Merge branch 'linus' into x86/urgent, to merge dependent patch
[linux-2.6-block.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/mm.h>
32 #include <linux/vmacache.h>
33 #include <linux/nsproxy.h>
34 #include <linux/capability.h>
35 #include <linux/cpu.h>
36 #include <linux/cgroup.h>
37 #include <linux/security.h>
38 #include <linux/hugetlb.h>
39 #include <linux/seccomp.h>
40 #include <linux/swap.h>
41 #include <linux/syscalls.h>
42 #include <linux/jiffies.h>
43 #include <linux/futex.h>
44 #include <linux/compat.h>
45 #include <linux/kthread.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/rcupdate.h>
48 #include <linux/ptrace.h>
49 #include <linux/mount.h>
50 #include <linux/audit.h>
51 #include <linux/memcontrol.h>
52 #include <linux/ftrace.h>
53 #include <linux/proc_fs.h>
54 #include <linux/profile.h>
55 #include <linux/rmap.h>
56 #include <linux/ksm.h>
57 #include <linux/acct.h>
58 #include <linux/tsacct_kern.h>
59 #include <linux/cn_proc.h>
60 #include <linux/freezer.h>
61 #include <linux/delayacct.h>
62 #include <linux/taskstats_kern.h>
63 #include <linux/random.h>
64 #include <linux/tty.h>
65 #include <linux/blkdev.h>
66 #include <linux/fs_struct.h>
67 #include <linux/magic.h>
68 #include <linux/perf_event.h>
69 #include <linux/posix-timers.h>
70 #include <linux/user-return-notifier.h>
71 #include <linux/oom.h>
72 #include <linux/khugepaged.h>
73 #include <linux/signalfd.h>
74 #include <linux/uprobes.h>
75 #include <linux/aio.h>
76 #include <linux/compiler.h>
77
78 #include <asm/pgtable.h>
79 #include <asm/pgalloc.h>
80 #include <asm/uaccess.h>
81 #include <asm/mmu_context.h>
82 #include <asm/cacheflush.h>
83 #include <asm/tlbflush.h>
84
85 #include <trace/events/sched.h>
86
87 #define CREATE_TRACE_POINTS
88 #include <trace/events/task.h>
89
90 /*
91  * Protected counters by write_lock_irq(&tasklist_lock)
92  */
93 unsigned long total_forks;      /* Handle normal Linux uptimes. */
94 int nr_threads;                 /* The idle threads do not count.. */
95
96 int max_threads;                /* tunable limit on nr_threads */
97
98 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
99
100 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
101
102 #ifdef CONFIG_PROVE_RCU
103 int lockdep_tasklist_lock_is_held(void)
104 {
105         return lockdep_is_held(&tasklist_lock);
106 }
107 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
108 #endif /* #ifdef CONFIG_PROVE_RCU */
109
110 int nr_processes(void)
111 {
112         int cpu;
113         int total = 0;
114
115         for_each_possible_cpu(cpu)
116                 total += per_cpu(process_counts, cpu);
117
118         return total;
119 }
120
121 void __weak arch_release_task_struct(struct task_struct *tsk)
122 {
123 }
124
125 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
126 static struct kmem_cache *task_struct_cachep;
127
128 static inline struct task_struct *alloc_task_struct_node(int node)
129 {
130         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
131 }
132
133 static inline void free_task_struct(struct task_struct *tsk)
134 {
135         kmem_cache_free(task_struct_cachep, tsk);
136 }
137 #endif
138
139 void __weak arch_release_thread_info(struct thread_info *ti)
140 {
141 }
142
143 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
144
145 /*
146  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
147  * kmemcache based allocator.
148  */
149 # if THREAD_SIZE >= PAGE_SIZE
150 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
151                                                   int node)
152 {
153         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
154                                                   THREAD_SIZE_ORDER);
155
156         return page ? page_address(page) : NULL;
157 }
158
159 static inline void free_thread_info(struct thread_info *ti)
160 {
161         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
162 }
163 # else
164 static struct kmem_cache *thread_info_cache;
165
166 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
167                                                   int node)
168 {
169         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
170 }
171
172 static void free_thread_info(struct thread_info *ti)
173 {
174         kmem_cache_free(thread_info_cache, ti);
175 }
176
177 void thread_info_cache_init(void)
178 {
179         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
180                                               THREAD_SIZE, 0, NULL);
181         BUG_ON(thread_info_cache == NULL);
182 }
183 # endif
184 #endif
185
186 /* SLAB cache for signal_struct structures (tsk->signal) */
187 static struct kmem_cache *signal_cachep;
188
189 /* SLAB cache for sighand_struct structures (tsk->sighand) */
190 struct kmem_cache *sighand_cachep;
191
192 /* SLAB cache for files_struct structures (tsk->files) */
193 struct kmem_cache *files_cachep;
194
195 /* SLAB cache for fs_struct structures (tsk->fs) */
196 struct kmem_cache *fs_cachep;
197
198 /* SLAB cache for vm_area_struct structures */
199 struct kmem_cache *vm_area_cachep;
200
201 /* SLAB cache for mm_struct structures (tsk->mm) */
202 static struct kmem_cache *mm_cachep;
203
204 static void account_kernel_stack(struct thread_info *ti, int account)
205 {
206         struct zone *zone = page_zone(virt_to_page(ti));
207
208         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
209 }
210
211 void free_task(struct task_struct *tsk)
212 {
213         account_kernel_stack(tsk->stack, -1);
214         arch_release_thread_info(tsk->stack);
215         free_thread_info(tsk->stack);
216         rt_mutex_debug_task_free(tsk);
217         ftrace_graph_exit_task(tsk);
218         put_seccomp_filter(tsk);
219         arch_release_task_struct(tsk);
220         free_task_struct(tsk);
221 }
222 EXPORT_SYMBOL(free_task);
223
224 static inline void free_signal_struct(struct signal_struct *sig)
225 {
226         taskstats_tgid_free(sig);
227         sched_autogroup_exit(sig);
228         kmem_cache_free(signal_cachep, sig);
229 }
230
231 static inline void put_signal_struct(struct signal_struct *sig)
232 {
233         if (atomic_dec_and_test(&sig->sigcnt))
234                 free_signal_struct(sig);
235 }
236
237 void __put_task_struct(struct task_struct *tsk)
238 {
239         WARN_ON(!tsk->exit_state);
240         WARN_ON(atomic_read(&tsk->usage));
241         WARN_ON(tsk == current);
242
243         task_numa_free(tsk);
244         security_task_free(tsk);
245         exit_creds(tsk);
246         delayacct_tsk_free(tsk);
247         put_signal_struct(tsk->signal);
248
249         if (!profile_handoff_task(tsk))
250                 free_task(tsk);
251 }
252 EXPORT_SYMBOL_GPL(__put_task_struct);
253
254 void __init __weak arch_task_cache_init(void) { }
255
256 void __init fork_init(unsigned long mempages)
257 {
258 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
259 #ifndef ARCH_MIN_TASKALIGN
260 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
261 #endif
262         /* create a slab on which task_structs can be allocated */
263         task_struct_cachep =
264                 kmem_cache_create("task_struct", sizeof(struct task_struct),
265                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
266 #endif
267
268         /* do the arch specific task caches init */
269         arch_task_cache_init();
270
271         /*
272          * The default maximum number of threads is set to a safe
273          * value: the thread structures can take up at most half
274          * of memory.
275          */
276         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
277
278         /*
279          * we need to allow at least 20 threads to boot a system
280          */
281         if (max_threads < 20)
282                 max_threads = 20;
283
284         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
285         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
286         init_task.signal->rlim[RLIMIT_SIGPENDING] =
287                 init_task.signal->rlim[RLIMIT_NPROC];
288 }
289
290 int __weak arch_dup_task_struct(struct task_struct *dst,
291                                                struct task_struct *src)
292 {
293         *dst = *src;
294         return 0;
295 }
296
297 void set_task_stack_end_magic(struct task_struct *tsk)
298 {
299         unsigned long *stackend;
300
301         stackend = end_of_stack(tsk);
302         *stackend = STACK_END_MAGIC;    /* for overflow detection */
303 }
304
305 static struct task_struct *dup_task_struct(struct task_struct *orig)
306 {
307         struct task_struct *tsk;
308         struct thread_info *ti;
309         int node = tsk_fork_get_node(orig);
310         int err;
311
312         tsk = alloc_task_struct_node(node);
313         if (!tsk)
314                 return NULL;
315
316         ti = alloc_thread_info_node(tsk, node);
317         if (!ti)
318                 goto free_tsk;
319
320         err = arch_dup_task_struct(tsk, orig);
321         if (err)
322                 goto free_ti;
323
324         tsk->stack = ti;
325 #ifdef CONFIG_SECCOMP
326         /*
327          * We must handle setting up seccomp filters once we're under
328          * the sighand lock in case orig has changed between now and
329          * then. Until then, filter must be NULL to avoid messing up
330          * the usage counts on the error path calling free_task.
331          */
332         tsk->seccomp.filter = NULL;
333 #endif
334
335         setup_thread_stack(tsk, orig);
336         clear_user_return_notifier(tsk);
337         clear_tsk_need_resched(tsk);
338         set_task_stack_end_magic(tsk);
339
340 #ifdef CONFIG_CC_STACKPROTECTOR
341         tsk->stack_canary = get_random_int();
342 #endif
343
344         /*
345          * One for us, one for whoever does the "release_task()" (usually
346          * parent)
347          */
348         atomic_set(&tsk->usage, 2);
349 #ifdef CONFIG_BLK_DEV_IO_TRACE
350         tsk->btrace_seq = 0;
351 #endif
352         tsk->splice_pipe = NULL;
353         tsk->task_frag.page = NULL;
354
355         account_kernel_stack(ti, 1);
356
357         return tsk;
358
359 free_ti:
360         free_thread_info(ti);
361 free_tsk:
362         free_task_struct(tsk);
363         return NULL;
364 }
365
366 #ifdef CONFIG_MMU
367 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
368 {
369         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
370         struct rb_node **rb_link, *rb_parent;
371         int retval;
372         unsigned long charge;
373
374         uprobe_start_dup_mmap();
375         down_write(&oldmm->mmap_sem);
376         flush_cache_dup_mm(oldmm);
377         uprobe_dup_mmap(oldmm, mm);
378         /*
379          * Not linked in yet - no deadlock potential:
380          */
381         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
382
383         mm->total_vm = oldmm->total_vm;
384         mm->shared_vm = oldmm->shared_vm;
385         mm->exec_vm = oldmm->exec_vm;
386         mm->stack_vm = oldmm->stack_vm;
387
388         rb_link = &mm->mm_rb.rb_node;
389         rb_parent = NULL;
390         pprev = &mm->mmap;
391         retval = ksm_fork(mm, oldmm);
392         if (retval)
393                 goto out;
394         retval = khugepaged_fork(mm, oldmm);
395         if (retval)
396                 goto out;
397
398         prev = NULL;
399         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
400                 struct file *file;
401
402                 if (mpnt->vm_flags & VM_DONTCOPY) {
403                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
404                                                         -vma_pages(mpnt));
405                         continue;
406                 }
407                 charge = 0;
408                 if (mpnt->vm_flags & VM_ACCOUNT) {
409                         unsigned long len = vma_pages(mpnt);
410
411                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
412                                 goto fail_nomem;
413                         charge = len;
414                 }
415                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
416                 if (!tmp)
417                         goto fail_nomem;
418                 *tmp = *mpnt;
419                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
420                 retval = vma_dup_policy(mpnt, tmp);
421                 if (retval)
422                         goto fail_nomem_policy;
423                 tmp->vm_mm = mm;
424                 if (anon_vma_fork(tmp, mpnt))
425                         goto fail_nomem_anon_vma_fork;
426                 tmp->vm_flags &= ~VM_LOCKED;
427                 tmp->vm_next = tmp->vm_prev = NULL;
428                 file = tmp->vm_file;
429                 if (file) {
430                         struct inode *inode = file_inode(file);
431                         struct address_space *mapping = file->f_mapping;
432
433                         get_file(file);
434                         if (tmp->vm_flags & VM_DENYWRITE)
435                                 atomic_dec(&inode->i_writecount);
436                         i_mmap_lock_write(mapping);
437                         if (tmp->vm_flags & VM_SHARED)
438                                 atomic_inc(&mapping->i_mmap_writable);
439                         flush_dcache_mmap_lock(mapping);
440                         /* insert tmp into the share list, just after mpnt */
441                         vma_interval_tree_insert_after(tmp, mpnt,
442                                         &mapping->i_mmap);
443                         flush_dcache_mmap_unlock(mapping);
444                         i_mmap_unlock_write(mapping);
445                 }
446
447                 /*
448                  * Clear hugetlb-related page reserves for children. This only
449                  * affects MAP_PRIVATE mappings. Faults generated by the child
450                  * are not guaranteed to succeed, even if read-only
451                  */
452                 if (is_vm_hugetlb_page(tmp))
453                         reset_vma_resv_huge_pages(tmp);
454
455                 /*
456                  * Link in the new vma and copy the page table entries.
457                  */
458                 *pprev = tmp;
459                 pprev = &tmp->vm_next;
460                 tmp->vm_prev = prev;
461                 prev = tmp;
462
463                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
464                 rb_link = &tmp->vm_rb.rb_right;
465                 rb_parent = &tmp->vm_rb;
466
467                 mm->map_count++;
468                 retval = copy_page_range(mm, oldmm, mpnt);
469
470                 if (tmp->vm_ops && tmp->vm_ops->open)
471                         tmp->vm_ops->open(tmp);
472
473                 if (retval)
474                         goto out;
475         }
476         /* a new mm has just been created */
477         arch_dup_mmap(oldmm, mm);
478         retval = 0;
479 out:
480         up_write(&mm->mmap_sem);
481         flush_tlb_mm(oldmm);
482         up_write(&oldmm->mmap_sem);
483         uprobe_end_dup_mmap();
484         return retval;
485 fail_nomem_anon_vma_fork:
486         mpol_put(vma_policy(tmp));
487 fail_nomem_policy:
488         kmem_cache_free(vm_area_cachep, tmp);
489 fail_nomem:
490         retval = -ENOMEM;
491         vm_unacct_memory(charge);
492         goto out;
493 }
494
495 static inline int mm_alloc_pgd(struct mm_struct *mm)
496 {
497         mm->pgd = pgd_alloc(mm);
498         if (unlikely(!mm->pgd))
499                 return -ENOMEM;
500         return 0;
501 }
502
503 static inline void mm_free_pgd(struct mm_struct *mm)
504 {
505         pgd_free(mm, mm->pgd);
506 }
507 #else
508 #define dup_mmap(mm, oldmm)     (0)
509 #define mm_alloc_pgd(mm)        (0)
510 #define mm_free_pgd(mm)
511 #endif /* CONFIG_MMU */
512
513 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
514
515 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
516 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
517
518 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
519
520 static int __init coredump_filter_setup(char *s)
521 {
522         default_dump_filter =
523                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
524                 MMF_DUMP_FILTER_MASK;
525         return 1;
526 }
527
528 __setup("coredump_filter=", coredump_filter_setup);
529
530 #include <linux/init_task.h>
531
532 static void mm_init_aio(struct mm_struct *mm)
533 {
534 #ifdef CONFIG_AIO
535         spin_lock_init(&mm->ioctx_lock);
536         mm->ioctx_table = NULL;
537 #endif
538 }
539
540 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
541 {
542 #ifdef CONFIG_MEMCG
543         mm->owner = p;
544 #endif
545 }
546
547 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
548 {
549         mm->mmap = NULL;
550         mm->mm_rb = RB_ROOT;
551         mm->vmacache_seqnum = 0;
552         atomic_set(&mm->mm_users, 1);
553         atomic_set(&mm->mm_count, 1);
554         init_rwsem(&mm->mmap_sem);
555         INIT_LIST_HEAD(&mm->mmlist);
556         mm->core_state = NULL;
557         atomic_long_set(&mm->nr_ptes, 0);
558         mm_nr_pmds_init(mm);
559         mm->map_count = 0;
560         mm->locked_vm = 0;
561         mm->pinned_vm = 0;
562         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
563         spin_lock_init(&mm->page_table_lock);
564         mm_init_cpumask(mm);
565         mm_init_aio(mm);
566         mm_init_owner(mm, p);
567         mmu_notifier_mm_init(mm);
568         clear_tlb_flush_pending(mm);
569 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
570         mm->pmd_huge_pte = NULL;
571 #endif
572
573         if (current->mm) {
574                 mm->flags = current->mm->flags & MMF_INIT_MASK;
575                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
576         } else {
577                 mm->flags = default_dump_filter;
578                 mm->def_flags = 0;
579         }
580
581         if (mm_alloc_pgd(mm))
582                 goto fail_nopgd;
583
584         if (init_new_context(p, mm))
585                 goto fail_nocontext;
586
587         return mm;
588
589 fail_nocontext:
590         mm_free_pgd(mm);
591 fail_nopgd:
592         free_mm(mm);
593         return NULL;
594 }
595
596 static void check_mm(struct mm_struct *mm)
597 {
598         int i;
599
600         for (i = 0; i < NR_MM_COUNTERS; i++) {
601                 long x = atomic_long_read(&mm->rss_stat.count[i]);
602
603                 if (unlikely(x))
604                         printk(KERN_ALERT "BUG: Bad rss-counter state "
605                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
606         }
607
608         if (atomic_long_read(&mm->nr_ptes))
609                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
610                                 atomic_long_read(&mm->nr_ptes));
611         if (mm_nr_pmds(mm))
612                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
613                                 mm_nr_pmds(mm));
614
615 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
616         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
617 #endif
618 }
619
620 /*
621  * Allocate and initialize an mm_struct.
622  */
623 struct mm_struct *mm_alloc(void)
624 {
625         struct mm_struct *mm;
626
627         mm = allocate_mm();
628         if (!mm)
629                 return NULL;
630
631         memset(mm, 0, sizeof(*mm));
632         return mm_init(mm, current);
633 }
634
635 /*
636  * Called when the last reference to the mm
637  * is dropped: either by a lazy thread or by
638  * mmput. Free the page directory and the mm.
639  */
640 void __mmdrop(struct mm_struct *mm)
641 {
642         BUG_ON(mm == &init_mm);
643         mm_free_pgd(mm);
644         destroy_context(mm);
645         mmu_notifier_mm_destroy(mm);
646         check_mm(mm);
647         free_mm(mm);
648 }
649 EXPORT_SYMBOL_GPL(__mmdrop);
650
651 /*
652  * Decrement the use count and release all resources for an mm.
653  */
654 void mmput(struct mm_struct *mm)
655 {
656         might_sleep();
657
658         if (atomic_dec_and_test(&mm->mm_users)) {
659                 uprobe_clear_state(mm);
660                 exit_aio(mm);
661                 ksm_exit(mm);
662                 khugepaged_exit(mm); /* must run before exit_mmap */
663                 exit_mmap(mm);
664                 set_mm_exe_file(mm, NULL);
665                 if (!list_empty(&mm->mmlist)) {
666                         spin_lock(&mmlist_lock);
667                         list_del(&mm->mmlist);
668                         spin_unlock(&mmlist_lock);
669                 }
670                 if (mm->binfmt)
671                         module_put(mm->binfmt->module);
672                 mmdrop(mm);
673         }
674 }
675 EXPORT_SYMBOL_GPL(mmput);
676
677 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
678 {
679         if (new_exe_file)
680                 get_file(new_exe_file);
681         if (mm->exe_file)
682                 fput(mm->exe_file);
683         mm->exe_file = new_exe_file;
684 }
685
686 struct file *get_mm_exe_file(struct mm_struct *mm)
687 {
688         struct file *exe_file;
689
690         /* We need mmap_sem to protect against races with removal of exe_file */
691         down_read(&mm->mmap_sem);
692         exe_file = mm->exe_file;
693         if (exe_file)
694                 get_file(exe_file);
695         up_read(&mm->mmap_sem);
696         return exe_file;
697 }
698
699 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
700 {
701         /* It's safe to write the exe_file pointer without exe_file_lock because
702          * this is called during fork when the task is not yet in /proc */
703         newmm->exe_file = get_mm_exe_file(oldmm);
704 }
705
706 /**
707  * get_task_mm - acquire a reference to the task's mm
708  *
709  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
710  * this kernel workthread has transiently adopted a user mm with use_mm,
711  * to do its AIO) is not set and if so returns a reference to it, after
712  * bumping up the use count.  User must release the mm via mmput()
713  * after use.  Typically used by /proc and ptrace.
714  */
715 struct mm_struct *get_task_mm(struct task_struct *task)
716 {
717         struct mm_struct *mm;
718
719         task_lock(task);
720         mm = task->mm;
721         if (mm) {
722                 if (task->flags & PF_KTHREAD)
723                         mm = NULL;
724                 else
725                         atomic_inc(&mm->mm_users);
726         }
727         task_unlock(task);
728         return mm;
729 }
730 EXPORT_SYMBOL_GPL(get_task_mm);
731
732 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
733 {
734         struct mm_struct *mm;
735         int err;
736
737         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
738         if (err)
739                 return ERR_PTR(err);
740
741         mm = get_task_mm(task);
742         if (mm && mm != current->mm &&
743                         !ptrace_may_access(task, mode)) {
744                 mmput(mm);
745                 mm = ERR_PTR(-EACCES);
746         }
747         mutex_unlock(&task->signal->cred_guard_mutex);
748
749         return mm;
750 }
751
752 static void complete_vfork_done(struct task_struct *tsk)
753 {
754         struct completion *vfork;
755
756         task_lock(tsk);
757         vfork = tsk->vfork_done;
758         if (likely(vfork)) {
759                 tsk->vfork_done = NULL;
760                 complete(vfork);
761         }
762         task_unlock(tsk);
763 }
764
765 static int wait_for_vfork_done(struct task_struct *child,
766                                 struct completion *vfork)
767 {
768         int killed;
769
770         freezer_do_not_count();
771         killed = wait_for_completion_killable(vfork);
772         freezer_count();
773
774         if (killed) {
775                 task_lock(child);
776                 child->vfork_done = NULL;
777                 task_unlock(child);
778         }
779
780         put_task_struct(child);
781         return killed;
782 }
783
784 /* Please note the differences between mmput and mm_release.
785  * mmput is called whenever we stop holding onto a mm_struct,
786  * error success whatever.
787  *
788  * mm_release is called after a mm_struct has been removed
789  * from the current process.
790  *
791  * This difference is important for error handling, when we
792  * only half set up a mm_struct for a new process and need to restore
793  * the old one.  Because we mmput the new mm_struct before
794  * restoring the old one. . .
795  * Eric Biederman 10 January 1998
796  */
797 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
798 {
799         /* Get rid of any futexes when releasing the mm */
800 #ifdef CONFIG_FUTEX
801         if (unlikely(tsk->robust_list)) {
802                 exit_robust_list(tsk);
803                 tsk->robust_list = NULL;
804         }
805 #ifdef CONFIG_COMPAT
806         if (unlikely(tsk->compat_robust_list)) {
807                 compat_exit_robust_list(tsk);
808                 tsk->compat_robust_list = NULL;
809         }
810 #endif
811         if (unlikely(!list_empty(&tsk->pi_state_list)))
812                 exit_pi_state_list(tsk);
813 #endif
814
815         uprobe_free_utask(tsk);
816
817         /* Get rid of any cached register state */
818         deactivate_mm(tsk, mm);
819
820         /*
821          * If we're exiting normally, clear a user-space tid field if
822          * requested.  We leave this alone when dying by signal, to leave
823          * the value intact in a core dump, and to save the unnecessary
824          * trouble, say, a killed vfork parent shouldn't touch this mm.
825          * Userland only wants this done for a sys_exit.
826          */
827         if (tsk->clear_child_tid) {
828                 if (!(tsk->flags & PF_SIGNALED) &&
829                     atomic_read(&mm->mm_users) > 1) {
830                         /*
831                          * We don't check the error code - if userspace has
832                          * not set up a proper pointer then tough luck.
833                          */
834                         put_user(0, tsk->clear_child_tid);
835                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
836                                         1, NULL, NULL, 0);
837                 }
838                 tsk->clear_child_tid = NULL;
839         }
840
841         /*
842          * All done, finally we can wake up parent and return this mm to him.
843          * Also kthread_stop() uses this completion for synchronization.
844          */
845         if (tsk->vfork_done)
846                 complete_vfork_done(tsk);
847 }
848
849 /*
850  * Allocate a new mm structure and copy contents from the
851  * mm structure of the passed in task structure.
852  */
853 static struct mm_struct *dup_mm(struct task_struct *tsk)
854 {
855         struct mm_struct *mm, *oldmm = current->mm;
856         int err;
857
858         mm = allocate_mm();
859         if (!mm)
860                 goto fail_nomem;
861
862         memcpy(mm, oldmm, sizeof(*mm));
863
864         if (!mm_init(mm, tsk))
865                 goto fail_nomem;
866
867         dup_mm_exe_file(oldmm, mm);
868
869         err = dup_mmap(mm, oldmm);
870         if (err)
871                 goto free_pt;
872
873         mm->hiwater_rss = get_mm_rss(mm);
874         mm->hiwater_vm = mm->total_vm;
875
876         if (mm->binfmt && !try_module_get(mm->binfmt->module))
877                 goto free_pt;
878
879         return mm;
880
881 free_pt:
882         /* don't put binfmt in mmput, we haven't got module yet */
883         mm->binfmt = NULL;
884         mmput(mm);
885
886 fail_nomem:
887         return NULL;
888 }
889
890 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
891 {
892         struct mm_struct *mm, *oldmm;
893         int retval;
894
895         tsk->min_flt = tsk->maj_flt = 0;
896         tsk->nvcsw = tsk->nivcsw = 0;
897 #ifdef CONFIG_DETECT_HUNG_TASK
898         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
899 #endif
900
901         tsk->mm = NULL;
902         tsk->active_mm = NULL;
903
904         /*
905          * Are we cloning a kernel thread?
906          *
907          * We need to steal a active VM for that..
908          */
909         oldmm = current->mm;
910         if (!oldmm)
911                 return 0;
912
913         /* initialize the new vmacache entries */
914         vmacache_flush(tsk);
915
916         if (clone_flags & CLONE_VM) {
917                 atomic_inc(&oldmm->mm_users);
918                 mm = oldmm;
919                 goto good_mm;
920         }
921
922         retval = -ENOMEM;
923         mm = dup_mm(tsk);
924         if (!mm)
925                 goto fail_nomem;
926
927 good_mm:
928         tsk->mm = mm;
929         tsk->active_mm = mm;
930         return 0;
931
932 fail_nomem:
933         return retval;
934 }
935
936 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
937 {
938         struct fs_struct *fs = current->fs;
939         if (clone_flags & CLONE_FS) {
940                 /* tsk->fs is already what we want */
941                 spin_lock(&fs->lock);
942                 if (fs->in_exec) {
943                         spin_unlock(&fs->lock);
944                         return -EAGAIN;
945                 }
946                 fs->users++;
947                 spin_unlock(&fs->lock);
948                 return 0;
949         }
950         tsk->fs = copy_fs_struct(fs);
951         if (!tsk->fs)
952                 return -ENOMEM;
953         return 0;
954 }
955
956 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
957 {
958         struct files_struct *oldf, *newf;
959         int error = 0;
960
961         /*
962          * A background process may not have any files ...
963          */
964         oldf = current->files;
965         if (!oldf)
966                 goto out;
967
968         if (clone_flags & CLONE_FILES) {
969                 atomic_inc(&oldf->count);
970                 goto out;
971         }
972
973         newf = dup_fd(oldf, &error);
974         if (!newf)
975                 goto out;
976
977         tsk->files = newf;
978         error = 0;
979 out:
980         return error;
981 }
982
983 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
984 {
985 #ifdef CONFIG_BLOCK
986         struct io_context *ioc = current->io_context;
987         struct io_context *new_ioc;
988
989         if (!ioc)
990                 return 0;
991         /*
992          * Share io context with parent, if CLONE_IO is set
993          */
994         if (clone_flags & CLONE_IO) {
995                 ioc_task_link(ioc);
996                 tsk->io_context = ioc;
997         } else if (ioprio_valid(ioc->ioprio)) {
998                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
999                 if (unlikely(!new_ioc))
1000                         return -ENOMEM;
1001
1002                 new_ioc->ioprio = ioc->ioprio;
1003                 put_io_context(new_ioc);
1004         }
1005 #endif
1006         return 0;
1007 }
1008
1009 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1010 {
1011         struct sighand_struct *sig;
1012
1013         if (clone_flags & CLONE_SIGHAND) {
1014                 atomic_inc(&current->sighand->count);
1015                 return 0;
1016         }
1017         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1018         rcu_assign_pointer(tsk->sighand, sig);
1019         if (!sig)
1020                 return -ENOMEM;
1021         atomic_set(&sig->count, 1);
1022         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1023         return 0;
1024 }
1025
1026 void __cleanup_sighand(struct sighand_struct *sighand)
1027 {
1028         if (atomic_dec_and_test(&sighand->count)) {
1029                 signalfd_cleanup(sighand);
1030                 /*
1031                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1032                  * without an RCU grace period, see __lock_task_sighand().
1033                  */
1034                 kmem_cache_free(sighand_cachep, sighand);
1035         }
1036 }
1037
1038 /*
1039  * Initialize POSIX timer handling for a thread group.
1040  */
1041 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1042 {
1043         unsigned long cpu_limit;
1044
1045         /* Thread group counters. */
1046         thread_group_cputime_init(sig);
1047
1048         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1049         if (cpu_limit != RLIM_INFINITY) {
1050                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1051                 sig->cputimer.running = 1;
1052         }
1053
1054         /* The timer lists. */
1055         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1056         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1057         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1058 }
1059
1060 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1061 {
1062         struct signal_struct *sig;
1063
1064         if (clone_flags & CLONE_THREAD)
1065                 return 0;
1066
1067         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1068         tsk->signal = sig;
1069         if (!sig)
1070                 return -ENOMEM;
1071
1072         sig->nr_threads = 1;
1073         atomic_set(&sig->live, 1);
1074         atomic_set(&sig->sigcnt, 1);
1075
1076         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1077         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1078         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1079
1080         init_waitqueue_head(&sig->wait_chldexit);
1081         sig->curr_target = tsk;
1082         init_sigpending(&sig->shared_pending);
1083         INIT_LIST_HEAD(&sig->posix_timers);
1084         seqlock_init(&sig->stats_lock);
1085
1086         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1087         sig->real_timer.function = it_real_fn;
1088
1089         task_lock(current->group_leader);
1090         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1091         task_unlock(current->group_leader);
1092
1093         posix_cpu_timers_init_group(sig);
1094
1095         tty_audit_fork(sig);
1096         sched_autogroup_fork(sig);
1097
1098 #ifdef CONFIG_CGROUPS
1099         init_rwsem(&sig->group_rwsem);
1100 #endif
1101
1102         sig->oom_score_adj = current->signal->oom_score_adj;
1103         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1104
1105         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1106                                    current->signal->is_child_subreaper;
1107
1108         mutex_init(&sig->cred_guard_mutex);
1109
1110         return 0;
1111 }
1112
1113 static void copy_seccomp(struct task_struct *p)
1114 {
1115 #ifdef CONFIG_SECCOMP
1116         /*
1117          * Must be called with sighand->lock held, which is common to
1118          * all threads in the group. Holding cred_guard_mutex is not
1119          * needed because this new task is not yet running and cannot
1120          * be racing exec.
1121          */
1122         assert_spin_locked(&current->sighand->siglock);
1123
1124         /* Ref-count the new filter user, and assign it. */
1125         get_seccomp_filter(current);
1126         p->seccomp = current->seccomp;
1127
1128         /*
1129          * Explicitly enable no_new_privs here in case it got set
1130          * between the task_struct being duplicated and holding the
1131          * sighand lock. The seccomp state and nnp must be in sync.
1132          */
1133         if (task_no_new_privs(current))
1134                 task_set_no_new_privs(p);
1135
1136         /*
1137          * If the parent gained a seccomp mode after copying thread
1138          * flags and between before we held the sighand lock, we have
1139          * to manually enable the seccomp thread flag here.
1140          */
1141         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1142                 set_tsk_thread_flag(p, TIF_SECCOMP);
1143 #endif
1144 }
1145
1146 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1147 {
1148         current->clear_child_tid = tidptr;
1149
1150         return task_pid_vnr(current);
1151 }
1152
1153 static void rt_mutex_init_task(struct task_struct *p)
1154 {
1155         raw_spin_lock_init(&p->pi_lock);
1156 #ifdef CONFIG_RT_MUTEXES
1157         p->pi_waiters = RB_ROOT;
1158         p->pi_waiters_leftmost = NULL;
1159         p->pi_blocked_on = NULL;
1160 #endif
1161 }
1162
1163 /*
1164  * Initialize POSIX timer handling for a single task.
1165  */
1166 static void posix_cpu_timers_init(struct task_struct *tsk)
1167 {
1168         tsk->cputime_expires.prof_exp = 0;
1169         tsk->cputime_expires.virt_exp = 0;
1170         tsk->cputime_expires.sched_exp = 0;
1171         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1172         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1173         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1174 }
1175
1176 static inline void
1177 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1178 {
1179          task->pids[type].pid = pid;
1180 }
1181
1182 /*
1183  * This creates a new process as a copy of the old one,
1184  * but does not actually start it yet.
1185  *
1186  * It copies the registers, and all the appropriate
1187  * parts of the process environment (as per the clone
1188  * flags). The actual kick-off is left to the caller.
1189  */
1190 static struct task_struct *copy_process(unsigned long clone_flags,
1191                                         unsigned long stack_start,
1192                                         unsigned long stack_size,
1193                                         int __user *child_tidptr,
1194                                         struct pid *pid,
1195                                         int trace)
1196 {
1197         int retval;
1198         struct task_struct *p;
1199
1200         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1201                 return ERR_PTR(-EINVAL);
1202
1203         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1204                 return ERR_PTR(-EINVAL);
1205
1206         /*
1207          * Thread groups must share signals as well, and detached threads
1208          * can only be started up within the thread group.
1209          */
1210         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1211                 return ERR_PTR(-EINVAL);
1212
1213         /*
1214          * Shared signal handlers imply shared VM. By way of the above,
1215          * thread groups also imply shared VM. Blocking this case allows
1216          * for various simplifications in other code.
1217          */
1218         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1219                 return ERR_PTR(-EINVAL);
1220
1221         /*
1222          * Siblings of global init remain as zombies on exit since they are
1223          * not reaped by their parent (swapper). To solve this and to avoid
1224          * multi-rooted process trees, prevent global and container-inits
1225          * from creating siblings.
1226          */
1227         if ((clone_flags & CLONE_PARENT) &&
1228                                 current->signal->flags & SIGNAL_UNKILLABLE)
1229                 return ERR_PTR(-EINVAL);
1230
1231         /*
1232          * If the new process will be in a different pid or user namespace
1233          * do not allow it to share a thread group or signal handlers or
1234          * parent with the forking task.
1235          */
1236         if (clone_flags & CLONE_SIGHAND) {
1237                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1238                     (task_active_pid_ns(current) !=
1239                                 current->nsproxy->pid_ns_for_children))
1240                         return ERR_PTR(-EINVAL);
1241         }
1242
1243         retval = security_task_create(clone_flags);
1244         if (retval)
1245                 goto fork_out;
1246
1247         retval = -ENOMEM;
1248         p = dup_task_struct(current);
1249         if (!p)
1250                 goto fork_out;
1251
1252         ftrace_graph_init_task(p);
1253
1254         rt_mutex_init_task(p);
1255
1256 #ifdef CONFIG_PROVE_LOCKING
1257         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1258         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1259 #endif
1260         retval = -EAGAIN;
1261         if (atomic_read(&p->real_cred->user->processes) >=
1262                         task_rlimit(p, RLIMIT_NPROC)) {
1263                 if (p->real_cred->user != INIT_USER &&
1264                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1265                         goto bad_fork_free;
1266         }
1267         current->flags &= ~PF_NPROC_EXCEEDED;
1268
1269         retval = copy_creds(p, clone_flags);
1270         if (retval < 0)
1271                 goto bad_fork_free;
1272
1273         /*
1274          * If multiple threads are within copy_process(), then this check
1275          * triggers too late. This doesn't hurt, the check is only there
1276          * to stop root fork bombs.
1277          */
1278         retval = -EAGAIN;
1279         if (nr_threads >= max_threads)
1280                 goto bad_fork_cleanup_count;
1281
1282         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1283                 goto bad_fork_cleanup_count;
1284
1285         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1286         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1287         p->flags |= PF_FORKNOEXEC;
1288         INIT_LIST_HEAD(&p->children);
1289         INIT_LIST_HEAD(&p->sibling);
1290         rcu_copy_process(p);
1291         p->vfork_done = NULL;
1292         spin_lock_init(&p->alloc_lock);
1293
1294         init_sigpending(&p->pending);
1295
1296         p->utime = p->stime = p->gtime = 0;
1297         p->utimescaled = p->stimescaled = 0;
1298 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1299         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1300 #endif
1301 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1302         seqlock_init(&p->vtime_seqlock);
1303         p->vtime_snap = 0;
1304         p->vtime_snap_whence = VTIME_SLEEPING;
1305 #endif
1306
1307 #if defined(SPLIT_RSS_COUNTING)
1308         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1309 #endif
1310
1311         p->default_timer_slack_ns = current->timer_slack_ns;
1312
1313         task_io_accounting_init(&p->ioac);
1314         acct_clear_integrals(p);
1315
1316         posix_cpu_timers_init(p);
1317
1318         p->start_time = ktime_get_ns();
1319         p->real_start_time = ktime_get_boot_ns();
1320         p->io_context = NULL;
1321         p->audit_context = NULL;
1322         if (clone_flags & CLONE_THREAD)
1323                 threadgroup_change_begin(current);
1324         cgroup_fork(p);
1325 #ifdef CONFIG_NUMA
1326         p->mempolicy = mpol_dup(p->mempolicy);
1327         if (IS_ERR(p->mempolicy)) {
1328                 retval = PTR_ERR(p->mempolicy);
1329                 p->mempolicy = NULL;
1330                 goto bad_fork_cleanup_threadgroup_lock;
1331         }
1332 #endif
1333 #ifdef CONFIG_CPUSETS
1334         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1335         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1336         seqcount_init(&p->mems_allowed_seq);
1337 #endif
1338 #ifdef CONFIG_TRACE_IRQFLAGS
1339         p->irq_events = 0;
1340         p->hardirqs_enabled = 0;
1341         p->hardirq_enable_ip = 0;
1342         p->hardirq_enable_event = 0;
1343         p->hardirq_disable_ip = _THIS_IP_;
1344         p->hardirq_disable_event = 0;
1345         p->softirqs_enabled = 1;
1346         p->softirq_enable_ip = _THIS_IP_;
1347         p->softirq_enable_event = 0;
1348         p->softirq_disable_ip = 0;
1349         p->softirq_disable_event = 0;
1350         p->hardirq_context = 0;
1351         p->softirq_context = 0;
1352 #endif
1353 #ifdef CONFIG_LOCKDEP
1354         p->lockdep_depth = 0; /* no locks held yet */
1355         p->curr_chain_key = 0;
1356         p->lockdep_recursion = 0;
1357 #endif
1358
1359 #ifdef CONFIG_DEBUG_MUTEXES
1360         p->blocked_on = NULL; /* not blocked yet */
1361 #endif
1362 #ifdef CONFIG_BCACHE
1363         p->sequential_io        = 0;
1364         p->sequential_io_avg    = 0;
1365 #endif
1366
1367         /* Perform scheduler related setup. Assign this task to a CPU. */
1368         retval = sched_fork(clone_flags, p);
1369         if (retval)
1370                 goto bad_fork_cleanup_policy;
1371
1372         retval = perf_event_init_task(p);
1373         if (retval)
1374                 goto bad_fork_cleanup_policy;
1375         retval = audit_alloc(p);
1376         if (retval)
1377                 goto bad_fork_cleanup_perf;
1378         /* copy all the process information */
1379         shm_init_task(p);
1380         retval = copy_semundo(clone_flags, p);
1381         if (retval)
1382                 goto bad_fork_cleanup_audit;
1383         retval = copy_files(clone_flags, p);
1384         if (retval)
1385                 goto bad_fork_cleanup_semundo;
1386         retval = copy_fs(clone_flags, p);
1387         if (retval)
1388                 goto bad_fork_cleanup_files;
1389         retval = copy_sighand(clone_flags, p);
1390         if (retval)
1391                 goto bad_fork_cleanup_fs;
1392         retval = copy_signal(clone_flags, p);
1393         if (retval)
1394                 goto bad_fork_cleanup_sighand;
1395         retval = copy_mm(clone_flags, p);
1396         if (retval)
1397                 goto bad_fork_cleanup_signal;
1398         retval = copy_namespaces(clone_flags, p);
1399         if (retval)
1400                 goto bad_fork_cleanup_mm;
1401         retval = copy_io(clone_flags, p);
1402         if (retval)
1403                 goto bad_fork_cleanup_namespaces;
1404         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1405         if (retval)
1406                 goto bad_fork_cleanup_io;
1407
1408         if (pid != &init_struct_pid) {
1409                 retval = -ENOMEM;
1410                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1411                 if (!pid)
1412                         goto bad_fork_cleanup_io;
1413         }
1414
1415         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1416         /*
1417          * Clear TID on mm_release()?
1418          */
1419         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1420 #ifdef CONFIG_BLOCK
1421         p->plug = NULL;
1422 #endif
1423 #ifdef CONFIG_FUTEX
1424         p->robust_list = NULL;
1425 #ifdef CONFIG_COMPAT
1426         p->compat_robust_list = NULL;
1427 #endif
1428         INIT_LIST_HEAD(&p->pi_state_list);
1429         p->pi_state_cache = NULL;
1430 #endif
1431         /*
1432          * sigaltstack should be cleared when sharing the same VM
1433          */
1434         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1435                 p->sas_ss_sp = p->sas_ss_size = 0;
1436
1437         /*
1438          * Syscall tracing and stepping should be turned off in the
1439          * child regardless of CLONE_PTRACE.
1440          */
1441         user_disable_single_step(p);
1442         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1443 #ifdef TIF_SYSCALL_EMU
1444         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1445 #endif
1446         clear_all_latency_tracing(p);
1447
1448         /* ok, now we should be set up.. */
1449         p->pid = pid_nr(pid);
1450         if (clone_flags & CLONE_THREAD) {
1451                 p->exit_signal = -1;
1452                 p->group_leader = current->group_leader;
1453                 p->tgid = current->tgid;
1454         } else {
1455                 if (clone_flags & CLONE_PARENT)
1456                         p->exit_signal = current->group_leader->exit_signal;
1457                 else
1458                         p->exit_signal = (clone_flags & CSIGNAL);
1459                 p->group_leader = p;
1460                 p->tgid = p->pid;
1461         }
1462
1463         p->nr_dirtied = 0;
1464         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1465         p->dirty_paused_when = 0;
1466
1467         p->pdeath_signal = 0;
1468         INIT_LIST_HEAD(&p->thread_group);
1469         p->task_works = NULL;
1470
1471         /*
1472          * Make it visible to the rest of the system, but dont wake it up yet.
1473          * Need tasklist lock for parent etc handling!
1474          */
1475         write_lock_irq(&tasklist_lock);
1476
1477         /* CLONE_PARENT re-uses the old parent */
1478         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1479                 p->real_parent = current->real_parent;
1480                 p->parent_exec_id = current->parent_exec_id;
1481         } else {
1482                 p->real_parent = current;
1483                 p->parent_exec_id = current->self_exec_id;
1484         }
1485
1486         spin_lock(&current->sighand->siglock);
1487
1488         /*
1489          * Copy seccomp details explicitly here, in case they were changed
1490          * before holding sighand lock.
1491          */
1492         copy_seccomp(p);
1493
1494         /*
1495          * Process group and session signals need to be delivered to just the
1496          * parent before the fork or both the parent and the child after the
1497          * fork. Restart if a signal comes in before we add the new process to
1498          * it's process group.
1499          * A fatal signal pending means that current will exit, so the new
1500          * thread can't slip out of an OOM kill (or normal SIGKILL).
1501         */
1502         recalc_sigpending();
1503         if (signal_pending(current)) {
1504                 spin_unlock(&current->sighand->siglock);
1505                 write_unlock_irq(&tasklist_lock);
1506                 retval = -ERESTARTNOINTR;
1507                 goto bad_fork_free_pid;
1508         }
1509
1510         if (likely(p->pid)) {
1511                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1512
1513                 init_task_pid(p, PIDTYPE_PID, pid);
1514                 if (thread_group_leader(p)) {
1515                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1516                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1517
1518                         if (is_child_reaper(pid)) {
1519                                 ns_of_pid(pid)->child_reaper = p;
1520                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1521                         }
1522
1523                         p->signal->leader_pid = pid;
1524                         p->signal->tty = tty_kref_get(current->signal->tty);
1525                         list_add_tail(&p->sibling, &p->real_parent->children);
1526                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1527                         attach_pid(p, PIDTYPE_PGID);
1528                         attach_pid(p, PIDTYPE_SID);
1529                         __this_cpu_inc(process_counts);
1530                 } else {
1531                         current->signal->nr_threads++;
1532                         atomic_inc(&current->signal->live);
1533                         atomic_inc(&current->signal->sigcnt);
1534                         list_add_tail_rcu(&p->thread_group,
1535                                           &p->group_leader->thread_group);
1536                         list_add_tail_rcu(&p->thread_node,
1537                                           &p->signal->thread_head);
1538                 }
1539                 attach_pid(p, PIDTYPE_PID);
1540                 nr_threads++;
1541         }
1542
1543         total_forks++;
1544         spin_unlock(&current->sighand->siglock);
1545         syscall_tracepoint_update(p);
1546         write_unlock_irq(&tasklist_lock);
1547
1548         proc_fork_connector(p);
1549         cgroup_post_fork(p);
1550         if (clone_flags & CLONE_THREAD)
1551                 threadgroup_change_end(current);
1552         perf_event_fork(p);
1553
1554         trace_task_newtask(p, clone_flags);
1555         uprobe_copy_process(p, clone_flags);
1556
1557         return p;
1558
1559 bad_fork_free_pid:
1560         if (pid != &init_struct_pid)
1561                 free_pid(pid);
1562 bad_fork_cleanup_io:
1563         if (p->io_context)
1564                 exit_io_context(p);
1565 bad_fork_cleanup_namespaces:
1566         exit_task_namespaces(p);
1567 bad_fork_cleanup_mm:
1568         if (p->mm)
1569                 mmput(p->mm);
1570 bad_fork_cleanup_signal:
1571         if (!(clone_flags & CLONE_THREAD))
1572                 free_signal_struct(p->signal);
1573 bad_fork_cleanup_sighand:
1574         __cleanup_sighand(p->sighand);
1575 bad_fork_cleanup_fs:
1576         exit_fs(p); /* blocking */
1577 bad_fork_cleanup_files:
1578         exit_files(p); /* blocking */
1579 bad_fork_cleanup_semundo:
1580         exit_sem(p);
1581 bad_fork_cleanup_audit:
1582         audit_free(p);
1583 bad_fork_cleanup_perf:
1584         perf_event_free_task(p);
1585 bad_fork_cleanup_policy:
1586 #ifdef CONFIG_NUMA
1587         mpol_put(p->mempolicy);
1588 bad_fork_cleanup_threadgroup_lock:
1589 #endif
1590         if (clone_flags & CLONE_THREAD)
1591                 threadgroup_change_end(current);
1592         delayacct_tsk_free(p);
1593         module_put(task_thread_info(p)->exec_domain->module);
1594 bad_fork_cleanup_count:
1595         atomic_dec(&p->cred->user->processes);
1596         exit_creds(p);
1597 bad_fork_free:
1598         free_task(p);
1599 fork_out:
1600         return ERR_PTR(retval);
1601 }
1602
1603 static inline void init_idle_pids(struct pid_link *links)
1604 {
1605         enum pid_type type;
1606
1607         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1608                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1609                 links[type].pid = &init_struct_pid;
1610         }
1611 }
1612
1613 struct task_struct *fork_idle(int cpu)
1614 {
1615         struct task_struct *task;
1616         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1617         if (!IS_ERR(task)) {
1618                 init_idle_pids(task->pids);
1619                 init_idle(task, cpu);
1620         }
1621
1622         return task;
1623 }
1624
1625 /*
1626  *  Ok, this is the main fork-routine.
1627  *
1628  * It copies the process, and if successful kick-starts
1629  * it and waits for it to finish using the VM if required.
1630  */
1631 long do_fork(unsigned long clone_flags,
1632               unsigned long stack_start,
1633               unsigned long stack_size,
1634               int __user *parent_tidptr,
1635               int __user *child_tidptr)
1636 {
1637         struct task_struct *p;
1638         int trace = 0;
1639         long nr;
1640
1641         /*
1642          * Determine whether and which event to report to ptracer.  When
1643          * called from kernel_thread or CLONE_UNTRACED is explicitly
1644          * requested, no event is reported; otherwise, report if the event
1645          * for the type of forking is enabled.
1646          */
1647         if (!(clone_flags & CLONE_UNTRACED)) {
1648                 if (clone_flags & CLONE_VFORK)
1649                         trace = PTRACE_EVENT_VFORK;
1650                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1651                         trace = PTRACE_EVENT_CLONE;
1652                 else
1653                         trace = PTRACE_EVENT_FORK;
1654
1655                 if (likely(!ptrace_event_enabled(current, trace)))
1656                         trace = 0;
1657         }
1658
1659         p = copy_process(clone_flags, stack_start, stack_size,
1660                          child_tidptr, NULL, trace);
1661         /*
1662          * Do this prior waking up the new thread - the thread pointer
1663          * might get invalid after that point, if the thread exits quickly.
1664          */
1665         if (!IS_ERR(p)) {
1666                 struct completion vfork;
1667                 struct pid *pid;
1668
1669                 trace_sched_process_fork(current, p);
1670
1671                 pid = get_task_pid(p, PIDTYPE_PID);
1672                 nr = pid_vnr(pid);
1673
1674                 if (clone_flags & CLONE_PARENT_SETTID)
1675                         put_user(nr, parent_tidptr);
1676
1677                 if (clone_flags & CLONE_VFORK) {
1678                         p->vfork_done = &vfork;
1679                         init_completion(&vfork);
1680                         get_task_struct(p);
1681                 }
1682
1683                 wake_up_new_task(p);
1684
1685                 /* forking complete and child started to run, tell ptracer */
1686                 if (unlikely(trace))
1687                         ptrace_event_pid(trace, pid);
1688
1689                 if (clone_flags & CLONE_VFORK) {
1690                         if (!wait_for_vfork_done(p, &vfork))
1691                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1692                 }
1693
1694                 put_pid(pid);
1695         } else {
1696                 nr = PTR_ERR(p);
1697         }
1698         return nr;
1699 }
1700
1701 /*
1702  * Create a kernel thread.
1703  */
1704 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1705 {
1706         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1707                 (unsigned long)arg, NULL, NULL);
1708 }
1709
1710 #ifdef __ARCH_WANT_SYS_FORK
1711 SYSCALL_DEFINE0(fork)
1712 {
1713 #ifdef CONFIG_MMU
1714         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1715 #else
1716         /* can not support in nommu mode */
1717         return -EINVAL;
1718 #endif
1719 }
1720 #endif
1721
1722 #ifdef __ARCH_WANT_SYS_VFORK
1723 SYSCALL_DEFINE0(vfork)
1724 {
1725         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1726                         0, NULL, NULL);
1727 }
1728 #endif
1729
1730 #ifdef __ARCH_WANT_SYS_CLONE
1731 #ifdef CONFIG_CLONE_BACKWARDS
1732 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1733                  int __user *, parent_tidptr,
1734                  int, tls_val,
1735                  int __user *, child_tidptr)
1736 #elif defined(CONFIG_CLONE_BACKWARDS2)
1737 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1738                  int __user *, parent_tidptr,
1739                  int __user *, child_tidptr,
1740                  int, tls_val)
1741 #elif defined(CONFIG_CLONE_BACKWARDS3)
1742 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1743                 int, stack_size,
1744                 int __user *, parent_tidptr,
1745                 int __user *, child_tidptr,
1746                 int, tls_val)
1747 #else
1748 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1749                  int __user *, parent_tidptr,
1750                  int __user *, child_tidptr,
1751                  int, tls_val)
1752 #endif
1753 {
1754         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1755 }
1756 #endif
1757
1758 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1759 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1760 #endif
1761
1762 static void sighand_ctor(void *data)
1763 {
1764         struct sighand_struct *sighand = data;
1765
1766         spin_lock_init(&sighand->siglock);
1767         init_waitqueue_head(&sighand->signalfd_wqh);
1768 }
1769
1770 void __init proc_caches_init(void)
1771 {
1772         sighand_cachep = kmem_cache_create("sighand_cache",
1773                         sizeof(struct sighand_struct), 0,
1774                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1775                         SLAB_NOTRACK, sighand_ctor);
1776         signal_cachep = kmem_cache_create("signal_cache",
1777                         sizeof(struct signal_struct), 0,
1778                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1779         files_cachep = kmem_cache_create("files_cache",
1780                         sizeof(struct files_struct), 0,
1781                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1782         fs_cachep = kmem_cache_create("fs_cache",
1783                         sizeof(struct fs_struct), 0,
1784                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1785         /*
1786          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1787          * whole struct cpumask for the OFFSTACK case. We could change
1788          * this to *only* allocate as much of it as required by the
1789          * maximum number of CPU's we can ever have.  The cpumask_allocation
1790          * is at the end of the structure, exactly for that reason.
1791          */
1792         mm_cachep = kmem_cache_create("mm_struct",
1793                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1794                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1795         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1796         mmap_init();
1797         nsproxy_cache_init();
1798 }
1799
1800 /*
1801  * Check constraints on flags passed to the unshare system call.
1802  */
1803 static int check_unshare_flags(unsigned long unshare_flags)
1804 {
1805         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1806                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1807                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1808                                 CLONE_NEWUSER|CLONE_NEWPID))
1809                 return -EINVAL;
1810         /*
1811          * Not implemented, but pretend it works if there is nothing to
1812          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1813          * needs to unshare vm.
1814          */
1815         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1816                 /* FIXME: get_task_mm() increments ->mm_users */
1817                 if (atomic_read(&current->mm->mm_users) > 1)
1818                         return -EINVAL;
1819         }
1820
1821         return 0;
1822 }
1823
1824 /*
1825  * Unshare the filesystem structure if it is being shared
1826  */
1827 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1828 {
1829         struct fs_struct *fs = current->fs;
1830
1831         if (!(unshare_flags & CLONE_FS) || !fs)
1832                 return 0;
1833
1834         /* don't need lock here; in the worst case we'll do useless copy */
1835         if (fs->users == 1)
1836                 return 0;
1837
1838         *new_fsp = copy_fs_struct(fs);
1839         if (!*new_fsp)
1840                 return -ENOMEM;
1841
1842         return 0;
1843 }
1844
1845 /*
1846  * Unshare file descriptor table if it is being shared
1847  */
1848 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1849 {
1850         struct files_struct *fd = current->files;
1851         int error = 0;
1852
1853         if ((unshare_flags & CLONE_FILES) &&
1854             (fd && atomic_read(&fd->count) > 1)) {
1855                 *new_fdp = dup_fd(fd, &error);
1856                 if (!*new_fdp)
1857                         return error;
1858         }
1859
1860         return 0;
1861 }
1862
1863 /*
1864  * unshare allows a process to 'unshare' part of the process
1865  * context which was originally shared using clone.  copy_*
1866  * functions used by do_fork() cannot be used here directly
1867  * because they modify an inactive task_struct that is being
1868  * constructed. Here we are modifying the current, active,
1869  * task_struct.
1870  */
1871 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1872 {
1873         struct fs_struct *fs, *new_fs = NULL;
1874         struct files_struct *fd, *new_fd = NULL;
1875         struct cred *new_cred = NULL;
1876         struct nsproxy *new_nsproxy = NULL;
1877         int do_sysvsem = 0;
1878         int err;
1879
1880         /*
1881          * If unsharing a user namespace must also unshare the thread.
1882          */
1883         if (unshare_flags & CLONE_NEWUSER)
1884                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1885         /*
1886          * If unsharing a thread from a thread group, must also unshare vm.
1887          */
1888         if (unshare_flags & CLONE_THREAD)
1889                 unshare_flags |= CLONE_VM;
1890         /*
1891          * If unsharing vm, must also unshare signal handlers.
1892          */
1893         if (unshare_flags & CLONE_VM)
1894                 unshare_flags |= CLONE_SIGHAND;
1895         /*
1896          * If unsharing namespace, must also unshare filesystem information.
1897          */
1898         if (unshare_flags & CLONE_NEWNS)
1899                 unshare_flags |= CLONE_FS;
1900
1901         err = check_unshare_flags(unshare_flags);
1902         if (err)
1903                 goto bad_unshare_out;
1904         /*
1905          * CLONE_NEWIPC must also detach from the undolist: after switching
1906          * to a new ipc namespace, the semaphore arrays from the old
1907          * namespace are unreachable.
1908          */
1909         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1910                 do_sysvsem = 1;
1911         err = unshare_fs(unshare_flags, &new_fs);
1912         if (err)
1913                 goto bad_unshare_out;
1914         err = unshare_fd(unshare_flags, &new_fd);
1915         if (err)
1916                 goto bad_unshare_cleanup_fs;
1917         err = unshare_userns(unshare_flags, &new_cred);
1918         if (err)
1919                 goto bad_unshare_cleanup_fd;
1920         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1921                                          new_cred, new_fs);
1922         if (err)
1923                 goto bad_unshare_cleanup_cred;
1924
1925         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1926                 if (do_sysvsem) {
1927                         /*
1928                          * CLONE_SYSVSEM is equivalent to sys_exit().
1929                          */
1930                         exit_sem(current);
1931                 }
1932                 if (unshare_flags & CLONE_NEWIPC) {
1933                         /* Orphan segments in old ns (see sem above). */
1934                         exit_shm(current);
1935                         shm_init_task(current);
1936                 }
1937
1938                 if (new_nsproxy)
1939                         switch_task_namespaces(current, new_nsproxy);
1940
1941                 task_lock(current);
1942
1943                 if (new_fs) {
1944                         fs = current->fs;
1945                         spin_lock(&fs->lock);
1946                         current->fs = new_fs;
1947                         if (--fs->users)
1948                                 new_fs = NULL;
1949                         else
1950                                 new_fs = fs;
1951                         spin_unlock(&fs->lock);
1952                 }
1953
1954                 if (new_fd) {
1955                         fd = current->files;
1956                         current->files = new_fd;
1957                         new_fd = fd;
1958                 }
1959
1960                 task_unlock(current);
1961
1962                 if (new_cred) {
1963                         /* Install the new user namespace */
1964                         commit_creds(new_cred);
1965                         new_cred = NULL;
1966                 }
1967         }
1968
1969 bad_unshare_cleanup_cred:
1970         if (new_cred)
1971                 put_cred(new_cred);
1972 bad_unshare_cleanup_fd:
1973         if (new_fd)
1974                 put_files_struct(new_fd);
1975
1976 bad_unshare_cleanup_fs:
1977         if (new_fs)
1978                 free_fs_struct(new_fs);
1979
1980 bad_unshare_out:
1981         return err;
1982 }
1983
1984 /*
1985  *      Helper to unshare the files of the current task.
1986  *      We don't want to expose copy_files internals to
1987  *      the exec layer of the kernel.
1988  */
1989
1990 int unshare_files(struct files_struct **displaced)
1991 {
1992         struct task_struct *task = current;
1993         struct files_struct *copy = NULL;
1994         int error;
1995
1996         error = unshare_fd(CLONE_FILES, &copy);
1997         if (error || !copy) {
1998                 *displaced = NULL;
1999                 return error;
2000         }
2001         *displaced = task->files;
2002         task_lock(task);
2003         task->files = copy;
2004         task_unlock(task);
2005         return 0;
2006 }