mm: correctly synchronize rss-counters at exit/exec
[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/seccomp.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/kthread.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/rmap.h>
54 #include <linux/ksm.h>
55 #include <linux/acct.h>
56 #include <linux/tsacct_kern.h>
57 #include <linux/cn_proc.h>
58 #include <linux/freezer.h>
59 #include <linux/delayacct.h>
60 #include <linux/taskstats_kern.h>
61 #include <linux/random.h>
62 #include <linux/tty.h>
63 #include <linux/blkdev.h>
64 #include <linux/fs_struct.h>
65 #include <linux/magic.h>
66 #include <linux/perf_event.h>
67 #include <linux/posix-timers.h>
68 #include <linux/user-return-notifier.h>
69 #include <linux/oom.h>
70 #include <linux/khugepaged.h>
71 #include <linux/signalfd.h>
72 #include <linux/uprobes.h>
73
74 #include <asm/pgtable.h>
75 #include <asm/pgalloc.h>
76 #include <asm/uaccess.h>
77 #include <asm/mmu_context.h>
78 #include <asm/cacheflush.h>
79 #include <asm/tlbflush.h>
80
81 #include <trace/events/sched.h>
82
83 #define CREATE_TRACE_POINTS
84 #include <trace/events/task.h>
85
86 /*
87  * Protected counters by write_lock_irq(&tasklist_lock)
88  */
89 unsigned long total_forks;      /* Handle normal Linux uptimes. */
90 int nr_threads;                 /* The idle threads do not count.. */
91
92 int max_threads;                /* tunable limit on nr_threads */
93
94 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
95
96 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
97
98 #ifdef CONFIG_PROVE_RCU
99 int lockdep_tasklist_lock_is_held(void)
100 {
101         return lockdep_is_held(&tasklist_lock);
102 }
103 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
104 #endif /* #ifdef CONFIG_PROVE_RCU */
105
106 int nr_processes(void)
107 {
108         int cpu;
109         int total = 0;
110
111         for_each_possible_cpu(cpu)
112                 total += per_cpu(process_counts, cpu);
113
114         return total;
115 }
116
117 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
118 static struct kmem_cache *task_struct_cachep;
119
120 static inline struct task_struct *alloc_task_struct_node(int node)
121 {
122         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
123 }
124
125 void __weak arch_release_task_struct(struct task_struct *tsk) { }
126
127 static inline void free_task_struct(struct task_struct *tsk)
128 {
129         arch_release_task_struct(tsk);
130         kmem_cache_free(task_struct_cachep, tsk);
131 }
132 #endif
133
134 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
135 void __weak arch_release_thread_info(struct thread_info *ti) { }
136
137 /*
138  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
139  * kmemcache based allocator.
140  */
141 # if THREAD_SIZE >= PAGE_SIZE
142 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
143                                                   int node)
144 {
145         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
146                                              THREAD_SIZE_ORDER);
147
148         return page ? page_address(page) : NULL;
149 }
150
151 static inline void free_thread_info(struct thread_info *ti)
152 {
153         arch_release_thread_info(ti);
154         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
155 }
156 # else
157 static struct kmem_cache *thread_info_cache;
158
159 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
160                                                   int node)
161 {
162         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
163 }
164
165 static void free_thread_info(struct thread_info *ti)
166 {
167         arch_release_thread_info(ti);
168         kmem_cache_free(thread_info_cache, ti);
169 }
170
171 void thread_info_cache_init(void)
172 {
173         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
174                                               THREAD_SIZE, 0, NULL);
175         BUG_ON(thread_info_cache == NULL);
176 }
177 # endif
178 #endif
179
180 /* SLAB cache for signal_struct structures (tsk->signal) */
181 static struct kmem_cache *signal_cachep;
182
183 /* SLAB cache for sighand_struct structures (tsk->sighand) */
184 struct kmem_cache *sighand_cachep;
185
186 /* SLAB cache for files_struct structures (tsk->files) */
187 struct kmem_cache *files_cachep;
188
189 /* SLAB cache for fs_struct structures (tsk->fs) */
190 struct kmem_cache *fs_cachep;
191
192 /* SLAB cache for vm_area_struct structures */
193 struct kmem_cache *vm_area_cachep;
194
195 /* SLAB cache for mm_struct structures (tsk->mm) */
196 static struct kmem_cache *mm_cachep;
197
198 static void account_kernel_stack(struct thread_info *ti, int account)
199 {
200         struct zone *zone = page_zone(virt_to_page(ti));
201
202         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
203 }
204
205 void free_task(struct task_struct *tsk)
206 {
207         account_kernel_stack(tsk->stack, -1);
208         free_thread_info(tsk->stack);
209         rt_mutex_debug_task_free(tsk);
210         ftrace_graph_exit_task(tsk);
211         put_seccomp_filter(tsk);
212         free_task_struct(tsk);
213 }
214 EXPORT_SYMBOL(free_task);
215
216 static inline void free_signal_struct(struct signal_struct *sig)
217 {
218         taskstats_tgid_free(sig);
219         sched_autogroup_exit(sig);
220         kmem_cache_free(signal_cachep, sig);
221 }
222
223 static inline void put_signal_struct(struct signal_struct *sig)
224 {
225         if (atomic_dec_and_test(&sig->sigcnt))
226                 free_signal_struct(sig);
227 }
228
229 void __put_task_struct(struct task_struct *tsk)
230 {
231         WARN_ON(!tsk->exit_state);
232         WARN_ON(atomic_read(&tsk->usage));
233         WARN_ON(tsk == current);
234
235         security_task_free(tsk);
236         exit_creds(tsk);
237         delayacct_tsk_free(tsk);
238         put_signal_struct(tsk->signal);
239
240         if (!profile_handoff_task(tsk))
241                 free_task(tsk);
242 }
243 EXPORT_SYMBOL_GPL(__put_task_struct);
244
245 void __init __weak arch_task_cache_init(void) { }
246
247 void __init fork_init(unsigned long mempages)
248 {
249 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
250 #ifndef ARCH_MIN_TASKALIGN
251 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
252 #endif
253         /* create a slab on which task_structs can be allocated */
254         task_struct_cachep =
255                 kmem_cache_create("task_struct", sizeof(struct task_struct),
256                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
257 #endif
258
259         /* do the arch specific task caches init */
260         arch_task_cache_init();
261
262         /*
263          * The default maximum number of threads is set to a safe
264          * value: the thread structures can take up at most half
265          * of memory.
266          */
267         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
268
269         /*
270          * we need to allow at least 20 threads to boot a system
271          */
272         if (max_threads < 20)
273                 max_threads = 20;
274
275         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
276         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
277         init_task.signal->rlim[RLIMIT_SIGPENDING] =
278                 init_task.signal->rlim[RLIMIT_NPROC];
279 }
280
281 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
282                                                struct task_struct *src)
283 {
284         *dst = *src;
285         return 0;
286 }
287
288 static struct task_struct *dup_task_struct(struct task_struct *orig)
289 {
290         struct task_struct *tsk;
291         struct thread_info *ti;
292         unsigned long *stackend;
293         int node = tsk_fork_get_node(orig);
294         int err;
295
296         tsk = alloc_task_struct_node(node);
297         if (!tsk)
298                 return NULL;
299
300         ti = alloc_thread_info_node(tsk, node);
301         if (!ti) {
302                 free_task_struct(tsk);
303                 return NULL;
304         }
305
306         err = arch_dup_task_struct(tsk, orig);
307         if (err)
308                 goto out;
309
310         tsk->stack = ti;
311
312         setup_thread_stack(tsk, orig);
313         clear_user_return_notifier(tsk);
314         clear_tsk_need_resched(tsk);
315         stackend = end_of_stack(tsk);
316         *stackend = STACK_END_MAGIC;    /* for overflow detection */
317
318 #ifdef CONFIG_CC_STACKPROTECTOR
319         tsk->stack_canary = get_random_int();
320 #endif
321
322         /*
323          * One for us, one for whoever does the "release_task()" (usually
324          * parent)
325          */
326         atomic_set(&tsk->usage, 2);
327 #ifdef CONFIG_BLK_DEV_IO_TRACE
328         tsk->btrace_seq = 0;
329 #endif
330         tsk->splice_pipe = NULL;
331
332         account_kernel_stack(ti, 1);
333
334         return tsk;
335
336 out:
337         free_thread_info(ti);
338         free_task_struct(tsk);
339         return NULL;
340 }
341
342 #ifdef CONFIG_MMU
343 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
344 {
345         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
346         struct rb_node **rb_link, *rb_parent;
347         int retval;
348         unsigned long charge;
349         struct mempolicy *pol;
350
351         down_write(&oldmm->mmap_sem);
352         flush_cache_dup_mm(oldmm);
353         /*
354          * Not linked in yet - no deadlock potential:
355          */
356         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
357
358         mm->locked_vm = 0;
359         mm->mmap = NULL;
360         mm->mmap_cache = NULL;
361         mm->free_area_cache = oldmm->mmap_base;
362         mm->cached_hole_size = ~0UL;
363         mm->map_count = 0;
364         cpumask_clear(mm_cpumask(mm));
365         mm->mm_rb = RB_ROOT;
366         rb_link = &mm->mm_rb.rb_node;
367         rb_parent = NULL;
368         pprev = &mm->mmap;
369         retval = ksm_fork(mm, oldmm);
370         if (retval)
371                 goto out;
372         retval = khugepaged_fork(mm, oldmm);
373         if (retval)
374                 goto out;
375
376         prev = NULL;
377         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
378                 struct file *file;
379
380                 if (mpnt->vm_flags & VM_DONTCOPY) {
381                         long pages = vma_pages(mpnt);
382                         mm->total_vm -= pages;
383                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
384                                                                 -pages);
385                         continue;
386                 }
387                 charge = 0;
388                 if (mpnt->vm_flags & VM_ACCOUNT) {
389                         unsigned long len;
390                         len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
391                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
392                                 goto fail_nomem;
393                         charge = len;
394                 }
395                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
396                 if (!tmp)
397                         goto fail_nomem;
398                 *tmp = *mpnt;
399                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
400                 pol = mpol_dup(vma_policy(mpnt));
401                 retval = PTR_ERR(pol);
402                 if (IS_ERR(pol))
403                         goto fail_nomem_policy;
404                 vma_set_policy(tmp, pol);
405                 tmp->vm_mm = mm;
406                 if (anon_vma_fork(tmp, mpnt))
407                         goto fail_nomem_anon_vma_fork;
408                 tmp->vm_flags &= ~VM_LOCKED;
409                 tmp->vm_next = tmp->vm_prev = NULL;
410                 file = tmp->vm_file;
411                 if (file) {
412                         struct inode *inode = file->f_path.dentry->d_inode;
413                         struct address_space *mapping = file->f_mapping;
414
415                         get_file(file);
416                         if (tmp->vm_flags & VM_DENYWRITE)
417                                 atomic_dec(&inode->i_writecount);
418                         mutex_lock(&mapping->i_mmap_mutex);
419                         if (tmp->vm_flags & VM_SHARED)
420                                 mapping->i_mmap_writable++;
421                         flush_dcache_mmap_lock(mapping);
422                         /* insert tmp into the share list, just after mpnt */
423                         vma_prio_tree_add(tmp, mpnt);
424                         flush_dcache_mmap_unlock(mapping);
425                         mutex_unlock(&mapping->i_mmap_mutex);
426                 }
427
428                 /*
429                  * Clear hugetlb-related page reserves for children. This only
430                  * affects MAP_PRIVATE mappings. Faults generated by the child
431                  * are not guaranteed to succeed, even if read-only
432                  */
433                 if (is_vm_hugetlb_page(tmp))
434                         reset_vma_resv_huge_pages(tmp);
435
436                 /*
437                  * Link in the new vma and copy the page table entries.
438                  */
439                 *pprev = tmp;
440                 pprev = &tmp->vm_next;
441                 tmp->vm_prev = prev;
442                 prev = tmp;
443
444                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
445                 rb_link = &tmp->vm_rb.rb_right;
446                 rb_parent = &tmp->vm_rb;
447
448                 mm->map_count++;
449                 retval = copy_page_range(mm, oldmm, mpnt);
450
451                 if (tmp->vm_ops && tmp->vm_ops->open)
452                         tmp->vm_ops->open(tmp);
453
454                 if (retval)
455                         goto out;
456
457                 if (file && uprobe_mmap(tmp))
458                         goto out;
459         }
460         /* a new mm has just been created */
461         arch_dup_mmap(oldmm, mm);
462         retval = 0;
463 out:
464         up_write(&mm->mmap_sem);
465         flush_tlb_mm(oldmm);
466         up_write(&oldmm->mmap_sem);
467         return retval;
468 fail_nomem_anon_vma_fork:
469         mpol_put(pol);
470 fail_nomem_policy:
471         kmem_cache_free(vm_area_cachep, tmp);
472 fail_nomem:
473         retval = -ENOMEM;
474         vm_unacct_memory(charge);
475         goto out;
476 }
477
478 static inline int mm_alloc_pgd(struct mm_struct *mm)
479 {
480         mm->pgd = pgd_alloc(mm);
481         if (unlikely(!mm->pgd))
482                 return -ENOMEM;
483         return 0;
484 }
485
486 static inline void mm_free_pgd(struct mm_struct *mm)
487 {
488         pgd_free(mm, mm->pgd);
489 }
490 #else
491 #define dup_mmap(mm, oldmm)     (0)
492 #define mm_alloc_pgd(mm)        (0)
493 #define mm_free_pgd(mm)
494 #endif /* CONFIG_MMU */
495
496 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
497
498 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
499 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
500
501 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
502
503 static int __init coredump_filter_setup(char *s)
504 {
505         default_dump_filter =
506                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
507                 MMF_DUMP_FILTER_MASK;
508         return 1;
509 }
510
511 __setup("coredump_filter=", coredump_filter_setup);
512
513 #include <linux/init_task.h>
514
515 static void mm_init_aio(struct mm_struct *mm)
516 {
517 #ifdef CONFIG_AIO
518         spin_lock_init(&mm->ioctx_lock);
519         INIT_HLIST_HEAD(&mm->ioctx_list);
520 #endif
521 }
522
523 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
524 {
525         atomic_set(&mm->mm_users, 1);
526         atomic_set(&mm->mm_count, 1);
527         init_rwsem(&mm->mmap_sem);
528         INIT_LIST_HEAD(&mm->mmlist);
529         mm->flags = (current->mm) ?
530                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
531         mm->core_state = NULL;
532         mm->nr_ptes = 0;
533         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
534         spin_lock_init(&mm->page_table_lock);
535         mm->free_area_cache = TASK_UNMAPPED_BASE;
536         mm->cached_hole_size = ~0UL;
537         mm_init_aio(mm);
538         mm_init_owner(mm, p);
539
540         if (likely(!mm_alloc_pgd(mm))) {
541                 mm->def_flags = 0;
542                 mmu_notifier_mm_init(mm);
543                 return mm;
544         }
545
546         free_mm(mm);
547         return NULL;
548 }
549
550 static void check_mm(struct mm_struct *mm)
551 {
552         int i;
553
554         for (i = 0; i < NR_MM_COUNTERS; i++) {
555                 long x = atomic_long_read(&mm->rss_stat.count[i]);
556
557                 if (unlikely(x))
558                         printk(KERN_ALERT "BUG: Bad rss-counter state "
559                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
560         }
561
562 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
563         VM_BUG_ON(mm->pmd_huge_pte);
564 #endif
565 }
566
567 /*
568  * Allocate and initialize an mm_struct.
569  */
570 struct mm_struct *mm_alloc(void)
571 {
572         struct mm_struct *mm;
573
574         mm = allocate_mm();
575         if (!mm)
576                 return NULL;
577
578         memset(mm, 0, sizeof(*mm));
579         mm_init_cpumask(mm);
580         return mm_init(mm, current);
581 }
582
583 /*
584  * Called when the last reference to the mm
585  * is dropped: either by a lazy thread or by
586  * mmput. Free the page directory and the mm.
587  */
588 void __mmdrop(struct mm_struct *mm)
589 {
590         BUG_ON(mm == &init_mm);
591         mm_free_pgd(mm);
592         destroy_context(mm);
593         mmu_notifier_mm_destroy(mm);
594         check_mm(mm);
595         free_mm(mm);
596 }
597 EXPORT_SYMBOL_GPL(__mmdrop);
598
599 /*
600  * Decrement the use count and release all resources for an mm.
601  */
602 void mmput(struct mm_struct *mm)
603 {
604         might_sleep();
605
606         if (atomic_dec_and_test(&mm->mm_users)) {
607                 uprobe_clear_state(mm);
608                 exit_aio(mm);
609                 ksm_exit(mm);
610                 khugepaged_exit(mm); /* must run before exit_mmap */
611                 exit_mmap(mm);
612                 set_mm_exe_file(mm, NULL);
613                 if (!list_empty(&mm->mmlist)) {
614                         spin_lock(&mmlist_lock);
615                         list_del(&mm->mmlist);
616                         spin_unlock(&mmlist_lock);
617                 }
618                 if (mm->binfmt)
619                         module_put(mm->binfmt->module);
620                 mmdrop(mm);
621         }
622
623         /*
624          * Final rss-counter synchronization. After this point there must be
625          * no pagefaults into this mm from the current context.  Otherwise
626          * mm->rss_stat will be inconsistent.
627          */
628         if (mm)
629                 sync_mm_rss(mm);
630 }
631 EXPORT_SYMBOL_GPL(mmput);
632
633 /*
634  * We added or removed a vma mapping the executable. The vmas are only mapped
635  * during exec and are not mapped with the mmap system call.
636  * Callers must hold down_write() on the mm's mmap_sem for these
637  */
638 void added_exe_file_vma(struct mm_struct *mm)
639 {
640         mm->num_exe_file_vmas++;
641 }
642
643 void removed_exe_file_vma(struct mm_struct *mm)
644 {
645         mm->num_exe_file_vmas--;
646         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
647                 fput(mm->exe_file);
648                 mm->exe_file = NULL;
649         }
650
651 }
652
653 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
654 {
655         if (new_exe_file)
656                 get_file(new_exe_file);
657         if (mm->exe_file)
658                 fput(mm->exe_file);
659         mm->exe_file = new_exe_file;
660         mm->num_exe_file_vmas = 0;
661 }
662
663 struct file *get_mm_exe_file(struct mm_struct *mm)
664 {
665         struct file *exe_file;
666
667         /* We need mmap_sem to protect against races with removal of
668          * VM_EXECUTABLE vmas */
669         down_read(&mm->mmap_sem);
670         exe_file = mm->exe_file;
671         if (exe_file)
672                 get_file(exe_file);
673         up_read(&mm->mmap_sem);
674         return exe_file;
675 }
676
677 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
678 {
679         /* It's safe to write the exe_file pointer without exe_file_lock because
680          * this is called during fork when the task is not yet in /proc */
681         newmm->exe_file = get_mm_exe_file(oldmm);
682 }
683
684 /**
685  * get_task_mm - acquire a reference to the task's mm
686  *
687  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
688  * this kernel workthread has transiently adopted a user mm with use_mm,
689  * to do its AIO) is not set and if so returns a reference to it, after
690  * bumping up the use count.  User must release the mm via mmput()
691  * after use.  Typically used by /proc and ptrace.
692  */
693 struct mm_struct *get_task_mm(struct task_struct *task)
694 {
695         struct mm_struct *mm;
696
697         task_lock(task);
698         mm = task->mm;
699         if (mm) {
700                 if (task->flags & PF_KTHREAD)
701                         mm = NULL;
702                 else
703                         atomic_inc(&mm->mm_users);
704         }
705         task_unlock(task);
706         return mm;
707 }
708 EXPORT_SYMBOL_GPL(get_task_mm);
709
710 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
711 {
712         struct mm_struct *mm;
713         int err;
714
715         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
716         if (err)
717                 return ERR_PTR(err);
718
719         mm = get_task_mm(task);
720         if (mm && mm != current->mm &&
721                         !ptrace_may_access(task, mode)) {
722                 mmput(mm);
723                 mm = ERR_PTR(-EACCES);
724         }
725         mutex_unlock(&task->signal->cred_guard_mutex);
726
727         return mm;
728 }
729
730 static void complete_vfork_done(struct task_struct *tsk)
731 {
732         struct completion *vfork;
733
734         task_lock(tsk);
735         vfork = tsk->vfork_done;
736         if (likely(vfork)) {
737                 tsk->vfork_done = NULL;
738                 complete(vfork);
739         }
740         task_unlock(tsk);
741 }
742
743 static int wait_for_vfork_done(struct task_struct *child,
744                                 struct completion *vfork)
745 {
746         int killed;
747
748         freezer_do_not_count();
749         killed = wait_for_completion_killable(vfork);
750         freezer_count();
751
752         if (killed) {
753                 task_lock(child);
754                 child->vfork_done = NULL;
755                 task_unlock(child);
756         }
757
758         put_task_struct(child);
759         return killed;
760 }
761
762 /* Please note the differences between mmput and mm_release.
763  * mmput is called whenever we stop holding onto a mm_struct,
764  * error success whatever.
765  *
766  * mm_release is called after a mm_struct has been removed
767  * from the current process.
768  *
769  * This difference is important for error handling, when we
770  * only half set up a mm_struct for a new process and need to restore
771  * the old one.  Because we mmput the new mm_struct before
772  * restoring the old one. . .
773  * Eric Biederman 10 January 1998
774  */
775 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
776 {
777         /* Get rid of any futexes when releasing the mm */
778 #ifdef CONFIG_FUTEX
779         if (unlikely(tsk->robust_list)) {
780                 exit_robust_list(tsk);
781                 tsk->robust_list = NULL;
782         }
783 #ifdef CONFIG_COMPAT
784         if (unlikely(tsk->compat_robust_list)) {
785                 compat_exit_robust_list(tsk);
786                 tsk->compat_robust_list = NULL;
787         }
788 #endif
789         if (unlikely(!list_empty(&tsk->pi_state_list)))
790                 exit_pi_state_list(tsk);
791 #endif
792
793         uprobe_free_utask(tsk);
794
795         /* Get rid of any cached register state */
796         deactivate_mm(tsk, mm);
797
798         /*
799          * If we're exiting normally, clear a user-space tid field if
800          * requested.  We leave this alone when dying by signal, to leave
801          * the value intact in a core dump, and to save the unnecessary
802          * trouble, say, a killed vfork parent shouldn't touch this mm.
803          * Userland only wants this done for a sys_exit.
804          */
805         if (tsk->clear_child_tid) {
806                 if (!(tsk->flags & PF_SIGNALED) &&
807                     atomic_read(&mm->mm_users) > 1) {
808                         /*
809                          * We don't check the error code - if userspace has
810                          * not set up a proper pointer then tough luck.
811                          */
812                         put_user(0, tsk->clear_child_tid);
813                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
814                                         1, NULL, NULL, 0);
815                 }
816                 tsk->clear_child_tid = NULL;
817         }
818
819         /*
820          * All done, finally we can wake up parent and return this mm to him.
821          * Also kthread_stop() uses this completion for synchronization.
822          */
823         if (tsk->vfork_done)
824                 complete_vfork_done(tsk);
825 }
826
827 /*
828  * Allocate a new mm structure and copy contents from the
829  * mm structure of the passed in task structure.
830  */
831 struct mm_struct *dup_mm(struct task_struct *tsk)
832 {
833         struct mm_struct *mm, *oldmm = current->mm;
834         int err;
835
836         if (!oldmm)
837                 return NULL;
838
839         mm = allocate_mm();
840         if (!mm)
841                 goto fail_nomem;
842
843         memcpy(mm, oldmm, sizeof(*mm));
844         mm_init_cpumask(mm);
845
846 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
847         mm->pmd_huge_pte = NULL;
848 #endif
849         uprobe_reset_state(mm);
850
851         if (!mm_init(mm, tsk))
852                 goto fail_nomem;
853
854         if (init_new_context(tsk, mm))
855                 goto fail_nocontext;
856
857         dup_mm_exe_file(oldmm, mm);
858
859         err = dup_mmap(mm, oldmm);
860         if (err)
861                 goto free_pt;
862
863         mm->hiwater_rss = get_mm_rss(mm);
864         mm->hiwater_vm = mm->total_vm;
865
866         if (mm->binfmt && !try_module_get(mm->binfmt->module))
867                 goto free_pt;
868
869         return mm;
870
871 free_pt:
872         /* don't put binfmt in mmput, we haven't got module yet */
873         mm->binfmt = NULL;
874         mmput(mm);
875
876 fail_nomem:
877         return NULL;
878
879 fail_nocontext:
880         /*
881          * If init_new_context() failed, we cannot use mmput() to free the mm
882          * because it calls destroy_context()
883          */
884         mm_free_pgd(mm);
885         free_mm(mm);
886         return NULL;
887 }
888
889 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
890 {
891         struct mm_struct *mm, *oldmm;
892         int retval;
893
894         tsk->min_flt = tsk->maj_flt = 0;
895         tsk->nvcsw = tsk->nivcsw = 0;
896 #ifdef CONFIG_DETECT_HUNG_TASK
897         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
898 #endif
899
900         tsk->mm = NULL;
901         tsk->active_mm = NULL;
902
903         /*
904          * Are we cloning a kernel thread?
905          *
906          * We need to steal a active VM for that..
907          */
908         oldmm = current->mm;
909         if (!oldmm)
910                 return 0;
911
912         if (clone_flags & CLONE_VM) {
913                 atomic_inc(&oldmm->mm_users);
914                 mm = oldmm;
915                 goto good_mm;
916         }
917
918         retval = -ENOMEM;
919         mm = dup_mm(tsk);
920         if (!mm)
921                 goto fail_nomem;
922
923 good_mm:
924         tsk->mm = mm;
925         tsk->active_mm = mm;
926         return 0;
927
928 fail_nomem:
929         return retval;
930 }
931
932 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
933 {
934         struct fs_struct *fs = current->fs;
935         if (clone_flags & CLONE_FS) {
936                 /* tsk->fs is already what we want */
937                 spin_lock(&fs->lock);
938                 if (fs->in_exec) {
939                         spin_unlock(&fs->lock);
940                         return -EAGAIN;
941                 }
942                 fs->users++;
943                 spin_unlock(&fs->lock);
944                 return 0;
945         }
946         tsk->fs = copy_fs_struct(fs);
947         if (!tsk->fs)
948                 return -ENOMEM;
949         return 0;
950 }
951
952 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
953 {
954         struct files_struct *oldf, *newf;
955         int error = 0;
956
957         /*
958          * A background process may not have any files ...
959          */
960         oldf = current->files;
961         if (!oldf)
962                 goto out;
963
964         if (clone_flags & CLONE_FILES) {
965                 atomic_inc(&oldf->count);
966                 goto out;
967         }
968
969         newf = dup_fd(oldf, &error);
970         if (!newf)
971                 goto out;
972
973         tsk->files = newf;
974         error = 0;
975 out:
976         return error;
977 }
978
979 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
980 {
981 #ifdef CONFIG_BLOCK
982         struct io_context *ioc = current->io_context;
983         struct io_context *new_ioc;
984
985         if (!ioc)
986                 return 0;
987         /*
988          * Share io context with parent, if CLONE_IO is set
989          */
990         if (clone_flags & CLONE_IO) {
991                 ioc_task_link(ioc);
992                 tsk->io_context = ioc;
993         } else if (ioprio_valid(ioc->ioprio)) {
994                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
995                 if (unlikely(!new_ioc))
996                         return -ENOMEM;
997
998                 new_ioc->ioprio = ioc->ioprio;
999                 put_io_context(new_ioc);
1000         }
1001 #endif
1002         return 0;
1003 }
1004
1005 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1006 {
1007         struct sighand_struct *sig;
1008
1009         if (clone_flags & CLONE_SIGHAND) {
1010                 atomic_inc(&current->sighand->count);
1011                 return 0;
1012         }
1013         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1014         rcu_assign_pointer(tsk->sighand, sig);
1015         if (!sig)
1016                 return -ENOMEM;
1017         atomic_set(&sig->count, 1);
1018         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1019         return 0;
1020 }
1021
1022 void __cleanup_sighand(struct sighand_struct *sighand)
1023 {
1024         if (atomic_dec_and_test(&sighand->count)) {
1025                 signalfd_cleanup(sighand);
1026                 kmem_cache_free(sighand_cachep, sighand);
1027         }
1028 }
1029
1030
1031 /*
1032  * Initialize POSIX timer handling for a thread group.
1033  */
1034 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1035 {
1036         unsigned long cpu_limit;
1037
1038         /* Thread group counters. */
1039         thread_group_cputime_init(sig);
1040
1041         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1042         if (cpu_limit != RLIM_INFINITY) {
1043                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1044                 sig->cputimer.running = 1;
1045         }
1046
1047         /* The timer lists. */
1048         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1049         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1050         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1051 }
1052
1053 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1054 {
1055         struct signal_struct *sig;
1056
1057         if (clone_flags & CLONE_THREAD)
1058                 return 0;
1059
1060         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1061         tsk->signal = sig;
1062         if (!sig)
1063                 return -ENOMEM;
1064
1065         sig->nr_threads = 1;
1066         atomic_set(&sig->live, 1);
1067         atomic_set(&sig->sigcnt, 1);
1068         init_waitqueue_head(&sig->wait_chldexit);
1069         if (clone_flags & CLONE_NEWPID)
1070                 sig->flags |= SIGNAL_UNKILLABLE;
1071         sig->curr_target = tsk;
1072         init_sigpending(&sig->shared_pending);
1073         INIT_LIST_HEAD(&sig->posix_timers);
1074
1075         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1076         sig->real_timer.function = it_real_fn;
1077
1078         task_lock(current->group_leader);
1079         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1080         task_unlock(current->group_leader);
1081
1082         posix_cpu_timers_init_group(sig);
1083
1084         tty_audit_fork(sig);
1085         sched_autogroup_fork(sig);
1086
1087 #ifdef CONFIG_CGROUPS
1088         init_rwsem(&sig->group_rwsem);
1089 #endif
1090
1091         sig->oom_adj = current->signal->oom_adj;
1092         sig->oom_score_adj = current->signal->oom_score_adj;
1093         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1094
1095         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1096                                    current->signal->is_child_subreaper;
1097
1098         mutex_init(&sig->cred_guard_mutex);
1099
1100         return 0;
1101 }
1102
1103 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1104 {
1105         unsigned long new_flags = p->flags;
1106
1107         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1108         new_flags |= PF_FORKNOEXEC;
1109         p->flags = new_flags;
1110 }
1111
1112 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1113 {
1114         current->clear_child_tid = tidptr;
1115
1116         return task_pid_vnr(current);
1117 }
1118
1119 static void rt_mutex_init_task(struct task_struct *p)
1120 {
1121         raw_spin_lock_init(&p->pi_lock);
1122 #ifdef CONFIG_RT_MUTEXES
1123         plist_head_init(&p->pi_waiters);
1124         p->pi_blocked_on = NULL;
1125 #endif
1126 }
1127
1128 #ifdef CONFIG_MM_OWNER
1129 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1130 {
1131         mm->owner = p;
1132 }
1133 #endif /* CONFIG_MM_OWNER */
1134
1135 /*
1136  * Initialize POSIX timer handling for a single task.
1137  */
1138 static void posix_cpu_timers_init(struct task_struct *tsk)
1139 {
1140         tsk->cputime_expires.prof_exp = 0;
1141         tsk->cputime_expires.virt_exp = 0;
1142         tsk->cputime_expires.sched_exp = 0;
1143         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1144         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1145         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1146 }
1147
1148 /*
1149  * This creates a new process as a copy of the old one,
1150  * but does not actually start it yet.
1151  *
1152  * It copies the registers, and all the appropriate
1153  * parts of the process environment (as per the clone
1154  * flags). The actual kick-off is left to the caller.
1155  */
1156 static struct task_struct *copy_process(unsigned long clone_flags,
1157                                         unsigned long stack_start,
1158                                         struct pt_regs *regs,
1159                                         unsigned long stack_size,
1160                                         int __user *child_tidptr,
1161                                         struct pid *pid,
1162                                         int trace)
1163 {
1164         int retval;
1165         struct task_struct *p;
1166         int cgroup_callbacks_done = 0;
1167
1168         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1169                 return ERR_PTR(-EINVAL);
1170
1171         /*
1172          * Thread groups must share signals as well, and detached threads
1173          * can only be started up within the thread group.
1174          */
1175         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1176                 return ERR_PTR(-EINVAL);
1177
1178         /*
1179          * Shared signal handlers imply shared VM. By way of the above,
1180          * thread groups also imply shared VM. Blocking this case allows
1181          * for various simplifications in other code.
1182          */
1183         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1184                 return ERR_PTR(-EINVAL);
1185
1186         /*
1187          * Siblings of global init remain as zombies on exit since they are
1188          * not reaped by their parent (swapper). To solve this and to avoid
1189          * multi-rooted process trees, prevent global and container-inits
1190          * from creating siblings.
1191          */
1192         if ((clone_flags & CLONE_PARENT) &&
1193                                 current->signal->flags & SIGNAL_UNKILLABLE)
1194                 return ERR_PTR(-EINVAL);
1195
1196         retval = security_task_create(clone_flags);
1197         if (retval)
1198                 goto fork_out;
1199
1200         retval = -ENOMEM;
1201         p = dup_task_struct(current);
1202         if (!p)
1203                 goto fork_out;
1204
1205         ftrace_graph_init_task(p);
1206         get_seccomp_filter(p);
1207
1208         rt_mutex_init_task(p);
1209
1210 #ifdef CONFIG_PROVE_LOCKING
1211         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1212         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1213 #endif
1214         retval = -EAGAIN;
1215         if (atomic_read(&p->real_cred->user->processes) >=
1216                         task_rlimit(p, RLIMIT_NPROC)) {
1217                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1218                     p->real_cred->user != INIT_USER)
1219                         goto bad_fork_free;
1220         }
1221         current->flags &= ~PF_NPROC_EXCEEDED;
1222
1223         retval = copy_creds(p, clone_flags);
1224         if (retval < 0)
1225                 goto bad_fork_free;
1226
1227         /*
1228          * If multiple threads are within copy_process(), then this check
1229          * triggers too late. This doesn't hurt, the check is only there
1230          * to stop root fork bombs.
1231          */
1232         retval = -EAGAIN;
1233         if (nr_threads >= max_threads)
1234                 goto bad_fork_cleanup_count;
1235
1236         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1237                 goto bad_fork_cleanup_count;
1238
1239         p->did_exec = 0;
1240         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1241         copy_flags(clone_flags, p);
1242         INIT_LIST_HEAD(&p->children);
1243         INIT_LIST_HEAD(&p->sibling);
1244         rcu_copy_process(p);
1245         p->vfork_done = NULL;
1246         spin_lock_init(&p->alloc_lock);
1247
1248         init_sigpending(&p->pending);
1249
1250         p->utime = p->stime = p->gtime = 0;
1251         p->utimescaled = p->stimescaled = 0;
1252 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1253         p->prev_utime = p->prev_stime = 0;
1254 #endif
1255 #if defined(SPLIT_RSS_COUNTING)
1256         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1257 #endif
1258
1259         p->default_timer_slack_ns = current->timer_slack_ns;
1260
1261         task_io_accounting_init(&p->ioac);
1262         acct_clear_integrals(p);
1263
1264         posix_cpu_timers_init(p);
1265
1266         do_posix_clock_monotonic_gettime(&p->start_time);
1267         p->real_start_time = p->start_time;
1268         monotonic_to_bootbased(&p->real_start_time);
1269         p->io_context = NULL;
1270         p->audit_context = NULL;
1271         if (clone_flags & CLONE_THREAD)
1272                 threadgroup_change_begin(current);
1273         cgroup_fork(p);
1274 #ifdef CONFIG_NUMA
1275         p->mempolicy = mpol_dup(p->mempolicy);
1276         if (IS_ERR(p->mempolicy)) {
1277                 retval = PTR_ERR(p->mempolicy);
1278                 p->mempolicy = NULL;
1279                 goto bad_fork_cleanup_cgroup;
1280         }
1281         mpol_fix_fork_child_flag(p);
1282 #endif
1283 #ifdef CONFIG_CPUSETS
1284         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1285         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1286         seqcount_init(&p->mems_allowed_seq);
1287 #endif
1288 #ifdef CONFIG_TRACE_IRQFLAGS
1289         p->irq_events = 0;
1290 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1291         p->hardirqs_enabled = 1;
1292 #else
1293         p->hardirqs_enabled = 0;
1294 #endif
1295         p->hardirq_enable_ip = 0;
1296         p->hardirq_enable_event = 0;
1297         p->hardirq_disable_ip = _THIS_IP_;
1298         p->hardirq_disable_event = 0;
1299         p->softirqs_enabled = 1;
1300         p->softirq_enable_ip = _THIS_IP_;
1301         p->softirq_enable_event = 0;
1302         p->softirq_disable_ip = 0;
1303         p->softirq_disable_event = 0;
1304         p->hardirq_context = 0;
1305         p->softirq_context = 0;
1306 #endif
1307 #ifdef CONFIG_LOCKDEP
1308         p->lockdep_depth = 0; /* no locks held yet */
1309         p->curr_chain_key = 0;
1310         p->lockdep_recursion = 0;
1311 #endif
1312
1313 #ifdef CONFIG_DEBUG_MUTEXES
1314         p->blocked_on = NULL; /* not blocked yet */
1315 #endif
1316 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1317         p->memcg_batch.do_batch = 0;
1318         p->memcg_batch.memcg = NULL;
1319 #endif
1320
1321         /* Perform scheduler related setup. Assign this task to a CPU. */
1322         sched_fork(p);
1323
1324         retval = perf_event_init_task(p);
1325         if (retval)
1326                 goto bad_fork_cleanup_policy;
1327         retval = audit_alloc(p);
1328         if (retval)
1329                 goto bad_fork_cleanup_policy;
1330         /* copy all the process information */
1331         retval = copy_semundo(clone_flags, p);
1332         if (retval)
1333                 goto bad_fork_cleanup_audit;
1334         retval = copy_files(clone_flags, p);
1335         if (retval)
1336                 goto bad_fork_cleanup_semundo;
1337         retval = copy_fs(clone_flags, p);
1338         if (retval)
1339                 goto bad_fork_cleanup_files;
1340         retval = copy_sighand(clone_flags, p);
1341         if (retval)
1342                 goto bad_fork_cleanup_fs;
1343         retval = copy_signal(clone_flags, p);
1344         if (retval)
1345                 goto bad_fork_cleanup_sighand;
1346         retval = copy_mm(clone_flags, p);
1347         if (retval)
1348                 goto bad_fork_cleanup_signal;
1349         retval = copy_namespaces(clone_flags, p);
1350         if (retval)
1351                 goto bad_fork_cleanup_mm;
1352         retval = copy_io(clone_flags, p);
1353         if (retval)
1354                 goto bad_fork_cleanup_namespaces;
1355         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1356         if (retval)
1357                 goto bad_fork_cleanup_io;
1358
1359         if (pid != &init_struct_pid) {
1360                 retval = -ENOMEM;
1361                 pid = alloc_pid(p->nsproxy->pid_ns);
1362                 if (!pid)
1363                         goto bad_fork_cleanup_io;
1364         }
1365
1366         p->pid = pid_nr(pid);
1367         p->tgid = p->pid;
1368         if (clone_flags & CLONE_THREAD)
1369                 p->tgid = current->tgid;
1370
1371         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1372         /*
1373          * Clear TID on mm_release()?
1374          */
1375         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1376 #ifdef CONFIG_BLOCK
1377         p->plug = NULL;
1378 #endif
1379 #ifdef CONFIG_FUTEX
1380         p->robust_list = NULL;
1381 #ifdef CONFIG_COMPAT
1382         p->compat_robust_list = NULL;
1383 #endif
1384         INIT_LIST_HEAD(&p->pi_state_list);
1385         p->pi_state_cache = NULL;
1386 #endif
1387         uprobe_copy_process(p);
1388         /*
1389          * sigaltstack should be cleared when sharing the same VM
1390          */
1391         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1392                 p->sas_ss_sp = p->sas_ss_size = 0;
1393
1394         /*
1395          * Syscall tracing and stepping should be turned off in the
1396          * child regardless of CLONE_PTRACE.
1397          */
1398         user_disable_single_step(p);
1399         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1400 #ifdef TIF_SYSCALL_EMU
1401         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1402 #endif
1403         clear_all_latency_tracing(p);
1404
1405         /* ok, now we should be set up.. */
1406         if (clone_flags & CLONE_THREAD)
1407                 p->exit_signal = -1;
1408         else if (clone_flags & CLONE_PARENT)
1409                 p->exit_signal = current->group_leader->exit_signal;
1410         else
1411                 p->exit_signal = (clone_flags & CSIGNAL);
1412
1413         p->pdeath_signal = 0;
1414         p->exit_state = 0;
1415
1416         p->nr_dirtied = 0;
1417         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1418         p->dirty_paused_when = 0;
1419
1420         /*
1421          * Ok, make it visible to the rest of the system.
1422          * We dont wake it up yet.
1423          */
1424         p->group_leader = p;
1425         INIT_LIST_HEAD(&p->thread_group);
1426         INIT_HLIST_HEAD(&p->task_works);
1427
1428         /* Now that the task is set up, run cgroup callbacks if
1429          * necessary. We need to run them before the task is visible
1430          * on the tasklist. */
1431         cgroup_fork_callbacks(p);
1432         cgroup_callbacks_done = 1;
1433
1434         /* Need tasklist lock for parent etc handling! */
1435         write_lock_irq(&tasklist_lock);
1436
1437         /* CLONE_PARENT re-uses the old parent */
1438         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1439                 p->real_parent = current->real_parent;
1440                 p->parent_exec_id = current->parent_exec_id;
1441         } else {
1442                 p->real_parent = current;
1443                 p->parent_exec_id = current->self_exec_id;
1444         }
1445
1446         spin_lock(&current->sighand->siglock);
1447
1448         /*
1449          * Process group and session signals need to be delivered to just the
1450          * parent before the fork or both the parent and the child after the
1451          * fork. Restart if a signal comes in before we add the new process to
1452          * it's process group.
1453          * A fatal signal pending means that current will exit, so the new
1454          * thread can't slip out of an OOM kill (or normal SIGKILL).
1455         */
1456         recalc_sigpending();
1457         if (signal_pending(current)) {
1458                 spin_unlock(&current->sighand->siglock);
1459                 write_unlock_irq(&tasklist_lock);
1460                 retval = -ERESTARTNOINTR;
1461                 goto bad_fork_free_pid;
1462         }
1463
1464         if (clone_flags & CLONE_THREAD) {
1465                 current->signal->nr_threads++;
1466                 atomic_inc(&current->signal->live);
1467                 atomic_inc(&current->signal->sigcnt);
1468                 p->group_leader = current->group_leader;
1469                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1470         }
1471
1472         if (likely(p->pid)) {
1473                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1474
1475                 if (thread_group_leader(p)) {
1476                         if (is_child_reaper(pid))
1477                                 p->nsproxy->pid_ns->child_reaper = p;
1478
1479                         p->signal->leader_pid = pid;
1480                         p->signal->tty = tty_kref_get(current->signal->tty);
1481                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1482                         attach_pid(p, PIDTYPE_SID, task_session(current));
1483                         list_add_tail(&p->sibling, &p->real_parent->children);
1484                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1485                         __this_cpu_inc(process_counts);
1486                 }
1487                 attach_pid(p, PIDTYPE_PID, pid);
1488                 nr_threads++;
1489         }
1490
1491         total_forks++;
1492         spin_unlock(&current->sighand->siglock);
1493         write_unlock_irq(&tasklist_lock);
1494         proc_fork_connector(p);
1495         cgroup_post_fork(p);
1496         if (clone_flags & CLONE_THREAD)
1497                 threadgroup_change_end(current);
1498         perf_event_fork(p);
1499
1500         trace_task_newtask(p, clone_flags);
1501
1502         return p;
1503
1504 bad_fork_free_pid:
1505         if (pid != &init_struct_pid)
1506                 free_pid(pid);
1507 bad_fork_cleanup_io:
1508         if (p->io_context)
1509                 exit_io_context(p);
1510 bad_fork_cleanup_namespaces:
1511         if (unlikely(clone_flags & CLONE_NEWPID))
1512                 pid_ns_release_proc(p->nsproxy->pid_ns);
1513         exit_task_namespaces(p);
1514 bad_fork_cleanup_mm:
1515         if (p->mm)
1516                 mmput(p->mm);
1517 bad_fork_cleanup_signal:
1518         if (!(clone_flags & CLONE_THREAD))
1519                 free_signal_struct(p->signal);
1520 bad_fork_cleanup_sighand:
1521         __cleanup_sighand(p->sighand);
1522 bad_fork_cleanup_fs:
1523         exit_fs(p); /* blocking */
1524 bad_fork_cleanup_files:
1525         exit_files(p); /* blocking */
1526 bad_fork_cleanup_semundo:
1527         exit_sem(p);
1528 bad_fork_cleanup_audit:
1529         audit_free(p);
1530 bad_fork_cleanup_policy:
1531         perf_event_free_task(p);
1532 #ifdef CONFIG_NUMA
1533         mpol_put(p->mempolicy);
1534 bad_fork_cleanup_cgroup:
1535 #endif
1536         if (clone_flags & CLONE_THREAD)
1537                 threadgroup_change_end(current);
1538         cgroup_exit(p, cgroup_callbacks_done);
1539         delayacct_tsk_free(p);
1540         module_put(task_thread_info(p)->exec_domain->module);
1541 bad_fork_cleanup_count:
1542         atomic_dec(&p->cred->user->processes);
1543         exit_creds(p);
1544 bad_fork_free:
1545         free_task(p);
1546 fork_out:
1547         return ERR_PTR(retval);
1548 }
1549
1550 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1551 {
1552         memset(regs, 0, sizeof(struct pt_regs));
1553         return regs;
1554 }
1555
1556 static inline void init_idle_pids(struct pid_link *links)
1557 {
1558         enum pid_type type;
1559
1560         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1561                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1562                 links[type].pid = &init_struct_pid;
1563         }
1564 }
1565
1566 struct task_struct * __cpuinit fork_idle(int cpu)
1567 {
1568         struct task_struct *task;
1569         struct pt_regs regs;
1570
1571         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1572                             &init_struct_pid, 0);
1573         if (!IS_ERR(task)) {
1574                 init_idle_pids(task->pids);
1575                 init_idle(task, cpu);
1576         }
1577
1578         return task;
1579 }
1580
1581 /*
1582  *  Ok, this is the main fork-routine.
1583  *
1584  * It copies the process, and if successful kick-starts
1585  * it and waits for it to finish using the VM if required.
1586  */
1587 long do_fork(unsigned long clone_flags,
1588               unsigned long stack_start,
1589               struct pt_regs *regs,
1590               unsigned long stack_size,
1591               int __user *parent_tidptr,
1592               int __user *child_tidptr)
1593 {
1594         struct task_struct *p;
1595         int trace = 0;
1596         long nr;
1597
1598         /*
1599          * Do some preliminary argument and permissions checking before we
1600          * actually start allocating stuff
1601          */
1602         if (clone_flags & CLONE_NEWUSER) {
1603                 if (clone_flags & CLONE_THREAD)
1604                         return -EINVAL;
1605                 /* hopefully this check will go away when userns support is
1606                  * complete
1607                  */
1608                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1609                                 !capable(CAP_SETGID))
1610                         return -EPERM;
1611         }
1612
1613         /*
1614          * Determine whether and which event to report to ptracer.  When
1615          * called from kernel_thread or CLONE_UNTRACED is explicitly
1616          * requested, no event is reported; otherwise, report if the event
1617          * for the type of forking is enabled.
1618          */
1619         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1620                 if (clone_flags & CLONE_VFORK)
1621                         trace = PTRACE_EVENT_VFORK;
1622                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1623                         trace = PTRACE_EVENT_CLONE;
1624                 else
1625                         trace = PTRACE_EVENT_FORK;
1626
1627                 if (likely(!ptrace_event_enabled(current, trace)))
1628                         trace = 0;
1629         }
1630
1631         p = copy_process(clone_flags, stack_start, regs, stack_size,
1632                          child_tidptr, NULL, trace);
1633         /*
1634          * Do this prior waking up the new thread - the thread pointer
1635          * might get invalid after that point, if the thread exits quickly.
1636          */
1637         if (!IS_ERR(p)) {
1638                 struct completion vfork;
1639
1640                 trace_sched_process_fork(current, p);
1641
1642                 nr = task_pid_vnr(p);
1643
1644                 if (clone_flags & CLONE_PARENT_SETTID)
1645                         put_user(nr, parent_tidptr);
1646
1647                 if (clone_flags & CLONE_VFORK) {
1648                         p->vfork_done = &vfork;
1649                         init_completion(&vfork);
1650                         get_task_struct(p);
1651                 }
1652
1653                 wake_up_new_task(p);
1654
1655                 /* forking complete and child started to run, tell ptracer */
1656                 if (unlikely(trace))
1657                         ptrace_event(trace, nr);
1658
1659                 if (clone_flags & CLONE_VFORK) {
1660                         if (!wait_for_vfork_done(p, &vfork))
1661                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1662                 }
1663         } else {
1664                 nr = PTR_ERR(p);
1665         }
1666         return nr;
1667 }
1668
1669 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1670 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1671 #endif
1672
1673 static void sighand_ctor(void *data)
1674 {
1675         struct sighand_struct *sighand = data;
1676
1677         spin_lock_init(&sighand->siglock);
1678         init_waitqueue_head(&sighand->signalfd_wqh);
1679 }
1680
1681 void __init proc_caches_init(void)
1682 {
1683         sighand_cachep = kmem_cache_create("sighand_cache",
1684                         sizeof(struct sighand_struct), 0,
1685                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1686                         SLAB_NOTRACK, sighand_ctor);
1687         signal_cachep = kmem_cache_create("signal_cache",
1688                         sizeof(struct signal_struct), 0,
1689                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1690         files_cachep = kmem_cache_create("files_cache",
1691                         sizeof(struct files_struct), 0,
1692                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1693         fs_cachep = kmem_cache_create("fs_cache",
1694                         sizeof(struct fs_struct), 0,
1695                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1696         /*
1697          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1698          * whole struct cpumask for the OFFSTACK case. We could change
1699          * this to *only* allocate as much of it as required by the
1700          * maximum number of CPU's we can ever have.  The cpumask_allocation
1701          * is at the end of the structure, exactly for that reason.
1702          */
1703         mm_cachep = kmem_cache_create("mm_struct",
1704                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1705                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1706         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1707         mmap_init();
1708         nsproxy_cache_init();
1709 }
1710
1711 /*
1712  * Check constraints on flags passed to the unshare system call.
1713  */
1714 static int check_unshare_flags(unsigned long unshare_flags)
1715 {
1716         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1717                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1718                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1719                 return -EINVAL;
1720         /*
1721          * Not implemented, but pretend it works if there is nothing to
1722          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1723          * needs to unshare vm.
1724          */
1725         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1726                 /* FIXME: get_task_mm() increments ->mm_users */
1727                 if (atomic_read(&current->mm->mm_users) > 1)
1728                         return -EINVAL;
1729         }
1730
1731         return 0;
1732 }
1733
1734 /*
1735  * Unshare the filesystem structure if it is being shared
1736  */
1737 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1738 {
1739         struct fs_struct *fs = current->fs;
1740
1741         if (!(unshare_flags & CLONE_FS) || !fs)
1742                 return 0;
1743
1744         /* don't need lock here; in the worst case we'll do useless copy */
1745         if (fs->users == 1)
1746                 return 0;
1747
1748         *new_fsp = copy_fs_struct(fs);
1749         if (!*new_fsp)
1750                 return -ENOMEM;
1751
1752         return 0;
1753 }
1754
1755 /*
1756  * Unshare file descriptor table if it is being shared
1757  */
1758 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1759 {
1760         struct files_struct *fd = current->files;
1761         int error = 0;
1762
1763         if ((unshare_flags & CLONE_FILES) &&
1764             (fd && atomic_read(&fd->count) > 1)) {
1765                 *new_fdp = dup_fd(fd, &error);
1766                 if (!*new_fdp)
1767                         return error;
1768         }
1769
1770         return 0;
1771 }
1772
1773 /*
1774  * unshare allows a process to 'unshare' part of the process
1775  * context which was originally shared using clone.  copy_*
1776  * functions used by do_fork() cannot be used here directly
1777  * because they modify an inactive task_struct that is being
1778  * constructed. Here we are modifying the current, active,
1779  * task_struct.
1780  */
1781 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1782 {
1783         struct fs_struct *fs, *new_fs = NULL;
1784         struct files_struct *fd, *new_fd = NULL;
1785         struct nsproxy *new_nsproxy = NULL;
1786         int do_sysvsem = 0;
1787         int err;
1788
1789         err = check_unshare_flags(unshare_flags);
1790         if (err)
1791                 goto bad_unshare_out;
1792
1793         /*
1794          * If unsharing namespace, must also unshare filesystem information.
1795          */
1796         if (unshare_flags & CLONE_NEWNS)
1797                 unshare_flags |= CLONE_FS;
1798         /*
1799          * CLONE_NEWIPC must also detach from the undolist: after switching
1800          * to a new ipc namespace, the semaphore arrays from the old
1801          * namespace are unreachable.
1802          */
1803         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1804                 do_sysvsem = 1;
1805         err = unshare_fs(unshare_flags, &new_fs);
1806         if (err)
1807                 goto bad_unshare_out;
1808         err = unshare_fd(unshare_flags, &new_fd);
1809         if (err)
1810                 goto bad_unshare_cleanup_fs;
1811         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1812         if (err)
1813                 goto bad_unshare_cleanup_fd;
1814
1815         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1816                 if (do_sysvsem) {
1817                         /*
1818                          * CLONE_SYSVSEM is equivalent to sys_exit().
1819                          */
1820                         exit_sem(current);
1821                 }
1822
1823                 if (new_nsproxy) {
1824                         switch_task_namespaces(current, new_nsproxy);
1825                         new_nsproxy = NULL;
1826                 }
1827
1828                 task_lock(current);
1829
1830                 if (new_fs) {
1831                         fs = current->fs;
1832                         spin_lock(&fs->lock);
1833                         current->fs = new_fs;
1834                         if (--fs->users)
1835                                 new_fs = NULL;
1836                         else
1837                                 new_fs = fs;
1838                         spin_unlock(&fs->lock);
1839                 }
1840
1841                 if (new_fd) {
1842                         fd = current->files;
1843                         current->files = new_fd;
1844                         new_fd = fd;
1845                 }
1846
1847                 task_unlock(current);
1848         }
1849
1850         if (new_nsproxy)
1851                 put_nsproxy(new_nsproxy);
1852
1853 bad_unshare_cleanup_fd:
1854         if (new_fd)
1855                 put_files_struct(new_fd);
1856
1857 bad_unshare_cleanup_fs:
1858         if (new_fs)
1859                 free_fs_struct(new_fs);
1860
1861 bad_unshare_out:
1862         return err;
1863 }
1864
1865 /*
1866  *      Helper to unshare the files of the current task.
1867  *      We don't want to expose copy_files internals to
1868  *      the exec layer of the kernel.
1869  */
1870
1871 int unshare_files(struct files_struct **displaced)
1872 {
1873         struct task_struct *task = current;
1874         struct files_struct *copy = NULL;
1875         int error;
1876
1877         error = unshare_fd(CLONE_FILES, &copy);
1878         if (error || !copy) {
1879                 *displaced = NULL;
1880                 return error;
1881         }
1882         *displaced = task->files;
1883         task_lock(task);
1884         task->files = copy;
1885         task_unlock(task);
1886         return 0;
1887 }