 0;
1468         long nr;
1469
1470         /*
1471          * Do some preliminary argument and permissions checking before we
1472          * actually start allocating stuff
1473          */
1474         if (clone_flags & CLONE_NEWUSER) {
1475                 if (clone_flags & CLONE_THREAD)
1476                         return -EINVAL;
1477                 /* hopefully this check will go away when userns support is
1478                  * complete
1479                  */
1480                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1481                                 !capable(CAP_SETGID))
1482                         return -EPERM;
1483         }
1484
1485         /*
1486          * Determine whether and which event to report to ptracer.  When
1487          * called from kernel_thread or CLONE_UNTRACED is explicitly
1488          * requested, no event is reported; otherwise, report if the event
1489          * for the type of forking is enabled.
1490          */
1491         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1492                 if (clone_flags & CLONE_VFORK)
1493                         trace = PTRACE_EVENT_VFORK;
1494                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1495                         trace = PTRACE_EVENT_CLONE;
1496                 else
1497                         trace = PTRACE_EVENT_FORK;
1498
1499                 if (likely(!ptrace_event_enabled(current, trace)))
1500                         trace = 0;
1501         }
1502
1503         p = copy_process(clone_flags, stack_start, regs, stack_size,
1504                          child_tidptr, NULL, trace);
1505         /*
1506          * Do this prior waking up the new thread - the thread pointer
1507          * might get invalid after that point, if the thread exits quickly.
1508          */
1509         if (!IS_ERR(p)) {
1510                 struct completion vfork;
1511
1512                 trace_sched_process_fork(current, p);
1513
1514                 nr = task_pid_vnr(p);
1515
1516                 if (clone_flags & CLONE_PARENT_SETTID)
1517                         put_user(nr, parent_tidptr);
1518
1519                 if (clone_flags & CLONE_VFORK) {
1520                         p->vfork_done = &vfork;
1521                         init_completion(&vfork);
1522                 }
1523
1524                 audit_finish_fork(p);
1525
1526                 /*
1527                  * We set PF_STARTING at creation in case tracing wants to
1528                  * use this to distinguish a fully live task from one that
1529                  * hasn't finished SIGSTOP raising yet.  Now we clear it
1530                  * and set the child going.
1531                  */
1532                 p->flags &= ~PF_STARTING;
1533
1534                 wake_up_new_task(p);
1535
1536                 /* forking complete and child started to run, tell ptracer */
1537                 if (unlikely(trace))
1538                         ptrace_event(trace, nr);
1539
1540                 if (clone_flags & CLONE_VFORK) {
1541                         freezer_do_not_count();
1542                         wait_for_completion(&vfork);
1543                         freezer_count();
1544                         ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1545                 }
1546         } else {
1547                 nr = PTR_ERR(p);
1548         }
1549         return nr;
1550 }
1551
1552 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1553 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1554 #endif
1555
1556 static void sighand_ctor(void *data)
1557 {
1558         struct sighand_struct *sighand = data;
1559
1560         spin_lock_init(&sighand->siglock);
1561         init_waitqueue_head(&sighand->signalfd_wqh);
1562 }
1563
1564 void __init proc_caches_init(void)
1565 {
1566         sighand_cachep = kmem_cache_create("sighand_cache",
1567                         sizeof(struct sighand_struct), 0,
1568                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1569                         SLAB_NOTRACK, sighand_ctor);
1570         signal_cachep = kmem_cache_create("signal_cache",
1571                         sizeof(struct signal_struct), 0,
1572                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1573         files_cachep = kmem_cache_create("files_cache",
1574                         sizeof(struct files_struct), 0,
1575                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1576         fs_cachep = kmem_cache_create("fs_cache",
1577                         sizeof(struct fs_struct), 0,
1578                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1579         /*
1580          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1581          * whole struct cpumask for the OFFSTACK case. We could change
1582          * this to *only* allocate as much of it as required by the
1583          * maximum number of CPU's we can ever have.  The cpumask_allocation
1584          * is at the end of the structure, exactly for that reason.
1585          */
1586         mm_cachep = kmem_cache_create("mm_struct",
1587                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1588                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1589         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1590         mmap_init();
1591         nsproxy_cache_init();
1592 }
1593
1594 /*
1595  * Check constraints on flags passed to the unshare system call.
1596  */
1597 static int check_unshare_flags(unsigned long unshare_flags)
1598 {
1599         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1600                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1601                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1602                 return -EINVAL;
1603         /*
1604          * Not implemented, but pretend it works if there is nothing to
1605          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1606          * needs to unshare vm.
1607          */
1608         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1609                 /* FIXME: get_task_mm() increments ->mm_users */
1610                 if (atomic_read(&current->mm->mm_users) > 1)
1611                         return -EINVAL;
1612         }
1613
1614         return 0;
1615 }
1616
1617 /*
1618  * Unshare the filesystem structure if it is being shared
1619  */
1620 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1621 {
1622         struct fs_struct *fs = current->fs;
1623
1624         if (!(unshare_flags & CLONE_FS) || !fs)
1625                 return 0;
1626
1627         /* don't need lock here; in the worst case we'll do useless copy */
1628         if (fs->users == 1)
1629                 return 0;
1630
1631         *new_fsp = copy_fs_struct(fs);
1632         if (!*new_fsp)
1633                 return -ENOMEM;
1634
1635         return 0;
1636 }
1637
1638 /*
1639  * Unshare file descriptor table if it is being shared
1640  */
1641 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1642 {
1643         struct files_struct *fd = current->files;
1644         int error = 0;
1645
1646         if ((unshare_flags & CLONE_FILES) &&
1647             (fd && atomic_read(&fd->count) > 1)) {
1648                 *new_fdp = dup_fd(fd, &error);
1649                 if (!*new_fdp)
1650                         return error;
1651         }
1652
1653         return 0;
1654 }
1655
1656 /*
1657  * unshare allows a process to 'unshare' part of the process
1658  * context which was originally shared using clone.  copy_*
1659  * functions used by do_fork() cannot be used here directly
1660  * because they modify an inactive task_struct that is being
1661  * constructed. Here we are modifying the current, active,
1662  * task_struct.
1663  */
1664 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1665 {
1666         struct fs_struct *fs, *new_fs = NULL;
1667         struct files_struct *fd, *new_fd = NULL;
1668         struct nsproxy *new_nsproxy = NULL;
1669         int do_sysvsem = 0;
1670         int err;
1671
1672         err = check_unshare_flags(unshare_flags);
1673         if (err)
1674                 goto bad_unshare_out;
1675
1676         /*
1677          * If unsharing namespace, must also unshare filesystem information.
1678          */
1679         if (unshare_flags & CLONE_NEWNS)
1680                 unshare_flags |= CLONE_FS;
1681         /*
1682          * CLONE_NEWIPC must also detach from the undolist: after switching
1683          * to a new ipc namespace, the semaphore arrays from the old
1684          * namespace are unreachable.
1685          */
1686         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1687                 do_sysvsem = 1;
1688         if ((err = unshare_fs(unshare_flags, &new_fs)))
1689                 goto bad_unshare_out;
1690         if ((err = unshare_fd(unshare_flags, &new_fd)))
1691                 goto bad_unshare_cleanup_fs;
1692         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1693                         new_fs)))
1694                 goto bad_unshare_cleanup_fd;
1695
1696         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1697                 if (do_sysvsem) {
1698                         /*
1699                          * CLONE_SYSVSEM is equivalent to sys_exit().
1700                          */
1701                         exit_sem(current);
1702                 }
1703
1704                 if (new_nsproxy) {
1705                         switch_task_namespaces(current, new_nsproxy);
1706                         new_nsproxy = NULL;
1707                 }
1708
1709                 task_lock(current);
1710
1711                 if (new_fs) {
1712                         fs = current->fs;
1713                         spin_lock(&fs->lock);
1714                         current->fs = new_fs;
1715                         if (--fs->users)
1716                                 new_fs = NULL;
1717                         else
1718                                 new_fs = fs;
1719                         spin_unlock(&fs->lock);
1720                 }
1721
1722                 if (new_fd) {
1723                         fd = current->files;
1724                         current->files = new_fd;
1725                         new_fd = fd;
1726                 }
1727
1728                 task_unlock(current);
1729         }
1730
1731         if (new_nsproxy)
1732                 put_nsproxy(new_nsproxy);
1733
1734 bad_unshare_cleanup_fd:
1735         if (new_fd)
1736                 put_files_struct(new_fd);
1737
1738 bad_unshare_cleanup_fs:
1739         if (new_fs)
1740                 free_fs_struct(new_fs);
1741
1742 bad_unshare_out:
1743         return err;
1744 }
1745
1746 /*
1747  *      Helper to unshare the files of the current task.
1748  *      We don't want to expose copy_files internals to
1749  *      the exec layer of the kernel.
1750  */
1751
1752 int unshare_files(struct files_struct **displaced)
1753 {
1754         struct task_struct *task = current;
1755         struct files_struct *copy = NULL;
1756         int error;
1757
1758         error = unshare_fd(CLONE_FILES, &copy);
1759         if (error || !copy) {
1760                 *displaced = NULL;
1761                 return error;
1762         }
1763         *displaced = task->files;
1764         task_lock(task);
1765         task->files = copy;
1766         task_unlock(task);
1767         return 0;
1768 }