Merge branch 'misc' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild
[firefly-linux-kernel-4.4.55.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 #include <linux/sysctl.h>
78
79 #include <asm/pgtable.h>
80 #include <asm/pgalloc.h>
81 #include <asm/uaccess.h>
82 #include <asm/mmu_context.h>
83 #include <asm/cacheflush.h>
84 #include <asm/tlbflush.h>
85
86 #include <trace/events/sched.h>
87
88 #define CREATE_TRACE_POINTS
89 #include <trace/events/task.h>
90
91 /*
92  * Minimum number of threads to boot the kernel
93  */
94 #define MIN_THREADS 20
95
96 /*
97  * Maximum number of threads
98  */
99 #define MAX_THREADS FUTEX_TID_MASK
100
101 /*
102  * Protected counters by write_lock_irq(&tasklist_lock)
103  */
104 unsigned long total_forks;      /* Handle normal Linux uptimes. */
105 int nr_threads;                 /* The idle threads do not count.. */
106
107 int max_threads;                /* tunable limit on nr_threads */
108
109 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
110
111 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
112
113 #ifdef CONFIG_PROVE_RCU
114 int lockdep_tasklist_lock_is_held(void)
115 {
116         return lockdep_is_held(&tasklist_lock);
117 }
118 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
119 #endif /* #ifdef CONFIG_PROVE_RCU */
120
121 int nr_processes(void)
122 {
123         int cpu;
124         int total = 0;
125
126         for_each_possible_cpu(cpu)
127                 total += per_cpu(process_counts, cpu);
128
129         return total;
130 }
131
132 void __weak arch_release_task_struct(struct task_struct *tsk)
133 {
134 }
135
136 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
137 static struct kmem_cache *task_struct_cachep;
138
139 static inline struct task_struct *alloc_task_struct_node(int node)
140 {
141         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
142 }
143
144 static inline void free_task_struct(struct task_struct *tsk)
145 {
146         kmem_cache_free(task_struct_cachep, tsk);
147 }
148 #endif
149
150 void __weak arch_release_thread_info(struct thread_info *ti)
151 {
152 }
153
154 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
155
156 /*
157  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
158  * kmemcache based allocator.
159  */
160 # if THREAD_SIZE >= PAGE_SIZE
161 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
162                                                   int node)
163 {
164         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
165                                                   THREAD_SIZE_ORDER);
166
167         return page ? page_address(page) : NULL;
168 }
169
170 static inline void free_thread_info(struct thread_info *ti)
171 {
172         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
173 }
174 # else
175 static struct kmem_cache *thread_info_cache;
176
177 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
178                                                   int node)
179 {
180         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
181 }
182
183 static void free_thread_info(struct thread_info *ti)
184 {
185         kmem_cache_free(thread_info_cache, ti);
186 }
187
188 void thread_info_cache_init(void)
189 {
190         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
191                                               THREAD_SIZE, 0, NULL);
192         BUG_ON(thread_info_cache == NULL);
193 }
194 # endif
195 #endif
196
197 /* SLAB cache for signal_struct structures (tsk->signal) */
198 static struct kmem_cache *signal_cachep;
199
200 /* SLAB cache for sighand_struct structures (tsk->sighand) */
201 struct kmem_cache *sighand_cachep;
202
203 /* SLAB cache for files_struct structures (tsk->files) */
204 struct kmem_cache *files_cachep;
205
206 /* SLAB cache for fs_struct structures (tsk->fs) */
207 struct kmem_cache *fs_cachep;
208
209 /* SLAB cache for vm_area_struct structures */
210 struct kmem_cache *vm_area_cachep;
211
212 /* SLAB cache for mm_struct structures (tsk->mm) */
213 static struct kmem_cache *mm_cachep;
214
215 static void account_kernel_stack(struct thread_info *ti, int account)
216 {
217         struct zone *zone = page_zone(virt_to_page(ti));
218
219         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
220 }
221
222 void free_task(struct task_struct *tsk)
223 {
224         account_kernel_stack(tsk->stack, -1);
225         arch_release_thread_info(tsk->stack);
226         free_thread_info(tsk->stack);
227         rt_mutex_debug_task_free(tsk);
228         ftrace_graph_exit_task(tsk);
229         put_seccomp_filter(tsk);
230         arch_release_task_struct(tsk);
231         free_task_struct(tsk);
232 }
233 EXPORT_SYMBOL(free_task);
234
235 static inline void free_signal_struct(struct signal_struct *sig)
236 {
237         taskstats_tgid_free(sig);
238         sched_autogroup_exit(sig);
239         kmem_cache_free(signal_cachep, sig);
240 }
241
242 static inline void put_signal_struct(struct signal_struct *sig)
243 {
244         if (atomic_dec_and_test(&sig->sigcnt))
245                 free_signal_struct(sig);
246 }
247
248 void __put_task_struct(struct task_struct *tsk)
249 {
250         WARN_ON(!tsk->exit_state);
251         WARN_ON(atomic_read(&tsk->usage));
252         WARN_ON(tsk == current);
253
254         cgroup_free(tsk);
255         task_numa_free(tsk);
256         security_task_free(tsk);
257         exit_creds(tsk);
258         delayacct_tsk_free(tsk);
259         put_signal_struct(tsk->signal);
260
261         if (!profile_handoff_task(tsk))
262                 free_task(tsk);
263 }
264 EXPORT_SYMBOL_GPL(__put_task_struct);
265
266 void __init __weak arch_task_cache_init(void) { }
267
268 /*
269  * set_max_threads
270  */
271 static void set_max_threads(unsigned int max_threads_suggested)
272 {
273         u64 threads;
274
275         /*
276          * The number of threads shall be limited such that the thread
277          * structures may only consume a small part of the available memory.
278          */
279         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
280                 threads = MAX_THREADS;
281         else
282                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
283                                     (u64) THREAD_SIZE * 8UL);
284
285         if (threads > max_threads_suggested)
286                 threads = max_threads_suggested;
287
288         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
289 }
290
291 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
292 /* Initialized by the architecture: */
293 int arch_task_struct_size __read_mostly;
294 #endif
295
296 void __init fork_init(void)
297 {
298 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
299 #ifndef ARCH_MIN_TASKALIGN
300 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
301 #endif
302         /* create a slab on which task_structs can be allocated */
303         task_struct_cachep =
304                 kmem_cache_create("task_struct", arch_task_struct_size,
305                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
306 #endif
307
308         /* do the arch specific task caches init */
309         arch_task_cache_init();
310
311         set_max_threads(MAX_THREADS);
312
313         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
314         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
315         init_task.signal->rlim[RLIMIT_SIGPENDING] =
316                 init_task.signal->rlim[RLIMIT_NPROC];
317 }
318
319 int __weak arch_dup_task_struct(struct task_struct *dst,
320                                                struct task_struct *src)
321 {
322         *dst = *src;
323         return 0;
324 }
325
326 void set_task_stack_end_magic(struct task_struct *tsk)
327 {
328         unsigned long *stackend;
329
330         stackend = end_of_stack(tsk);
331         *stackend = STACK_END_MAGIC;    /* for overflow detection */
332 }
333
334 static struct task_struct *dup_task_struct(struct task_struct *orig)
335 {
336         struct task_struct *tsk;
337         struct thread_info *ti;
338         int node = tsk_fork_get_node(orig);
339         int err;
340
341         tsk = alloc_task_struct_node(node);
342         if (!tsk)
343                 return NULL;
344
345         ti = alloc_thread_info_node(tsk, node);
346         if (!ti)
347                 goto free_tsk;
348
349         err = arch_dup_task_struct(tsk, orig);
350         if (err)
351                 goto free_ti;
352
353         tsk->stack = ti;
354 #ifdef CONFIG_SECCOMP
355         /*
356          * We must handle setting up seccomp filters once we're under
357          * the sighand lock in case orig has changed between now and
358          * then. Until then, filter must be NULL to avoid messing up
359          * the usage counts on the error path calling free_task.
360          */
361         tsk->seccomp.filter = NULL;
362 #endif
363
364         setup_thread_stack(tsk, orig);
365         clear_user_return_notifier(tsk);
366         clear_tsk_need_resched(tsk);
367         set_task_stack_end_magic(tsk);
368
369 #ifdef CONFIG_CC_STACKPROTECTOR
370         tsk->stack_canary = get_random_int();
371 #endif
372
373         /*
374          * One for us, one for whoever does the "release_task()" (usually
375          * parent)
376          */
377         atomic_set(&tsk->usage, 2);
378 #ifdef CONFIG_BLK_DEV_IO_TRACE
379         tsk->btrace_seq = 0;
380 #endif
381         tsk->splice_pipe = NULL;
382         tsk->task_frag.page = NULL;
383
384         account_kernel_stack(ti, 1);
385
386         return tsk;
387
388 free_ti:
389         free_thread_info(ti);
390 free_tsk:
391         free_task_struct(tsk);
392         return NULL;
393 }
394
395 #ifdef CONFIG_MMU
396 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
397 {
398         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
399         struct rb_node **rb_link, *rb_parent;
400         int retval;
401         unsigned long charge;
402
403         uprobe_start_dup_mmap();
404         down_write(&oldmm->mmap_sem);
405         flush_cache_dup_mm(oldmm);
406         uprobe_dup_mmap(oldmm, mm);
407         /*
408          * Not linked in yet - no deadlock potential:
409          */
410         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
411
412         /* No ordering required: file already has been exposed. */
413         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
414
415         mm->total_vm = oldmm->total_vm;
416         mm->shared_vm = oldmm->shared_vm;
417         mm->exec_vm = oldmm->exec_vm;
418         mm->stack_vm = oldmm->stack_vm;
419
420         rb_link = &mm->mm_rb.rb_node;
421         rb_parent = NULL;
422         pprev = &mm->mmap;
423         retval = ksm_fork(mm, oldmm);
424         if (retval)
425                 goto out;
426         retval = khugepaged_fork(mm, oldmm);
427         if (retval)
428                 goto out;
429
430         prev = NULL;
431         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
432                 struct file *file;
433
434                 if (mpnt->vm_flags & VM_DONTCOPY) {
435                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
436                                                         -vma_pages(mpnt));
437                         continue;
438                 }
439                 charge = 0;
440                 if (mpnt->vm_flags & VM_ACCOUNT) {
441                         unsigned long len = vma_pages(mpnt);
442
443                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
444                                 goto fail_nomem;
445                         charge = len;
446                 }
447                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
448                 if (!tmp)
449                         goto fail_nomem;
450                 *tmp = *mpnt;
451                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
452                 retval = vma_dup_policy(mpnt, tmp);
453                 if (retval)
454                         goto fail_nomem_policy;
455                 tmp->vm_mm = mm;
456                 if (anon_vma_fork(tmp, mpnt))
457                         goto fail_nomem_anon_vma_fork;
458                 tmp->vm_flags &=
459                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
460                 tmp->vm_next = tmp->vm_prev = NULL;
461                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
462                 file = tmp->vm_file;
463                 if (file) {
464                         struct inode *inode = file_inode(file);
465                         struct address_space *mapping = file->f_mapping;
466
467                         get_file(file);
468                         if (tmp->vm_flags & VM_DENYWRITE)
469                                 atomic_dec(&inode->i_writecount);
470                         i_mmap_lock_write(mapping);
471                         if (tmp->vm_flags & VM_SHARED)
472                                 atomic_inc(&mapping->i_mmap_writable);
473                         flush_dcache_mmap_lock(mapping);
474                         /* insert tmp into the share list, just after mpnt */
475                         vma_interval_tree_insert_after(tmp, mpnt,
476                                         &mapping->i_mmap);
477                         flush_dcache_mmap_unlock(mapping);
478                         i_mmap_unlock_write(mapping);
479                 }
480
481                 /*
482                  * Clear hugetlb-related page reserves for children. This only
483                  * affects MAP_PRIVATE mappings. Faults generated by the child
484                  * are not guaranteed to succeed, even if read-only
485                  */
486                 if (is_vm_hugetlb_page(tmp))
487                         reset_vma_resv_huge_pages(tmp);
488
489                 /*
490                  * Link in the new vma and copy the page table entries.
491                  */
492                 *pprev = tmp;
493                 pprev = &tmp->vm_next;
494                 tmp->vm_prev = prev;
495                 prev = tmp;
496
497                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
498                 rb_link = &tmp->vm_rb.rb_right;
499                 rb_parent = &tmp->vm_rb;
500
501                 mm->map_count++;
502                 retval = copy_page_range(mm, oldmm, mpnt);
503
504                 if (tmp->vm_ops && tmp->vm_ops->open)
505                         tmp->vm_ops->open(tmp);
506
507                 if (retval)
508                         goto out;
509         }
510         /* a new mm has just been created */
511         arch_dup_mmap(oldmm, mm);
512         retval = 0;
513 out:
514         up_write(&mm->mmap_sem);
515         flush_tlb_mm(oldmm);
516         up_write(&oldmm->mmap_sem);
517         uprobe_end_dup_mmap();
518         return retval;
519 fail_nomem_anon_vma_fork:
520         mpol_put(vma_policy(tmp));
521 fail_nomem_policy:
522         kmem_cache_free(vm_area_cachep, tmp);
523 fail_nomem:
524         retval = -ENOMEM;
525         vm_unacct_memory(charge);
526         goto out;
527 }
528
529 static inline int mm_alloc_pgd(struct mm_struct *mm)
530 {
531         mm->pgd = pgd_alloc(mm);
532         if (unlikely(!mm->pgd))
533                 return -ENOMEM;
534         return 0;
535 }
536
537 static inline void mm_free_pgd(struct mm_struct *mm)
538 {
539         pgd_free(mm, mm->pgd);
540 }
541 #else
542 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
543 {
544         down_write(&oldmm->mmap_sem);
545         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
546         up_write(&oldmm->mmap_sem);
547         return 0;
548 }
549 #define mm_alloc_pgd(mm)        (0)
550 #define mm_free_pgd(mm)
551 #endif /* CONFIG_MMU */
552
553 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
554
555 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
556 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
557
558 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
559
560 static int __init coredump_filter_setup(char *s)
561 {
562         default_dump_filter =
563                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
564                 MMF_DUMP_FILTER_MASK;
565         return 1;
566 }
567
568 __setup("coredump_filter=", coredump_filter_setup);
569
570 #include <linux/init_task.h>
571
572 static void mm_init_aio(struct mm_struct *mm)
573 {
574 #ifdef CONFIG_AIO
575         spin_lock_init(&mm->ioctx_lock);
576         mm->ioctx_table = NULL;
577 #endif
578 }
579
580 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
581 {
582 #ifdef CONFIG_MEMCG
583         mm->owner = p;
584 #endif
585 }
586
587 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
588 {
589         mm->mmap = NULL;
590         mm->mm_rb = RB_ROOT;
591         mm->vmacache_seqnum = 0;
592         atomic_set(&mm->mm_users, 1);
593         atomic_set(&mm->mm_count, 1);
594         init_rwsem(&mm->mmap_sem);
595         INIT_LIST_HEAD(&mm->mmlist);
596         mm->core_state = NULL;
597         atomic_long_set(&mm->nr_ptes, 0);
598         mm_nr_pmds_init(mm);
599         mm->map_count = 0;
600         mm->locked_vm = 0;
601         mm->pinned_vm = 0;
602         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
603         spin_lock_init(&mm->page_table_lock);
604         mm_init_cpumask(mm);
605         mm_init_aio(mm);
606         mm_init_owner(mm, p);
607         mmu_notifier_mm_init(mm);
608         clear_tlb_flush_pending(mm);
609 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
610         mm->pmd_huge_pte = NULL;
611 #endif
612
613         if (current->mm) {
614                 mm->flags = current->mm->flags & MMF_INIT_MASK;
615                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
616         } else {
617                 mm->flags = default_dump_filter;
618                 mm->def_flags = 0;
619         }
620
621         if (mm_alloc_pgd(mm))
622                 goto fail_nopgd;
623
624         if (init_new_context(p, mm))
625                 goto fail_nocontext;
626
627         return mm;
628
629 fail_nocontext:
630         mm_free_pgd(mm);
631 fail_nopgd:
632         free_mm(mm);
633         return NULL;
634 }
635
636 static void check_mm(struct mm_struct *mm)
637 {
638         int i;
639
640         for (i = 0; i < NR_MM_COUNTERS; i++) {
641                 long x = atomic_long_read(&mm->rss_stat.count[i]);
642
643                 if (unlikely(x))
644                         printk(KERN_ALERT "BUG: Bad rss-counter state "
645                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
646         }
647
648         if (atomic_long_read(&mm->nr_ptes))
649                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
650                                 atomic_long_read(&mm->nr_ptes));
651         if (mm_nr_pmds(mm))
652                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
653                                 mm_nr_pmds(mm));
654
655 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
656         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
657 #endif
658 }
659
660 /*
661  * Allocate and initialize an mm_struct.
662  */
663 struct mm_struct *mm_alloc(void)
664 {
665         struct mm_struct *mm;
666
667         mm = allocate_mm();
668         if (!mm)
669                 return NULL;
670
671         memset(mm, 0, sizeof(*mm));
672         return mm_init(mm, current);
673 }
674
675 /*
676  * Called when the last reference to the mm
677  * is dropped: either by a lazy thread or by
678  * mmput. Free the page directory and the mm.
679  */
680 void __mmdrop(struct mm_struct *mm)
681 {
682         BUG_ON(mm == &init_mm);
683         mm_free_pgd(mm);
684         destroy_context(mm);
685         mmu_notifier_mm_destroy(mm);
686         check_mm(mm);
687         free_mm(mm);
688 }
689 EXPORT_SYMBOL_GPL(__mmdrop);
690
691 /*
692  * Decrement the use count and release all resources for an mm.
693  */
694 void mmput(struct mm_struct *mm)
695 {
696         might_sleep();
697
698         if (atomic_dec_and_test(&mm->mm_users)) {
699                 uprobe_clear_state(mm);
700                 exit_aio(mm);
701                 ksm_exit(mm);
702                 khugepaged_exit(mm); /* must run before exit_mmap */
703                 exit_mmap(mm);
704                 set_mm_exe_file(mm, NULL);
705                 if (!list_empty(&mm->mmlist)) {
706                         spin_lock(&mmlist_lock);
707                         list_del(&mm->mmlist);
708                         spin_unlock(&mmlist_lock);
709                 }
710                 if (mm->binfmt)
711                         module_put(mm->binfmt->module);
712                 mmdrop(mm);
713         }
714 }
715 EXPORT_SYMBOL_GPL(mmput);
716
717 /**
718  * set_mm_exe_file - change a reference to the mm's executable file
719  *
720  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
721  *
722  * Main users are mmput() and sys_execve(). Callers prevent concurrent
723  * invocations: in mmput() nobody alive left, in execve task is single
724  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
725  * mm->exe_file, but does so without using set_mm_exe_file() in order
726  * to do avoid the need for any locks.
727  */
728 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
729 {
730         struct file *old_exe_file;
731
732         /*
733          * It is safe to dereference the exe_file without RCU as
734          * this function is only called if nobody else can access
735          * this mm -- see comment above for justification.
736          */
737         old_exe_file = rcu_dereference_raw(mm->exe_file);
738
739         if (new_exe_file)
740                 get_file(new_exe_file);
741         rcu_assign_pointer(mm->exe_file, new_exe_file);
742         if (old_exe_file)
743                 fput(old_exe_file);
744 }
745
746 /**
747  * get_mm_exe_file - acquire a reference to the mm's executable file
748  *
749  * Returns %NULL if mm has no associated executable file.
750  * User must release file via fput().
751  */
752 struct file *get_mm_exe_file(struct mm_struct *mm)
753 {
754         struct file *exe_file;
755
756         rcu_read_lock();
757         exe_file = rcu_dereference(mm->exe_file);
758         if (exe_file && !get_file_rcu(exe_file))
759                 exe_file = NULL;
760         rcu_read_unlock();
761         return exe_file;
762 }
763 EXPORT_SYMBOL(get_mm_exe_file);
764
765 /**
766  * get_task_mm - acquire a reference to the task's mm
767  *
768  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
769  * this kernel workthread has transiently adopted a user mm with use_mm,
770  * to do its AIO) is not set and if so returns a reference to it, after
771  * bumping up the use count.  User must release the mm via mmput()
772  * after use.  Typically used by /proc and ptrace.
773  */
774 struct mm_struct *get_task_mm(struct task_struct *task)
775 {
776         struct mm_struct *mm;
777
778         task_lock(task);
779         mm = task->mm;
780         if (mm) {
781                 if (task->flags & PF_KTHREAD)
782                         mm = NULL;
783                 else
784                         atomic_inc(&mm->mm_users);
785         }
786         task_unlock(task);
787         return mm;
788 }
789 EXPORT_SYMBOL_GPL(get_task_mm);
790
791 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
792 {
793         struct mm_struct *mm;
794         int err;
795
796         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
797         if (err)
798                 return ERR_PTR(err);
799
800         mm = get_task_mm(task);
801         if (mm && mm != current->mm &&
802                         !ptrace_may_access(task, mode)) {
803                 mmput(mm);
804                 mm = ERR_PTR(-EACCES);
805         }
806         mutex_unlock(&task->signal->cred_guard_mutex);
807
808         return mm;
809 }
810
811 static void complete_vfork_done(struct task_struct *tsk)
812 {
813         struct completion *vfork;
814
815         task_lock(tsk);
816         vfork = tsk->vfork_done;
817         if (likely(vfork)) {
818                 tsk->vfork_done = NULL;
819                 complete(vfork);
820         }
821         task_unlock(tsk);
822 }
823
824 static int wait_for_vfork_done(struct task_struct *child,
825                                 struct completion *vfork)
826 {
827         int killed;
828
829         freezer_do_not_count();
830         killed = wait_for_completion_killable(vfork);
831         freezer_count();
832
833         if (killed) {
834                 task_lock(child);
835                 child->vfork_done = NULL;
836                 task_unlock(child);
837         }
838
839         put_task_struct(child);
840         return killed;
841 }
842
843 /* Please note the differences between mmput and mm_release.
844  * mmput is called whenever we stop holding onto a mm_struct,
845  * error success whatever.
846  *
847  * mm_release is called after a mm_struct has been removed
848  * from the current process.
849  *
850  * This difference is important for error handling, when we
851  * only half set up a mm_struct for a new process and need to restore
852  * the old one.  Because we mmput the new mm_struct before
853  * restoring the old one. . .
854  * Eric Biederman 10 January 1998
855  */
856 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
857 {
858         /* Get rid of any futexes when releasing the mm */
859 #ifdef CONFIG_FUTEX
860         if (unlikely(tsk->robust_list)) {
861                 exit_robust_list(tsk);
862                 tsk->robust_list = NULL;
863         }
864 #ifdef CONFIG_COMPAT
865         if (unlikely(tsk->compat_robust_list)) {
866                 compat_exit_robust_list(tsk);
867                 tsk->compat_robust_list = NULL;
868         }
869 #endif
870         if (unlikely(!list_empty(&tsk->pi_state_list)))
871                 exit_pi_state_list(tsk);
872 #endif
873
874         uprobe_free_utask(tsk);
875
876         /* Get rid of any cached register state */
877         deactivate_mm(tsk, mm);
878
879         /*
880          * If we're exiting normally, clear a user-space tid field if
881          * requested.  We leave this alone when dying by signal, to leave
882          * the value intact in a core dump, and to save the unnecessary
883          * trouble, say, a killed vfork parent shouldn't touch this mm.
884          * Userland only wants this done for a sys_exit.
885          */
886         if (tsk->clear_child_tid) {
887                 if (!(tsk->flags & PF_SIGNALED) &&
888                     atomic_read(&mm->mm_users) > 1) {
889                         /*
890                          * We don't check the error code - if userspace has
891                          * not set up a proper pointer then tough luck.
892                          */
893                         put_user(0, tsk->clear_child_tid);
894                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
895                                         1, NULL, NULL, 0);
896                 }
897                 tsk->clear_child_tid = NULL;
898         }
899
900         /*
901          * All done, finally we can wake up parent and return this mm to him.
902          * Also kthread_stop() uses this completion for synchronization.
903          */
904         if (tsk->vfork_done)
905                 complete_vfork_done(tsk);
906 }
907
908 /*
909  * Allocate a new mm structure and copy contents from the
910  * mm structure of the passed in task structure.
911  */
912 static struct mm_struct *dup_mm(struct task_struct *tsk)
913 {
914         struct mm_struct *mm, *oldmm = current->mm;
915         int err;
916
917         mm = allocate_mm();
918         if (!mm)
919                 goto fail_nomem;
920
921         memcpy(mm, oldmm, sizeof(*mm));
922
923         if (!mm_init(mm, tsk))
924                 goto fail_nomem;
925
926         err = dup_mmap(mm, oldmm);
927         if (err)
928                 goto free_pt;
929
930         mm->hiwater_rss = get_mm_rss(mm);
931         mm->hiwater_vm = mm->total_vm;
932
933         if (mm->binfmt && !try_module_get(mm->binfmt->module))
934                 goto free_pt;
935
936         return mm;
937
938 free_pt:
939         /* don't put binfmt in mmput, we haven't got module yet */
940         mm->binfmt = NULL;
941         mmput(mm);
942
943 fail_nomem:
944         return NULL;
945 }
946
947 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
948 {
949         struct mm_struct *mm, *oldmm;
950         int retval;
951
952         tsk->min_flt = tsk->maj_flt = 0;
953         tsk->nvcsw = tsk->nivcsw = 0;
954 #ifdef CONFIG_DETECT_HUNG_TASK
955         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
956 #endif
957
958         tsk->mm = NULL;
959         tsk->active_mm = NULL;
960
961         /*
962          * Are we cloning a kernel thread?
963          *
964          * We need to steal a active VM for that..
965          */
966         oldmm = current->mm;
967         if (!oldmm)
968                 return 0;
969
970         /* initialize the new vmacache entries */
971         vmacache_flush(tsk);
972
973         if (clone_flags & CLONE_VM) {
974                 atomic_inc(&oldmm->mm_users);
975                 mm = oldmm;
976                 goto good_mm;
977         }
978
979         retval = -ENOMEM;
980         mm = dup_mm(tsk);
981         if (!mm)
982                 goto fail_nomem;
983
984 good_mm:
985         tsk->mm = mm;
986         tsk->active_mm = mm;
987         return 0;
988
989 fail_nomem:
990         return retval;
991 }
992
993 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
994 {
995         struct fs_struct *fs = current->fs;
996         if (clone_flags & CLONE_FS) {
997                 /* tsk->fs is already what we want */
998                 spin_lock(&fs->lock);
999                 if (fs->in_exec) {
1000                         spin_unlock(&fs->lock);
1001                         return -EAGAIN;
1002                 }
1003                 fs->users++;
1004                 spin_unlock(&fs->lock);
1005                 return 0;
1006         }
1007         tsk->fs = copy_fs_struct(fs);
1008         if (!tsk->fs)
1009                 return -ENOMEM;
1010         return 0;
1011 }
1012
1013 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1014 {
1015         struct files_struct *oldf, *newf;
1016         int error = 0;
1017
1018         /*
1019          * A background process may not have any files ...
1020          */
1021         oldf = current->files;
1022         if (!oldf)
1023                 goto out;
1024
1025         if (clone_flags & CLONE_FILES) {
1026                 atomic_inc(&oldf->count);
1027                 goto out;
1028         }
1029
1030         newf = dup_fd(oldf, &error);
1031         if (!newf)
1032                 goto out;
1033
1034         tsk->files = newf;
1035         error = 0;
1036 out:
1037         return error;
1038 }
1039
1040 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1041 {
1042 #ifdef CONFIG_BLOCK
1043         struct io_context *ioc = current->io_context;
1044         struct io_context *new_ioc;
1045
1046         if (!ioc)
1047                 return 0;
1048         /*
1049          * Share io context with parent, if CLONE_IO is set
1050          */
1051         if (clone_flags & CLONE_IO) {
1052                 ioc_task_link(ioc);
1053                 tsk->io_context = ioc;
1054         } else if (ioprio_valid(ioc->ioprio)) {
1055                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1056                 if (unlikely(!new_ioc))
1057                         return -ENOMEM;
1058
1059                 new_ioc->ioprio = ioc->ioprio;
1060                 put_io_context(new_ioc);
1061         }
1062 #endif
1063         return 0;
1064 }
1065
1066 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1067 {
1068         struct sighand_struct *sig;
1069
1070         if (clone_flags & CLONE_SIGHAND) {
1071                 atomic_inc(&current->sighand->count);
1072                 return 0;
1073         }
1074         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1075         rcu_assign_pointer(tsk->sighand, sig);
1076         if (!sig)
1077                 return -ENOMEM;
1078
1079         atomic_set(&sig->count, 1);
1080         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1081         return 0;
1082 }
1083
1084 void __cleanup_sighand(struct sighand_struct *sighand)
1085 {
1086         if (atomic_dec_and_test(&sighand->count)) {
1087                 signalfd_cleanup(sighand);
1088                 /*
1089                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1090                  * without an RCU grace period, see __lock_task_sighand().
1091                  */
1092                 kmem_cache_free(sighand_cachep, sighand);
1093         }
1094 }
1095
1096 /*
1097  * Initialize POSIX timer handling for a thread group.
1098  */
1099 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1100 {
1101         unsigned long cpu_limit;
1102
1103         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1104         if (cpu_limit != RLIM_INFINITY) {
1105                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1106                 sig->cputimer.running = true;
1107         }
1108
1109         /* The timer lists. */
1110         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1111         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1112         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1113 }
1114
1115 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1116 {
1117         struct signal_struct *sig;
1118
1119         if (clone_flags & CLONE_THREAD)
1120                 return 0;
1121
1122         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1123         tsk->signal = sig;
1124         if (!sig)
1125                 return -ENOMEM;
1126
1127         sig->nr_threads = 1;
1128         atomic_set(&sig->live, 1);
1129         atomic_set(&sig->sigcnt, 1);
1130
1131         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1132         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1133         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1134
1135         init_waitqueue_head(&sig->wait_chldexit);
1136         sig->curr_target = tsk;
1137         init_sigpending(&sig->shared_pending);
1138         INIT_LIST_HEAD(&sig->posix_timers);
1139         seqlock_init(&sig->stats_lock);
1140         prev_cputime_init(&sig->prev_cputime);
1141
1142         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1143         sig->real_timer.function = it_real_fn;
1144
1145         task_lock(current->group_leader);
1146         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1147         task_unlock(current->group_leader);
1148
1149         posix_cpu_timers_init_group(sig);
1150
1151         tty_audit_fork(sig);
1152         sched_autogroup_fork(sig);
1153
1154         sig->oom_score_adj = current->signal->oom_score_adj;
1155         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1156
1157         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1158                                    current->signal->is_child_subreaper;
1159
1160         mutex_init(&sig->cred_guard_mutex);
1161
1162         return 0;
1163 }
1164
1165 static void copy_seccomp(struct task_struct *p)
1166 {
1167 #ifdef CONFIG_SECCOMP
1168         /*
1169          * Must be called with sighand->lock held, which is common to
1170          * all threads in the group. Holding cred_guard_mutex is not
1171          * needed because this new task is not yet running and cannot
1172          * be racing exec.
1173          */
1174         assert_spin_locked(&current->sighand->siglock);
1175
1176         /* Ref-count the new filter user, and assign it. */
1177         get_seccomp_filter(current);
1178         p->seccomp = current->seccomp;
1179
1180         /*
1181          * Explicitly enable no_new_privs here in case it got set
1182          * between the task_struct being duplicated and holding the
1183          * sighand lock. The seccomp state and nnp must be in sync.
1184          */
1185         if (task_no_new_privs(current))
1186                 task_set_no_new_privs(p);
1187
1188         /*
1189          * If the parent gained a seccomp mode after copying thread
1190          * flags and between before we held the sighand lock, we have
1191          * to manually enable the seccomp thread flag here.
1192          */
1193         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1194                 set_tsk_thread_flag(p, TIF_SECCOMP);
1195 #endif
1196 }
1197
1198 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1199 {
1200         current->clear_child_tid = tidptr;
1201
1202         return task_pid_vnr(current);
1203 }
1204
1205 static void rt_mutex_init_task(struct task_struct *p)
1206 {
1207         raw_spin_lock_init(&p->pi_lock);
1208 #ifdef CONFIG_RT_MUTEXES
1209         p->pi_waiters = RB_ROOT;
1210         p->pi_waiters_leftmost = NULL;
1211         p->pi_blocked_on = NULL;
1212 #endif
1213 }
1214
1215 /*
1216  * Initialize POSIX timer handling for a single task.
1217  */
1218 static void posix_cpu_timers_init(struct task_struct *tsk)
1219 {
1220         tsk->cputime_expires.prof_exp = 0;
1221         tsk->cputime_expires.virt_exp = 0;
1222         tsk->cputime_expires.sched_exp = 0;
1223         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1224         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1225         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1226 }
1227
1228 static inline void
1229 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1230 {
1231          task->pids[type].pid = pid;
1232 }
1233
1234 /*
1235  * This creates a new process as a copy of the old one,
1236  * but does not actually start it yet.
1237  *
1238  * It copies the registers, and all the appropriate
1239  * parts of the process environment (as per the clone
1240  * flags). The actual kick-off is left to the caller.
1241  */
1242 static struct task_struct *copy_process(unsigned long clone_flags,
1243                                         unsigned long stack_start,
1244                                         unsigned long stack_size,
1245                                         int __user *child_tidptr,
1246                                         struct pid *pid,
1247                                         int trace,
1248                                         unsigned long tls)
1249 {
1250         int retval;
1251         struct task_struct *p;
1252         void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
1253
1254         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1255                 return ERR_PTR(-EINVAL);
1256
1257         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1258                 return ERR_PTR(-EINVAL);
1259
1260         /*
1261          * Thread groups must share signals as well, and detached threads
1262          * can only be started up within the thread group.
1263          */
1264         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1265                 return ERR_PTR(-EINVAL);
1266
1267         /*
1268          * Shared signal handlers imply shared VM. By way of the above,
1269          * thread groups also imply shared VM. Blocking this case allows
1270          * for various simplifications in other code.
1271          */
1272         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1273                 return ERR_PTR(-EINVAL);
1274
1275         /*
1276          * Siblings of global init remain as zombies on exit since they are
1277          * not reaped by their parent (swapper). To solve this and to avoid
1278          * multi-rooted process trees, prevent global and container-inits
1279          * from creating siblings.
1280          */
1281         if ((clone_flags & CLONE_PARENT) &&
1282                                 current->signal->flags & SIGNAL_UNKILLABLE)
1283                 return ERR_PTR(-EINVAL);
1284
1285         /*
1286          * If the new process will be in a different pid or user namespace
1287          * do not allow it to share a thread group with the forking task.
1288          */
1289         if (clone_flags & CLONE_THREAD) {
1290                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1291                     (task_active_pid_ns(current) !=
1292                                 current->nsproxy->pid_ns_for_children))
1293                         return ERR_PTR(-EINVAL);
1294         }
1295
1296         retval = security_task_create(clone_flags);
1297         if (retval)
1298                 goto fork_out;
1299
1300         retval = -ENOMEM;
1301         p = dup_task_struct(current);
1302         if (!p)
1303                 goto fork_out;
1304
1305         ftrace_graph_init_task(p);
1306
1307         rt_mutex_init_task(p);
1308
1309 #ifdef CONFIG_PROVE_LOCKING
1310         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1311         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1312 #endif
1313         retval = -EAGAIN;
1314         if (atomic_read(&p->real_cred->user->processes) >=
1315                         task_rlimit(p, RLIMIT_NPROC)) {
1316                 if (p->real_cred->user != INIT_USER &&
1317                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1318                         goto bad_fork_free;
1319         }
1320         current->flags &= ~PF_NPROC_EXCEEDED;
1321
1322         retval = copy_creds(p, clone_flags);
1323         if (retval < 0)
1324                 goto bad_fork_free;
1325
1326         /*
1327          * If multiple threads are within copy_process(), then this check
1328          * triggers too late. This doesn't hurt, the check is only there
1329          * to stop root fork bombs.
1330          */
1331         retval = -EAGAIN;
1332         if (nr_threads >= max_threads)
1333                 goto bad_fork_cleanup_count;
1334
1335         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1336         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1337         p->flags |= PF_FORKNOEXEC;
1338         INIT_LIST_HEAD(&p->children);
1339         INIT_LIST_HEAD(&p->sibling);
1340         rcu_copy_process(p);
1341         p->vfork_done = NULL;
1342         spin_lock_init(&p->alloc_lock);
1343
1344         init_sigpending(&p->pending);
1345
1346         p->utime = p->stime = p->gtime = 0;
1347         p->utimescaled = p->stimescaled = 0;
1348         prev_cputime_init(&p->prev_cputime);
1349
1350 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1351         seqlock_init(&p->vtime_seqlock);
1352         p->vtime_snap = 0;
1353         p->vtime_snap_whence = VTIME_SLEEPING;
1354 #endif
1355
1356 #if defined(SPLIT_RSS_COUNTING)
1357         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1358 #endif
1359
1360         p->default_timer_slack_ns = current->timer_slack_ns;
1361
1362         task_io_accounting_init(&p->ioac);
1363         acct_clear_integrals(p);
1364
1365         posix_cpu_timers_init(p);
1366
1367         p->start_time = ktime_get_ns();
1368         p->real_start_time = ktime_get_boot_ns();
1369         p->io_context = NULL;
1370         p->audit_context = NULL;
1371         if (clone_flags & CLONE_THREAD)
1372                 threadgroup_change_begin(current);
1373         cgroup_fork(p);
1374 #ifdef CONFIG_NUMA
1375         p->mempolicy = mpol_dup(p->mempolicy);
1376         if (IS_ERR(p->mempolicy)) {
1377                 retval = PTR_ERR(p->mempolicy);
1378                 p->mempolicy = NULL;
1379                 goto bad_fork_cleanup_threadgroup_lock;
1380         }
1381 #endif
1382 #ifdef CONFIG_CPUSETS
1383         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1384         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1385         seqcount_init(&p->mems_allowed_seq);
1386 #endif
1387 #ifdef CONFIG_TRACE_IRQFLAGS
1388         p->irq_events = 0;
1389         p->hardirqs_enabled = 0;
1390         p->hardirq_enable_ip = 0;
1391         p->hardirq_enable_event = 0;
1392         p->hardirq_disable_ip = _THIS_IP_;
1393         p->hardirq_disable_event = 0;
1394         p->softirqs_enabled = 1;
1395         p->softirq_enable_ip = _THIS_IP_;
1396         p->softirq_enable_event = 0;
1397         p->softirq_disable_ip = 0;
1398         p->softirq_disable_event = 0;
1399         p->hardirq_context = 0;
1400         p->softirq_context = 0;
1401 #endif
1402
1403         p->pagefault_disabled = 0;
1404
1405 #ifdef CONFIG_LOCKDEP
1406         p->lockdep_depth = 0; /* no locks held yet */
1407         p->curr_chain_key = 0;
1408         p->lockdep_recursion = 0;
1409 #endif
1410
1411 #ifdef CONFIG_DEBUG_MUTEXES
1412         p->blocked_on = NULL; /* not blocked yet */
1413 #endif
1414 #ifdef CONFIG_BCACHE
1415         p->sequential_io        = 0;
1416         p->sequential_io_avg    = 0;
1417 #endif
1418
1419         /* Perform scheduler related setup. Assign this task to a CPU. */
1420         retval = sched_fork(clone_flags, p);
1421         if (retval)
1422                 goto bad_fork_cleanup_policy;
1423
1424         retval = perf_event_init_task(p);
1425         if (retval)
1426                 goto bad_fork_cleanup_policy;
1427         retval = audit_alloc(p);
1428         if (retval)
1429                 goto bad_fork_cleanup_perf;
1430         /* copy all the process information */
1431         shm_init_task(p);
1432         retval = copy_semundo(clone_flags, p);
1433         if (retval)
1434                 goto bad_fork_cleanup_audit;
1435         retval = copy_files(clone_flags, p);
1436         if (retval)
1437                 goto bad_fork_cleanup_semundo;
1438         retval = copy_fs(clone_flags, p);
1439         if (retval)
1440                 goto bad_fork_cleanup_files;
1441         retval = copy_sighand(clone_flags, p);
1442         if (retval)
1443                 goto bad_fork_cleanup_fs;
1444         retval = copy_signal(clone_flags, p);
1445         if (retval)
1446                 goto bad_fork_cleanup_sighand;
1447         retval = copy_mm(clone_flags, p);
1448         if (retval)
1449                 goto bad_fork_cleanup_signal;
1450         retval = copy_namespaces(clone_flags, p);
1451         if (retval)
1452                 goto bad_fork_cleanup_mm;
1453         retval = copy_io(clone_flags, p);
1454         if (retval)
1455                 goto bad_fork_cleanup_namespaces;
1456         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1457         if (retval)
1458                 goto bad_fork_cleanup_io;
1459
1460         if (pid != &init_struct_pid) {
1461                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1462                 if (IS_ERR(pid)) {
1463                         retval = PTR_ERR(pid);
1464                         goto bad_fork_cleanup_io;
1465                 }
1466         }
1467
1468         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1469         /*
1470          * Clear TID on mm_release()?
1471          */
1472         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1473 #ifdef CONFIG_BLOCK
1474         p->plug = NULL;
1475 #endif
1476 #ifdef CONFIG_FUTEX
1477         p->robust_list = NULL;
1478 #ifdef CONFIG_COMPAT
1479         p->compat_robust_list = NULL;
1480 #endif
1481         INIT_LIST_HEAD(&p->pi_state_list);
1482         p->pi_state_cache = NULL;
1483 #endif
1484         /*
1485          * sigaltstack should be cleared when sharing the same VM
1486          */
1487         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1488                 p->sas_ss_sp = p->sas_ss_size = 0;
1489
1490         /*
1491          * Syscall tracing and stepping should be turned off in the
1492          * child regardless of CLONE_PTRACE.
1493          */
1494         user_disable_single_step(p);
1495         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1496 #ifdef TIF_SYSCALL_EMU
1497         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1498 #endif
1499         clear_all_latency_tracing(p);
1500
1501         /* ok, now we should be set up.. */
1502         p->pid = pid_nr(pid);
1503         if (clone_flags & CLONE_THREAD) {
1504                 p->exit_signal = -1;
1505                 p->group_leader = current->group_leader;
1506                 p->tgid = current->tgid;
1507         } else {
1508                 if (clone_flags & CLONE_PARENT)
1509                         p->exit_signal = current->group_leader->exit_signal;
1510                 else
1511                         p->exit_signal = (clone_flags & CSIGNAL);
1512                 p->group_leader = p;
1513                 p->tgid = p->pid;
1514         }
1515
1516         p->nr_dirtied = 0;
1517         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1518         p->dirty_paused_when = 0;
1519
1520         p->pdeath_signal = 0;
1521         INIT_LIST_HEAD(&p->thread_group);
1522         p->task_works = NULL;
1523
1524         /*
1525          * Ensure that the cgroup subsystem policies allow the new process to be
1526          * forked. It should be noted the the new process's css_set can be changed
1527          * between here and cgroup_post_fork() if an organisation operation is in
1528          * progress.
1529          */
1530         retval = cgroup_can_fork(p, cgrp_ss_priv);
1531         if (retval)
1532                 goto bad_fork_free_pid;
1533
1534         /*
1535          * Make it visible to the rest of the system, but dont wake it up yet.
1536          * Need tasklist lock for parent etc handling!
1537          */
1538         write_lock_irq(&tasklist_lock);
1539
1540         /* CLONE_PARENT re-uses the old parent */
1541         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1542                 p->real_parent = current->real_parent;
1543                 p->parent_exec_id = current->parent_exec_id;
1544         } else {
1545                 p->real_parent = current;
1546                 p->parent_exec_id = current->self_exec_id;
1547         }
1548
1549         spin_lock(&current->sighand->siglock);
1550
1551         /*
1552          * Copy seccomp details explicitly here, in case they were changed
1553          * before holding sighand lock.
1554          */
1555         copy_seccomp(p);
1556
1557         /*
1558          * Process group and session signals need to be delivered to just the
1559          * parent before the fork or both the parent and the child after the
1560          * fork. Restart if a signal comes in before we add the new process to
1561          * it's process group.
1562          * A fatal signal pending means that current will exit, so the new
1563          * thread can't slip out of an OOM kill (or normal SIGKILL).
1564         */
1565         recalc_sigpending();
1566         if (signal_pending(current)) {
1567                 spin_unlock(&current->sighand->siglock);
1568                 write_unlock_irq(&tasklist_lock);
1569                 retval = -ERESTARTNOINTR;
1570                 goto bad_fork_cancel_cgroup;
1571         }
1572
1573         if (likely(p->pid)) {
1574                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1575
1576                 init_task_pid(p, PIDTYPE_PID, pid);
1577                 if (thread_group_leader(p)) {
1578                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1579                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1580
1581                         if (is_child_reaper(pid)) {
1582                                 ns_of_pid(pid)->child_reaper = p;
1583                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1584                         }
1585
1586                         p->signal->leader_pid = pid;
1587                         p->signal->tty = tty_kref_get(current->signal->tty);
1588                         list_add_tail(&p->sibling, &p->real_parent->children);
1589                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1590                         attach_pid(p, PIDTYPE_PGID);
1591                         attach_pid(p, PIDTYPE_SID);
1592                         __this_cpu_inc(process_counts);
1593                 } else {
1594                         current->signal->nr_threads++;
1595                         atomic_inc(&current->signal->live);
1596                         atomic_inc(&current->signal->sigcnt);
1597                         list_add_tail_rcu(&p->thread_group,
1598                                           &p->group_leader->thread_group);
1599                         list_add_tail_rcu(&p->thread_node,
1600                                           &p->signal->thread_head);
1601                 }
1602                 attach_pid(p, PIDTYPE_PID);
1603                 nr_threads++;
1604         }
1605
1606         total_forks++;
1607         spin_unlock(&current->sighand->siglock);
1608         syscall_tracepoint_update(p);
1609         write_unlock_irq(&tasklist_lock);
1610
1611         proc_fork_connector(p);
1612         cgroup_post_fork(p, cgrp_ss_priv);
1613         if (clone_flags & CLONE_THREAD)
1614                 threadgroup_change_end(current);
1615         perf_event_fork(p);
1616
1617         trace_task_newtask(p, clone_flags);
1618         uprobe_copy_process(p, clone_flags);
1619
1620         return p;
1621
1622 bad_fork_cancel_cgroup:
1623         cgroup_cancel_fork(p, cgrp_ss_priv);
1624 bad_fork_free_pid:
1625         if (pid != &init_struct_pid)
1626                 free_pid(pid);
1627 bad_fork_cleanup_io:
1628         if (p->io_context)
1629                 exit_io_context(p);
1630 bad_fork_cleanup_namespaces:
1631         exit_task_namespaces(p);
1632 bad_fork_cleanup_mm:
1633         if (p->mm)
1634                 mmput(p->mm);
1635 bad_fork_cleanup_signal:
1636         if (!(clone_flags & CLONE_THREAD))
1637                 free_signal_struct(p->signal);
1638 bad_fork_cleanup_sighand:
1639         __cleanup_sighand(p->sighand);
1640 bad_fork_cleanup_fs:
1641         exit_fs(p); /* blocking */
1642 bad_fork_cleanup_files:
1643         exit_files(p); /* blocking */
1644 bad_fork_cleanup_semundo:
1645         exit_sem(p);
1646 bad_fork_cleanup_audit:
1647         audit_free(p);
1648 bad_fork_cleanup_perf:
1649         perf_event_free_task(p);
1650 bad_fork_cleanup_policy:
1651 #ifdef CONFIG_NUMA
1652         mpol_put(p->mempolicy);
1653 bad_fork_cleanup_threadgroup_lock:
1654 #endif
1655         if (clone_flags & CLONE_THREAD)
1656                 threadgroup_change_end(current);
1657         delayacct_tsk_free(p);
1658 bad_fork_cleanup_count:
1659         atomic_dec(&p->cred->user->processes);
1660         exit_creds(p);
1661 bad_fork_free:
1662         free_task(p);
1663 fork_out:
1664         return ERR_PTR(retval);
1665 }
1666
1667 static inline void init_idle_pids(struct pid_link *links)
1668 {
1669         enum pid_type type;
1670
1671         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1672                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1673                 links[type].pid = &init_struct_pid;
1674         }
1675 }
1676
1677 struct task_struct *fork_idle(int cpu)
1678 {
1679         struct task_struct *task;
1680         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1681         if (!IS_ERR(task)) {
1682                 init_idle_pids(task->pids);
1683                 init_idle(task, cpu);
1684         }
1685
1686         return task;
1687 }
1688
1689 /*
1690  *  Ok, this is the main fork-routine.
1691  *
1692  * It copies the process, and if successful kick-starts
1693  * it and waits for it to finish using the VM if required.
1694  */
1695 long _do_fork(unsigned long clone_flags,
1696               unsigned long stack_start,
1697               unsigned long stack_size,
1698               int __user *parent_tidptr,
1699               int __user *child_tidptr,
1700               unsigned long tls)
1701 {
1702         struct task_struct *p;
1703         int trace = 0;
1704         long nr;
1705
1706         /*
1707          * Determine whether and which event to report to ptracer.  When
1708          * called from kernel_thread or CLONE_UNTRACED is explicitly
1709          * requested, no event is reported; otherwise, report if the event
1710          * for the type of forking is enabled.
1711          */
1712         if (!(clone_flags & CLONE_UNTRACED)) {
1713                 if (clone_flags & CLONE_VFORK)
1714                         trace = PTRACE_EVENT_VFORK;
1715                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1716                         trace = PTRACE_EVENT_CLONE;
1717                 else
1718                         trace = PTRACE_EVENT_FORK;
1719
1720                 if (likely(!ptrace_event_enabled(current, trace)))
1721                         trace = 0;
1722         }
1723
1724         p = copy_process(clone_flags, stack_start, stack_size,
1725                          child_tidptr, NULL, trace, tls);
1726         /*
1727          * Do this prior waking up the new thread - the thread pointer
1728          * might get invalid after that point, if the thread exits quickly.
1729          */
1730         if (!IS_ERR(p)) {
1731                 struct completion vfork;
1732                 struct pid *pid;
1733
1734                 trace_sched_process_fork(current, p);
1735
1736                 pid = get_task_pid(p, PIDTYPE_PID);
1737                 nr = pid_vnr(pid);
1738
1739                 if (clone_flags & CLONE_PARENT_SETTID)
1740                         put_user(nr, parent_tidptr);
1741
1742                 if (clone_flags & CLONE_VFORK) {
1743                         p->vfork_done = &vfork;
1744                         init_completion(&vfork);
1745                         get_task_struct(p);
1746                 }
1747
1748                 wake_up_new_task(p);
1749
1750                 /* forking complete and child started to run, tell ptracer */
1751                 if (unlikely(trace))
1752                         ptrace_event_pid(trace, pid);
1753
1754                 if (clone_flags & CLONE_VFORK) {
1755                         if (!wait_for_vfork_done(p, &vfork))
1756                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1757                 }
1758
1759                 put_pid(pid);
1760         } else {
1761                 nr = PTR_ERR(p);
1762         }
1763         return nr;
1764 }
1765
1766 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1767 /* For compatibility with architectures that call do_fork directly rather than
1768  * using the syscall entry points below. */
1769 long do_fork(unsigned long clone_flags,
1770               unsigned long stack_start,
1771               unsigned long stack_size,
1772               int __user *parent_tidptr,
1773               int __user *child_tidptr)
1774 {
1775         return _do_fork(clone_flags, stack_start, stack_size,
1776                         parent_tidptr, child_tidptr, 0);
1777 }
1778 #endif
1779
1780 /*
1781  * Create a kernel thread.
1782  */
1783 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1784 {
1785         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1786                 (unsigned long)arg, NULL, NULL, 0);
1787 }
1788
1789 #ifdef __ARCH_WANT_SYS_FORK
1790 SYSCALL_DEFINE0(fork)
1791 {
1792 #ifdef CONFIG_MMU
1793         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1794 #else
1795         /* can not support in nommu mode */
1796         return -EINVAL;
1797 #endif
1798 }
1799 #endif
1800
1801 #ifdef __ARCH_WANT_SYS_VFORK
1802 SYSCALL_DEFINE0(vfork)
1803 {
1804         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1805                         0, NULL, NULL, 0);
1806 }
1807 #endif
1808
1809 #ifdef __ARCH_WANT_SYS_CLONE
1810 #ifdef CONFIG_CLONE_BACKWARDS
1811 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1812                  int __user *, parent_tidptr,
1813                  unsigned long, tls,
1814                  int __user *, child_tidptr)
1815 #elif defined(CONFIG_CLONE_BACKWARDS2)
1816 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1817                  int __user *, parent_tidptr,
1818                  int __user *, child_tidptr,
1819                  unsigned long, tls)
1820 #elif defined(CONFIG_CLONE_BACKWARDS3)
1821 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1822                 int, stack_size,
1823                 int __user *, parent_tidptr,
1824                 int __user *, child_tidptr,
1825                 unsigned long, tls)
1826 #else
1827 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1828                  int __user *, parent_tidptr,
1829                  int __user *, child_tidptr,
1830                  unsigned long, tls)
1831 #endif
1832 {
1833         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1834 }
1835 #endif
1836
1837 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1838 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1839 #endif
1840
1841 static void sighand_ctor(void *data)
1842 {
1843         struct sighand_struct *sighand = data;
1844
1845         spin_lock_init(&sighand->siglock);
1846         init_waitqueue_head(&sighand->signalfd_wqh);
1847 }
1848
1849 void __init proc_caches_init(void)
1850 {
1851         sighand_cachep = kmem_cache_create("sighand_cache",
1852                         sizeof(struct sighand_struct), 0,
1853                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1854                         SLAB_NOTRACK, sighand_ctor);
1855         signal_cachep = kmem_cache_create("signal_cache",
1856                         sizeof(struct signal_struct), 0,
1857                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1858         files_cachep = kmem_cache_create("files_cache",
1859                         sizeof(struct files_struct), 0,
1860                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1861         fs_cachep = kmem_cache_create("fs_cache",
1862                         sizeof(struct fs_struct), 0,
1863                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1864         /*
1865          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1866          * whole struct cpumask for the OFFSTACK case. We could change
1867          * this to *only* allocate as much of it as required by the
1868          * maximum number of CPU's we can ever have.  The cpumask_allocation
1869          * is at the end of the structure, exactly for that reason.
1870          */
1871         mm_cachep = kmem_cache_create("mm_struct",
1872                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1873                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1874         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1875         mmap_init();
1876         nsproxy_cache_init();
1877 }
1878
1879 /*
1880  * Check constraints on flags passed to the unshare system call.
1881  */
1882 static int check_unshare_flags(unsigned long unshare_flags)
1883 {
1884         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1885                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1886                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1887                                 CLONE_NEWUSER|CLONE_NEWPID))
1888                 return -EINVAL;
1889         /*
1890          * Not implemented, but pretend it works if there is nothing
1891          * to unshare.  Note that unsharing the address space or the
1892          * signal handlers also need to unshare the signal queues (aka
1893          * CLONE_THREAD).
1894          */
1895         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1896                 if (!thread_group_empty(current))
1897                         return -EINVAL;
1898         }
1899         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1900                 if (atomic_read(&current->sighand->count) > 1)
1901                         return -EINVAL;
1902         }
1903         if (unshare_flags & CLONE_VM) {
1904                 if (!current_is_single_threaded())
1905                         return -EINVAL;
1906         }
1907
1908         return 0;
1909 }
1910
1911 /*
1912  * Unshare the filesystem structure if it is being shared
1913  */
1914 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1915 {
1916         struct fs_struct *fs = current->fs;
1917
1918         if (!(unshare_flags & CLONE_FS) || !fs)
1919                 return 0;
1920
1921         /* don't need lock here; in the worst case we'll do useless copy */
1922         if (fs->users == 1)
1923                 return 0;
1924
1925         *new_fsp = copy_fs_struct(fs);
1926         if (!*new_fsp)
1927                 return -ENOMEM;
1928
1929         return 0;
1930 }
1931
1932 /*
1933  * Unshare file descriptor table if it is being shared
1934  */
1935 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1936 {
1937         struct files_struct *fd = current->files;
1938         int error = 0;
1939
1940         if ((unshare_flags & CLONE_FILES) &&
1941             (fd && atomic_read(&fd->count) > 1)) {
1942                 *new_fdp = dup_fd(fd, &error);
1943                 if (!*new_fdp)
1944                         return error;
1945         }
1946
1947         return 0;
1948 }
1949
1950 /*
1951  * unshare allows a process to 'unshare' part of the process
1952  * context which was originally shared using clone.  copy_*
1953  * functions used by do_fork() cannot be used here directly
1954  * because they modify an inactive task_struct that is being
1955  * constructed. Here we are modifying the current, active,
1956  * task_struct.
1957  */
1958 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1959 {
1960         struct fs_struct *fs, *new_fs = NULL;
1961         struct files_struct *fd, *new_fd = NULL;
1962         struct cred *new_cred = NULL;
1963         struct nsproxy *new_nsproxy = NULL;
1964         int do_sysvsem = 0;
1965         int err;
1966
1967         /*
1968          * If unsharing a user namespace must also unshare the thread group
1969          * and unshare the filesystem root and working directories.
1970          */
1971         if (unshare_flags & CLONE_NEWUSER)
1972                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1973         /*
1974          * If unsharing vm, must also unshare signal handlers.
1975          */
1976         if (unshare_flags & CLONE_VM)
1977                 unshare_flags |= CLONE_SIGHAND;
1978         /*
1979          * If unsharing a signal handlers, must also unshare the signal queues.
1980          */
1981         if (unshare_flags & CLONE_SIGHAND)
1982                 unshare_flags |= CLONE_THREAD;
1983         /*
1984          * If unsharing namespace, must also unshare filesystem information.
1985          */
1986         if (unshare_flags & CLONE_NEWNS)
1987                 unshare_flags |= CLONE_FS;
1988
1989         err = check_unshare_flags(unshare_flags);
1990         if (err)
1991                 goto bad_unshare_out;
1992         /*
1993          * CLONE_NEWIPC must also detach from the undolist: after switching
1994          * to a new ipc namespace, the semaphore arrays from the old
1995          * namespace are unreachable.
1996          */
1997         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1998                 do_sysvsem = 1;
1999         err = unshare_fs(unshare_flags, &new_fs);
2000         if (err)
2001                 goto bad_unshare_out;
2002         err = unshare_fd(unshare_flags, &new_fd);
2003         if (err)
2004                 goto bad_unshare_cleanup_fs;
2005         err = unshare_userns(unshare_flags, &new_cred);
2006         if (err)
2007                 goto bad_unshare_cleanup_fd;
2008         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2009                                          new_cred, new_fs);
2010         if (err)
2011                 goto bad_unshare_cleanup_cred;
2012
2013         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2014                 if (do_sysvsem) {
2015                         /*
2016                          * CLONE_SYSVSEM is equivalent to sys_exit().
2017                          */
2018                         exit_sem(current);
2019                 }
2020                 if (unshare_flags & CLONE_NEWIPC) {
2021                         /* Orphan segments in old ns (see sem above). */
2022                         exit_shm(current);
2023                         shm_init_task(current);
2024                 }
2025
2026                 if (new_nsproxy)
2027                         switch_task_namespaces(current, new_nsproxy);
2028
2029                 task_lock(current);
2030
2031                 if (new_fs) {
2032                         fs = current->fs;
2033                         spin_lock(&fs->lock);
2034                         current->fs = new_fs;
2035                         if (--fs->users)
2036                                 new_fs = NULL;
2037                         else
2038                                 new_fs = fs;
2039                         spin_unlock(&fs->lock);
2040                 }
2041
2042                 if (new_fd) {
2043                         fd = current->files;
2044                         current->files = new_fd;
2045                         new_fd = fd;
2046                 }
2047
2048                 task_unlock(current);
2049
2050                 if (new_cred) {
2051                         /* Install the new user namespace */
2052                         commit_creds(new_cred);
2053                         new_cred = NULL;
2054                 }
2055         }
2056
2057 bad_unshare_cleanup_cred:
2058         if (new_cred)
2059                 put_cred(new_cred);
2060 bad_unshare_cleanup_fd:
2061         if (new_fd)
2062                 put_files_struct(new_fd);
2063
2064 bad_unshare_cleanup_fs:
2065         if (new_fs)
2066                 free_fs_struct(new_fs);
2067
2068 bad_unshare_out:
2069         return err;
2070 }
2071
2072 /*
2073  *      Helper to unshare the files of the current task.
2074  *      We don't want to expose copy_files internals to
2075  *      the exec layer of the kernel.
2076  */
2077
2078 int unshare_files(struct files_struct **displaced)
2079 {
2080         struct task_struct *task = current;
2081         struct files_struct *copy = NULL;
2082         int error;
2083
2084         error = unshare_fd(CLONE_FILES, &copy);
2085         if (error || !copy) {
2086                 *displaced = NULL;
2087                 return error;
2088         }
2089         *displaced = task->files;
2090         task_lock(task);
2091         task->files = copy;
2092         task_unlock(task);
2093         return 0;
2094 }
2095
2096 int sysctl_max_threads(struct ctl_table *table, int write,
2097                        void __user *buffer, size_t *lenp, loff_t *ppos)
2098 {
2099         struct ctl_table t;
2100         int ret;
2101         int threads = max_threads;
2102         int min = MIN_THREADS;
2103         int max = MAX_THREADS;
2104
2105         t = *table;
2106         t.data = &threads;
2107         t.extra1 = &min;
2108         t.extra2 = &max;
2109
2110         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2111         if (ret || !write)
2112                 return ret;
2113
2114         set_max_threads(threads);
2115
2116         return 0;
2117 }