fork: Remove the weak insanity
[linux-3.10.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 #include <linux/signalfd.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 #define CREATE_TRACE_POINTS
81 #include <trace/events/task.h>
82
83 /*
84  * Protected counters by write_lock_irq(&tasklist_lock)
85  */
86 unsigned long total_forks;      /* Handle normal Linux uptimes. */
87 int nr_threads;                 /* The idle threads do not count.. */
88
89 int max_threads;                /* tunable limit on nr_threads */
90
91 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
92
93 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
94
95 #ifdef CONFIG_PROVE_RCU
96 int lockdep_tasklist_lock_is_held(void)
97 {
98         return lockdep_is_held(&tasklist_lock);
99 }
100 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
101 #endif /* #ifdef CONFIG_PROVE_RCU */
102
103 int nr_processes(void)
104 {
105         int cpu;
106         int total = 0;
107
108         for_each_possible_cpu(cpu)
109                 total += per_cpu(process_counts, cpu);
110
111         return total;
112 }
113
114 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
115 # define alloc_task_struct_node(node)           \
116                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
117 # define free_task_struct(tsk)                  \
118                 kmem_cache_free(task_struct_cachep, (tsk))
119 static struct kmem_cache *task_struct_cachep;
120 #endif
121
122 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
123 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
124                                                   int node)
125 {
126 #ifdef CONFIG_DEBUG_STACK_USAGE
127         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
128 #else
129         gfp_t mask = GFP_KERNEL;
130 #endif
131         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
132
133         return page ? page_address(page) : NULL;
134 }
135
136 static inline void free_thread_info(struct thread_info *ti)
137 {
138         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
139 }
140 #endif
141
142 /* SLAB cache for signal_struct structures (tsk->signal) */
143 static struct kmem_cache *signal_cachep;
144
145 /* SLAB cache for sighand_struct structures (tsk->sighand) */
146 struct kmem_cache *sighand_cachep;
147
148 /* SLAB cache for files_struct structures (tsk->files) */
149 struct kmem_cache *files_cachep;
150
151 /* SLAB cache for fs_struct structures (tsk->fs) */
152 struct kmem_cache *fs_cachep;
153
154 /* SLAB cache for vm_area_struct structures */
155 struct kmem_cache *vm_area_cachep;
156
157 /* SLAB cache for mm_struct structures (tsk->mm) */
158 static struct kmem_cache *mm_cachep;
159
160 static void account_kernel_stack(struct thread_info *ti, int account)
161 {
162         struct zone *zone = page_zone(virt_to_page(ti));
163
164         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
165 }
166
167 void free_task(struct task_struct *tsk)
168 {
169         account_kernel_stack(tsk->stack, -1);
170         free_thread_info(tsk->stack);
171         rt_mutex_debug_task_free(tsk);
172         ftrace_graph_exit_task(tsk);
173         free_task_struct(tsk);
174 }
175 EXPORT_SYMBOL(free_task);
176
177 static inline void free_signal_struct(struct signal_struct *sig)
178 {
179         taskstats_tgid_free(sig);
180         sched_autogroup_exit(sig);
181         kmem_cache_free(signal_cachep, sig);
182 }
183
184 static inline void put_signal_struct(struct signal_struct *sig)
185 {
186         if (atomic_dec_and_test(&sig->sigcnt))
187                 free_signal_struct(sig);
188 }
189
190 void __put_task_struct(struct task_struct *tsk)
191 {
192         WARN_ON(!tsk->exit_state);
193         WARN_ON(atomic_read(&tsk->usage));
194         WARN_ON(tsk == current);
195
196         security_task_free(tsk);
197         exit_creds(tsk);
198         delayacct_tsk_free(tsk);
199         put_signal_struct(tsk->signal);
200
201         if (!profile_handoff_task(tsk))
202                 free_task(tsk);
203 }
204 EXPORT_SYMBOL_GPL(__put_task_struct);
205
206 void __init __weak arch_task_cache_init(void) { }
207
208 void __init fork_init(unsigned long mempages)
209 {
210 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
211 #ifndef ARCH_MIN_TASKALIGN
212 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
213 #endif
214         /* create a slab on which task_structs can be allocated */
215         task_struct_cachep =
216                 kmem_cache_create("task_struct", sizeof(struct task_struct),
217                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
218 #endif
219
220         /* do the arch specific task caches init */
221         arch_task_cache_init();
222
223         /*
224          * The default maximum number of threads is set to a safe
225          * value: the thread structures can take up at most half
226          * of memory.
227          */
228         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
229
230         /*
231          * we need to allow at least 20 threads to boot a system
232          */
233         if (max_threads < 20)
234                 max_threads = 20;
235
236         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
237         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
238         init_task.signal->rlim[RLIMIT_SIGPENDING] =
239                 init_task.signal->rlim[RLIMIT_NPROC];
240 }
241
242 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
243                                                struct task_struct *src)
244 {
245         *dst = *src;
246         return 0;
247 }
248
249 static struct task_struct *dup_task_struct(struct task_struct *orig)
250 {
251         struct task_struct *tsk;
252         struct thread_info *ti;
253         unsigned long *stackend;
254         int node = tsk_fork_get_node(orig);
255         int err;
256
257         prepare_to_copy(orig);
258
259         tsk = alloc_task_struct_node(node);
260         if (!tsk)
261                 return NULL;
262
263         ti = alloc_thread_info_node(tsk, node);
264         if (!ti) {
265                 free_task_struct(tsk);
266                 return NULL;
267         }
268
269         err = arch_dup_task_struct(tsk, orig);
270         if (err)
271                 goto out;
272
273         tsk->stack = ti;
274
275         setup_thread_stack(tsk, orig);
276         clear_user_return_notifier(tsk);
277         clear_tsk_need_resched(tsk);
278         stackend = end_of_stack(tsk);
279         *stackend = STACK_END_MAGIC;    /* for overflow detection */
280
281 #ifdef CONFIG_CC_STACKPROTECTOR
282         tsk->stack_canary = get_random_int();
283 #endif
284
285         /*
286          * One for us, one for whoever does the "release_task()" (usually
287          * parent)
288          */
289         atomic_set(&tsk->usage, 2);
290 #ifdef CONFIG_BLK_DEV_IO_TRACE
291         tsk->btrace_seq = 0;
292 #endif
293         tsk->splice_pipe = NULL;
294
295         account_kernel_stack(ti, 1);
296
297         return tsk;
298
299 out:
300         free_thread_info(ti);
301         free_task_struct(tsk);
302         return NULL;
303 }
304
305 #ifdef CONFIG_MMU
306 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
307 {
308         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
309         struct rb_node **rb_link, *rb_parent;
310         int retval;
311         unsigned long charge;
312         struct mempolicy *pol;
313
314         down_write(&oldmm->mmap_sem);
315         flush_cache_dup_mm(oldmm);
316         /*
317          * Not linked in yet - no deadlock potential:
318          */
319         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
320
321         mm->locked_vm = 0;
322         mm->mmap = NULL;
323         mm->mmap_cache = NULL;
324         mm->free_area_cache = oldmm->mmap_base;
325         mm->cached_hole_size = ~0UL;
326         mm->map_count = 0;
327         cpumask_clear(mm_cpumask(mm));
328         mm->mm_rb = RB_ROOT;
329         rb_link = &mm->mm_rb.rb_node;
330         rb_parent = NULL;
331         pprev = &mm->mmap;
332         retval = ksm_fork(mm, oldmm);
333         if (retval)
334                 goto out;
335         retval = khugepaged_fork(mm, oldmm);
336         if (retval)
337                 goto out;
338
339         prev = NULL;
340         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
341                 struct file *file;
342
343                 if (mpnt->vm_flags & VM_DONTCOPY) {
344                         long pages = vma_pages(mpnt);
345                         mm->total_vm -= pages;
346                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
347                                                                 -pages);
348                         continue;
349                 }
350                 charge = 0;
351                 if (mpnt->vm_flags & VM_ACCOUNT) {
352                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
353                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
354                                 goto fail_nomem;
355                         charge = len;
356                 }
357                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
358                 if (!tmp)
359                         goto fail_nomem;
360                 *tmp = *mpnt;
361                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
362                 pol = mpol_dup(vma_policy(mpnt));
363                 retval = PTR_ERR(pol);
364                 if (IS_ERR(pol))
365                         goto fail_nomem_policy;
366                 vma_set_policy(tmp, pol);
367                 tmp->vm_mm = mm;
368                 if (anon_vma_fork(tmp, mpnt))
369                         goto fail_nomem_anon_vma_fork;
370                 tmp->vm_flags &= ~VM_LOCKED;
371                 tmp->vm_next = tmp->vm_prev = NULL;
372                 file = tmp->vm_file;
373                 if (file) {
374                         struct inode *inode = file->f_path.dentry->d_inode;
375                         struct address_space *mapping = file->f_mapping;
376
377                         get_file(file);
378                         if (tmp->vm_flags & VM_DENYWRITE)
379                                 atomic_dec(&inode->i_writecount);
380                         mutex_lock(&mapping->i_mmap_mutex);
381                         if (tmp->vm_flags & VM_SHARED)
382                                 mapping->i_mmap_writable++;
383                         flush_dcache_mmap_lock(mapping);
384                         /* insert tmp into the share list, just after mpnt */
385                         vma_prio_tree_add(tmp, mpnt);
386                         flush_dcache_mmap_unlock(mapping);
387                         mutex_unlock(&mapping->i_mmap_mutex);
388                 }
389
390                 /*
391                  * Clear hugetlb-related page reserves for children. This only
392                  * affects MAP_PRIVATE mappings. Faults generated by the child
393                  * are not guaranteed to succeed, even if read-only
394                  */
395                 if (is_vm_hugetlb_page(tmp))
396                         reset_vma_resv_huge_pages(tmp);
397
398                 /*
399                  * Link in the new vma and copy the page table entries.
400                  */
401                 *pprev = tmp;
402                 pprev = &tmp->vm_next;
403                 tmp->vm_prev = prev;
404                 prev = tmp;
405
406                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
407                 rb_link = &tmp->vm_rb.rb_right;
408                 rb_parent = &tmp->vm_rb;
409
410                 mm->map_count++;
411                 retval = copy_page_range(mm, oldmm, mpnt);
412
413                 if (tmp->vm_ops && tmp->vm_ops->open)
414                         tmp->vm_ops->open(tmp);
415
416                 if (retval)
417                         goto out;
418         }
419         /* a new mm has just been created */
420         arch_dup_mmap(oldmm, mm);
421         retval = 0;
422 out:
423         up_write(&mm->mmap_sem);
424         flush_tlb_mm(oldmm);
425         up_write(&oldmm->mmap_sem);
426         return retval;
427 fail_nomem_anon_vma_fork:
428         mpol_put(pol);
429 fail_nomem_policy:
430         kmem_cache_free(vm_area_cachep, tmp);
431 fail_nomem:
432         retval = -ENOMEM;
433         vm_unacct_memory(charge);
434         goto out;
435 }
436
437 static inline int mm_alloc_pgd(struct mm_struct *mm)
438 {
439         mm->pgd = pgd_alloc(mm);
440         if (unlikely(!mm->pgd))
441                 return -ENOMEM;
442         return 0;
443 }
444
445 static inline void mm_free_pgd(struct mm_struct *mm)
446 {
447         pgd_free(mm, mm->pgd);
448 }
449 #else
450 #define dup_mmap(mm, oldmm)     (0)
451 #define mm_alloc_pgd(mm)        (0)
452 #define mm_free_pgd(mm)
453 #endif /* CONFIG_MMU */
454
455 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
456
457 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
458 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
459
460 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
461
462 static int __init coredump_filter_setup(char *s)
463 {
464         default_dump_filter =
465                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
466                 MMF_DUMP_FILTER_MASK;
467         return 1;
468 }
469
470 __setup("coredump_filter=", coredump_filter_setup);
471
472 #include <linux/init_task.h>
473
474 static void mm_init_aio(struct mm_struct *mm)
475 {
476 #ifdef CONFIG_AIO
477         spin_lock_init(&mm->ioctx_lock);
478         INIT_HLIST_HEAD(&mm->ioctx_list);
479 #endif
480 }
481
482 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
483 {
484         atomic_set(&mm->mm_users, 1);
485         atomic_set(&mm->mm_count, 1);
486         init_rwsem(&mm->mmap_sem);
487         INIT_LIST_HEAD(&mm->mmlist);
488         mm->flags = (current->mm) ?
489                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
490         mm->core_state = NULL;
491         mm->nr_ptes = 0;
492         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
493         spin_lock_init(&mm->page_table_lock);
494         mm->free_area_cache = TASK_UNMAPPED_BASE;
495         mm->cached_hole_size = ~0UL;
496         mm_init_aio(mm);
497         mm_init_owner(mm, p);
498
499         if (likely(!mm_alloc_pgd(mm))) {
500                 mm->def_flags = 0;
501                 mmu_notifier_mm_init(mm);
502                 return mm;
503         }
504
505         free_mm(mm);
506         return NULL;
507 }
508
509 static void check_mm(struct mm_struct *mm)
510 {
511         int i;
512
513         for (i = 0; i < NR_MM_COUNTERS; i++) {
514                 long x = atomic_long_read(&mm->rss_stat.count[i]);
515
516                 if (unlikely(x))
517                         printk(KERN_ALERT "BUG: Bad rss-counter state "
518                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
519         }
520
521 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
522         VM_BUG_ON(mm->pmd_huge_pte);
523 #endif
524 }
525
526 /*
527  * Allocate and initialize an mm_struct.
528  */
529 struct mm_struct *mm_alloc(void)
530 {
531         struct mm_struct *mm;
532
533         mm = allocate_mm();
534         if (!mm)
535                 return NULL;
536
537         memset(mm, 0, sizeof(*mm));
538         mm_init_cpumask(mm);
539         return mm_init(mm, current);
540 }
541
542 /*
543  * Called when the last reference to the mm
544  * is dropped: either by a lazy thread or by
545  * mmput. Free the page directory and the mm.
546  */
547 void __mmdrop(struct mm_struct *mm)
548 {
549         BUG_ON(mm == &init_mm);
550         mm_free_pgd(mm);
551         destroy_context(mm);
552         mmu_notifier_mm_destroy(mm);
553         check_mm(mm);
554         free_mm(mm);
555 }
556 EXPORT_SYMBOL_GPL(__mmdrop);
557
558 /*
559  * Decrement the use count and release all resources for an mm.
560  */
561 void mmput(struct mm_struct *mm)
562 {
563         might_sleep();
564
565         if (atomic_dec_and_test(&mm->mm_users)) {
566                 exit_aio(mm);
567                 ksm_exit(mm);
568                 khugepaged_exit(mm); /* must run before exit_mmap */
569                 exit_mmap(mm);
570                 set_mm_exe_file(mm, NULL);
571                 if (!list_empty(&mm->mmlist)) {
572                         spin_lock(&mmlist_lock);
573                         list_del(&mm->mmlist);
574                         spin_unlock(&mmlist_lock);
575                 }
576                 put_swap_token(mm);
577                 if (mm->binfmt)
578                         module_put(mm->binfmt->module);
579                 mmdrop(mm);
580         }
581 }
582 EXPORT_SYMBOL_GPL(mmput);
583
584 /*
585  * We added or removed a vma mapping the executable. The vmas are only mapped
586  * during exec and are not mapped with the mmap system call.
587  * Callers must hold down_write() on the mm's mmap_sem for these
588  */
589 void added_exe_file_vma(struct mm_struct *mm)
590 {
591         mm->num_exe_file_vmas++;
592 }
593
594 void removed_exe_file_vma(struct mm_struct *mm)
595 {
596         mm->num_exe_file_vmas--;
597         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
598                 fput(mm->exe_file);
599                 mm->exe_file = NULL;
600         }
601
602 }
603
604 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
605 {
606         if (new_exe_file)
607                 get_file(new_exe_file);
608         if (mm->exe_file)
609                 fput(mm->exe_file);
610         mm->exe_file = new_exe_file;
611         mm->num_exe_file_vmas = 0;
612 }
613
614 struct file *get_mm_exe_file(struct mm_struct *mm)
615 {
616         struct file *exe_file;
617
618         /* We need mmap_sem to protect against races with removal of
619          * VM_EXECUTABLE vmas */
620         down_read(&mm->mmap_sem);
621         exe_file = mm->exe_file;
622         if (exe_file)
623                 get_file(exe_file);
624         up_read(&mm->mmap_sem);
625         return exe_file;
626 }
627
628 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
629 {
630         /* It's safe to write the exe_file pointer without exe_file_lock because
631          * this is called during fork when the task is not yet in /proc */
632         newmm->exe_file = get_mm_exe_file(oldmm);
633 }
634
635 /**
636  * get_task_mm - acquire a reference to the task's mm
637  *
638  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
639  * this kernel workthread has transiently adopted a user mm with use_mm,
640  * to do its AIO) is not set and if so returns a reference to it, after
641  * bumping up the use count.  User must release the mm via mmput()
642  * after use.  Typically used by /proc and ptrace.
643  */
644 struct mm_struct *get_task_mm(struct task_struct *task)
645 {
646         struct mm_struct *mm;
647
648         task_lock(task);
649         mm = task->mm;
650         if (mm) {
651                 if (task->flags & PF_KTHREAD)
652                         mm = NULL;
653                 else
654                         atomic_inc(&mm->mm_users);
655         }
656         task_unlock(task);
657         return mm;
658 }
659 EXPORT_SYMBOL_GPL(get_task_mm);
660
661 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
662 {
663         struct mm_struct *mm;
664         int err;
665
666         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
667         if (err)
668                 return ERR_PTR(err);
669
670         mm = get_task_mm(task);
671         if (mm && mm != current->mm &&
672                         !ptrace_may_access(task, mode)) {
673                 mmput(mm);
674                 mm = ERR_PTR(-EACCES);
675         }
676         mutex_unlock(&task->signal->cred_guard_mutex);
677
678         return mm;
679 }
680
681 static void complete_vfork_done(struct task_struct *tsk)
682 {
683         struct completion *vfork;
684
685         task_lock(tsk);
686         vfork = tsk->vfork_done;
687         if (likely(vfork)) {
688                 tsk->vfork_done = NULL;
689                 complete(vfork);
690         }
691         task_unlock(tsk);
692 }
693
694 static int wait_for_vfork_done(struct task_struct *child,
695                                 struct completion *vfork)
696 {
697         int killed;
698
699         freezer_do_not_count();
700         killed = wait_for_completion_killable(vfork);
701         freezer_count();
702
703         if (killed) {
704                 task_lock(child);
705                 child->vfork_done = NULL;
706                 task_unlock(child);
707         }
708
709         put_task_struct(child);
710         return killed;
711 }
712
713 /* Please note the differences between mmput and mm_release.
714  * mmput is called whenever we stop holding onto a mm_struct,
715  * error success whatever.
716  *
717  * mm_release is called after a mm_struct has been removed
718  * from the current process.
719  *
720  * This difference is important for error handling, when we
721  * only half set up a mm_struct for a new process and need to restore
722  * the old one.  Because we mmput the new mm_struct before
723  * restoring the old one. . .
724  * Eric Biederman 10 January 1998
725  */
726 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
727 {
728         /* Get rid of any futexes when releasing the mm */
729 #ifdef CONFIG_FUTEX
730         if (unlikely(tsk->robust_list)) {
731                 exit_robust_list(tsk);
732                 tsk->robust_list = NULL;
733         }
734 #ifdef CONFIG_COMPAT
735         if (unlikely(tsk->compat_robust_list)) {
736                 compat_exit_robust_list(tsk);
737                 tsk->compat_robust_list = NULL;
738         }
739 #endif
740         if (unlikely(!list_empty(&tsk->pi_state_list)))
741                 exit_pi_state_list(tsk);
742 #endif
743
744         /* Get rid of any cached register state */
745         deactivate_mm(tsk, mm);
746
747         if (tsk->vfork_done)
748                 complete_vfork_done(tsk);
749
750         /*
751          * If we're exiting normally, clear a user-space tid field if
752          * requested.  We leave this alone when dying by signal, to leave
753          * the value intact in a core dump, and to save the unnecessary
754          * trouble, say, a killed vfork parent shouldn't touch this mm.
755          * Userland only wants this done for a sys_exit.
756          */
757         if (tsk->clear_child_tid) {
758                 if (!(tsk->flags & PF_SIGNALED) &&
759                     atomic_read(&mm->mm_users) > 1) {
760                         /*
761                          * We don't check the error code - if userspace has
762                          * not set up a proper pointer then tough luck.
763                          */
764                         put_user(0, tsk->clear_child_tid);
765                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
766                                         1, NULL, NULL, 0);
767                 }
768                 tsk->clear_child_tid = NULL;
769         }
770 }
771
772 /*
773  * Allocate a new mm structure and copy contents from the
774  * mm structure of the passed in task structure.
775  */
776 struct mm_struct *dup_mm(struct task_struct *tsk)
777 {
778         struct mm_struct *mm, *oldmm = current->mm;
779         int err;
780
781         if (!oldmm)
782                 return NULL;
783
784         mm = allocate_mm();
785         if (!mm)
786                 goto fail_nomem;
787
788         memcpy(mm, oldmm, sizeof(*mm));
789         mm_init_cpumask(mm);
790
791         /* Initializing for Swap token stuff */
792         mm->token_priority = 0;
793         mm->last_interval = 0;
794
795 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
796         mm->pmd_huge_pte = NULL;
797 #endif
798
799         if (!mm_init(mm, tsk))
800                 goto fail_nomem;
801
802         if (init_new_context(tsk, mm))
803                 goto fail_nocontext;
804
805         dup_mm_exe_file(oldmm, mm);
806
807         err = dup_mmap(mm, oldmm);
808         if (err)
809                 goto free_pt;
810
811         mm->hiwater_rss = get_mm_rss(mm);
812         mm->hiwater_vm = mm->total_vm;
813
814         if (mm->binfmt && !try_module_get(mm->binfmt->module))
815                 goto free_pt;
816
817         return mm;
818
819 free_pt:
820         /* don't put binfmt in mmput, we haven't got module yet */
821         mm->binfmt = NULL;
822         mmput(mm);
823
824 fail_nomem:
825         return NULL;
826
827 fail_nocontext:
828         /*
829          * If init_new_context() failed, we cannot use mmput() to free the mm
830          * because it calls destroy_context()
831          */
832         mm_free_pgd(mm);
833         free_mm(mm);
834         return NULL;
835 }
836
837 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
838 {
839         struct mm_struct *mm, *oldmm;
840         int retval;
841
842         tsk->min_flt = tsk->maj_flt = 0;
843         tsk->nvcsw = tsk->nivcsw = 0;
844 #ifdef CONFIG_DETECT_HUNG_TASK
845         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
846 #endif
847
848         tsk->mm = NULL;
849         tsk->active_mm = NULL;
850
851         /*
852          * Are we cloning a kernel thread?
853          *
854          * We need to steal a active VM for that..
855          */
856         oldmm = current->mm;
857         if (!oldmm)
858                 return 0;
859
860         if (clone_flags & CLONE_VM) {
861                 atomic_inc(&oldmm->mm_users);
862                 mm = oldmm;
863                 goto good_mm;
864         }
865
866         retval = -ENOMEM;
867         mm = dup_mm(tsk);
868         if (!mm)
869                 goto fail_nomem;
870
871 good_mm:
872         /* Initializing for Swap token stuff */
873         mm->token_priority = 0;
874         mm->last_interval = 0;
875
876         tsk->mm = mm;
877         tsk->active_mm = mm;
878         return 0;
879
880 fail_nomem:
881         return retval;
882 }
883
884 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
885 {
886         struct fs_struct *fs = current->fs;
887         if (clone_flags & CLONE_FS) {
888                 /* tsk->fs is already what we want */
889                 spin_lock(&fs->lock);
890                 if (fs->in_exec) {
891                         spin_unlock(&fs->lock);
892                         return -EAGAIN;
893                 }
894                 fs->users++;
895                 spin_unlock(&fs->lock);
896                 return 0;
897         }
898         tsk->fs = copy_fs_struct(fs);
899         if (!tsk->fs)
900                 return -ENOMEM;
901         return 0;
902 }
903
904 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
905 {
906         struct files_struct *oldf, *newf;
907         int error = 0;
908
909         /*
910          * A background process may not have any files ...
911          */
912         oldf = current->files;
913         if (!oldf)
914                 goto out;
915
916         if (clone_flags & CLONE_FILES) {
917                 atomic_inc(&oldf->count);
918                 goto out;
919         }
920
921         newf = dup_fd(oldf, &error);
922         if (!newf)
923                 goto out;
924
925         tsk->files = newf;
926         error = 0;
927 out:
928         return error;
929 }
930
931 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
932 {
933 #ifdef CONFIG_BLOCK
934         struct io_context *ioc = current->io_context;
935         struct io_context *new_ioc;
936
937         if (!ioc)
938                 return 0;
939         /*
940          * Share io context with parent, if CLONE_IO is set
941          */
942         if (clone_flags & CLONE_IO) {
943                 tsk->io_context = ioc_task_link(ioc);
944                 if (unlikely(!tsk->io_context))
945                         return -ENOMEM;
946         } else if (ioprio_valid(ioc->ioprio)) {
947                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
948                 if (unlikely(!new_ioc))
949                         return -ENOMEM;
950
951                 new_ioc->ioprio = ioc->ioprio;
952                 put_io_context(new_ioc);
953         }
954 #endif
955         return 0;
956 }
957
958 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
959 {
960         struct sighand_struct *sig;
961
962         if (clone_flags & CLONE_SIGHAND) {
963                 atomic_inc(&current->sighand->count);
964                 return 0;
965         }
966         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
967         rcu_assign_pointer(tsk->sighand, sig);
968         if (!sig)
969                 return -ENOMEM;
970         atomic_set(&sig->count, 1);
971         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
972         return 0;
973 }
974
975 void __cleanup_sighand(struct sighand_struct *sighand)
976 {
977         if (atomic_dec_and_test(&sighand->count)) {
978                 signalfd_cleanup(sighand);
979                 kmem_cache_free(sighand_cachep, sighand);
980         }
981 }
982
983
984 /*
985  * Initialize POSIX timer handling for a thread group.
986  */
987 static void posix_cpu_timers_init_group(struct signal_struct *sig)
988 {
989         unsigned long cpu_limit;
990
991         /* Thread group counters. */
992         thread_group_cputime_init(sig);
993
994         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
995         if (cpu_limit != RLIM_INFINITY) {
996                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
997                 sig->cputimer.running = 1;
998         }
999
1000         /* The timer lists. */
1001         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1002         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1003         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1004 }
1005
1006 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1007 {
1008         struct signal_struct *sig;
1009
1010         if (clone_flags & CLONE_THREAD)
1011                 return 0;
1012
1013         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1014         tsk->signal = sig;
1015         if (!sig)
1016                 return -ENOMEM;
1017
1018         sig->nr_threads = 1;
1019         atomic_set(&sig->live, 1);
1020         atomic_set(&sig->sigcnt, 1);
1021         init_waitqueue_head(&sig->wait_chldexit);
1022         if (clone_flags & CLONE_NEWPID)
1023                 sig->flags |= SIGNAL_UNKILLABLE;
1024         sig->curr_target = tsk;
1025         init_sigpending(&sig->shared_pending);
1026         INIT_LIST_HEAD(&sig->posix_timers);
1027
1028         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1029         sig->real_timer.function = it_real_fn;
1030
1031         task_lock(current->group_leader);
1032         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1033         task_unlock(current->group_leader);
1034
1035         posix_cpu_timers_init_group(sig);
1036
1037         tty_audit_fork(sig);
1038         sched_autogroup_fork(sig);
1039
1040 #ifdef CONFIG_CGROUPS
1041         init_rwsem(&sig->group_rwsem);
1042 #endif
1043
1044         sig->oom_adj = current->signal->oom_adj;
1045         sig->oom_score_adj = current->signal->oom_score_adj;
1046         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1047
1048         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1049                                    current->signal->is_child_subreaper;
1050
1051         mutex_init(&sig->cred_guard_mutex);
1052
1053         return 0;
1054 }
1055
1056 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1057 {
1058         unsigned long new_flags = p->flags;
1059
1060         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1061         new_flags |= PF_FORKNOEXEC;
1062         p->flags = new_flags;
1063 }
1064
1065 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1066 {
1067         current->clear_child_tid = tidptr;
1068
1069         return task_pid_vnr(current);
1070 }
1071
1072 static void rt_mutex_init_task(struct task_struct *p)
1073 {
1074         raw_spin_lock_init(&p->pi_lock);
1075 #ifdef CONFIG_RT_MUTEXES
1076         plist_head_init(&p->pi_waiters);
1077         p->pi_blocked_on = NULL;
1078 #endif
1079 }
1080
1081 #ifdef CONFIG_MM_OWNER
1082 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1083 {
1084         mm->owner = p;
1085 }
1086 #endif /* CONFIG_MM_OWNER */
1087
1088 /*
1089  * Initialize POSIX timer handling for a single task.
1090  */
1091 static void posix_cpu_timers_init(struct task_struct *tsk)
1092 {
1093         tsk->cputime_expires.prof_exp = 0;
1094         tsk->cputime_expires.virt_exp = 0;
1095         tsk->cputime_expires.sched_exp = 0;
1096         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1097         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1098         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1099 }
1100
1101 /*
1102  * This creates a new process as a copy of the old one,
1103  * but does not actually start it yet.
1104  *
1105  * It copies the registers, and all the appropriate
1106  * parts of the process environment (as per the clone
1107  * flags). The actual kick-off is left to the caller.
1108  */
1109 static struct task_struct *copy_process(unsigned long clone_flags,
1110                                         unsigned long stack_start,
1111                                         struct pt_regs *regs,
1112                                         unsigned long stack_size,
1113                                         int __user *child_tidptr,
1114                                         struct pid *pid,
1115                                         int trace)
1116 {
1117         int retval;
1118         struct task_struct *p;
1119         int cgroup_callbacks_done = 0;
1120
1121         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1122                 return ERR_PTR(-EINVAL);
1123
1124         /*
1125          * Thread groups must share signals as well, and detached threads
1126          * can only be started up within the thread group.
1127          */
1128         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1129                 return ERR_PTR(-EINVAL);
1130
1131         /*
1132          * Shared signal handlers imply shared VM. By way of the above,
1133          * thread groups also imply shared VM. Blocking this case allows
1134          * for various simplifications in other code.
1135          */
1136         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1137                 return ERR_PTR(-EINVAL);
1138
1139         /*
1140          * Siblings of global init remain as zombies on exit since they are
1141          * not reaped by their parent (swapper). To solve this and to avoid
1142          * multi-rooted process trees, prevent global and container-inits
1143          * from creating siblings.
1144          */
1145         if ((clone_flags & CLONE_PARENT) &&
1146                                 current->signal->flags & SIGNAL_UNKILLABLE)
1147                 return ERR_PTR(-EINVAL);
1148
1149         retval = security_task_create(clone_flags);
1150         if (retval)
1151                 goto fork_out;
1152
1153         retval = -ENOMEM;
1154         p = dup_task_struct(current);
1155         if (!p)
1156                 goto fork_out;
1157
1158         ftrace_graph_init_task(p);
1159
1160         rt_mutex_init_task(p);
1161
1162 #ifdef CONFIG_PROVE_LOCKING
1163         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1164         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1165 #endif
1166         retval = -EAGAIN;
1167         if (atomic_read(&p->real_cred->user->processes) >=
1168                         task_rlimit(p, RLIMIT_NPROC)) {
1169                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1170                     p->real_cred->user != INIT_USER)
1171                         goto bad_fork_free;
1172         }
1173         current->flags &= ~PF_NPROC_EXCEEDED;
1174
1175         retval = copy_creds(p, clone_flags);
1176         if (retval < 0)
1177                 goto bad_fork_free;
1178
1179         /*
1180          * If multiple threads are within copy_process(), then this check
1181          * triggers too late. This doesn't hurt, the check is only there
1182          * to stop root fork bombs.
1183          */
1184         retval = -EAGAIN;
1185         if (nr_threads >= max_threads)
1186                 goto bad_fork_cleanup_count;
1187
1188         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1189                 goto bad_fork_cleanup_count;
1190
1191         p->did_exec = 0;
1192         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1193         copy_flags(clone_flags, p);
1194         INIT_LIST_HEAD(&p->children);
1195         INIT_LIST_HEAD(&p->sibling);
1196         rcu_copy_process(p);
1197         p->vfork_done = NULL;
1198         spin_lock_init(&p->alloc_lock);
1199
1200         init_sigpending(&p->pending);
1201
1202         p->utime = p->stime = p->gtime = 0;
1203         p->utimescaled = p->stimescaled = 0;
1204 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1205         p->prev_utime = p->prev_stime = 0;
1206 #endif
1207 #if defined(SPLIT_RSS_COUNTING)
1208         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1209 #endif
1210
1211         p->default_timer_slack_ns = current->timer_slack_ns;
1212
1213         task_io_accounting_init(&p->ioac);
1214         acct_clear_integrals(p);
1215
1216         posix_cpu_timers_init(p);
1217
1218         do_posix_clock_monotonic_gettime(&p->start_time);
1219         p->real_start_time = p->start_time;
1220         monotonic_to_bootbased(&p->real_start_time);
1221         p->io_context = NULL;
1222         p->audit_context = NULL;
1223         if (clone_flags & CLONE_THREAD)
1224                 threadgroup_change_begin(current);
1225         cgroup_fork(p);
1226 #ifdef CONFIG_NUMA
1227         p->mempolicy = mpol_dup(p->mempolicy);
1228         if (IS_ERR(p->mempolicy)) {
1229                 retval = PTR_ERR(p->mempolicy);
1230                 p->mempolicy = NULL;
1231                 goto bad_fork_cleanup_cgroup;
1232         }
1233         mpol_fix_fork_child_flag(p);
1234 #endif
1235 #ifdef CONFIG_CPUSETS
1236         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1237         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1238         seqcount_init(&p->mems_allowed_seq);
1239 #endif
1240 #ifdef CONFIG_TRACE_IRQFLAGS
1241         p->irq_events = 0;
1242 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1243         p->hardirqs_enabled = 1;
1244 #else
1245         p->hardirqs_enabled = 0;
1246 #endif
1247         p->hardirq_enable_ip = 0;
1248         p->hardirq_enable_event = 0;
1249         p->hardirq_disable_ip = _THIS_IP_;
1250         p->hardirq_disable_event = 0;
1251         p->softirqs_enabled = 1;
1252         p->softirq_enable_ip = _THIS_IP_;
1253         p->softirq_enable_event = 0;
1254         p->softirq_disable_ip = 0;
1255         p->softirq_disable_event = 0;
1256         p->hardirq_context = 0;
1257         p->softirq_context = 0;
1258 #endif
1259 #ifdef CONFIG_LOCKDEP
1260         p->lockdep_depth = 0; /* no locks held yet */
1261         p->curr_chain_key = 0;
1262         p->lockdep_recursion = 0;
1263 #endif
1264
1265 #ifdef CONFIG_DEBUG_MUTEXES
1266         p->blocked_on = NULL; /* not blocked yet */
1267 #endif
1268 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1269         p->memcg_batch.do_batch = 0;
1270         p->memcg_batch.memcg = NULL;
1271 #endif
1272
1273         /* Perform scheduler related setup. Assign this task to a CPU. */
1274         sched_fork(p);
1275
1276         retval = perf_event_init_task(p);
1277         if (retval)
1278                 goto bad_fork_cleanup_policy;
1279         retval = audit_alloc(p);
1280         if (retval)
1281                 goto bad_fork_cleanup_policy;
1282         /* copy all the process information */
1283         retval = copy_semundo(clone_flags, p);
1284         if (retval)
1285                 goto bad_fork_cleanup_audit;
1286         retval = copy_files(clone_flags, p);
1287         if (retval)
1288                 goto bad_fork_cleanup_semundo;
1289         retval = copy_fs(clone_flags, p);
1290         if (retval)
1291                 goto bad_fork_cleanup_files;
1292         retval = copy_sighand(clone_flags, p);
1293         if (retval)
1294                 goto bad_fork_cleanup_fs;
1295         retval = copy_signal(clone_flags, p);
1296         if (retval)
1297                 goto bad_fork_cleanup_sighand;
1298         retval = copy_mm(clone_flags, p);
1299         if (retval)
1300                 goto bad_fork_cleanup_signal;
1301         retval = copy_namespaces(clone_flags, p);
1302         if (retval)
1303                 goto bad_fork_cleanup_mm;
1304         retval = copy_io(clone_flags, p);
1305         if (retval)
1306                 goto bad_fork_cleanup_namespaces;
1307         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1308         if (retval)
1309                 goto bad_fork_cleanup_io;
1310
1311         if (pid != &init_struct_pid) {
1312                 retval = -ENOMEM;
1313                 pid = alloc_pid(p->nsproxy->pid_ns);
1314                 if (!pid)
1315                         goto bad_fork_cleanup_io;
1316         }
1317
1318         p->pid = pid_nr(pid);
1319         p->tgid = p->pid;
1320         if (clone_flags & CLONE_THREAD)
1321                 p->tgid = current->tgid;
1322
1323         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1324         /*
1325          * Clear TID on mm_release()?
1326          */
1327         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1328 #ifdef CONFIG_BLOCK
1329         p->plug = NULL;
1330 #endif
1331 #ifdef CONFIG_FUTEX
1332         p->robust_list = NULL;
1333 #ifdef CONFIG_COMPAT
1334         p->compat_robust_list = NULL;
1335 #endif
1336         INIT_LIST_HEAD(&p->pi_state_list);
1337         p->pi_state_cache = NULL;
1338 #endif
1339         /*
1340          * sigaltstack should be cleared when sharing the same VM
1341          */
1342         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1343                 p->sas_ss_sp = p->sas_ss_size = 0;
1344
1345         /*
1346          * Syscall tracing and stepping should be turned off in the
1347          * child regardless of CLONE_PTRACE.
1348          */
1349         user_disable_single_step(p);
1350         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1351 #ifdef TIF_SYSCALL_EMU
1352         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1353 #endif
1354         clear_all_latency_tracing(p);
1355
1356         /* ok, now we should be set up.. */
1357         if (clone_flags & CLONE_THREAD)
1358                 p->exit_signal = -1;
1359         else if (clone_flags & CLONE_PARENT)
1360                 p->exit_signal = current->group_leader->exit_signal;
1361         else
1362                 p->exit_signal = (clone_flags & CSIGNAL);
1363
1364         p->pdeath_signal = 0;
1365         p->exit_state = 0;
1366
1367         p->nr_dirtied = 0;
1368         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1369         p->dirty_paused_when = 0;
1370
1371         /*
1372          * Ok, make it visible to the rest of the system.
1373          * We dont wake it up yet.
1374          */
1375         p->group_leader = p;
1376         INIT_LIST_HEAD(&p->thread_group);
1377
1378         /* Now that the task is set up, run cgroup callbacks if
1379          * necessary. We need to run them before the task is visible
1380          * on the tasklist. */
1381         cgroup_fork_callbacks(p);
1382         cgroup_callbacks_done = 1;
1383
1384         /* Need tasklist lock for parent etc handling! */
1385         write_lock_irq(&tasklist_lock);
1386
1387         /* CLONE_PARENT re-uses the old parent */
1388         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1389                 p->real_parent = current->real_parent;
1390                 p->parent_exec_id = current->parent_exec_id;
1391         } else {
1392                 p->real_parent = current;
1393                 p->parent_exec_id = current->self_exec_id;
1394         }
1395
1396         spin_lock(&current->sighand->siglock);
1397
1398         /*
1399          * Process group and session signals need to be delivered to just the
1400          * parent before the fork or both the parent and the child after the
1401          * fork. Restart if a signal comes in before we add the new process to
1402          * it's process group.
1403          * A fatal signal pending means that current will exit, so the new
1404          * thread can't slip out of an OOM kill (or normal SIGKILL).
1405         */
1406         recalc_sigpending();
1407         if (signal_pending(current)) {
1408                 spin_unlock(&current->sighand->siglock);
1409                 write_unlock_irq(&tasklist_lock);
1410                 retval = -ERESTARTNOINTR;
1411                 goto bad_fork_free_pid;
1412         }
1413
1414         if (clone_flags & CLONE_THREAD) {
1415                 current->signal->nr_threads++;
1416                 atomic_inc(&current->signal->live);
1417                 atomic_inc(&current->signal->sigcnt);
1418                 p->group_leader = current->group_leader;
1419                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1420         }
1421
1422         if (likely(p->pid)) {
1423                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1424
1425                 if (thread_group_leader(p)) {
1426                         if (is_child_reaper(pid))
1427                                 p->nsproxy->pid_ns->child_reaper = p;
1428
1429                         p->signal->leader_pid = pid;
1430                         p->signal->tty = tty_kref_get(current->signal->tty);
1431                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1432                         attach_pid(p, PIDTYPE_SID, task_session(current));
1433                         list_add_tail(&p->sibling, &p->real_parent->children);
1434                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1435                         __this_cpu_inc(process_counts);
1436                 }
1437                 attach_pid(p, PIDTYPE_PID, pid);
1438                 nr_threads++;
1439         }
1440
1441         total_forks++;
1442         spin_unlock(&current->sighand->siglock);
1443         write_unlock_irq(&tasklist_lock);
1444         proc_fork_connector(p);
1445         cgroup_post_fork(p);
1446         if (clone_flags & CLONE_THREAD)
1447                 threadgroup_change_end(current);
1448         perf_event_fork(p);
1449
1450         trace_task_newtask(p, clone_flags);
1451
1452         return p;
1453
1454 bad_fork_free_pid:
1455         if (pid != &init_struct_pid)
1456                 free_pid(pid);
1457 bad_fork_cleanup_io:
1458         if (p->io_context)
1459                 exit_io_context(p);
1460 bad_fork_cleanup_namespaces:
1461         exit_task_namespaces(p);
1462 bad_fork_cleanup_mm:
1463         if (p->mm)
1464                 mmput(p->mm);
1465 bad_fork_cleanup_signal:
1466         if (!(clone_flags & CLONE_THREAD))
1467                 free_signal_struct(p->signal);
1468 bad_fork_cleanup_sighand:
1469         __cleanup_sighand(p->sighand);
1470 bad_fork_cleanup_fs:
1471         exit_fs(p); /* blocking */
1472 bad_fork_cleanup_files:
1473         exit_files(p); /* blocking */
1474 bad_fork_cleanup_semundo:
1475         exit_sem(p);
1476 bad_fork_cleanup_audit:
1477         audit_free(p);
1478 bad_fork_cleanup_policy:
1479         perf_event_free_task(p);
1480 #ifdef CONFIG_NUMA
1481         mpol_put(p->mempolicy);
1482 bad_fork_cleanup_cgroup:
1483 #endif
1484         if (clone_flags & CLONE_THREAD)
1485                 threadgroup_change_end(current);
1486         cgroup_exit(p, cgroup_callbacks_done);
1487         delayacct_tsk_free(p);
1488         module_put(task_thread_info(p)->exec_domain->module);
1489 bad_fork_cleanup_count:
1490         atomic_dec(&p->cred->user->processes);
1491         exit_creds(p);
1492 bad_fork_free:
1493         free_task(p);
1494 fork_out:
1495         return ERR_PTR(retval);
1496 }
1497
1498 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1499 {
1500         memset(regs, 0, sizeof(struct pt_regs));
1501         return regs;
1502 }
1503
1504 static inline void init_idle_pids(struct pid_link *links)
1505 {
1506         enum pid_type type;
1507
1508         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1509                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1510                 links[type].pid = &init_struct_pid;
1511         }
1512 }
1513
1514 struct task_struct * __cpuinit fork_idle(int cpu)
1515 {
1516         struct task_struct *task;
1517         struct pt_regs regs;
1518
1519         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1520                             &init_struct_pid, 0);
1521         if (!IS_ERR(task)) {
1522                 init_idle_pids(task->pids);
1523                 init_idle(task, cpu);
1524         }
1525
1526         return task;
1527 }
1528
1529 /*
1530  *  Ok, this is the main fork-routine.
1531  *
1532  * It copies the process, and if successful kick-starts
1533  * it and waits for it to finish using the VM if required.
1534  */
1535 long do_fork(unsigned long clone_flags,
1536               unsigned long stack_start,
1537               struct pt_regs *regs,
1538               unsigned long stack_size,
1539               int __user *parent_tidptr,
1540               int __user *child_tidptr)
1541 {
1542         struct task_struct *p;
1543         int trace = 0;
1544         long nr;
1545
1546         /*
1547          * Do some preliminary argument and permissions checking before we
1548          * actually start allocating stuff
1549          */
1550         if (clone_flags & CLONE_NEWUSER) {
1551                 if (clone_flags & CLONE_THREAD)
1552                         return -EINVAL;
1553                 /* hopefully this check will go away when userns support is
1554                  * complete
1555                  */
1556                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1557                                 !capable(CAP_SETGID))
1558                         return -EPERM;
1559         }
1560
1561         /*
1562          * Determine whether and which event to report to ptracer.  When
1563          * called from kernel_thread or CLONE_UNTRACED is explicitly
1564          * requested, no event is reported; otherwise, report if the event
1565          * for the type of forking is enabled.
1566          */
1567         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1568                 if (clone_flags & CLONE_VFORK)
1569                         trace = PTRACE_EVENT_VFORK;
1570                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1571                         trace = PTRACE_EVENT_CLONE;
1572                 else
1573                         trace = PTRACE_EVENT_FORK;
1574
1575                 if (likely(!ptrace_event_enabled(current, trace)))
1576                         trace = 0;
1577         }
1578
1579         p = copy_process(clone_flags, stack_start, regs, stack_size,
1580                          child_tidptr, NULL, trace);
1581         /*
1582          * Do this prior waking up the new thread - the thread pointer
1583          * might get invalid after that point, if the thread exits quickly.
1584          */
1585         if (!IS_ERR(p)) {
1586                 struct completion vfork;
1587
1588                 trace_sched_process_fork(current, p);
1589
1590                 nr = task_pid_vnr(p);
1591
1592                 if (clone_flags & CLONE_PARENT_SETTID)
1593                         put_user(nr, parent_tidptr);
1594
1595                 if (clone_flags & CLONE_VFORK) {
1596                         p->vfork_done = &vfork;
1597                         init_completion(&vfork);
1598                         get_task_struct(p);
1599                 }
1600
1601                 wake_up_new_task(p);
1602
1603                 /* forking complete and child started to run, tell ptracer */
1604                 if (unlikely(trace))
1605                         ptrace_event(trace, nr);
1606
1607                 if (clone_flags & CLONE_VFORK) {
1608                         if (!wait_for_vfork_done(p, &vfork))
1609                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1610                 }
1611         } else {
1612                 nr = PTR_ERR(p);
1613         }
1614         return nr;
1615 }
1616
1617 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1618 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1619 #endif
1620
1621 static void sighand_ctor(void *data)
1622 {
1623         struct sighand_struct *sighand = data;
1624
1625         spin_lock_init(&sighand->siglock);
1626         init_waitqueue_head(&sighand->signalfd_wqh);
1627 }
1628
1629 void __init proc_caches_init(void)
1630 {
1631         sighand_cachep = kmem_cache_create("sighand_cache",
1632                         sizeof(struct sighand_struct), 0,
1633                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1634                         SLAB_NOTRACK, sighand_ctor);
1635         signal_cachep = kmem_cache_create("signal_cache",
1636                         sizeof(struct signal_struct), 0,
1637                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1638         files_cachep = kmem_cache_create("files_cache",
1639                         sizeof(struct files_struct), 0,
1640                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1641         fs_cachep = kmem_cache_create("fs_cache",
1642                         sizeof(struct fs_struct), 0,
1643                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1644         /*
1645          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1646          * whole struct cpumask for the OFFSTACK case. We could change
1647          * this to *only* allocate as much of it as required by the
1648          * maximum number of CPU's we can ever have.  The cpumask_allocation
1649          * is at the end of the structure, exactly for that reason.
1650          */
1651         mm_cachep = kmem_cache_create("mm_struct",
1652                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1653                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1654         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1655         mmap_init();
1656         nsproxy_cache_init();
1657 }
1658
1659 /*
1660  * Check constraints on flags passed to the unshare system call.
1661  */
1662 static int check_unshare_flags(unsigned long unshare_flags)
1663 {
1664         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1665                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1666                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1667                 return -EINVAL;
1668         /*
1669          * Not implemented, but pretend it works if there is nothing to
1670          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1671          * needs to unshare vm.
1672          */
1673         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1674                 /* FIXME: get_task_mm() increments ->mm_users */
1675                 if (atomic_read(&current->mm->mm_users) > 1)
1676                         return -EINVAL;
1677         }
1678
1679         return 0;
1680 }
1681
1682 /*
1683  * Unshare the filesystem structure if it is being shared
1684  */
1685 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1686 {
1687         struct fs_struct *fs = current->fs;
1688
1689         if (!(unshare_flags & CLONE_FS) || !fs)
1690                 return 0;
1691
1692         /* don't need lock here; in the worst case we'll do useless copy */
1693         if (fs->users == 1)
1694                 return 0;
1695
1696         *new_fsp = copy_fs_struct(fs);
1697         if (!*new_fsp)
1698                 return -ENOMEM;
1699
1700         return 0;
1701 }
1702
1703 /*
1704  * Unshare file descriptor table if it is being shared
1705  */
1706 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1707 {
1708         struct files_struct *fd = current->files;
1709         int error = 0;
1710
1711         if ((unshare_flags & CLONE_FILES) &&
1712             (fd && atomic_read(&fd->count) > 1)) {
1713                 *new_fdp = dup_fd(fd, &error);
1714                 if (!*new_fdp)
1715                         return error;
1716         }
1717
1718         return 0;
1719 }
1720
1721 /*
1722  * unshare allows a process to 'unshare' part of the process
1723  * context which was originally shared using clone.  copy_*
1724  * functions used by do_fork() cannot be used here directly
1725  * because they modify an inactive task_struct that is being
1726  * constructed. Here we are modifying the current, active,
1727  * task_struct.
1728  */
1729 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1730 {
1731         struct fs_struct *fs, *new_fs = NULL;
1732         struct files_struct *fd, *new_fd = NULL;
1733         struct nsproxy *new_nsproxy = NULL;
1734         int do_sysvsem = 0;
1735         int err;
1736
1737         err = check_unshare_flags(unshare_flags);
1738         if (err)
1739                 goto bad_unshare_out;
1740
1741         /*
1742          * If unsharing namespace, must also unshare filesystem information.
1743          */
1744         if (unshare_flags & CLONE_NEWNS)
1745                 unshare_flags |= CLONE_FS;
1746         /*
1747          * CLONE_NEWIPC must also detach from the undolist: after switching
1748          * to a new ipc namespace, the semaphore arrays from the old
1749          * namespace are unreachable.
1750          */
1751         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1752                 do_sysvsem = 1;
1753         err = unshare_fs(unshare_flags, &new_fs);
1754         if (err)
1755                 goto bad_unshare_out;
1756         err = unshare_fd(unshare_flags, &new_fd);
1757         if (err)
1758                 goto bad_unshare_cleanup_fs;
1759         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1760         if (err)
1761                 goto bad_unshare_cleanup_fd;
1762
1763         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1764                 if (do_sysvsem) {
1765                         /*
1766                          * CLONE_SYSVSEM is equivalent to sys_exit().
1767                          */
1768                         exit_sem(current);
1769                 }
1770
1771                 if (new_nsproxy) {
1772                         switch_task_namespaces(current, new_nsproxy);
1773                         new_nsproxy = NULL;
1774                 }
1775
1776                 task_lock(current);
1777
1778                 if (new_fs) {
1779                         fs = current->fs;
1780                         spin_lock(&fs->lock);
1781                         current->fs = new_fs;
1782                         if (--fs->users)
1783                                 new_fs = NULL;
1784                         else
1785                                 new_fs = fs;
1786                         spin_unlock(&fs->lock);
1787                 }
1788
1789                 if (new_fd) {
1790                         fd = current->files;
1791                         current->files = new_fd;
1792                         new_fd = fd;
1793                 }
1794
1795                 task_unlock(current);
1796         }
1797
1798         if (new_nsproxy)
1799                 put_nsproxy(new_nsproxy);
1800
1801 bad_unshare_cleanup_fd:
1802         if (new_fd)
1803                 put_files_struct(new_fd);
1804
1805 bad_unshare_cleanup_fs:
1806         if (new_fs)
1807                 free_fs_struct(new_fs);
1808
1809 bad_unshare_out:
1810         return err;
1811 }
1812
1813 /*
1814  *      Helper to unshare the files of the current task.
1815  *      We don't want to expose copy_files internals to
1816  *      the exec layer of the kernel.
1817  */
1818
1819 int unshare_files(struct files_struct **displaced)
1820 {
1821         struct task_struct *task = current;
1822         struct files_struct *copy = NULL;
1823         int error;
1824
1825         error = unshare_fd(CLONE_FILES, &copy);
1826         if (error || !copy) {
1827                 *displaced = NULL;
1828                 return error;
1829         }
1830         *displaced = task->files;
1831         task_lock(task);
1832         task->files = copy;
1833         task_unlock(task);
1834         return 0;
1835 }