31fa13e63b709e917f5165ba5a422cc68c3e812b
[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/tracehook.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/kthread.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/profile.h>
52 #include <linux/rmap.h>
53 #include <linux/ksm.h>
54 #include <linux/acct.h>
55 #include <linux/tsacct_kern.h>
56 #include <linux/cn_proc.h>
57 #include <linux/freezer.h>
58 #include <linux/delayacct.h>
59 #include <linux/taskstats_kern.h>
60 #include <linux/random.h>
61 #include <linux/tty.h>
62 #include <linux/blkdev.h>
63 #include <linux/fs_struct.h>
64 #include <linux/magic.h>
65 #include <linux/perf_event.h>
66 #include <linux/posix-timers.h>
67 #include <linux/user-return-notifier.h>
68 #include <linux/oom.h>
69 #include <linux/khugepaged.h>
70
71 #include <asm/pgtable.h>
72 #include <asm/pgalloc.h>
73 #include <asm/uaccess.h>
74 #include <asm/mmu_context.h>
75 #include <asm/cacheflush.h>
76 #include <asm/tlbflush.h>
77
78 #include <trace/events/sched.h>
79
80 /*
81  * Protected counters by write_lock_irq(&tasklist_lock)
82  */
83 unsigned long total_forks;      /* Handle normal Linux uptimes. */
84 int nr_threads;                 /* The idle threads do not count.. */
85
86 int max_threads;                /* tunable limit on nr_threads */
87
88 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
89
90 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
91
92 #ifdef CONFIG_PROVE_RCU
93 int lockdep_tasklist_lock_is_held(void)
94 {
95         return lockdep_is_held(&tasklist_lock);
96 }
97 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
98 #endif /* #ifdef CONFIG_PROVE_RCU */
99
100 int nr_processes(void)
101 {
102         int cpu;
103         int total = 0;
104
105         for_each_possible_cpu(cpu)
106                 total += per_cpu(process_counts, cpu);
107
108         return total;
109 }
110
111 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
112 # define alloc_task_struct_node(node)           \
113                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
114 # define free_task_struct(tsk)                  \
115                 kmem_cache_free(task_struct_cachep, (tsk))
116 static struct kmem_cache *task_struct_cachep;
117 #endif
118
119 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
120 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
121                                                   int node)
122 {
123 #ifdef CONFIG_DEBUG_STACK_USAGE
124         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
125 #else
126         gfp_t mask = GFP_KERNEL;
127 #endif
128         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
129
130         return page ? page_address(page) : NULL;
131 }
132
133 static inline void free_thread_info(struct thread_info *ti)
134 {
135         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
136 }
137 #endif
138
139 /* SLAB cache for signal_struct structures (tsk->signal) */
140 static struct kmem_cache *signal_cachep;
141
142 /* SLAB cache for sighand_struct structures (tsk->sighand) */
143 struct kmem_cache *sighand_cachep;
144
145 /* SLAB cache for files_struct structures (tsk->files) */
146 struct kmem_cache *files_cachep;
147
148 /* SLAB cache for fs_struct structures (tsk->fs) */
149 struct kmem_cache *fs_cachep;
150
151 /* SLAB cache for vm_area_struct structures */
152 struct kmem_cache *vm_area_cachep;
153
154 /* SLAB cache for mm_struct structures (tsk->mm) */
155 static struct kmem_cache *mm_cachep;
156
157 static void account_kernel_stack(struct thread_info *ti, int account)
158 {
159         struct zone *zone = page_zone(virt_to_page(ti));
160
161         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
162 }
163
164 void free_task(struct task_struct *tsk)
165 {
166         prop_local_destroy_single(&tsk->dirties);
167         account_kernel_stack(tsk->stack, -1);
168         free_thread_info(tsk->stack);
169         rt_mutex_debug_task_free(tsk);
170         ftrace_graph_exit_task(tsk);
171         free_task_struct(tsk);
172 }
173 EXPORT_SYMBOL(free_task);
174
175 static inline void free_signal_struct(struct signal_struct *sig)
176 {
177         taskstats_tgid_free(sig);
178         sched_autogroup_exit(sig);
179         kmem_cache_free(signal_cachep, sig);
180 }
181
182 static inline void put_signal_struct(struct signal_struct *sig)
183 {
184         if (atomic_dec_and_test(&sig->sigcnt))
185                 free_signal_struct(sig);
186 }
187
188 void __put_task_struct(struct task_struct *tsk)
189 {
190         WARN_ON(!tsk->exit_state);
191         WARN_ON(atomic_read(&tsk->usage));
192         WARN_ON(tsk == current);
193
194         exit_creds(tsk);
195         delayacct_tsk_free(tsk);
196         put_signal_struct(tsk->signal);
197
198         if (!profile_handoff_task(tsk))
199                 free_task(tsk);
200 }
201 EXPORT_SYMBOL_GPL(__put_task_struct);
202
203 /*
204  * macro override instead of weak attribute alias, to workaround
205  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
206  */
207 #ifndef arch_task_cache_init
208 #define arch_task_cache_init()
209 #endif
210
211 void __init fork_init(unsigned long mempages)
212 {
213 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
214 #ifndef ARCH_MIN_TASKALIGN
215 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
216 #endif
217         /* create a slab on which task_structs can be allocated */
218         task_struct_cachep =
219                 kmem_cache_create("task_struct", sizeof(struct task_struct),
220                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
221 #endif
222
223         /* do the arch specific task caches init */
224         arch_task_cache_init();
225
226         /*
227          * The default maximum number of threads is set to a safe
228          * value: the thread structures can take up at most half
229          * of memory.
230          */
231         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
232
233         /*
234          * we need to allow at least 20 threads to boot a system
235          */
236         if(max_threads < 20)
237                 max_threads = 20;
238
239         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
240         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
241         init_task.signal->rlim[RLIMIT_SIGPENDING] =
242                 init_task.signal->rlim[RLIMIT_NPROC];
243 }
244
245 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
246                                                struct task_struct *src)
247 {
248         *dst = *src;
249         return 0;
250 }
251
252 static struct task_struct *dup_task_struct(struct task_struct *orig)
253 {
254         struct task_struct *tsk;
255         struct thread_info *ti;
256         unsigned long *stackend;
257         int node = tsk_fork_get_node(orig);
258         int err;
259
260         prepare_to_copy(orig);
261
262         tsk = alloc_task_struct_node(node);
263         if (!tsk)
264                 return NULL;
265
266         ti = alloc_thread_info_node(tsk, node);
267         if (!ti) {
268                 free_task_struct(tsk);
269                 return NULL;
270         }
271
272         err = arch_dup_task_struct(tsk, orig);
273         if (err)
274                 goto out;
275
276         tsk->stack = ti;
277
278         err = prop_local_init_single(&tsk->dirties);
279         if (err)
280                 goto out;
281
282         setup_thread_stack(tsk, orig);
283         clear_user_return_notifier(tsk);
284         clear_tsk_need_resched(tsk);
285         stackend = end_of_stack(tsk);
286         *stackend = STACK_END_MAGIC;    /* for overflow detection */
287
288 #ifdef CONFIG_CC_STACKPROTECTOR
289         tsk->stack_canary = get_random_int();
290 #endif
291
292         /* One for us, one for whoever does the "release_task()" (usually parent) */
293         atomic_set(&tsk->usage,2);
294         atomic_set(&tsk->fs_excl, 0);
295 #ifdef CONFIG_BLK_DEV_IO_TRACE
296         tsk->btrace_seq = 0;
297 #endif
298         tsk->splice_pipe = NULL;
299
300         account_kernel_stack(ti, 1);
301
302         return tsk;
303
304 out:
305         free_thread_info(ti);
306         free_task_struct(tsk);
307         return NULL;
308 }
309
310 #ifdef CONFIG_MMU
311 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
312 {
313         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
314         struct rb_node **rb_link, *rb_parent;
315         int retval;
316         unsigned long charge;
317         struct mempolicy *pol;
318
319         down_write(&oldmm->mmap_sem);
320         flush_cache_dup_mm(oldmm);
321         /*
322          * Not linked in yet - no deadlock potential:
323          */
324         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
325
326         mm->locked_vm = 0;
327         mm->mmap = NULL;
328         mm->mmap_cache = NULL;
329         mm->free_area_cache = oldmm->mmap_base;
330         mm->cached_hole_size = ~0UL;
331         mm->map_count = 0;
332         cpumask_clear(mm_cpumask(mm));
333         mm->mm_rb = RB_ROOT;
334         rb_link = &mm->mm_rb.rb_node;
335         rb_parent = NULL;
336         pprev = &mm->mmap;
337         retval = ksm_fork(mm, oldmm);
338         if (retval)
339                 goto out;
340         retval = khugepaged_fork(mm, oldmm);
341         if (retval)
342                 goto out;
343
344         prev = NULL;
345         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
346                 struct file *file;
347
348                 if (mpnt->vm_flags & VM_DONTCOPY) {
349                         long pages = vma_pages(mpnt);
350                         mm->total_vm -= pages;
351                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
352                                                                 -pages);
353                         continue;
354                 }
355                 charge = 0;
356                 if (mpnt->vm_flags & VM_ACCOUNT) {
357                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
358                         if (security_vm_enough_memory(len))
359                                 goto fail_nomem;
360                         charge = len;
361                 }
362                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
363                 if (!tmp)
364                         goto fail_nomem;
365                 *tmp = *mpnt;
366                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
367                 pol = mpol_dup(vma_policy(mpnt));
368                 retval = PTR_ERR(pol);
369                 if (IS_ERR(pol))
370                         goto fail_nomem_policy;
371                 vma_set_policy(tmp, pol);
372                 tmp->vm_mm = mm;
373                 if (anon_vma_fork(tmp, mpnt))
374                         goto fail_nomem_anon_vma_fork;
375                 tmp->vm_flags &= ~VM_LOCKED;
376                 tmp->vm_next = tmp->vm_prev = NULL;
377                 file = tmp->vm_file;
378                 if (file) {
379                         struct inode *inode = file->f_path.dentry->d_inode;
380                         struct address_space *mapping = file->f_mapping;
381
382                         get_file(file);
383                         if (tmp->vm_flags & VM_DENYWRITE)
384                                 atomic_dec(&inode->i_writecount);
385                         mutex_lock(&mapping->i_mmap_mutex);
386                         if (tmp->vm_flags & VM_SHARED)
387                                 mapping->i_mmap_writable++;
388                         flush_dcache_mmap_lock(mapping);
389                         /* insert tmp into the share list, just after mpnt */
390                         vma_prio_tree_add(tmp, mpnt);
391                         flush_dcache_mmap_unlock(mapping);
392                         mutex_unlock(&mapping->i_mmap_mutex);
393                 }
394
395                 /*
396                  * Clear hugetlb-related page reserves for children. This only
397                  * affects MAP_PRIVATE mappings. Faults generated by the child
398                  * are not guaranteed to succeed, even if read-only
399                  */
400                 if (is_vm_hugetlb_page(tmp))
401                         reset_vma_resv_huge_pages(tmp);
402
403                 /*
404                  * Link in the new vma and copy the page table entries.
405                  */
406                 *pprev = tmp;
407                 pprev = &tmp->vm_next;
408                 tmp->vm_prev = prev;
409                 prev = tmp;
410
411                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
412                 rb_link = &tmp->vm_rb.rb_right;
413                 rb_parent = &tmp->vm_rb;
414
415                 mm->map_count++;
416                 retval = copy_page_range(mm, oldmm, mpnt);
417
418                 if (tmp->vm_ops && tmp->vm_ops->open)
419                         tmp->vm_ops->open(tmp);
420
421                 if (retval)
422                         goto out;
423         }
424         /* a new mm has just been created */
425         arch_dup_mmap(oldmm, mm);
426         retval = 0;
427 out:
428         up_write(&mm->mmap_sem);
429         flush_tlb_mm(oldmm);
430         up_write(&oldmm->mmap_sem);
431         return retval;
432 fail_nomem_anon_vma_fork:
433         mpol_put(pol);
434 fail_nomem_policy:
435         kmem_cache_free(vm_area_cachep, tmp);
436 fail_nomem:
437         retval = -ENOMEM;
438         vm_unacct_memory(charge);
439         goto out;
440 }
441
442 static inline int mm_alloc_pgd(struct mm_struct * mm)
443 {
444         mm->pgd = pgd_alloc(mm);
445         if (unlikely(!mm->pgd))
446                 return -ENOMEM;
447         return 0;
448 }
449
450 static inline void mm_free_pgd(struct mm_struct * mm)
451 {
452         pgd_free(mm, mm->pgd);
453 }
454 #else
455 #define dup_mmap(mm, oldmm)     (0)
456 #define mm_alloc_pgd(mm)        (0)
457 #define mm_free_pgd(mm)
458 #endif /* CONFIG_MMU */
459
460 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
461
462 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
463 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
464
465 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
466
467 static int __init coredump_filter_setup(char *s)
468 {
469         default_dump_filter =
470                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
471                 MMF_DUMP_FILTER_MASK;
472         return 1;
473 }
474
475 __setup("coredump_filter=", coredump_filter_setup);
476
477 #include <linux/init_task.h>
478
479 static void mm_init_aio(struct mm_struct *mm)
480 {
481 #ifdef CONFIG_AIO
482         spin_lock_init(&mm->ioctx_lock);
483         INIT_HLIST_HEAD(&mm->ioctx_list);
484 #endif
485 }
486
487 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
488 {
489         atomic_set(&mm->mm_users, 1);
490         atomic_set(&mm->mm_count, 1);
491         init_rwsem(&mm->mmap_sem);
492         INIT_LIST_HEAD(&mm->mmlist);
493         mm->flags = (current->mm) ?
494                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
495         mm->core_state = NULL;
496         mm->nr_ptes = 0;
497         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
498         spin_lock_init(&mm->page_table_lock);
499         mm->free_area_cache = TASK_UNMAPPED_BASE;
500         mm->cached_hole_size = ~0UL;
501         mm_init_aio(mm);
502         mm_init_owner(mm, p);
503         atomic_set(&mm->oom_disable_count, 0);
504
505         if (likely(!mm_alloc_pgd(mm))) {
506                 mm->def_flags = 0;
507                 mmu_notifier_mm_init(mm);
508                 return mm;
509         }
510
511         free_mm(mm);
512         return NULL;
513 }
514
515 /*
516  * Allocate and initialize an mm_struct.
517  */
518 struct mm_struct * mm_alloc(void)
519 {
520         struct mm_struct * mm;
521
522         mm = allocate_mm();
523         if (!mm)
524                 return NULL;
525
526         memset(mm, 0, sizeof(*mm));
527         mm_init_cpumask(mm);
528         return mm_init(mm, current);
529 }
530
531 /*
532  * Called when the last reference to the mm
533  * is dropped: either by a lazy thread or by
534  * mmput. Free the page directory and the mm.
535  */
536 void __mmdrop(struct mm_struct *mm)
537 {
538         BUG_ON(mm == &init_mm);
539         mm_free_pgd(mm);
540         destroy_context(mm);
541         mmu_notifier_mm_destroy(mm);
542 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
543         VM_BUG_ON(mm->pmd_huge_pte);
544 #endif
545         free_mm(mm);
546 }
547 EXPORT_SYMBOL_GPL(__mmdrop);
548
549 /*
550  * Decrement the use count and release all resources for an mm.
551  */
552 void mmput(struct mm_struct *mm)
553 {
554         might_sleep();
555
556         if (atomic_dec_and_test(&mm->mm_users)) {
557                 exit_aio(mm);
558                 ksm_exit(mm);
559                 khugepaged_exit(mm); /* must run before exit_mmap */
560                 exit_mmap(mm);
561                 set_mm_exe_file(mm, NULL);
562                 if (!list_empty(&mm->mmlist)) {
563                         spin_lock(&mmlist_lock);
564                         list_del(&mm->mmlist);
565                         spin_unlock(&mmlist_lock);
566                 }
567                 put_swap_token(mm);
568                 if (mm->binfmt)
569                         module_put(mm->binfmt->module);
570                 mmdrop(mm);
571         }
572 }
573 EXPORT_SYMBOL_GPL(mmput);
574
575 /*
576  * We added or removed a vma mapping the executable. The vmas are only mapped
577  * during exec and are not mapped with the mmap system call.
578  * Callers must hold down_write() on the mm's mmap_sem for these
579  */
580 void added_exe_file_vma(struct mm_struct *mm)
581 {
582         mm->num_exe_file_vmas++;
583 }
584
585 void removed_exe_file_vma(struct mm_struct *mm)
586 {
587         mm->num_exe_file_vmas--;
588         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
589                 fput(mm->exe_file);
590                 mm->exe_file = NULL;
591         }
592
593 }
594
595 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
596 {
597         if (new_exe_file)
598                 get_file(new_exe_file);
599         if (mm->exe_file)
600                 fput(mm->exe_file);
601         mm->exe_file = new_exe_file;
602         mm->num_exe_file_vmas = 0;
603 }
604
605 struct file *get_mm_exe_file(struct mm_struct *mm)
606 {
607         struct file *exe_file;
608
609         /* We need mmap_sem to protect against races with removal of
610          * VM_EXECUTABLE vmas */
611         down_read(&mm->mmap_sem);
612         exe_file = mm->exe_file;
613         if (exe_file)
614                 get_file(exe_file);
615         up_read(&mm->mmap_sem);
616         return exe_file;
617 }
618
619 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
620 {
621         /* It's safe to write the exe_file pointer without exe_file_lock because
622          * this is called during fork when the task is not yet in /proc */
623         newmm->exe_file = get_mm_exe_file(oldmm);
624 }
625
626 /**
627  * get_task_mm - acquire a reference to the task's mm
628  *
629  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
630  * this kernel workthread has transiently adopted a user mm with use_mm,
631  * to do its AIO) is not set and if so returns a reference to it, after
632  * bumping up the use count.  User must release the mm via mmput()
633  * after use.  Typically used by /proc and ptrace.
634  */
635 struct mm_struct *get_task_mm(struct task_struct *task)
636 {
637         struct mm_struct *mm;
638
639         task_lock(task);
640         mm = task->mm;
641         if (mm) {
642                 if (task->flags & PF_KTHREAD)
643                         mm = NULL;
644                 else
645                         atomic_inc(&mm->mm_users);
646         }
647         task_unlock(task);
648         return mm;
649 }
650 EXPORT_SYMBOL_GPL(get_task_mm);
651
652 /* Please note the differences between mmput and mm_release.
653  * mmput is called whenever we stop holding onto a mm_struct,
654  * error success whatever.
655  *
656  * mm_release is called after a mm_struct has been removed
657  * from the current process.
658  *
659  * This difference is important for error handling, when we
660  * only half set up a mm_struct for a new process and need to restore
661  * the old one.  Because we mmput the new mm_struct before
662  * restoring the old one. . .
663  * Eric Biederman 10 January 1998
664  */
665 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
666 {
667         struct completion *vfork_done = tsk->vfork_done;
668
669         /* Get rid of any futexes when releasing the mm */
670 #ifdef CONFIG_FUTEX
671         if (unlikely(tsk->robust_list)) {
672                 exit_robust_list(tsk);
673                 tsk->robust_list = NULL;
674         }
675 #ifdef CONFIG_COMPAT
676         if (unlikely(tsk->compat_robust_list)) {
677                 compat_exit_robust_list(tsk);
678                 tsk->compat_robust_list = NULL;
679         }
680 #endif
681         if (unlikely(!list_empty(&tsk->pi_state_list)))
682                 exit_pi_state_list(tsk);
683 #endif
684
685         /* Get rid of any cached register state */
686         deactivate_mm(tsk, mm);
687
688         /* notify parent sleeping on vfork() */
689         if (vfork_done) {
690                 tsk->vfork_done = NULL;
691                 complete(vfork_done);
692         }
693
694         /*
695          * If we're exiting normally, clear a user-space tid field if
696          * requested.  We leave this alone when dying by signal, to leave
697          * the value intact in a core dump, and to save the unnecessary
698          * trouble otherwise.  Userland only wants this done for a sys_exit.
699          */
700         if (tsk->clear_child_tid) {
701                 if (!(tsk->flags & PF_SIGNALED) &&
702                     atomic_read(&mm->mm_users) > 1) {
703                         /*
704                          * We don't check the error code - if userspace has
705                          * not set up a proper pointer then tough luck.
706                          */
707                         put_user(0, tsk->clear_child_tid);
708                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
709                                         1, NULL, NULL, 0);
710                 }
711                 tsk->clear_child_tid = NULL;
712         }
713 }
714
715 /*
716  * Allocate a new mm structure and copy contents from the
717  * mm structure of the passed in task structure.
718  */
719 struct mm_struct *dup_mm(struct task_struct *tsk)
720 {
721         struct mm_struct *mm, *oldmm = current->mm;
722         int err;
723
724         if (!oldmm)
725                 return NULL;
726
727         mm = allocate_mm();
728         if (!mm)
729                 goto fail_nomem;
730
731         memcpy(mm, oldmm, sizeof(*mm));
732         mm_init_cpumask(mm);
733
734         /* Initializing for Swap token stuff */
735         mm->token_priority = 0;
736         mm->last_interval = 0;
737
738 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
739         mm->pmd_huge_pte = NULL;
740 #endif
741
742         if (!mm_init(mm, tsk))
743                 goto fail_nomem;
744
745         if (init_new_context(tsk, mm))
746                 goto fail_nocontext;
747
748         dup_mm_exe_file(oldmm, mm);
749
750         err = dup_mmap(mm, oldmm);
751         if (err)
752                 goto free_pt;
753
754         mm->hiwater_rss = get_mm_rss(mm);
755         mm->hiwater_vm = mm->total_vm;
756
757         if (mm->binfmt && !try_module_get(mm->binfmt->module))
758                 goto free_pt;
759
760         return mm;
761
762 free_pt:
763         /* don't put binfmt in mmput, we haven't got module yet */
764         mm->binfmt = NULL;
765         mmput(mm);
766
767 fail_nomem:
768         return NULL;
769
770 fail_nocontext:
771         /*
772          * If init_new_context() failed, we cannot use mmput() to free the mm
773          * because it calls destroy_context()
774          */
775         mm_free_pgd(mm);
776         free_mm(mm);
777         return NULL;
778 }
779
780 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
781 {
782         struct mm_struct * mm, *oldmm;
783         int retval;
784
785         tsk->min_flt = tsk->maj_flt = 0;
786         tsk->nvcsw = tsk->nivcsw = 0;
787 #ifdef CONFIG_DETECT_HUNG_TASK
788         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
789 #endif
790
791         tsk->mm = NULL;
792         tsk->active_mm = NULL;
793
794         /*
795          * Are we cloning a kernel thread?
796          *
797          * We need to steal a active VM for that..
798          */
799         oldmm = current->mm;
800         if (!oldmm)
801                 return 0;
802
803         if (clone_flags & CLONE_VM) {
804                 atomic_inc(&oldmm->mm_users);
805                 mm = oldmm;
806                 goto good_mm;
807         }
808
809         retval = -ENOMEM;
810         mm = dup_mm(tsk);
811         if (!mm)
812                 goto fail_nomem;
813
814 good_mm:
815         /* Initializing for Swap token stuff */
816         mm->token_priority = 0;
817         mm->last_interval = 0;
818         if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
819                 atomic_inc(&mm->oom_disable_count);
820
821         tsk->mm = mm;
822         tsk->active_mm = mm;
823         return 0;
824
825 fail_nomem:
826         return retval;
827 }
828
829 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
830 {
831         struct fs_struct *fs = current->fs;
832         if (clone_flags & CLONE_FS) {
833                 /* tsk->fs is already what we want */
834                 spin_lock(&fs->lock);
835                 if (fs->in_exec) {
836                         spin_unlock(&fs->lock);
837                         return -EAGAIN;
838                 }
839                 fs->users++;
840                 spin_unlock(&fs->lock);
841                 return 0;
842         }
843         tsk->fs = copy_fs_struct(fs);
844         if (!tsk->fs)
845                 return -ENOMEM;
846         return 0;
847 }
848
849 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
850 {
851         struct files_struct *oldf, *newf;
852         int error = 0;
853
854         /*
855          * A background process may not have any files ...
856          */
857         oldf = current->files;
858         if (!oldf)
859                 goto out;
860
861         if (clone_flags & CLONE_FILES) {
862                 atomic_inc(&oldf->count);
863                 goto out;
864         }
865
866         newf = dup_fd(oldf, &error);
867         if (!newf)
868                 goto out;
869
870         tsk->files = newf;
871         error = 0;
872 out:
873         return error;
874 }
875
876 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
877 {
878 #ifdef CONFIG_BLOCK
879         struct io_context *ioc = current->io_context;
880
881         if (!ioc)
882                 return 0;
883         /*
884          * Share io context with parent, if CLONE_IO is set
885          */
886         if (clone_flags & CLONE_IO) {
887                 tsk->io_context = ioc_task_link(ioc);
888                 if (unlikely(!tsk->io_context))
889                         return -ENOMEM;
890         } else if (ioprio_valid(ioc->ioprio)) {
891                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
892                 if (unlikely(!tsk->io_context))
893                         return -ENOMEM;
894
895                 tsk->io_context->ioprio = ioc->ioprio;
896         }
897 #endif
898         return 0;
899 }
900
901 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
902 {
903         struct sighand_struct *sig;
904
905         if (clone_flags & CLONE_SIGHAND) {
906                 atomic_inc(&current->sighand->count);
907                 return 0;
908         }
909         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
910         rcu_assign_pointer(tsk->sighand, sig);
911         if (!sig)
912                 return -ENOMEM;
913         atomic_set(&sig->count, 1);
914         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
915         return 0;
916 }
917
918 void __cleanup_sighand(struct sighand_struct *sighand)
919 {
920         if (atomic_dec_and_test(&sighand->count))
921                 kmem_cache_free(sighand_cachep, sighand);
922 }
923
924
925 /*
926  * Initialize POSIX timer handling for a thread group.
927  */
928 static void posix_cpu_timers_init_group(struct signal_struct *sig)
929 {
930         unsigned long cpu_limit;
931
932         /* Thread group counters. */
933         thread_group_cputime_init(sig);
934
935         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
936         if (cpu_limit != RLIM_INFINITY) {
937                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
938                 sig->cputimer.running = 1;
939         }
940
941         /* The timer lists. */
942         INIT_LIST_HEAD(&sig->cpu_timers[0]);
943         INIT_LIST_HEAD(&sig->cpu_timers[1]);
944         INIT_LIST_HEAD(&sig->cpu_timers[2]);
945 }
946
947 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
948 {
949         struct signal_struct *sig;
950
951         if (clone_flags & CLONE_THREAD)
952                 return 0;
953
954         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
955         tsk->signal = sig;
956         if (!sig)
957                 return -ENOMEM;
958
959         sig->nr_threads = 1;
960         atomic_set(&sig->live, 1);
961         atomic_set(&sig->sigcnt, 1);
962         init_waitqueue_head(&sig->wait_chldexit);
963         if (clone_flags & CLONE_NEWPID)
964                 sig->flags |= SIGNAL_UNKILLABLE;
965         sig->curr_target = tsk;
966         init_sigpending(&sig->shared_pending);
967         INIT_LIST_HEAD(&sig->posix_timers);
968
969         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
970         sig->real_timer.function = it_real_fn;
971
972         task_lock(current->group_leader);
973         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
974         task_unlock(current->group_leader);
975
976         posix_cpu_timers_init_group(sig);
977
978         tty_audit_fork(sig);
979         sched_autogroup_fork(sig);
980
981 #ifdef CONFIG_CGROUPS
982         init_rwsem(&sig->threadgroup_fork_lock);
983 #endif
984
985         sig->oom_adj = current->signal->oom_adj;
986         sig->oom_score_adj = current->signal->oom_score_adj;
987         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
988
989         mutex_init(&sig->cred_guard_mutex);
990
991         return 0;
992 }
993
994 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
995 {
996         unsigned long new_flags = p->flags;
997
998         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
999         new_flags |= PF_FORKNOEXEC;
1000         new_flags |= PF_STARTING;
1001         p->flags = new_flags;
1002         clear_freeze_flag(p);
1003 }
1004
1005 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1006 {
1007         current->clear_child_tid = tidptr;
1008
1009         return task_pid_vnr(current);
1010 }
1011
1012 static void rt_mutex_init_task(struct task_struct *p)
1013 {
1014         raw_spin_lock_init(&p->pi_lock);
1015 #ifdef CONFIG_RT_MUTEXES
1016         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
1017         p->pi_blocked_on = NULL;
1018 #endif
1019 }
1020
1021 #ifdef CONFIG_MM_OWNER
1022 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1023 {
1024         mm->owner = p;
1025 }
1026 #endif /* CONFIG_MM_OWNER */
1027
1028 /*
1029  * Initialize POSIX timer handling for a single task.
1030  */
1031 static void posix_cpu_timers_init(struct task_struct *tsk)
1032 {
1033         tsk->cputime_expires.prof_exp = cputime_zero;
1034         tsk->cputime_expires.virt_exp = cputime_zero;
1035         tsk->cputime_expires.sched_exp = 0;
1036         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1037         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1038         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1039 }
1040
1041 /*
1042  * This creates a new process as a copy of the old one,
1043  * but does not actually start it yet.
1044  *
1045  * It copies the registers, and all the appropriate
1046  * parts of the process environment (as per the clone
1047  * flags). The actual kick-off is left to the caller.
1048  */
1049 static struct task_struct *copy_process(unsigned long clone_flags,
1050                                         unsigned long stack_start,
1051                                         struct pt_regs *regs,
1052                                         unsigned long stack_size,
1053                                         int __user *child_tidptr,
1054                                         struct pid *pid,
1055                                         int trace)
1056 {
1057         int retval;
1058         struct task_struct *p;
1059         int cgroup_callbacks_done = 0;
1060
1061         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1062                 return ERR_PTR(-EINVAL);
1063
1064         /*
1065          * Thread groups must share signals as well, and detached threads
1066          * can only be started up within the thread group.
1067          */
1068         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1069                 return ERR_PTR(-EINVAL);
1070
1071         /*
1072          * Shared signal handlers imply shared VM. By way of the above,
1073          * thread groups also imply shared VM. Blocking this case allows
1074          * for various simplifications in other code.
1075          */
1076         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1077                 return ERR_PTR(-EINVAL);
1078
1079         /*
1080          * Siblings of global init remain as zombies on exit since they are
1081          * not reaped by their parent (swapper). To solve this and to avoid
1082          * multi-rooted process trees, prevent global and container-inits
1083          * from creating siblings.
1084          */
1085         if ((clone_flags & CLONE_PARENT) &&
1086                                 current->signal->flags & SIGNAL_UNKILLABLE)
1087                 return ERR_PTR(-EINVAL);
1088
1089         retval = security_task_create(clone_flags);
1090         if (retval)
1091                 goto fork_out;
1092
1093         retval = -ENOMEM;
1094         p = dup_task_struct(current);
1095         if (!p)
1096                 goto fork_out;
1097
1098         ftrace_graph_init_task(p);
1099
1100         rt_mutex_init_task(p);
1101
1102 #ifdef CONFIG_PROVE_LOCKING
1103         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1104         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1105 #endif
1106         retval = -EAGAIN;
1107         if (atomic_read(&p->real_cred->user->processes) >=
1108                         task_rlimit(p, RLIMIT_NPROC)) {
1109                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1110                     p->real_cred->user != INIT_USER)
1111                         goto bad_fork_free;
1112         }
1113
1114         retval = copy_creds(p, clone_flags);
1115         if (retval < 0)
1116                 goto bad_fork_free;
1117
1118         /*
1119          * If multiple threads are within copy_process(), then this check
1120          * triggers too late. This doesn't hurt, the check is only there
1121          * to stop root fork bombs.
1122          */
1123         retval = -EAGAIN;
1124         if (nr_threads >= max_threads)
1125                 goto bad_fork_cleanup_count;
1126
1127         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1128                 goto bad_fork_cleanup_count;
1129
1130         p->did_exec = 0;
1131         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1132         copy_flags(clone_flags, p);
1133         INIT_LIST_HEAD(&p->children);
1134         INIT_LIST_HEAD(&p->sibling);
1135         rcu_copy_process(p);
1136         p->vfork_done = NULL;
1137         spin_lock_init(&p->alloc_lock);
1138
1139         init_sigpending(&p->pending);
1140
1141         p->utime = cputime_zero;
1142         p->stime = cputime_zero;
1143         p->gtime = cputime_zero;
1144         p->utimescaled = cputime_zero;
1145         p->stimescaled = cputime_zero;
1146 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1147         p->prev_utime = cputime_zero;
1148         p->prev_stime = cputime_zero;
1149 #endif
1150 #if defined(SPLIT_RSS_COUNTING)
1151         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1152 #endif
1153
1154         p->default_timer_slack_ns = current->timer_slack_ns;
1155
1156         task_io_accounting_init(&p->ioac);
1157         acct_clear_integrals(p);
1158
1159         posix_cpu_timers_init(p);
1160
1161         do_posix_clock_monotonic_gettime(&p->start_time);
1162         p->real_start_time = p->start_time;
1163         monotonic_to_bootbased(&p->real_start_time);
1164         p->io_context = NULL;
1165         p->audit_context = NULL;
1166         if (clone_flags & CLONE_THREAD)
1167                 threadgroup_fork_read_lock(current);
1168         cgroup_fork(p);
1169 #ifdef CONFIG_NUMA
1170         p->mempolicy = mpol_dup(p->mempolicy);
1171         if (IS_ERR(p->mempolicy)) {
1172                 retval = PTR_ERR(p->mempolicy);
1173                 p->mempolicy = NULL;
1174                 goto bad_fork_cleanup_cgroup;
1175         }
1176         mpol_fix_fork_child_flag(p);
1177 #endif
1178 #ifdef CONFIG_TRACE_IRQFLAGS
1179         p->irq_events = 0;
1180 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1181         p->hardirqs_enabled = 1;
1182 #else
1183         p->hardirqs_enabled = 0;
1184 #endif
1185         p->hardirq_enable_ip = 0;
1186         p->hardirq_enable_event = 0;
1187         p->hardirq_disable_ip = _THIS_IP_;
1188         p->hardirq_disable_event = 0;
1189         p->softirqs_enabled = 1;
1190         p->softirq_enable_ip = _THIS_IP_;
1191         p->softirq_enable_event = 0;
1192         p->softirq_disable_ip = 0;
1193         p->softirq_disable_event = 0;
1194         p->hardirq_context = 0;
1195         p->softirq_context = 0;
1196 #endif
1197 #ifdef CONFIG_LOCKDEP
1198         p->lockdep_depth = 0; /* no locks held yet */
1199         p->curr_chain_key = 0;
1200         p->lockdep_recursion = 0;
1201 #endif
1202
1203 #ifdef CONFIG_DEBUG_MUTEXES
1204         p->blocked_on = NULL; /* not blocked yet */
1205 #endif
1206 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1207         p->memcg_batch.do_batch = 0;
1208         p->memcg_batch.memcg = NULL;
1209 #endif
1210
1211         /* Perform scheduler related setup. Assign this task to a CPU. */
1212         sched_fork(p);
1213
1214         retval = perf_event_init_task(p);
1215         if (retval)
1216                 goto bad_fork_cleanup_policy;
1217
1218         if ((retval = audit_alloc(p)))
1219                 goto bad_fork_cleanup_policy;
1220         /* copy all the process information */
1221         if ((retval = copy_semundo(clone_flags, p)))
1222                 goto bad_fork_cleanup_audit;
1223         if ((retval = copy_files(clone_flags, p)))
1224                 goto bad_fork_cleanup_semundo;
1225         if ((retval = copy_fs(clone_flags, p)))
1226                 goto bad_fork_cleanup_files;
1227         if ((retval = copy_sighand(clone_flags, p)))
1228                 goto bad_fork_cleanup_fs;
1229         if ((retval = copy_signal(clone_flags, p)))
1230                 goto bad_fork_cleanup_sighand;
1231         if ((retval = copy_mm(clone_flags, p)))
1232                 goto bad_fork_cleanup_signal;
1233         if ((retval = copy_namespaces(clone_flags, p)))
1234                 goto bad_fork_cleanup_mm;
1235         if ((retval = copy_io(clone_flags, p)))
1236                 goto bad_fork_cleanup_namespaces;
1237         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1238         if (retval)
1239                 goto bad_fork_cleanup_io;
1240
1241         if (pid != &init_struct_pid) {
1242                 retval = -ENOMEM;
1243                 pid = alloc_pid(p->nsproxy->pid_ns);
1244                 if (!pid)
1245                         goto bad_fork_cleanup_io;
1246         }
1247
1248         p->pid = pid_nr(pid);
1249         p->tgid = p->pid;
1250         if (clone_flags & CLONE_THREAD)
1251                 p->tgid = current->tgid;
1252
1253         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1254         /*
1255          * Clear TID on mm_release()?
1256          */
1257         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1258 #ifdef CONFIG_BLOCK
1259         p->plug = NULL;
1260 #endif
1261 #ifdef CONFIG_FUTEX
1262         p->robust_list = NULL;
1263 #ifdef CONFIG_COMPAT
1264         p->compat_robust_list = NULL;
1265 #endif
1266         INIT_LIST_HEAD(&p->pi_state_list);
1267         p->pi_state_cache = NULL;
1268 #endif
1269         /*
1270          * sigaltstack should be cleared when sharing the same VM
1271          */
1272         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1273                 p->sas_ss_sp = p->sas_ss_size = 0;
1274
1275         /*
1276          * Syscall tracing and stepping should be turned off in the
1277          * child regardless of CLONE_PTRACE.
1278          */
1279         user_disable_single_step(p);
1280         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1281 #ifdef TIF_SYSCALL_EMU
1282         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1283 #endif
1284         clear_all_latency_tracing(p);
1285
1286         /* ok, now we should be set up.. */
1287         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1288         p->pdeath_signal = 0;
1289         p->exit_state = 0;
1290
1291         /*
1292          * Ok, make it visible to the rest of the system.
1293          * We dont wake it up yet.
1294          */
1295         p->group_leader = p;
1296         INIT_LIST_HEAD(&p->thread_group);
1297
1298         /* Now that the task is set up, run cgroup callbacks if
1299          * necessary. We need to run them before the task is visible
1300          * on the tasklist. */
1301         cgroup_fork_callbacks(p);
1302         cgroup_callbacks_done = 1;
1303
1304         /* Need tasklist lock for parent etc handling! */
1305         write_lock_irq(&tasklist_lock);
1306
1307         /* CLONE_PARENT re-uses the old parent */
1308         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1309                 p->real_parent = current->real_parent;
1310                 p->parent_exec_id = current->parent_exec_id;
1311         } else {
1312                 p->real_parent = current;
1313                 p->parent_exec_id = current->self_exec_id;
1314         }
1315
1316         spin_lock(&current->sighand->siglock);
1317
1318         /*
1319          * Process group and session signals need to be delivered to just the
1320          * parent before the fork or both the parent and the child after the
1321          * fork. Restart if a signal comes in before we add the new process to
1322          * it's process group.
1323          * A fatal signal pending means that current will exit, so the new
1324          * thread can't slip out of an OOM kill (or normal SIGKILL).
1325          */
1326         recalc_sigpending();
1327         if (signal_pending(current)) {
1328                 spin_unlock(&current->sighand->siglock);
1329                 write_unlock_irq(&tasklist_lock);
1330                 retval = -ERESTARTNOINTR;
1331                 goto bad_fork_free_pid;
1332         }
1333
1334         if (clone_flags & CLONE_THREAD) {
1335                 current->signal->nr_threads++;
1336                 atomic_inc(&current->signal->live);
1337                 atomic_inc(&current->signal->sigcnt);
1338                 p->group_leader = current->group_leader;
1339                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1340         }
1341
1342         if (likely(p->pid)) {
1343                 tracehook_finish_clone(p, clone_flags, trace);
1344
1345                 if (thread_group_leader(p)) {
1346                         if (is_child_reaper(pid))
1347                                 p->nsproxy->pid_ns->child_reaper = p;
1348
1349                         p->signal->leader_pid = pid;
1350                         p->signal->tty = tty_kref_get(current->signal->tty);
1351                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1352                         attach_pid(p, PIDTYPE_SID, task_session(current));
1353                         list_add_tail(&p->sibling, &p->real_parent->children);
1354                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1355                         __this_cpu_inc(process_counts);
1356                 }
1357                 attach_pid(p, PIDTYPE_PID, pid);
1358                 nr_threads++;
1359         }
1360
1361         total_forks++;
1362         spin_unlock(&current->sighand->siglock);
1363         write_unlock_irq(&tasklist_lock);
1364         proc_fork_connector(p);
1365         cgroup_post_fork(p);
1366         if (clone_flags & CLONE_THREAD)
1367                 threadgroup_fork_read_unlock(current);
1368         perf_event_fork(p);
1369         return p;
1370
1371 bad_fork_free_pid:
1372         if (pid != &init_struct_pid)
1373                 free_pid(pid);
1374 bad_fork_cleanup_io:
1375         if (p->io_context)
1376                 exit_io_context(p);
1377 bad_fork_cleanup_namespaces:
1378         exit_task_namespaces(p);
1379 bad_fork_cleanup_mm:
1380         if (p->mm) {
1381                 task_lock(p);
1382                 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1383                         atomic_dec(&p->mm->oom_disable_count);
1384                 task_unlock(p);
1385                 mmput(p->mm);
1386         }
1387 bad_fork_cleanup_signal:
1388         if (!(clone_flags & CLONE_THREAD))
1389                 free_signal_struct(p->signal);
1390 bad_fork_cleanup_sighand:
1391         __cleanup_sighand(p->sighand);
1392 bad_fork_cleanup_fs:
1393         exit_fs(p); /* blocking */
1394 bad_fork_cleanup_files:
1395         exit_files(p); /* blocking */
1396 bad_fork_cleanup_semundo:
1397         exit_sem(p);
1398 bad_fork_cleanup_audit:
1399         audit_free(p);
1400 bad_fork_cleanup_policy:
1401         perf_event_free_task(p);
1402 #ifdef CONFIG_NUMA
1403         mpol_put(p->mempolicy);
1404 bad_fork_cleanup_cgroup:
1405 #endif
1406         if (clone_flags & CLONE_THREAD)
1407                 threadgroup_fork_read_unlock(current);
1408         cgroup_exit(p, cgroup_callbacks_done);
1409         delayacct_tsk_free(p);
1410         module_put(task_thread_info(p)->exec_domain->module);
1411 bad_fork_cleanup_count:
1412         atomic_dec(&p->cred->user->processes);
1413         exit_creds(p);
1414 bad_fork_free:
1415         free_task(p);
1416 fork_out:
1417         return ERR_PTR(retval);
1418 }
1419
1420 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1421 {
1422         memset(regs, 0, sizeof(struct pt_regs));
1423         return regs;
1424 }
1425
1426 static inline void init_idle_pids(struct pid_link *links)
1427 {
1428         enum pid_type type;
1429
1430         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1431                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1432                 links[type].pid = &init_struct_pid;
1433         }
1434 }
1435
1436 struct task_struct * __cpuinit fork_idle(int cpu)
1437 {
1438         struct task_struct *task;
1439         struct pt_regs regs;
1440
1441         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1442                             &init_struct_pid, 0);
1443         if (!IS_ERR(task)) {
1444                 init_idle_pids(task->pids);
1445                 init_idle(task, cpu);
1446         }
1447
1448         return task;
1449 }
1450
1451 /*
1452  *  Ok, this is the main fork-routine.
1453  *
1454  * It copies the process, and if successful kick-starts
1455  * it and waits for it to finish using the VM if required.
1456  */
1457 long do_fork(unsigned long clone_flags,
1458               unsigned long stack_start,
1459               struct pt_regs *regs,
1460               unsigned long stack_size,
1461               int __user *parent_tidptr,
1462               int __user *child_tidptr)
1463 {
1464         struct task_struct *p;
1465         int trace = 0;
1466         long nr;
1467
1468         /*
1469          * Do some preliminary argument and permissions checking before we
1470          * actually start allocating stuff
1471          */
1472         if (clone_flags & CLONE_NEWUSER) {
1473                 if (clone_flags & CLONE_THREAD)
1474                         return -EINVAL;
1475                 /* hopefully this check will go away when userns support is
1476                  * complete
1477                  */
1478                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1479                                 !capable(CAP_SETGID))
1480                         return -EPERM;
1481         }
1482
1483         /*
1484          * When called from kernel_thread, don't do user tracing stuff.
1485          */
1486         if (likely(user_mode(regs)))
1487                 trace = tracehook_prepare_clone(clone_flags);
1488
1489         p = copy_process(clone_flags, stack_start, regs, stack_size,
1490                          child_tidptr, NULL, trace);
1491         /*
1492          * Do this prior waking up the new thread - the thread pointer
1493          * might get invalid after that point, if the thread exits quickly.
1494          */
1495         if (!IS_ERR(p)) {
1496                 struct completion vfork;
1497
1498                 trace_sched_process_fork(current, p);
1499
1500                 nr = task_pid_vnr(p);
1501
1502                 if (clone_flags & CLONE_PARENT_SETTID)
1503                         put_user(nr, parent_tidptr);
1504
1505                 if (clone_flags & CLONE_VFORK) {
1506                         p->vfork_done = &vfork;
1507                         init_completion(&vfork);
1508                 }
1509
1510                 audit_finish_fork(p);
1511                 tracehook_report_clone(regs, clone_flags, nr, p);
1512
1513                 /*
1514                  * We set PF_STARTING at creation in case tracing wants to
1515                  * use this to distinguish a fully live task from one that
1516                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1517                  * clear it and set the child going.
1518                  */
1519                 p->flags &= ~PF_STARTING;
1520
1521                 wake_up_new_task(p);
1522
1523                 tracehook_report_clone_complete(trace, regs,
1524                                                 clone_flags, nr, p);
1525
1526                 if (clone_flags & CLONE_VFORK) {
1527                         freezer_do_not_count();
1528                         wait_for_completion(&vfork);
1529                         freezer_count();
1530                         tracehook_report_vfork_done(p, nr);
1531                 }
1532         } else {
1533                 nr = PTR_ERR(p);
1534         }
1535         return nr;
1536 }
1537
1538 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1539 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1540 #endif
1541
1542 static void sighand_ctor(void *data)
1543 {
1544         struct sighand_struct *sighand = data;
1545
1546         spin_lock_init(&sighand->siglock);
1547         init_waitqueue_head(&sighand->signalfd_wqh);
1548 }
1549
1550 void __init proc_caches_init(void)
1551 {
1552         sighand_cachep = kmem_cache_create("sighand_cache",
1553                         sizeof(struct sighand_struct), 0,
1554                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1555                         SLAB_NOTRACK, sighand_ctor);
1556         signal_cachep = kmem_cache_create("signal_cache",
1557                         sizeof(struct signal_struct), 0,
1558                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1559         files_cachep = kmem_cache_create("files_cache",
1560                         sizeof(struct files_struct), 0,
1561                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1562         fs_cachep = kmem_cache_create("fs_cache",
1563                         sizeof(struct fs_struct), 0,
1564                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1565         /*
1566          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1567          * whole struct cpumask for the OFFSTACK case. We could change
1568          * this to *only* allocate as much of it as required by the
1569          * maximum number of CPU's we can ever have.  The cpumask_allocation
1570          * is at the end of the structure, exactly for that reason.
1571          */
1572         mm_cachep = kmem_cache_create("mm_struct",
1573                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1574                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1575         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1576         mmap_init();
1577         nsproxy_cache_init();
1578 }
1579
1580 /*
1581  * Check constraints on flags passed to the unshare system call.
1582  */
1583 static int check_unshare_flags(unsigned long unshare_flags)
1584 {
1585         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1586                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1587                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1588                 return -EINVAL;
1589         /*
1590          * Not implemented, but pretend it works if there is nothing to
1591          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1592          * needs to unshare vm.
1593          */
1594         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1595                 /* FIXME: get_task_mm() increments ->mm_users */
1596                 if (atomic_read(&current->mm->mm_users) > 1)
1597                         return -EINVAL;
1598         }
1599
1600         return 0;
1601 }
1602
1603 /*
1604  * Unshare the filesystem structure if it is being shared
1605  */
1606 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1607 {
1608         struct fs_struct *fs = current->fs;
1609
1610         if (!(unshare_flags & CLONE_FS) || !fs)
1611                 return 0;
1612
1613         /* don't need lock here; in the worst case we'll do useless copy */
1614         if (fs->users == 1)
1615                 return 0;
1616
1617         *new_fsp = copy_fs_struct(fs);
1618         if (!*new_fsp)
1619                 return -ENOMEM;
1620
1621         return 0;
1622 }
1623
1624 /*
1625  * Unshare file descriptor table if it is being shared
1626  */
1627 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1628 {
1629         struct files_struct *fd = current->files;
1630         int error = 0;
1631
1632         if ((unshare_flags & CLONE_FILES) &&
1633             (fd && atomic_read(&fd->count) > 1)) {
1634                 *new_fdp = dup_fd(fd, &error);
1635                 if (!*new_fdp)
1636                         return error;
1637         }
1638
1639         return 0;
1640 }
1641
1642 /*
1643  * unshare allows a process to 'unshare' part of the process
1644  * context which was originally shared using clone.  copy_*
1645  * functions used by do_fork() cannot be used here directly
1646  * because they modify an inactive task_struct that is being
1647  * constructed. Here we are modifying the current, active,
1648  * task_struct.
1649  */
1650 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1651 {
1652         struct fs_struct *fs, *new_fs = NULL;
1653         struct files_struct *fd, *new_fd = NULL;
1654         struct nsproxy *new_nsproxy = NULL;
1655         int do_sysvsem = 0;
1656         int err;
1657
1658         err = check_unshare_flags(unshare_flags);
1659         if (err)
1660                 goto bad_unshare_out;
1661
1662         /*
1663          * If unsharing namespace, must also unshare filesystem information.
1664          */
1665         if (unshare_flags & CLONE_NEWNS)
1666                 unshare_flags |= CLONE_FS;
1667         /*
1668          * CLONE_NEWIPC must also detach from the undolist: after switching
1669          * to a new ipc namespace, the semaphore arrays from the old
1670          * namespace are unreachable.
1671          */
1672         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1673                 do_sysvsem = 1;
1674         if ((err = unshare_fs(unshare_flags, &new_fs)))
1675                 goto bad_unshare_out;
1676         if ((err = unshare_fd(unshare_flags, &new_fd)))
1677                 goto bad_unshare_cleanup_fs;
1678         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1679                         new_fs)))
1680                 goto bad_unshare_cleanup_fd;
1681
1682         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1683                 if (do_sysvsem) {
1684                         /*
1685                          * CLONE_SYSVSEM is equivalent to sys_exit().
1686                          */
1687                         exit_sem(current);
1688                 }
1689
1690                 if (new_nsproxy) {
1691                         switch_task_namespaces(current, new_nsproxy);
1692                         new_nsproxy = NULL;
1693                 }
1694
1695                 task_lock(current);
1696
1697                 if (new_fs) {
1698                         fs = current->fs;
1699                         spin_lock(&fs->lock);
1700                         current->fs = new_fs;
1701                         if (--fs->users)
1702                                 new_fs = NULL;
1703                         else
1704                                 new_fs = fs;
1705                         spin_unlock(&fs->lock);
1706                 }
1707
1708                 if (new_fd) {
1709                         fd = current->files;
1710                         current->files = new_fd;
1711                         new_fd = fd;
1712                 }
1713
1714                 task_unlock(current);
1715         }
1716
1717         if (new_nsproxy)
1718                 put_nsproxy(new_nsproxy);
1719
1720 bad_unshare_cleanup_fd:
1721         if (new_fd)
1722                 put_files_struct(new_fd);
1723
1724 bad_unshare_cleanup_fs:
1725         if (new_fs)
1726                 free_fs_struct(new_fs);
1727
1728 bad_unshare_out:
1729         return err;
1730 }
1731
1732 /*
1733  *      Helper to unshare the files of the current task.
1734  *      We don't want to expose copy_files internals to
1735  *      the exec layer of the kernel.
1736  */
1737
1738 int unshare_files(struct files_struct **displaced)
1739 {
1740         struct task_struct *task = current;
1741         struct files_struct *copy = NULL;
1742         int error;
1743
1744         error = unshare_fd(CLONE_FILES, &copy);
1745         if (error || !copy) {
1746                 *displaced = NULL;
1747                 return error;
1748         }
1749         *displaced = task->files;
1750         task_lock(task);
1751         task->files = copy;
1752         task_unlock(task);
1753         return 0;
1754 }