eb7b63334009833d9fa7ccede3f4a3426066bbfb
[linux-3.10.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/futex.h>
41 #include <linux/compat.h>
42 #include <linux/kthread.h>
43 #include <linux/task_io_accounting_ops.h>
44 #include <linux/rcupdate.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/audit.h>
48 #include <linux/memcontrol.h>
49 #include <linux/ftrace.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/ksm.h>
53 #include <linux/acct.h>
54 #include <linux/tsacct_kern.h>
55 #include <linux/cn_proc.h>
56 #include <linux/freezer.h>
57 #include <linux/delayacct.h>
58 #include <linux/taskstats_kern.h>
59 #include <linux/random.h>
60 #include <linux/tty.h>
61 #include <linux/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_event.h>
65 #include <linux/posix-timers.h>
66 #include <linux/user-return-notifier.h>
67 #include <linux/oom.h>
68 #include <linux/khugepaged.h>
69 #include <linux/signalfd.h>
70 #include <linux/uprobes.h>
71
72 #include <asm/pgtable.h>
73 #include <asm/pgalloc.h>
74 #include <asm/uaccess.h>
75 #include <asm/mmu_context.h>
76 #include <asm/cacheflush.h>
77 #include <asm/tlbflush.h>
78
79 #include <trace/events/sched.h>
80
81 #define CREATE_TRACE_POINTS
82 #include <trace/events/task.h>
83
84 /*
85  * Protected counters by write_lock_irq(&tasklist_lock)
86  */
87 unsigned long total_forks;      /* Handle normal Linux uptimes. */
88 int nr_threads;                 /* The idle threads do not count.. */
89
90 int max_threads;                /* tunable limit on nr_threads */
91
92 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
93
94 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
95
96 #ifdef CONFIG_PROVE_RCU
97 int lockdep_tasklist_lock_is_held(void)
98 {
99         return lockdep_is_held(&tasklist_lock);
100 }
101 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
102 #endif /* #ifdef CONFIG_PROVE_RCU */
103
104 int nr_processes(void)
105 {
106         int cpu;
107         int total = 0;
108
109         for_each_possible_cpu(cpu)
110                 total += per_cpu(process_counts, cpu);
111
112         return total;
113 }
114
115 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
116 # define alloc_task_struct_node(node)           \
117                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
118 # define free_task_struct(tsk)                  \
119                 kmem_cache_free(task_struct_cachep, (tsk))
120 static struct kmem_cache *task_struct_cachep;
121 #endif
122
123 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
124 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
125                                                   int node)
126 {
127 #ifdef CONFIG_DEBUG_STACK_USAGE
128         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
129 #else
130         gfp_t mask = GFP_KERNEL;
131 #endif
132         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
133
134         return page ? page_address(page) : NULL;
135 }
136
137 static inline void free_thread_info(struct thread_info *ti)
138 {
139         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
140 }
141 #endif
142
143 /* SLAB cache for signal_struct structures (tsk->signal) */
144 static struct kmem_cache *signal_cachep;
145
146 /* SLAB cache for sighand_struct structures (tsk->sighand) */
147 struct kmem_cache *sighand_cachep;
148
149 /* SLAB cache for files_struct structures (tsk->files) */
150 struct kmem_cache *files_cachep;
151
152 /* SLAB cache for fs_struct structures (tsk->fs) */
153 struct kmem_cache *fs_cachep;
154
155 /* SLAB cache for vm_area_struct structures */
156 struct kmem_cache *vm_area_cachep;
157
158 /* SLAB cache for mm_struct structures (tsk->mm) */
159 static struct kmem_cache *mm_cachep;
160
161 static void account_kernel_stack(struct thread_info *ti, int account)
162 {
163         struct zone *zone = page_zone(virt_to_page(ti));
164
165         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
166 }
167
168 void free_task(struct task_struct *tsk)
169 {
170         account_kernel_stack(tsk->stack, -1);
171         free_thread_info(tsk->stack);
172         rt_mutex_debug_task_free(tsk);
173         ftrace_graph_exit_task(tsk);
174         free_task_struct(tsk);
175 }
176 EXPORT_SYMBOL(free_task);
177
178 static inline void free_signal_struct(struct signal_struct *sig)
179 {
180         taskstats_tgid_free(sig);
181         sched_autogroup_exit(sig);
182         kmem_cache_free(signal_cachep, sig);
183 }
184
185 static inline void put_signal_struct(struct signal_struct *sig)
186 {
187         if (atomic_dec_and_test(&sig->sigcnt))
188                 free_signal_struct(sig);
189 }
190
191 void __put_task_struct(struct task_struct *tsk)
192 {
193         WARN_ON(!tsk->exit_state);
194         WARN_ON(atomic_read(&tsk->usage));
195         WARN_ON(tsk == current);
196
197         exit_creds(tsk);
198         delayacct_tsk_free(tsk);
199         put_signal_struct(tsk->signal);
200
201         if (!profile_handoff_task(tsk))
202                 free_task(tsk);
203 }
204 EXPORT_SYMBOL_GPL(__put_task_struct);
205
206 /*
207  * macro override instead of weak attribute alias, to workaround
208  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
209  */
210 #ifndef arch_task_cache_init
211 #define arch_task_cache_init()
212 #endif
213
214 void __init fork_init(unsigned long mempages)
215 {
216 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
217 #ifndef ARCH_MIN_TASKALIGN
218 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
219 #endif
220         /* create a slab on which task_structs can be allocated */
221         task_struct_cachep =
222                 kmem_cache_create("task_struct", sizeof(struct task_struct),
223                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
224 #endif
225
226         /* do the arch specific task caches init */
227         arch_task_cache_init();
228
229         /*
230          * The default maximum number of threads is set to a safe
231          * value: the thread structures can take up at most half
232          * of memory.
233          */
234         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
235
236         /*
237          * we need to allow at least 20 threads to boot a system
238          */
239         if (max_threads < 20)
240                 max_threads = 20;
241
242         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
243         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
244         init_task.signal->rlim[RLIMIT_SIGPENDING] =
245                 init_task.signal->rlim[RLIMIT_NPROC];
246 }
247
248 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
249                                                struct task_struct *src)
250 {
251         *dst = *src;
252         return 0;
253 }
254
255 static struct task_struct *dup_task_struct(struct task_struct *orig)
256 {
257         struct task_struct *tsk;
258         struct thread_info *ti;
259         unsigned long *stackend;
260         int node = tsk_fork_get_node(orig);
261         int err;
262
263         prepare_to_copy(orig);
264
265         tsk = alloc_task_struct_node(node);
266         if (!tsk)
267                 return NULL;
268
269         ti = alloc_thread_info_node(tsk, node);
270         if (!ti) {
271                 free_task_struct(tsk);
272                 return NULL;
273         }
274
275         err = arch_dup_task_struct(tsk, orig);
276         if (err)
277                 goto out;
278
279         tsk->stack = ti;
280
281         setup_thread_stack(tsk, orig);
282         clear_user_return_notifier(tsk);
283         clear_tsk_need_resched(tsk);
284         stackend = end_of_stack(tsk);
285         *stackend = STACK_END_MAGIC;    /* for overflow detection */
286
287 #ifdef CONFIG_CC_STACKPROTECTOR
288         tsk->stack_canary = get_random_int();
289 #endif
290
291         /*
292          * One for us, one for whoever does the "release_task()" (usually
293          * parent)
294          */
295         atomic_set(&tsk->usage, 2);
296 #ifdef CONFIG_BLK_DEV_IO_TRACE
297         tsk->btrace_seq = 0;
298 #endif
299         tsk->splice_pipe = NULL;
300
301         account_kernel_stack(ti, 1);
302
303         return tsk;
304
305 out:
306         free_thread_info(ti);
307         free_task_struct(tsk);
308         return NULL;
309 }
310
311 #ifdef CONFIG_MMU
312 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
313 {
314         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
315         struct rb_node **rb_link, *rb_parent;
316         int retval;
317         unsigned long charge;
318         struct mempolicy *pol;
319
320         down_write(&oldmm->mmap_sem);
321         flush_cache_dup_mm(oldmm);
322         /*
323          * Not linked in yet - no deadlock potential:
324          */
325         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
326
327         mm->locked_vm = 0;
328         mm->mmap = NULL;
329         mm->mmap_cache = NULL;
330         mm->free_area_cache = oldmm->mmap_base;
331         mm->cached_hole_size = ~0UL;
332         mm->map_count = 0;
333         cpumask_clear(mm_cpumask(mm));
334         mm->mm_rb = RB_ROOT;
335         rb_link = &mm->mm_rb.rb_node;
336         rb_parent = NULL;
337         pprev = &mm->mmap;
338         retval = ksm_fork(mm, oldmm);
339         if (retval)
340                 goto out;
341         retval = khugepaged_fork(mm, oldmm);
342         if (retval)
343                 goto out;
344
345         prev = NULL;
346         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
347                 struct file *file;
348
349                 if (mpnt->vm_flags & VM_DONTCOPY) {
350                         long pages = vma_pages(mpnt);
351                         mm->total_vm -= pages;
352                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
353                                                                 -pages);
354                         continue;
355                 }
356                 charge = 0;
357                 if (mpnt->vm_flags & VM_ACCOUNT) {
358                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
359                         if (security_vm_enough_memory(len))
360                                 goto fail_nomem;
361                         charge = len;
362                 }
363                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
364                 if (!tmp)
365                         goto fail_nomem;
366                 *tmp = *mpnt;
367                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
368                 pol = mpol_dup(vma_policy(mpnt));
369                 retval = PTR_ERR(pol);
370                 if (IS_ERR(pol))
371                         goto fail_nomem_policy;
372                 vma_set_policy(tmp, pol);
373                 tmp->vm_mm = mm;
374                 if (anon_vma_fork(tmp, mpnt))
375                         goto fail_nomem_anon_vma_fork;
376                 tmp->vm_flags &= ~VM_LOCKED;
377                 tmp->vm_next = tmp->vm_prev = NULL;
378                 file = tmp->vm_file;
379                 if (file) {
380                         struct inode *inode = file->f_path.dentry->d_inode;
381                         struct address_space *mapping = file->f_mapping;
382
383                         get_file(file);
384                         if (tmp->vm_flags & VM_DENYWRITE)
385                                 atomic_dec(&inode->i_writecount);
386                         mutex_lock(&mapping->i_mmap_mutex);
387                         if (tmp->vm_flags & VM_SHARED)
388                                 mapping->i_mmap_writable++;
389                         flush_dcache_mmap_lock(mapping);
390                         /* insert tmp into the share list, just after mpnt */
391                         vma_prio_tree_add(tmp, mpnt);
392                         flush_dcache_mmap_unlock(mapping);
393                         mutex_unlock(&mapping->i_mmap_mutex);
394                 }
395
396                 /*
397                  * Clear hugetlb-related page reserves for children. This only
398                  * affects MAP_PRIVATE mappings. Faults generated by the child
399                  * are not guaranteed to succeed, even if read-only
400                  */
401                 if (is_vm_hugetlb_page(tmp))
402                         reset_vma_resv_huge_pages(tmp);
403
404                 /*
405                  * Link in the new vma and copy the page table entries.
406                  */
407                 *pprev = tmp;
408                 pprev = &tmp->vm_next;
409                 tmp->vm_prev = prev;
410                 prev = tmp;
411
412                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
413                 rb_link = &tmp->vm_rb.rb_right;
414                 rb_parent = &tmp->vm_rb;
415
416                 mm->map_count++;
417                 retval = copy_page_range(mm, oldmm, mpnt);
418
419                 if (tmp->vm_ops && tmp->vm_ops->open)
420                         tmp->vm_ops->open(tmp);
421
422                 if (retval)
423                         goto out;
424         }
425         /* a new mm has just been created */
426         arch_dup_mmap(oldmm, mm);
427         retval = 0;
428 out:
429         up_write(&mm->mmap_sem);
430         flush_tlb_mm(oldmm);
431         up_write(&oldmm->mmap_sem);
432         return retval;
433 fail_nomem_anon_vma_fork:
434         mpol_put(pol);
435 fail_nomem_policy:
436         kmem_cache_free(vm_area_cachep, tmp);
437 fail_nomem:
438         retval = -ENOMEM;
439         vm_unacct_memory(charge);
440         goto out;
441 }
442
443 static inline int mm_alloc_pgd(struct mm_struct *mm)
444 {
445         mm->pgd = pgd_alloc(mm);
446         if (unlikely(!mm->pgd))
447                 return -ENOMEM;
448         return 0;
449 }
450
451 static inline void mm_free_pgd(struct mm_struct *mm)
452 {
453         pgd_free(mm, mm->pgd);
454 }
455 #else
456 #define dup_mmap(mm, oldmm)     (0)
457 #define mm_alloc_pgd(mm)        (0)
458 #define mm_free_pgd(mm)
459 #endif /* CONFIG_MMU */
460
461 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
462
463 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
464 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
465
466 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
467
468 static int __init coredump_filter_setup(char *s)
469 {
470         default_dump_filter =
471                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
472                 MMF_DUMP_FILTER_MASK;
473         return 1;
474 }
475
476 __setup("coredump_filter=", coredump_filter_setup);
477
478 #include <linux/init_task.h>
479
480 static void mm_init_aio(struct mm_struct *mm)
481 {
482 #ifdef CONFIG_AIO
483         spin_lock_init(&mm->ioctx_lock);
484         INIT_HLIST_HEAD(&mm->ioctx_list);
485 #endif
486 }
487
488 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
489 {
490         atomic_set(&mm->mm_users, 1);
491         atomic_set(&mm->mm_count, 1);
492         init_rwsem(&mm->mmap_sem);
493         INIT_LIST_HEAD(&mm->mmlist);
494         mm->flags = (current->mm) ?
495                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
496         mm->core_state = NULL;
497         mm->nr_ptes = 0;
498         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
499         spin_lock_init(&mm->page_table_lock);
500         mm->free_area_cache = TASK_UNMAPPED_BASE;
501         mm->cached_hole_size = ~0UL;
502         mm_init_aio(mm);
503         mm_init_owner(mm, p);
504
505         if (likely(!mm_alloc_pgd(mm))) {
506                 mm->def_flags = 0;
507                 mmu_notifier_mm_init(mm);
508                 return mm;
509         }
510
511         free_mm(mm);
512         return NULL;
513 }
514
515 /*
516  * Allocate and initialize an mm_struct.
517  */
518 struct mm_struct *mm_alloc(void)
519 {
520         struct mm_struct *mm;
521
522         mm = allocate_mm();
523         if (!mm)
524                 return NULL;
525
526         memset(mm, 0, sizeof(*mm));
527         mm_init_cpumask(mm);
528         return mm_init(mm, current);
529 }
530
531 /*
532  * Called when the last reference to the mm
533  * is dropped: either by a lazy thread or by
534  * mmput. Free the page directory and the mm.
535  */
536 void __mmdrop(struct mm_struct *mm)
537 {
538         BUG_ON(mm == &init_mm);
539         mm_free_pgd(mm);
540         destroy_context(mm);
541         mmu_notifier_mm_destroy(mm);
542 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
543         VM_BUG_ON(mm->pmd_huge_pte);
544 #endif
545         free_mm(mm);
546 }
547 EXPORT_SYMBOL_GPL(__mmdrop);
548
549 /*
550  * Decrement the use count and release all resources for an mm.
551  */
552 void mmput(struct mm_struct *mm)
553 {
554         might_sleep();
555
556         if (atomic_dec_and_test(&mm->mm_users)) {
557                 exit_aio(mm);
558                 ksm_exit(mm);
559                 khugepaged_exit(mm); /* must run before exit_mmap */
560                 exit_mmap(mm);
561                 set_mm_exe_file(mm, NULL);
562                 if (!list_empty(&mm->mmlist)) {
563                         spin_lock(&mmlist_lock);
564                         list_del(&mm->mmlist);
565                         spin_unlock(&mmlist_lock);
566                 }
567                 put_swap_token(mm);
568                 if (mm->binfmt)
569                         module_put(mm->binfmt->module);
570                 mmdrop(mm);
571         }
572 }
573 EXPORT_SYMBOL_GPL(mmput);
574
575 /*
576  * We added or removed a vma mapping the executable. The vmas are only mapped
577  * during exec and are not mapped with the mmap system call.
578  * Callers must hold down_write() on the mm's mmap_sem for these
579  */
580 void added_exe_file_vma(struct mm_struct *mm)
581 {
582         mm->num_exe_file_vmas++;
583 }
584
585 void removed_exe_file_vma(struct mm_struct *mm)
586 {
587         mm->num_exe_file_vmas--;
588         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
589                 fput(mm->exe_file);
590                 mm->exe_file = NULL;
591         }
592
593 }
594
595 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
596 {
597         if (new_exe_file)
598                 get_file(new_exe_file);
599         if (mm->exe_file)
600                 fput(mm->exe_file);
601         mm->exe_file = new_exe_file;
602         mm->num_exe_file_vmas = 0;
603 }
604
605 struct file *get_mm_exe_file(struct mm_struct *mm)
606 {
607         struct file *exe_file;
608
609         /* We need mmap_sem to protect against races with removal of
610          * VM_EXECUTABLE vmas */
611         down_read(&mm->mmap_sem);
612         exe_file = mm->exe_file;
613         if (exe_file)
614                 get_file(exe_file);
615         up_read(&mm->mmap_sem);
616         return exe_file;
617 }
618
619 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
620 {
621         /* It's safe to write the exe_file pointer without exe_file_lock because
622          * this is called during fork when the task is not yet in /proc */
623         newmm->exe_file = get_mm_exe_file(oldmm);
624 }
625
626 /**
627  * get_task_mm - acquire a reference to the task's mm
628  *
629  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
630  * this kernel workthread has transiently adopted a user mm with use_mm,
631  * to do its AIO) is not set and if so returns a reference to it, after
632  * bumping up the use count.  User must release the mm via mmput()
633  * after use.  Typically used by /proc and ptrace.
634  */
635 struct mm_struct *get_task_mm(struct task_struct *task)
636 {
637         struct mm_struct *mm;
638
639         task_lock(task);
640         mm = task->mm;
641         if (mm) {
642                 if (task->flags & PF_KTHREAD)
643                         mm = NULL;
644                 else
645                         atomic_inc(&mm->mm_users);
646         }
647         task_unlock(task);
648         return mm;
649 }
650 EXPORT_SYMBOL_GPL(get_task_mm);
651
652 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
653 {
654         struct mm_struct *mm;
655         int err;
656
657         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
658         if (err)
659                 return ERR_PTR(err);
660
661         mm = get_task_mm(task);
662         if (mm && mm != current->mm &&
663                         !ptrace_may_access(task, mode)) {
664                 mmput(mm);
665                 mm = ERR_PTR(-EACCES);
666         }
667         mutex_unlock(&task->signal->cred_guard_mutex);
668
669         return mm;
670 }
671
672 /* Please note the differences between mmput and mm_release.
673  * mmput is called whenever we stop holding onto a mm_struct,
674  * error success whatever.
675  *
676  * mm_release is called after a mm_struct has been removed
677  * from the current process.
678  *
679  * This difference is important for error handling, when we
680  * only half set up a mm_struct for a new process and need to restore
681  * the old one.  Because we mmput the new mm_struct before
682  * restoring the old one. . .
683  * Eric Biederman 10 January 1998
684  */
685 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
686 {
687         struct completion *vfork_done = tsk->vfork_done;
688
689         /* Get rid of any futexes when releasing the mm */
690 #ifdef CONFIG_FUTEX
691         if (unlikely(tsk->robust_list)) {
692                 exit_robust_list(tsk);
693                 tsk->robust_list = NULL;
694         }
695 #ifdef CONFIG_COMPAT
696         if (unlikely(tsk->compat_robust_list)) {
697                 compat_exit_robust_list(tsk);
698                 tsk->compat_robust_list = NULL;
699         }
700 #endif
701         if (unlikely(!list_empty(&tsk->pi_state_list)))
702                 exit_pi_state_list(tsk);
703 #endif
704
705         uprobe_free_utask(tsk);
706
707         /* Get rid of any cached register state */
708         deactivate_mm(tsk, mm);
709
710         /* notify parent sleeping on vfork() */
711         if (vfork_done) {
712                 tsk->vfork_done = NULL;
713                 complete(vfork_done);
714         }
715
716         /*
717          * If we're exiting normally, clear a user-space tid field if
718          * requested.  We leave this alone when dying by signal, to leave
719          * the value intact in a core dump, and to save the unnecessary
720          * trouble otherwise.  Userland only wants this done for a sys_exit.
721          */
722         if (tsk->clear_child_tid) {
723                 if (!(tsk->flags & PF_SIGNALED) &&
724                     atomic_read(&mm->mm_users) > 1) {
725                         /*
726                          * We don't check the error code - if userspace has
727                          * not set up a proper pointer then tough luck.
728                          */
729                         put_user(0, tsk->clear_child_tid);
730                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
731                                         1, NULL, NULL, 0);
732                 }
733                 tsk->clear_child_tid = NULL;
734         }
735 }
736
737 /*
738  * Allocate a new mm structure and copy contents from the
739  * mm structure of the passed in task structure.
740  */
741 struct mm_struct *dup_mm(struct task_struct *tsk)
742 {
743         struct mm_struct *mm, *oldmm = current->mm;
744         int err;
745
746         if (!oldmm)
747                 return NULL;
748
749         mm = allocate_mm();
750         if (!mm)
751                 goto fail_nomem;
752
753         memcpy(mm, oldmm, sizeof(*mm));
754         mm_init_cpumask(mm);
755
756         /* Initializing for Swap token stuff */
757         mm->token_priority = 0;
758         mm->last_interval = 0;
759
760 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
761         mm->pmd_huge_pte = NULL;
762 #endif
763
764         if (!mm_init(mm, tsk))
765                 goto fail_nomem;
766
767         if (init_new_context(tsk, mm))
768                 goto fail_nocontext;
769
770         dup_mm_exe_file(oldmm, mm);
771
772         err = dup_mmap(mm, oldmm);
773         if (err)
774                 goto free_pt;
775
776         mm->hiwater_rss = get_mm_rss(mm);
777         mm->hiwater_vm = mm->total_vm;
778
779         if (mm->binfmt && !try_module_get(mm->binfmt->module))
780                 goto free_pt;
781
782         return mm;
783
784 free_pt:
785         /* don't put binfmt in mmput, we haven't got module yet */
786         mm->binfmt = NULL;
787         mmput(mm);
788
789 fail_nomem:
790         return NULL;
791
792 fail_nocontext:
793         /*
794          * If init_new_context() failed, we cannot use mmput() to free the mm
795          * because it calls destroy_context()
796          */
797         mm_free_pgd(mm);
798         free_mm(mm);
799         return NULL;
800 }
801
802 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
803 {
804         struct mm_struct *mm, *oldmm;
805         int retval;
806
807         tsk->min_flt = tsk->maj_flt = 0;
808         tsk->nvcsw = tsk->nivcsw = 0;
809 #ifdef CONFIG_DETECT_HUNG_TASK
810         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
811 #endif
812
813         tsk->mm = NULL;
814         tsk->active_mm = NULL;
815
816         /*
817          * Are we cloning a kernel thread?
818          *
819          * We need to steal a active VM for that..
820          */
821         oldmm = current->mm;
822         if (!oldmm)
823                 return 0;
824
825         if (clone_flags & CLONE_VM) {
826                 atomic_inc(&oldmm->mm_users);
827                 mm = oldmm;
828                 goto good_mm;
829         }
830
831         retval = -ENOMEM;
832         mm = dup_mm(tsk);
833         if (!mm)
834                 goto fail_nomem;
835
836 good_mm:
837         /* Initializing for Swap token stuff */
838         mm->token_priority = 0;
839         mm->last_interval = 0;
840
841         tsk->mm = mm;
842         tsk->active_mm = mm;
843         return 0;
844
845 fail_nomem:
846         return retval;
847 }
848
849 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
850 {
851         struct fs_struct *fs = current->fs;
852         if (clone_flags & CLONE_FS) {
853                 /* tsk->fs is already what we want */
854                 spin_lock(&fs->lock);
855                 if (fs->in_exec) {
856                         spin_unlock(&fs->lock);
857                         return -EAGAIN;
858                 }
859                 fs->users++;
860                 spin_unlock(&fs->lock);
861                 return 0;
862         }
863         tsk->fs = copy_fs_struct(fs);
864         if (!tsk->fs)
865                 return -ENOMEM;
866         return 0;
867 }
868
869 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
870 {
871         struct files_struct *oldf, *newf;
872         int error = 0;
873
874         /*
875          * A background process may not have any files ...
876          */
877         oldf = current->files;
878         if (!oldf)
879                 goto out;
880
881         if (clone_flags & CLONE_FILES) {
882                 atomic_inc(&oldf->count);
883                 goto out;
884         }
885
886         newf = dup_fd(oldf, &error);
887         if (!newf)
888                 goto out;
889
890         tsk->files = newf;
891         error = 0;
892 out:
893         return error;
894 }
895
896 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
897 {
898 #ifdef CONFIG_BLOCK
899         struct io_context *ioc = current->io_context;
900         struct io_context *new_ioc;
901
902         if (!ioc)
903                 return 0;
904         /*
905          * Share io context with parent, if CLONE_IO is set
906          */
907         if (clone_flags & CLONE_IO) {
908                 tsk->io_context = ioc_task_link(ioc);
909                 if (unlikely(!tsk->io_context))
910                         return -ENOMEM;
911         } else if (ioprio_valid(ioc->ioprio)) {
912                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
913                 if (unlikely(!new_ioc))
914                         return -ENOMEM;
915
916                 new_ioc->ioprio = ioc->ioprio;
917                 put_io_context(new_ioc);
918         }
919 #endif
920         return 0;
921 }
922
923 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
924 {
925         struct sighand_struct *sig;
926
927         if (clone_flags & CLONE_SIGHAND) {
928                 atomic_inc(&current->sighand->count);
929                 return 0;
930         }
931         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
932         rcu_assign_pointer(tsk->sighand, sig);
933         if (!sig)
934                 return -ENOMEM;
935         atomic_set(&sig->count, 1);
936         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
937         return 0;
938 }
939
940 void __cleanup_sighand(struct sighand_struct *sighand)
941 {
942         if (atomic_dec_and_test(&sighand->count)) {
943                 signalfd_cleanup(sighand);
944                 kmem_cache_free(sighand_cachep, sighand);
945         }
946 }
947
948
949 /*
950  * Initialize POSIX timer handling for a thread group.
951  */
952 static void posix_cpu_timers_init_group(struct signal_struct *sig)
953 {
954         unsigned long cpu_limit;
955
956         /* Thread group counters. */
957         thread_group_cputime_init(sig);
958
959         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
960         if (cpu_limit != RLIM_INFINITY) {
961                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
962                 sig->cputimer.running = 1;
963         }
964
965         /* The timer lists. */
966         INIT_LIST_HEAD(&sig->cpu_timers[0]);
967         INIT_LIST_HEAD(&sig->cpu_timers[1]);
968         INIT_LIST_HEAD(&sig->cpu_timers[2]);
969 }
970
971 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
972 {
973         struct signal_struct *sig;
974
975         if (clone_flags & CLONE_THREAD)
976                 return 0;
977
978         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
979         tsk->signal = sig;
980         if (!sig)
981                 return -ENOMEM;
982
983         sig->nr_threads = 1;
984         atomic_set(&sig->live, 1);
985         atomic_set(&sig->sigcnt, 1);
986         init_waitqueue_head(&sig->wait_chldexit);
987         if (clone_flags & CLONE_NEWPID)
988                 sig->flags |= SIGNAL_UNKILLABLE;
989         sig->curr_target = tsk;
990         init_sigpending(&sig->shared_pending);
991         INIT_LIST_HEAD(&sig->posix_timers);
992
993         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
994         sig->real_timer.function = it_real_fn;
995
996         task_lock(current->group_leader);
997         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
998         task_unlock(current->group_leader);
999
1000         posix_cpu_timers_init_group(sig);
1001
1002         tty_audit_fork(sig);
1003         sched_autogroup_fork(sig);
1004
1005 #ifdef CONFIG_CGROUPS
1006         init_rwsem(&sig->group_rwsem);
1007 #endif
1008
1009         sig->oom_adj = current->signal->oom_adj;
1010         sig->oom_score_adj = current->signal->oom_score_adj;
1011         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1012
1013         mutex_init(&sig->cred_guard_mutex);
1014
1015         return 0;
1016 }
1017
1018 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1019 {
1020         unsigned long new_flags = p->flags;
1021
1022         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1023         new_flags |= PF_FORKNOEXEC;
1024         new_flags |= PF_STARTING;
1025         p->flags = new_flags;
1026 }
1027
1028 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1029 {
1030         current->clear_child_tid = tidptr;
1031
1032         return task_pid_vnr(current);
1033 }
1034
1035 static void rt_mutex_init_task(struct task_struct *p)
1036 {
1037         raw_spin_lock_init(&p->pi_lock);
1038 #ifdef CONFIG_RT_MUTEXES
1039         plist_head_init(&p->pi_waiters);
1040         p->pi_blocked_on = NULL;
1041 #endif
1042 }
1043
1044 #ifdef CONFIG_MM_OWNER
1045 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1046 {
1047         mm->owner = p;
1048 }
1049 #endif /* CONFIG_MM_OWNER */
1050
1051 /*
1052  * Initialize POSIX timer handling for a single task.
1053  */
1054 static void posix_cpu_timers_init(struct task_struct *tsk)
1055 {
1056         tsk->cputime_expires.prof_exp = 0;
1057         tsk->cputime_expires.virt_exp = 0;
1058         tsk->cputime_expires.sched_exp = 0;
1059         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1060         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1061         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1062 }
1063
1064 /*
1065  * This creates a new process as a copy of the old one,
1066  * but does not actually start it yet.
1067  *
1068  * It copies the registers, and all the appropriate
1069  * parts of the process environment (as per the clone
1070  * flags). The actual kick-off is left to the caller.
1071  */
1072 static struct task_struct *copy_process(unsigned long clone_flags,
1073                                         unsigned long stack_start,
1074                                         struct pt_regs *regs,
1075                                         unsigned long stack_size,
1076                                         int __user *child_tidptr,
1077                                         struct pid *pid,
1078                                         int trace)
1079 {
1080         int retval;
1081         struct task_struct *p;
1082         int cgroup_callbacks_done = 0;
1083
1084         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1085                 return ERR_PTR(-EINVAL);
1086
1087         /*
1088          * Thread groups must share signals as well, and detached threads
1089          * can only be started up within the thread group.
1090          */
1091         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1092                 return ERR_PTR(-EINVAL);
1093
1094         /*
1095          * Shared signal handlers imply shared VM. By way of the above,
1096          * thread groups also imply shared VM. Blocking this case allows
1097          * for various simplifications in other code.
1098          */
1099         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1100                 return ERR_PTR(-EINVAL);
1101
1102         /*
1103          * Siblings of global init remain as zombies on exit since they are
1104          * not reaped by their parent (swapper). To solve this and to avoid
1105          * multi-rooted process trees, prevent global and container-inits
1106          * from creating siblings.
1107          */
1108         if ((clone_flags & CLONE_PARENT) &&
1109                                 current->signal->flags & SIGNAL_UNKILLABLE)
1110                 return ERR_PTR(-EINVAL);
1111
1112         retval = security_task_create(clone_flags);
1113         if (retval)
1114                 goto fork_out;
1115
1116         retval = -ENOMEM;
1117         p = dup_task_struct(current);
1118         if (!p)
1119                 goto fork_out;
1120
1121         ftrace_graph_init_task(p);
1122
1123         rt_mutex_init_task(p);
1124
1125 #ifdef CONFIG_PROVE_LOCKING
1126         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1127         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1128 #endif
1129         retval = -EAGAIN;
1130         if (atomic_read(&p->real_cred->user->processes) >=
1131                         task_rlimit(p, RLIMIT_NPROC)) {
1132                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1133                     p->real_cred->user != INIT_USER)
1134                         goto bad_fork_free;
1135         }
1136         current->flags &= ~PF_NPROC_EXCEEDED;
1137
1138         retval = copy_creds(p, clone_flags);
1139         if (retval < 0)
1140                 goto bad_fork_free;
1141
1142         /*
1143          * If multiple threads are within copy_process(), then this check
1144          * triggers too late. This doesn't hurt, the check is only there
1145          * to stop root fork bombs.
1146          */
1147         retval = -EAGAIN;
1148         if (nr_threads >= max_threads)
1149                 goto bad_fork_cleanup_count;
1150
1151         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1152                 goto bad_fork_cleanup_count;
1153
1154         p->did_exec = 0;
1155         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1156         copy_flags(clone_flags, p);
1157         INIT_LIST_HEAD(&p->children);
1158         INIT_LIST_HEAD(&p->sibling);
1159         rcu_copy_process(p);
1160         p->vfork_done = NULL;
1161         spin_lock_init(&p->alloc_lock);
1162
1163         init_sigpending(&p->pending);
1164
1165         p->utime = p->stime = p->gtime = 0;
1166         p->utimescaled = p->stimescaled = 0;
1167 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1168         p->prev_utime = p->prev_stime = 0;
1169 #endif
1170 #if defined(SPLIT_RSS_COUNTING)
1171         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1172 #endif
1173
1174         p->default_timer_slack_ns = current->timer_slack_ns;
1175
1176         task_io_accounting_init(&p->ioac);
1177         acct_clear_integrals(p);
1178
1179         posix_cpu_timers_init(p);
1180
1181         do_posix_clock_monotonic_gettime(&p->start_time);
1182         p->real_start_time = p->start_time;
1183         monotonic_to_bootbased(&p->real_start_time);
1184         p->io_context = NULL;
1185         p->audit_context = NULL;
1186         if (clone_flags & CLONE_THREAD)
1187                 threadgroup_change_begin(current);
1188         cgroup_fork(p);
1189 #ifdef CONFIG_NUMA
1190         p->mempolicy = mpol_dup(p->mempolicy);
1191         if (IS_ERR(p->mempolicy)) {
1192                 retval = PTR_ERR(p->mempolicy);
1193                 p->mempolicy = NULL;
1194                 goto bad_fork_cleanup_cgroup;
1195         }
1196         mpol_fix_fork_child_flag(p);
1197 #endif
1198 #ifdef CONFIG_CPUSETS
1199         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1200         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1201 #endif
1202 #ifdef CONFIG_TRACE_IRQFLAGS
1203         p->irq_events = 0;
1204 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1205         p->hardirqs_enabled = 1;
1206 #else
1207         p->hardirqs_enabled = 0;
1208 #endif
1209         p->hardirq_enable_ip = 0;
1210         p->hardirq_enable_event = 0;
1211         p->hardirq_disable_ip = _THIS_IP_;
1212         p->hardirq_disable_event = 0;
1213         p->softirqs_enabled = 1;
1214         p->softirq_enable_ip = _THIS_IP_;
1215         p->softirq_enable_event = 0;
1216         p->softirq_disable_ip = 0;
1217         p->softirq_disable_event = 0;
1218         p->hardirq_context = 0;
1219         p->softirq_context = 0;
1220 #endif
1221 #ifdef CONFIG_LOCKDEP
1222         p->lockdep_depth = 0; /* no locks held yet */
1223         p->curr_chain_key = 0;
1224         p->lockdep_recursion = 0;
1225 #endif
1226
1227 #ifdef CONFIG_DEBUG_MUTEXES
1228         p->blocked_on = NULL; /* not blocked yet */
1229 #endif
1230 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1231         p->memcg_batch.do_batch = 0;
1232         p->memcg_batch.memcg = NULL;
1233 #endif
1234
1235         /* Perform scheduler related setup. Assign this task to a CPU. */
1236         sched_fork(p);
1237
1238         retval = perf_event_init_task(p);
1239         if (retval)
1240                 goto bad_fork_cleanup_policy;
1241         retval = audit_alloc(p);
1242         if (retval)
1243                 goto bad_fork_cleanup_policy;
1244         /* copy all the process information */
1245         retval = copy_semundo(clone_flags, p);
1246         if (retval)
1247                 goto bad_fork_cleanup_audit;
1248         retval = copy_files(clone_flags, p);
1249         if (retval)
1250                 goto bad_fork_cleanup_semundo;
1251         retval = copy_fs(clone_flags, p);
1252         if (retval)
1253                 goto bad_fork_cleanup_files;
1254         retval = copy_sighand(clone_flags, p);
1255         if (retval)
1256                 goto bad_fork_cleanup_fs;
1257         retval = copy_signal(clone_flags, p);
1258         if (retval)
1259                 goto bad_fork_cleanup_sighand;
1260         retval = copy_mm(clone_flags, p);
1261         if (retval)
1262                 goto bad_fork_cleanup_signal;
1263         retval = copy_namespaces(clone_flags, p);
1264         if (retval)
1265                 goto bad_fork_cleanup_mm;
1266         retval = copy_io(clone_flags, p);
1267         if (retval)
1268                 goto bad_fork_cleanup_namespaces;
1269         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1270         if (retval)
1271                 goto bad_fork_cleanup_io;
1272
1273         if (pid != &init_struct_pid) {
1274                 retval = -ENOMEM;
1275                 pid = alloc_pid(p->nsproxy->pid_ns);
1276                 if (!pid)
1277                         goto bad_fork_cleanup_io;
1278         }
1279
1280         p->pid = pid_nr(pid);
1281         p->tgid = p->pid;
1282         if (clone_flags & CLONE_THREAD)
1283                 p->tgid = current->tgid;
1284
1285         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1286         /*
1287          * Clear TID on mm_release()?
1288          */
1289         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1290 #ifdef CONFIG_BLOCK
1291         p->plug = NULL;
1292 #endif
1293 #ifdef CONFIG_FUTEX
1294         p->robust_list = NULL;
1295 #ifdef CONFIG_COMPAT
1296         p->compat_robust_list = NULL;
1297 #endif
1298         INIT_LIST_HEAD(&p->pi_state_list);
1299         p->pi_state_cache = NULL;
1300 #endif
1301         uprobe_copy_process(p);
1302         /*
1303          * sigaltstack should be cleared when sharing the same VM
1304          */
1305         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1306                 p->sas_ss_sp = p->sas_ss_size = 0;
1307
1308         /*
1309          * Syscall tracing and stepping should be turned off in the
1310          * child regardless of CLONE_PTRACE.
1311          */
1312         user_disable_single_step(p);
1313         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1314 #ifdef TIF_SYSCALL_EMU
1315         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1316 #endif
1317         clear_all_latency_tracing(p);
1318
1319         /* ok, now we should be set up.. */
1320         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1321         p->pdeath_signal = 0;
1322         p->exit_state = 0;
1323
1324         p->nr_dirtied = 0;
1325         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1326         p->dirty_paused_when = 0;
1327
1328         /*
1329          * Ok, make it visible to the rest of the system.
1330          * We dont wake it up yet.
1331          */
1332         p->group_leader = p;
1333         INIT_LIST_HEAD(&p->thread_group);
1334
1335         /* Now that the task is set up, run cgroup callbacks if
1336          * necessary. We need to run them before the task is visible
1337          * on the tasklist. */
1338         cgroup_fork_callbacks(p);
1339         cgroup_callbacks_done = 1;
1340
1341         /* Need tasklist lock for parent etc handling! */
1342         write_lock_irq(&tasklist_lock);
1343
1344         /* CLONE_PARENT re-uses the old parent */
1345         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1346                 p->real_parent = current->real_parent;
1347                 p->parent_exec_id = current->parent_exec_id;
1348         } else {
1349                 p->real_parent = current;
1350                 p->parent_exec_id = current->self_exec_id;
1351         }
1352
1353         spin_lock(&current->sighand->siglock);
1354
1355         /*
1356          * Process group and session signals need to be delivered to just the
1357          * parent before the fork or both the parent and the child after the
1358          * fork. Restart if a signal comes in before we add the new process to
1359          * it's process group.
1360          * A fatal signal pending means that current will exit, so the new
1361          * thread can't slip out of an OOM kill (or normal SIGKILL).
1362         */
1363         recalc_sigpending();
1364         if (signal_pending(current)) {
1365                 spin_unlock(&current->sighand->siglock);
1366                 write_unlock_irq(&tasklist_lock);
1367                 retval = -ERESTARTNOINTR;
1368                 goto bad_fork_free_pid;
1369         }
1370
1371         if (clone_flags & CLONE_THREAD) {
1372                 current->signal->nr_threads++;
1373                 atomic_inc(&current->signal->live);
1374                 atomic_inc(&current->signal->sigcnt);
1375                 p->group_leader = current->group_leader;
1376                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1377         }
1378
1379         if (likely(p->pid)) {
1380                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1381
1382                 if (thread_group_leader(p)) {
1383                         if (is_child_reaper(pid))
1384                                 p->nsproxy->pid_ns->child_reaper = p;
1385
1386                         p->signal->leader_pid = pid;
1387                         p->signal->tty = tty_kref_get(current->signal->tty);
1388                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1389                         attach_pid(p, PIDTYPE_SID, task_session(current));
1390                         list_add_tail(&p->sibling, &p->real_parent->children);
1391                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1392                         __this_cpu_inc(process_counts);
1393                 }
1394                 attach_pid(p, PIDTYPE_PID, pid);
1395                 nr_threads++;
1396         }
1397
1398         total_forks++;
1399         spin_unlock(&current->sighand->siglock);
1400         write_unlock_irq(&tasklist_lock);
1401         proc_fork_connector(p);
1402         cgroup_post_fork(p);
1403         if (clone_flags & CLONE_THREAD)
1404                 threadgroup_change_end(current);
1405         perf_event_fork(p);
1406
1407         trace_task_newtask(p, clone_flags);
1408
1409         return p;
1410
1411 bad_fork_free_pid:
1412         if (pid != &init_struct_pid)
1413                 free_pid(pid);
1414 bad_fork_cleanup_io:
1415         if (p->io_context)
1416                 exit_io_context(p);
1417 bad_fork_cleanup_namespaces:
1418         exit_task_namespaces(p);
1419 bad_fork_cleanup_mm:
1420         if (p->mm)
1421                 mmput(p->mm);
1422 bad_fork_cleanup_signal:
1423         if (!(clone_flags & CLONE_THREAD))
1424                 free_signal_struct(p->signal);
1425 bad_fork_cleanup_sighand:
1426         __cleanup_sighand(p->sighand);
1427 bad_fork_cleanup_fs:
1428         exit_fs(p); /* blocking */
1429 bad_fork_cleanup_files:
1430         exit_files(p); /* blocking */
1431 bad_fork_cleanup_semundo:
1432         exit_sem(p);
1433 bad_fork_cleanup_audit:
1434         audit_free(p);
1435 bad_fork_cleanup_policy:
1436         perf_event_free_task(p);
1437 #ifdef CONFIG_NUMA
1438         mpol_put(p->mempolicy);
1439 bad_fork_cleanup_cgroup:
1440 #endif
1441         if (clone_flags & CLONE_THREAD)
1442                 threadgroup_change_end(current);
1443         cgroup_exit(p, cgroup_callbacks_done);
1444         delayacct_tsk_free(p);
1445         module_put(task_thread_info(p)->exec_domain->module);
1446 bad_fork_cleanup_count:
1447         atomic_dec(&p->cred->user->processes);
1448         exit_creds(p);
1449 bad_fork_free:
1450         free_task(p);
1451 fork_out:
1452         return ERR_PTR(retval);
1453 }
1454
1455 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1456 {
1457         memset(regs, 0, sizeof(struct pt_regs));
1458         return regs;
1459 }
1460
1461 static inline void init_idle_pids(struct pid_link *links)
1462 {
1463         enum pid_type type;
1464
1465         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1466                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1467                 links[type].pid = &init_struct_pid;
1468         }
1469 }
1470
1471 struct task_struct * __cpuinit fork_idle(int cpu)
1472 {
1473         struct task_struct *task;
1474         struct pt_regs regs;
1475
1476         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1477                             &init_struct_pid, 0);
1478         if (!IS_ERR(task)) {
1479                 init_idle_pids(task->pids);
1480                 init_idle(task, cpu);
1481         }
1482
1483         return task;
1484 }
1485
1486 /*
1487  *  Ok, this is the main fork-routine.
1488  *
1489  * It copies the process, and if successful kick-starts
1490  * it and waits for it to finish using the VM if required.
1491  */
1492 long do_fork(unsigned long clone_flags,
1493               unsigned long stack_start,
1494               struct pt_regs *regs,
1495               unsigned long stack_size,
1496               int __user *parent_tidptr,
1497               int __user *child_tidptr)
1498 {
1499         struct task_struct *p;
1500         int trace = 0;
1501         long nr;
1502
1503         /*
1504          * Do some preliminary argument and permissions checking before we
1505          * actually start allocating stuff
1506          */
1507         if (clone_flags & CLONE_NEWUSER) {
1508                 if (clone_flags & CLONE_THREAD)
1509                         return -EINVAL;
1510                 /* hopefully this check will go away when userns support is
1511                  * complete
1512                  */
1513                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1514                                 !capable(CAP_SETGID))
1515                         return -EPERM;
1516         }
1517
1518         /*
1519          * Determine whether and which event to report to ptracer.  When
1520          * called from kernel_thread or CLONE_UNTRACED is explicitly
1521          * requested, no event is reported; otherwise, report if the event
1522          * for the type of forking is enabled.
1523          */
1524         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1525                 if (clone_flags & CLONE_VFORK)
1526                         trace = PTRACE_EVENT_VFORK;
1527                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1528                         trace = PTRACE_EVENT_CLONE;
1529                 else
1530                         trace = PTRACE_EVENT_FORK;
1531
1532                 if (likely(!ptrace_event_enabled(current, trace)))
1533                         trace = 0;
1534         }
1535
1536         p = copy_process(clone_flags, stack_start, regs, stack_size,
1537                          child_tidptr, NULL, trace);
1538         /*
1539          * Do this prior waking up the new thread - the thread pointer
1540          * might get invalid after that point, if the thread exits quickly.
1541          */
1542         if (!IS_ERR(p)) {
1543                 struct completion vfork;
1544
1545                 trace_sched_process_fork(current, p);
1546
1547                 nr = task_pid_vnr(p);
1548
1549                 if (clone_flags & CLONE_PARENT_SETTID)
1550                         put_user(nr, parent_tidptr);
1551
1552                 if (clone_flags & CLONE_VFORK) {
1553                         p->vfork_done = &vfork;
1554                         init_completion(&vfork);
1555                 }
1556
1557                 /*
1558                  * We set PF_STARTING at creation in case tracing wants to
1559                  * use this to distinguish a fully live task from one that
1560                  * hasn't finished SIGSTOP raising yet.  Now we clear it
1561                  * and set the child going.
1562                  */
1563                 p->flags &= ~PF_STARTING;
1564
1565                 wake_up_new_task(p);
1566
1567                 /* forking complete and child started to run, tell ptracer */
1568                 if (unlikely(trace))
1569                         ptrace_event(trace, nr);
1570
1571                 if (clone_flags & CLONE_VFORK) {
1572                         freezer_do_not_count();
1573                         wait_for_completion(&vfork);
1574                         freezer_count();
1575                         ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1576                 }
1577         } else {
1578                 nr = PTR_ERR(p);
1579         }
1580         return nr;
1581 }
1582
1583 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1584 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1585 #endif
1586
1587 static void sighand_ctor(void *data)
1588 {
1589         struct sighand_struct *sighand = data;
1590
1591         spin_lock_init(&sighand->siglock);
1592         init_waitqueue_head(&sighand->signalfd_wqh);
1593 }
1594
1595 void __init proc_caches_init(void)
1596 {
1597         sighand_cachep = kmem_cache_create("sighand_cache",
1598                         sizeof(struct sighand_struct), 0,
1599                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1600                         SLAB_NOTRACK, sighand_ctor);
1601         signal_cachep = kmem_cache_create("signal_cache",
1602                         sizeof(struct signal_struct), 0,
1603                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1604         files_cachep = kmem_cache_create("files_cache",
1605                         sizeof(struct files_struct), 0,
1606                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1607         fs_cachep = kmem_cache_create("fs_cache",
1608                         sizeof(struct fs_struct), 0,
1609                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1610         /*
1611          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1612          * whole struct cpumask for the OFFSTACK case. We could change
1613          * this to *only* allocate as much of it as required by the
1614          * maximum number of CPU's we can ever have.  The cpumask_allocation
1615          * is at the end of the structure, exactly for that reason.
1616          */
1617         mm_cachep = kmem_cache_create("mm_struct",
1618                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1619                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1620         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1621         mmap_init();
1622         nsproxy_cache_init();
1623 }
1624
1625 /*
1626  * Check constraints on flags passed to the unshare system call.
1627  */
1628 static int check_unshare_flags(unsigned long unshare_flags)
1629 {
1630         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1631                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1632                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1633                 return -EINVAL;
1634         /*
1635          * Not implemented, but pretend it works if there is nothing to
1636          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1637          * needs to unshare vm.
1638          */
1639         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1640                 /* FIXME: get_task_mm() increments ->mm_users */
1641                 if (atomic_read(&current->mm->mm_users) > 1)
1642                         return -EINVAL;
1643         }
1644
1645         return 0;
1646 }
1647
1648 /*
1649  * Unshare the filesystem structure if it is being shared
1650  */
1651 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1652 {
1653         struct fs_struct *fs = current->fs;
1654
1655         if (!(unshare_flags & CLONE_FS) || !fs)
1656                 return 0;
1657
1658         /* don't need lock here; in the worst case we'll do useless copy */
1659         if (fs->users == 1)
1660                 return 0;
1661
1662         *new_fsp = copy_fs_struct(fs);
1663         if (!*new_fsp)
1664                 return -ENOMEM;
1665
1666         return 0;
1667 }
1668
1669 /*
1670  * Unshare file descriptor table if it is being shared
1671  */
1672 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1673 {
1674         struct files_struct *fd = current->files;
1675         int error = 0;
1676
1677         if ((unshare_flags & CLONE_FILES) &&
1678             (fd && atomic_read(&fd->count) > 1)) {
1679                 *new_fdp = dup_fd(fd, &error);
1680                 if (!*new_fdp)
1681                         return error;
1682         }
1683
1684         return 0;
1685 }
1686
1687 /*
1688  * unshare allows a process to 'unshare' part of the process
1689  * context which was originally shared using clone.  copy_*
1690  * functions used by do_fork() cannot be used here directly
1691  * because they modify an inactive task_struct that is being
1692  * constructed. Here we are modifying the current, active,
1693  * task_struct.
1694  */
1695 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1696 {
1697         struct fs_struct *fs, *new_fs = NULL;
1698         struct files_struct *fd, *new_fd = NULL;
1699         struct nsproxy *new_nsproxy = NULL;
1700         int do_sysvsem = 0;
1701         int err;
1702
1703         err = check_unshare_flags(unshare_flags);
1704         if (err)
1705                 goto bad_unshare_out;
1706
1707         /*
1708          * If unsharing namespace, must also unshare filesystem information.
1709          */
1710         if (unshare_flags & CLONE_NEWNS)
1711                 unshare_flags |= CLONE_FS;
1712         /*
1713          * CLONE_NEWIPC must also detach from the undolist: after switching
1714          * to a new ipc namespace, the semaphore arrays from the old
1715          * namespace are unreachable.
1716          */
1717         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1718                 do_sysvsem = 1;
1719         err = unshare_fs(unshare_flags, &new_fs);
1720         if (err)
1721                 goto bad_unshare_out;
1722         err = unshare_fd(unshare_flags, &new_fd);
1723         if (err)
1724                 goto bad_unshare_cleanup_fs;
1725         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1726         if (err)
1727                 goto bad_unshare_cleanup_fd;
1728
1729         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1730                 if (do_sysvsem) {
1731                         /*
1732                          * CLONE_SYSVSEM is equivalent to sys_exit().
1733                          */
1734                         exit_sem(current);
1735                 }
1736
1737                 if (new_nsproxy) {
1738                         switch_task_namespaces(current, new_nsproxy);
1739                         new_nsproxy = NULL;
1740                 }
1741
1742                 task_lock(current);
1743
1744                 if (new_fs) {
1745                         fs = current->fs;
1746                         spin_lock(&fs->lock);
1747                         current->fs = new_fs;
1748                         if (--fs->users)
1749                                 new_fs = NULL;
1750                         else
1751                                 new_fs = fs;
1752                         spin_unlock(&fs->lock);
1753                 }
1754
1755                 if (new_fd) {
1756                         fd = current->files;
1757                         current->files = new_fd;
1758                         new_fd = fd;
1759                 }
1760
1761                 task_unlock(current);
1762         }
1763
1764         if (new_nsproxy)
1765                 put_nsproxy(new_nsproxy);
1766
1767 bad_unshare_cleanup_fd:
1768         if (new_fd)
1769                 put_files_struct(new_fd);
1770
1771 bad_unshare_cleanup_fs:
1772         if (new_fs)
1773                 free_fs_struct(new_fs);
1774
1775 bad_unshare_out:
1776         return err;
1777 }
1778
1779 /*
1780  *      Helper to unshare the files of the current task.
1781  *      We don't want to expose copy_files internals to
1782  *      the exec layer of the kernel.
1783  */
1784
1785 int unshare_files(struct files_struct **displaced)
1786 {
1787         struct task_struct *task = current;
1788         struct files_struct *copy = NULL;
1789         int error;
1790
1791         error = unshare_fd(CLONE_FILES, &copy);
1792         if (error || !copy) {
1793                 *displaced = NULL;
1794                 return error;
1795         }
1796         *displaced = task->files;
1797         task_lock(task);
1798         task->files = copy;
1799         task_unlock(task);
1800         return 0;
1801 }