mm: NUMA aware alloc_thread_info_node()
[linux-2.6.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/tracehook.h>
41 #include <linux/futex.h>
42 #include <linux/compat.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/proc_fs.h>
62 #include <linux/blkdev.h>
63 #include <linux/fs_struct.h>
64 #include <linux/magic.h>
65 #include <linux/perf_event.h>
66 #include <linux/posix-timers.h>
67 #include <linux/user-return-notifier.h>
68 #include <linux/oom.h>
69 #include <linux/khugepaged.h>
70
71 #include <asm/pgtable.h>
72 #include <asm/pgalloc.h>
73 #include <asm/uaccess.h>
74 #include <asm/mmu_context.h>
75 #include <asm/cacheflush.h>
76 #include <asm/tlbflush.h>
77
78 #include <trace/events/sched.h>
79
80 /*
81  * Protected counters by write_lock_irq(&tasklist_lock)
82  */
83 unsigned long total_forks;      /* Handle normal Linux uptimes. */
84 int nr_threads;                 /* The idle threads do not count.. */
85
86 int max_threads;                /* tunable limit on nr_threads */
87
88 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
89
90 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
91
92 #ifdef CONFIG_PROVE_RCU
93 int lockdep_tasklist_lock_is_held(void)
94 {
95         return lockdep_is_held(&tasklist_lock);
96 }
97 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
98 #endif /* #ifdef CONFIG_PROVE_RCU */
99
100 int nr_processes(void)
101 {
102         int cpu;
103         int total = 0;
104
105         for_each_possible_cpu(cpu)
106                 total += per_cpu(process_counts, cpu);
107
108         return total;
109 }
110
111 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
112 # define alloc_task_struct_node(node)           \
113                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
114 # define free_task_struct(tsk)                  \
115                 kmem_cache_free(task_struct_cachep, (tsk))
116 static struct kmem_cache *task_struct_cachep;
117 #endif
118
119 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
120 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
121                                                   int node)
122 {
123 #ifdef CONFIG_DEBUG_STACK_USAGE
124         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
125 #else
126         gfp_t mask = GFP_KERNEL;
127 #endif
128         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
129
130         return page ? page_address(page) : NULL;
131 }
132
133 static inline void free_thread_info(struct thread_info *ti)
134 {
135         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
136 }
137 #endif
138
139 /* SLAB cache for signal_struct structures (tsk->signal) */
140 static struct kmem_cache *signal_cachep;
141
142 /* SLAB cache for sighand_struct structures (tsk->sighand) */
143 struct kmem_cache *sighand_cachep;
144
145 /* SLAB cache for files_struct structures (tsk->files) */
146 struct kmem_cache *files_cachep;
147
148 /* SLAB cache for fs_struct structures (tsk->fs) */
149 struct kmem_cache *fs_cachep;
150
151 /* SLAB cache for vm_area_struct structures */
152 struct kmem_cache *vm_area_cachep;
153
154 /* SLAB cache for mm_struct structures (tsk->mm) */
155 static struct kmem_cache *mm_cachep;
156
157 static void account_kernel_stack(struct thread_info *ti, int account)
158 {
159         struct zone *zone = page_zone(virt_to_page(ti));
160
161         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
162 }
163
164 void free_task(struct task_struct *tsk)
165 {
166         prop_local_destroy_single(&tsk->dirties);
167         account_kernel_stack(tsk->stack, -1);
168         free_thread_info(tsk->stack);
169         rt_mutex_debug_task_free(tsk);
170         ftrace_graph_exit_task(tsk);
171         free_task_struct(tsk);
172 }
173 EXPORT_SYMBOL(free_task);
174
175 static inline void free_signal_struct(struct signal_struct *sig)
176 {
177         taskstats_tgid_free(sig);
178         sched_autogroup_exit(sig);
179         kmem_cache_free(signal_cachep, sig);
180 }
181
182 static inline void put_signal_struct(struct signal_struct *sig)
183 {
184         if (atomic_dec_and_test(&sig->sigcnt))
185                 free_signal_struct(sig);
186 }
187
188 void __put_task_struct(struct task_struct *tsk)
189 {
190         WARN_ON(!tsk->exit_state);
191         WARN_ON(atomic_read(&tsk->usage));
192         WARN_ON(tsk == current);
193
194         exit_creds(tsk);
195         delayacct_tsk_free(tsk);
196         put_signal_struct(tsk->signal);
197
198         if (!profile_handoff_task(tsk))
199                 free_task(tsk);
200 }
201 EXPORT_SYMBOL_GPL(__put_task_struct);
202
203 /*
204  * macro override instead of weak attribute alias, to workaround
205  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
206  */
207 #ifndef arch_task_cache_init
208 #define arch_task_cache_init()
209 #endif
210
211 void __init fork_init(unsigned long mempages)
212 {
213 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
214 #ifndef ARCH_MIN_TASKALIGN
215 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
216 #endif
217         /* create a slab on which task_structs can be allocated */
218         task_struct_cachep =
219                 kmem_cache_create("task_struct", sizeof(struct task_struct),
220                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
221 #endif
222
223         /* do the arch specific task caches init */
224         arch_task_cache_init();
225
226         /*
227          * The default maximum number of threads is set to a safe
228          * value: the thread structures can take up at most half
229          * of memory.
230          */
231         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
232
233         /*
234          * we need to allow at least 20 threads to boot a system
235          */
236         if(max_threads < 20)
237                 max_threads = 20;
238
239         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
240         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
241         init_task.signal->rlim[RLIMIT_SIGPENDING] =
242                 init_task.signal->rlim[RLIMIT_NPROC];
243 }
244
245 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
246                                                struct task_struct *src)
247 {
248         *dst = *src;
249         return 0;
250 }
251
252 static struct task_struct *dup_task_struct(struct task_struct *orig)
253 {
254         struct task_struct *tsk;
255         struct thread_info *ti;
256         unsigned long *stackend;
257         int node = numa_node_id();
258         int err;
259
260         prepare_to_copy(orig);
261
262         tsk = alloc_task_struct_node(node);
263         if (!tsk)
264                 return NULL;
265
266         ti = alloc_thread_info_node(tsk, node);
267         if (!ti) {
268                 free_task_struct(tsk);
269                 return NULL;
270         }
271
272         err = arch_dup_task_struct(tsk, orig);
273         if (err)
274                 goto out;
275
276         tsk->stack = ti;
277
278         err = prop_local_init_single(&tsk->dirties);
279         if (err)
280                 goto out;
281
282         setup_thread_stack(tsk, orig);
283         clear_user_return_notifier(tsk);
284         clear_tsk_need_resched(tsk);
285         stackend = end_of_stack(tsk);
286         *stackend = STACK_END_MAGIC;    /* for overflow detection */
287
288 #ifdef CONFIG_CC_STACKPROTECTOR
289         tsk->stack_canary = get_random_int();
290 #endif
291
292         /* One for us, one for whoever does the "release_task()" (usually parent) */
293         atomic_set(&tsk->usage,2);
294         atomic_set(&tsk->fs_excl, 0);
295 #ifdef CONFIG_BLK_DEV_IO_TRACE
296         tsk->btrace_seq = 0;
297 #endif
298         tsk->splice_pipe = NULL;
299
300         account_kernel_stack(ti, 1);
301
302         return tsk;
303
304 out:
305         free_thread_info(ti);
306         free_task_struct(tsk);
307         return NULL;
308 }
309
310 #ifdef CONFIG_MMU
311 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
312 {
313         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
314         struct rb_node **rb_link, *rb_parent;
315         int retval;
316         unsigned long charge;
317         struct mempolicy *pol;
318
319         down_write(&oldmm->mmap_sem);
320         flush_cache_dup_mm(oldmm);
321         /*
322          * Not linked in yet - no deadlock potential:
323          */
324         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
325
326         mm->locked_vm = 0;
327         mm->mmap = NULL;
328         mm->mmap_cache = NULL;
329         mm->free_area_cache = oldmm->mmap_base;
330         mm->cached_hole_size = ~0UL;
331         mm->map_count = 0;
332         cpumask_clear(mm_cpumask(mm));
333         mm->mm_rb = RB_ROOT;
334         rb_link = &mm->mm_rb.rb_node;
335         rb_parent = NULL;
336         pprev = &mm->mmap;
337         retval = ksm_fork(mm, oldmm);
338         if (retval)
339                 goto out;
340         retval = khugepaged_fork(mm, oldmm);
341         if (retval)
342                 goto out;
343
344         prev = NULL;
345         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
346                 struct file *file;
347
348                 if (mpnt->vm_flags & VM_DONTCOPY) {
349                         long pages = vma_pages(mpnt);
350                         mm->total_vm -= pages;
351                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
352                                                                 -pages);
353                         continue;
354                 }
355                 charge = 0;
356                 if (mpnt->vm_flags & VM_ACCOUNT) {
357                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
358                         if (security_vm_enough_memory(len))
359                                 goto fail_nomem;
360                         charge = len;
361                 }
362                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
363                 if (!tmp)
364                         goto fail_nomem;
365                 *tmp = *mpnt;
366                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
367                 pol = mpol_dup(vma_policy(mpnt));
368                 retval = PTR_ERR(pol);
369                 if (IS_ERR(pol))
370                         goto fail_nomem_policy;
371                 vma_set_policy(tmp, pol);
372                 tmp->vm_mm = mm;
373                 if (anon_vma_fork(tmp, mpnt))
374                         goto fail_nomem_anon_vma_fork;
375                 tmp->vm_flags &= ~VM_LOCKED;
376                 tmp->vm_next = tmp->vm_prev = NULL;
377                 file = tmp->vm_file;
378                 if (file) {
379                         struct inode *inode = file->f_path.dentry->d_inode;
380                         struct address_space *mapping = file->f_mapping;
381
382                         get_file(file);
383                         if (tmp->vm_flags & VM_DENYWRITE)
384                                 atomic_dec(&inode->i_writecount);
385                         spin_lock(&mapping->i_mmap_lock);
386                         if (tmp->vm_flags & VM_SHARED)
387                                 mapping->i_mmap_writable++;
388                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
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                         spin_unlock(&mapping->i_mmap_lock);
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         atomic_set(&mm->oom_disable_count, 0);
505
506         if (likely(!mm_alloc_pgd(mm))) {
507                 mm->def_flags = 0;
508                 mmu_notifier_mm_init(mm);
509                 return mm;
510         }
511
512         free_mm(mm);
513         return NULL;
514 }
515
516 /*
517  * Allocate and initialize an mm_struct.
518  */
519 struct mm_struct * mm_alloc(void)
520 {
521         struct mm_struct * mm;
522
523         mm = allocate_mm();
524         if (mm) {
525                 memset(mm, 0, sizeof(*mm));
526                 mm = mm_init(mm, current);
527         }
528         return mm;
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  * get_task_mm - acquire a reference to the task's mm
577  *
578  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
579  * this kernel workthread has transiently adopted a user mm with use_mm,
580  * to do its AIO) is not set and if so returns a reference to it, after
581  * bumping up the use count.  User must release the mm via mmput()
582  * after use.  Typically used by /proc and ptrace.
583  */
584 struct mm_struct *get_task_mm(struct task_struct *task)
585 {
586         struct mm_struct *mm;
587
588         task_lock(task);
589         mm = task->mm;
590         if (mm) {
591                 if (task->flags & PF_KTHREAD)
592                         mm = NULL;
593                 else
594                         atomic_inc(&mm->mm_users);
595         }
596         task_unlock(task);
597         return mm;
598 }
599 EXPORT_SYMBOL_GPL(get_task_mm);
600
601 /* Please note the differences between mmput and mm_release.
602  * mmput is called whenever we stop holding onto a mm_struct,
603  * error success whatever.
604  *
605  * mm_release is called after a mm_struct has been removed
606  * from the current process.
607  *
608  * This difference is important for error handling, when we
609  * only half set up a mm_struct for a new process and need to restore
610  * the old one.  Because we mmput the new mm_struct before
611  * restoring the old one. . .
612  * Eric Biederman 10 January 1998
613  */
614 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
615 {
616         struct completion *vfork_done = tsk->vfork_done;
617
618         /* Get rid of any futexes when releasing the mm */
619 #ifdef CONFIG_FUTEX
620         if (unlikely(tsk->robust_list)) {
621                 exit_robust_list(tsk);
622                 tsk->robust_list = NULL;
623         }
624 #ifdef CONFIG_COMPAT
625         if (unlikely(tsk->compat_robust_list)) {
626                 compat_exit_robust_list(tsk);
627                 tsk->compat_robust_list = NULL;
628         }
629 #endif
630         if (unlikely(!list_empty(&tsk->pi_state_list)))
631                 exit_pi_state_list(tsk);
632 #endif
633
634         /* Get rid of any cached register state */
635         deactivate_mm(tsk, mm);
636
637         /* notify parent sleeping on vfork() */
638         if (vfork_done) {
639                 tsk->vfork_done = NULL;
640                 complete(vfork_done);
641         }
642
643         /*
644          * If we're exiting normally, clear a user-space tid field if
645          * requested.  We leave this alone when dying by signal, to leave
646          * the value intact in a core dump, and to save the unnecessary
647          * trouble otherwise.  Userland only wants this done for a sys_exit.
648          */
649         if (tsk->clear_child_tid) {
650                 if (!(tsk->flags & PF_SIGNALED) &&
651                     atomic_read(&mm->mm_users) > 1) {
652                         /*
653                          * We don't check the error code - if userspace has
654                          * not set up a proper pointer then tough luck.
655                          */
656                         put_user(0, tsk->clear_child_tid);
657                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
658                                         1, NULL, NULL, 0);
659                 }
660                 tsk->clear_child_tid = NULL;
661         }
662 }
663
664 /*
665  * Allocate a new mm structure and copy contents from the
666  * mm structure of the passed in task structure.
667  */
668 struct mm_struct *dup_mm(struct task_struct *tsk)
669 {
670         struct mm_struct *mm, *oldmm = current->mm;
671         int err;
672
673         if (!oldmm)
674                 return NULL;
675
676         mm = allocate_mm();
677         if (!mm)
678                 goto fail_nomem;
679
680         memcpy(mm, oldmm, sizeof(*mm));
681
682         /* Initializing for Swap token stuff */
683         mm->token_priority = 0;
684         mm->last_interval = 0;
685
686 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
687         mm->pmd_huge_pte = NULL;
688 #endif
689
690         if (!mm_init(mm, tsk))
691                 goto fail_nomem;
692
693         if (init_new_context(tsk, mm))
694                 goto fail_nocontext;
695
696         dup_mm_exe_file(oldmm, mm);
697
698         err = dup_mmap(mm, oldmm);
699         if (err)
700                 goto free_pt;
701
702         mm->hiwater_rss = get_mm_rss(mm);
703         mm->hiwater_vm = mm->total_vm;
704
705         if (mm->binfmt && !try_module_get(mm->binfmt->module))
706                 goto free_pt;
707
708         return mm;
709
710 free_pt:
711         /* don't put binfmt in mmput, we haven't got module yet */
712         mm->binfmt = NULL;
713         mmput(mm);
714
715 fail_nomem:
716         return NULL;
717
718 fail_nocontext:
719         /*
720          * If init_new_context() failed, we cannot use mmput() to free the mm
721          * because it calls destroy_context()
722          */
723         mm_free_pgd(mm);
724         free_mm(mm);
725         return NULL;
726 }
727
728 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
729 {
730         struct mm_struct * mm, *oldmm;
731         int retval;
732
733         tsk->min_flt = tsk->maj_flt = 0;
734         tsk->nvcsw = tsk->nivcsw = 0;
735 #ifdef CONFIG_DETECT_HUNG_TASK
736         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
737 #endif
738
739         tsk->mm = NULL;
740         tsk->active_mm = NULL;
741
742         /*
743          * Are we cloning a kernel thread?
744          *
745          * We need to steal a active VM for that..
746          */
747         oldmm = current->mm;
748         if (!oldmm)
749                 return 0;
750
751         if (clone_flags & CLONE_VM) {
752                 atomic_inc(&oldmm->mm_users);
753                 mm = oldmm;
754                 goto good_mm;
755         }
756
757         retval = -ENOMEM;
758         mm = dup_mm(tsk);
759         if (!mm)
760                 goto fail_nomem;
761
762 good_mm:
763         /* Initializing for Swap token stuff */
764         mm->token_priority = 0;
765         mm->last_interval = 0;
766         if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
767                 atomic_inc(&mm->oom_disable_count);
768
769         tsk->mm = mm;
770         tsk->active_mm = mm;
771         return 0;
772
773 fail_nomem:
774         return retval;
775 }
776
777 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
778 {
779         struct fs_struct *fs = current->fs;
780         if (clone_flags & CLONE_FS) {
781                 /* tsk->fs is already what we want */
782                 spin_lock(&fs->lock);
783                 if (fs->in_exec) {
784                         spin_unlock(&fs->lock);
785                         return -EAGAIN;
786                 }
787                 fs->users++;
788                 spin_unlock(&fs->lock);
789                 return 0;
790         }
791         tsk->fs = copy_fs_struct(fs);
792         if (!tsk->fs)
793                 return -ENOMEM;
794         return 0;
795 }
796
797 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
798 {
799         struct files_struct *oldf, *newf;
800         int error = 0;
801
802         /*
803          * A background process may not have any files ...
804          */
805         oldf = current->files;
806         if (!oldf)
807                 goto out;
808
809         if (clone_flags & CLONE_FILES) {
810                 atomic_inc(&oldf->count);
811                 goto out;
812         }
813
814         newf = dup_fd(oldf, &error);
815         if (!newf)
816                 goto out;
817
818         tsk->files = newf;
819         error = 0;
820 out:
821         return error;
822 }
823
824 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
825 {
826 #ifdef CONFIG_BLOCK
827         struct io_context *ioc = current->io_context;
828
829         if (!ioc)
830                 return 0;
831         /*
832          * Share io context with parent, if CLONE_IO is set
833          */
834         if (clone_flags & CLONE_IO) {
835                 tsk->io_context = ioc_task_link(ioc);
836                 if (unlikely(!tsk->io_context))
837                         return -ENOMEM;
838         } else if (ioprio_valid(ioc->ioprio)) {
839                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
840                 if (unlikely(!tsk->io_context))
841                         return -ENOMEM;
842
843                 tsk->io_context->ioprio = ioc->ioprio;
844         }
845 #endif
846         return 0;
847 }
848
849 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
850 {
851         struct sighand_struct *sig;
852
853         if (clone_flags & CLONE_SIGHAND) {
854                 atomic_inc(&current->sighand->count);
855                 return 0;
856         }
857         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
858         rcu_assign_pointer(tsk->sighand, sig);
859         if (!sig)
860                 return -ENOMEM;
861         atomic_set(&sig->count, 1);
862         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
863         return 0;
864 }
865
866 void __cleanup_sighand(struct sighand_struct *sighand)
867 {
868         if (atomic_dec_and_test(&sighand->count))
869                 kmem_cache_free(sighand_cachep, sighand);
870 }
871
872
873 /*
874  * Initialize POSIX timer handling for a thread group.
875  */
876 static void posix_cpu_timers_init_group(struct signal_struct *sig)
877 {
878         unsigned long cpu_limit;
879
880         /* Thread group counters. */
881         thread_group_cputime_init(sig);
882
883         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
884         if (cpu_limit != RLIM_INFINITY) {
885                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
886                 sig->cputimer.running = 1;
887         }
888
889         /* The timer lists. */
890         INIT_LIST_HEAD(&sig->cpu_timers[0]);
891         INIT_LIST_HEAD(&sig->cpu_timers[1]);
892         INIT_LIST_HEAD(&sig->cpu_timers[2]);
893 }
894
895 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
896 {
897         struct signal_struct *sig;
898
899         if (clone_flags & CLONE_THREAD)
900                 return 0;
901
902         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
903         tsk->signal = sig;
904         if (!sig)
905                 return -ENOMEM;
906
907         sig->nr_threads = 1;
908         atomic_set(&sig->live, 1);
909         atomic_set(&sig->sigcnt, 1);
910         init_waitqueue_head(&sig->wait_chldexit);
911         if (clone_flags & CLONE_NEWPID)
912                 sig->flags |= SIGNAL_UNKILLABLE;
913         sig->curr_target = tsk;
914         init_sigpending(&sig->shared_pending);
915         INIT_LIST_HEAD(&sig->posix_timers);
916
917         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
918         sig->real_timer.function = it_real_fn;
919
920         task_lock(current->group_leader);
921         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
922         task_unlock(current->group_leader);
923
924         posix_cpu_timers_init_group(sig);
925
926         tty_audit_fork(sig);
927         sched_autogroup_fork(sig);
928
929         sig->oom_adj = current->signal->oom_adj;
930         sig->oom_score_adj = current->signal->oom_score_adj;
931         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
932
933         mutex_init(&sig->cred_guard_mutex);
934
935         return 0;
936 }
937
938 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
939 {
940         unsigned long new_flags = p->flags;
941
942         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
943         new_flags |= PF_FORKNOEXEC;
944         new_flags |= PF_STARTING;
945         p->flags = new_flags;
946         clear_freeze_flag(p);
947 }
948
949 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
950 {
951         current->clear_child_tid = tidptr;
952
953         return task_pid_vnr(current);
954 }
955
956 static void rt_mutex_init_task(struct task_struct *p)
957 {
958         raw_spin_lock_init(&p->pi_lock);
959 #ifdef CONFIG_RT_MUTEXES
960         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
961         p->pi_blocked_on = NULL;
962 #endif
963 }
964
965 #ifdef CONFIG_MM_OWNER
966 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
967 {
968         mm->owner = p;
969 }
970 #endif /* CONFIG_MM_OWNER */
971
972 /*
973  * Initialize POSIX timer handling for a single task.
974  */
975 static void posix_cpu_timers_init(struct task_struct *tsk)
976 {
977         tsk->cputime_expires.prof_exp = cputime_zero;
978         tsk->cputime_expires.virt_exp = cputime_zero;
979         tsk->cputime_expires.sched_exp = 0;
980         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
981         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
982         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
983 }
984
985 /*
986  * This creates a new process as a copy of the old one,
987  * but does not actually start it yet.
988  *
989  * It copies the registers, and all the appropriate
990  * parts of the process environment (as per the clone
991  * flags). The actual kick-off is left to the caller.
992  */
993 static struct task_struct *copy_process(unsigned long clone_flags,
994                                         unsigned long stack_start,
995                                         struct pt_regs *regs,
996                                         unsigned long stack_size,
997                                         int __user *child_tidptr,
998                                         struct pid *pid,
999                                         int trace)
1000 {
1001         int retval;
1002         struct task_struct *p;
1003         int cgroup_callbacks_done = 0;
1004
1005         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1006                 return ERR_PTR(-EINVAL);
1007
1008         /*
1009          * Thread groups must share signals as well, and detached threads
1010          * can only be started up within the thread group.
1011          */
1012         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1013                 return ERR_PTR(-EINVAL);
1014
1015         /*
1016          * Shared signal handlers imply shared VM. By way of the above,
1017          * thread groups also imply shared VM. Blocking this case allows
1018          * for various simplifications in other code.
1019          */
1020         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1021                 return ERR_PTR(-EINVAL);
1022
1023         /*
1024          * Siblings of global init remain as zombies on exit since they are
1025          * not reaped by their parent (swapper). To solve this and to avoid
1026          * multi-rooted process trees, prevent global and container-inits
1027          * from creating siblings.
1028          */
1029         if ((clone_flags & CLONE_PARENT) &&
1030                                 current->signal->flags & SIGNAL_UNKILLABLE)
1031                 return ERR_PTR(-EINVAL);
1032
1033         retval = security_task_create(clone_flags);
1034         if (retval)
1035                 goto fork_out;
1036
1037         retval = -ENOMEM;
1038         p = dup_task_struct(current);
1039         if (!p)
1040                 goto fork_out;
1041
1042         ftrace_graph_init_task(p);
1043
1044         rt_mutex_init_task(p);
1045
1046 #ifdef CONFIG_PROVE_LOCKING
1047         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1048         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1049 #endif
1050         retval = -EAGAIN;
1051         if (atomic_read(&p->real_cred->user->processes) >=
1052                         task_rlimit(p, RLIMIT_NPROC)) {
1053                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1054                     p->real_cred->user != INIT_USER)
1055                         goto bad_fork_free;
1056         }
1057
1058         retval = copy_creds(p, clone_flags);
1059         if (retval < 0)
1060                 goto bad_fork_free;
1061
1062         /*
1063          * If multiple threads are within copy_process(), then this check
1064          * triggers too late. This doesn't hurt, the check is only there
1065          * to stop root fork bombs.
1066          */
1067         retval = -EAGAIN;
1068         if (nr_threads >= max_threads)
1069                 goto bad_fork_cleanup_count;
1070
1071         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1072                 goto bad_fork_cleanup_count;
1073
1074         p->did_exec = 0;
1075         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1076         copy_flags(clone_flags, p);
1077         INIT_LIST_HEAD(&p->children);
1078         INIT_LIST_HEAD(&p->sibling);
1079         rcu_copy_process(p);
1080         p->vfork_done = NULL;
1081         spin_lock_init(&p->alloc_lock);
1082
1083         init_sigpending(&p->pending);
1084
1085         p->utime = cputime_zero;
1086         p->stime = cputime_zero;
1087         p->gtime = cputime_zero;
1088         p->utimescaled = cputime_zero;
1089         p->stimescaled = cputime_zero;
1090 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1091         p->prev_utime = cputime_zero;
1092         p->prev_stime = cputime_zero;
1093 #endif
1094 #if defined(SPLIT_RSS_COUNTING)
1095         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1096 #endif
1097
1098         p->default_timer_slack_ns = current->timer_slack_ns;
1099
1100         task_io_accounting_init(&p->ioac);
1101         acct_clear_integrals(p);
1102
1103         posix_cpu_timers_init(p);
1104
1105         p->lock_depth = -1;             /* -1 = no lock */
1106         do_posix_clock_monotonic_gettime(&p->start_time);
1107         p->real_start_time = p->start_time;
1108         monotonic_to_bootbased(&p->real_start_time);
1109         p->io_context = NULL;
1110         p->audit_context = NULL;
1111         cgroup_fork(p);
1112 #ifdef CONFIG_NUMA
1113         p->mempolicy = mpol_dup(p->mempolicy);
1114         if (IS_ERR(p->mempolicy)) {
1115                 retval = PTR_ERR(p->mempolicy);
1116                 p->mempolicy = NULL;
1117                 goto bad_fork_cleanup_cgroup;
1118         }
1119         mpol_fix_fork_child_flag(p);
1120 #endif
1121 #ifdef CONFIG_TRACE_IRQFLAGS
1122         p->irq_events = 0;
1123 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1124         p->hardirqs_enabled = 1;
1125 #else
1126         p->hardirqs_enabled = 0;
1127 #endif
1128         p->hardirq_enable_ip = 0;
1129         p->hardirq_enable_event = 0;
1130         p->hardirq_disable_ip = _THIS_IP_;
1131         p->hardirq_disable_event = 0;
1132         p->softirqs_enabled = 1;
1133         p->softirq_enable_ip = _THIS_IP_;
1134         p->softirq_enable_event = 0;
1135         p->softirq_disable_ip = 0;
1136         p->softirq_disable_event = 0;
1137         p->hardirq_context = 0;
1138         p->softirq_context = 0;
1139 #endif
1140 #ifdef CONFIG_LOCKDEP
1141         p->lockdep_depth = 0; /* no locks held yet */
1142         p->curr_chain_key = 0;
1143         p->lockdep_recursion = 0;
1144 #endif
1145
1146 #ifdef CONFIG_DEBUG_MUTEXES
1147         p->blocked_on = NULL; /* not blocked yet */
1148 #endif
1149 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1150         p->memcg_batch.do_batch = 0;
1151         p->memcg_batch.memcg = NULL;
1152 #endif
1153
1154         /* Perform scheduler related setup. Assign this task to a CPU. */
1155         sched_fork(p, clone_flags);
1156
1157         retval = perf_event_init_task(p);
1158         if (retval)
1159                 goto bad_fork_cleanup_policy;
1160
1161         if ((retval = audit_alloc(p)))
1162                 goto bad_fork_cleanup_policy;
1163         /* copy all the process information */
1164         if ((retval = copy_semundo(clone_flags, p)))
1165                 goto bad_fork_cleanup_audit;
1166         if ((retval = copy_files(clone_flags, p)))
1167                 goto bad_fork_cleanup_semundo;
1168         if ((retval = copy_fs(clone_flags, p)))
1169                 goto bad_fork_cleanup_files;
1170         if ((retval = copy_sighand(clone_flags, p)))
1171                 goto bad_fork_cleanup_fs;
1172         if ((retval = copy_signal(clone_flags, p)))
1173                 goto bad_fork_cleanup_sighand;
1174         if ((retval = copy_mm(clone_flags, p)))
1175                 goto bad_fork_cleanup_signal;
1176         if ((retval = copy_namespaces(clone_flags, p)))
1177                 goto bad_fork_cleanup_mm;
1178         if ((retval = copy_io(clone_flags, p)))
1179                 goto bad_fork_cleanup_namespaces;
1180         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1181         if (retval)
1182                 goto bad_fork_cleanup_io;
1183
1184         if (pid != &init_struct_pid) {
1185                 retval = -ENOMEM;
1186                 pid = alloc_pid(p->nsproxy->pid_ns);
1187                 if (!pid)
1188                         goto bad_fork_cleanup_io;
1189
1190                 if (clone_flags & CLONE_NEWPID) {
1191                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1192                         if (retval < 0)
1193                                 goto bad_fork_free_pid;
1194                 }
1195         }
1196
1197         p->pid = pid_nr(pid);
1198         p->tgid = p->pid;
1199         if (clone_flags & CLONE_THREAD)
1200                 p->tgid = current->tgid;
1201
1202         if (current->nsproxy != p->nsproxy) {
1203                 retval = ns_cgroup_clone(p, pid);
1204                 if (retval)
1205                         goto bad_fork_free_pid;
1206         }
1207
1208         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1209         /*
1210          * Clear TID on mm_release()?
1211          */
1212         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1213 #ifdef CONFIG_FUTEX
1214         p->robust_list = NULL;
1215 #ifdef CONFIG_COMPAT
1216         p->compat_robust_list = NULL;
1217 #endif
1218         INIT_LIST_HEAD(&p->pi_state_list);
1219         p->pi_state_cache = NULL;
1220 #endif
1221         /*
1222          * sigaltstack should be cleared when sharing the same VM
1223          */
1224         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1225                 p->sas_ss_sp = p->sas_ss_size = 0;
1226
1227         /*
1228          * Syscall tracing and stepping should be turned off in the
1229          * child regardless of CLONE_PTRACE.
1230          */
1231         user_disable_single_step(p);
1232         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1233 #ifdef TIF_SYSCALL_EMU
1234         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1235 #endif
1236         clear_all_latency_tracing(p);
1237
1238         /* ok, now we should be set up.. */
1239         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1240         p->pdeath_signal = 0;
1241         p->exit_state = 0;
1242
1243         /*
1244          * Ok, make it visible to the rest of the system.
1245          * We dont wake it up yet.
1246          */
1247         p->group_leader = p;
1248         INIT_LIST_HEAD(&p->thread_group);
1249
1250         /* Now that the task is set up, run cgroup callbacks if
1251          * necessary. We need to run them before the task is visible
1252          * on the tasklist. */
1253         cgroup_fork_callbacks(p);
1254         cgroup_callbacks_done = 1;
1255
1256         /* Need tasklist lock for parent etc handling! */
1257         write_lock_irq(&tasklist_lock);
1258
1259         /* CLONE_PARENT re-uses the old parent */
1260         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1261                 p->real_parent = current->real_parent;
1262                 p->parent_exec_id = current->parent_exec_id;
1263         } else {
1264                 p->real_parent = current;
1265                 p->parent_exec_id = current->self_exec_id;
1266         }
1267
1268         spin_lock(&current->sighand->siglock);
1269
1270         /*
1271          * Process group and session signals need to be delivered to just the
1272          * parent before the fork or both the parent and the child after the
1273          * fork. Restart if a signal comes in before we add the new process to
1274          * it's process group.
1275          * A fatal signal pending means that current will exit, so the new
1276          * thread can't slip out of an OOM kill (or normal SIGKILL).
1277          */
1278         recalc_sigpending();
1279         if (signal_pending(current)) {
1280                 spin_unlock(&current->sighand->siglock);
1281                 write_unlock_irq(&tasklist_lock);
1282                 retval = -ERESTARTNOINTR;
1283                 goto bad_fork_free_pid;
1284         }
1285
1286         if (clone_flags & CLONE_THREAD) {
1287                 current->signal->nr_threads++;
1288                 atomic_inc(&current->signal->live);
1289                 atomic_inc(&current->signal->sigcnt);
1290                 p->group_leader = current->group_leader;
1291                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1292         }
1293
1294         if (likely(p->pid)) {
1295                 tracehook_finish_clone(p, clone_flags, trace);
1296
1297                 if (thread_group_leader(p)) {
1298                         if (clone_flags & CLONE_NEWPID)
1299                                 p->nsproxy->pid_ns->child_reaper = p;
1300
1301                         p->signal->leader_pid = pid;
1302                         p->signal->tty = tty_kref_get(current->signal->tty);
1303                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1304                         attach_pid(p, PIDTYPE_SID, task_session(current));
1305                         list_add_tail(&p->sibling, &p->real_parent->children);
1306                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1307                         __this_cpu_inc(process_counts);
1308                 }
1309                 attach_pid(p, PIDTYPE_PID, pid);
1310                 nr_threads++;
1311         }
1312
1313         total_forks++;
1314         spin_unlock(&current->sighand->siglock);
1315         write_unlock_irq(&tasklist_lock);
1316         proc_fork_connector(p);
1317         cgroup_post_fork(p);
1318         perf_event_fork(p);
1319         return p;
1320
1321 bad_fork_free_pid:
1322         if (pid != &init_struct_pid)
1323                 free_pid(pid);
1324 bad_fork_cleanup_io:
1325         if (p->io_context)
1326                 exit_io_context(p);
1327 bad_fork_cleanup_namespaces:
1328         exit_task_namespaces(p);
1329 bad_fork_cleanup_mm:
1330         if (p->mm) {
1331                 task_lock(p);
1332                 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1333                         atomic_dec(&p->mm->oom_disable_count);
1334                 task_unlock(p);
1335                 mmput(p->mm);
1336         }
1337 bad_fork_cleanup_signal:
1338         if (!(clone_flags & CLONE_THREAD))
1339                 free_signal_struct(p->signal);
1340 bad_fork_cleanup_sighand:
1341         __cleanup_sighand(p->sighand);
1342 bad_fork_cleanup_fs:
1343         exit_fs(p); /* blocking */
1344 bad_fork_cleanup_files:
1345         exit_files(p); /* blocking */
1346 bad_fork_cleanup_semundo:
1347         exit_sem(p);
1348 bad_fork_cleanup_audit:
1349         audit_free(p);
1350 bad_fork_cleanup_policy:
1351         perf_event_free_task(p);
1352 #ifdef CONFIG_NUMA
1353         mpol_put(p->mempolicy);
1354 bad_fork_cleanup_cgroup:
1355 #endif
1356         cgroup_exit(p, cgroup_callbacks_done);
1357         delayacct_tsk_free(p);
1358         module_put(task_thread_info(p)->exec_domain->module);
1359 bad_fork_cleanup_count:
1360         atomic_dec(&p->cred->user->processes);
1361         exit_creds(p);
1362 bad_fork_free:
1363         free_task(p);
1364 fork_out:
1365         return ERR_PTR(retval);
1366 }
1367
1368 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1369 {
1370         memset(regs, 0, sizeof(struct pt_regs));
1371         return regs;
1372 }
1373
1374 static inline void init_idle_pids(struct pid_link *links)
1375 {
1376         enum pid_type type;
1377
1378         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1379                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1380                 links[type].pid = &init_struct_pid;
1381         }
1382 }
1383
1384 struct task_struct * __cpuinit fork_idle(int cpu)
1385 {
1386         struct task_struct *task;
1387         struct pt_regs regs;
1388
1389         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1390                             &init_struct_pid, 0);
1391         if (!IS_ERR(task)) {
1392                 init_idle_pids(task->pids);
1393                 init_idle(task, cpu);
1394         }
1395
1396         return task;
1397 }
1398
1399 /*
1400  *  Ok, this is the main fork-routine.
1401  *
1402  * It copies the process, and if successful kick-starts
1403  * it and waits for it to finish using the VM if required.
1404  */
1405 long do_fork(unsigned long clone_flags,
1406               unsigned long stack_start,
1407               struct pt_regs *regs,
1408               unsigned long stack_size,
1409               int __user *parent_tidptr,
1410               int __user *child_tidptr)
1411 {
1412         struct task_struct *p;
1413         int trace = 0;
1414         long nr;
1415
1416         /*
1417          * Do some preliminary argument and permissions checking before we
1418          * actually start allocating stuff
1419          */
1420         if (clone_flags & CLONE_NEWUSER) {
1421                 if (clone_flags & CLONE_THREAD)
1422                         return -EINVAL;
1423                 /* hopefully this check will go away when userns support is
1424                  * complete
1425                  */
1426                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1427                                 !capable(CAP_SETGID))
1428                         return -EPERM;
1429         }
1430
1431         /*
1432          * When called from kernel_thread, don't do user tracing stuff.
1433          */
1434         if (likely(user_mode(regs)))
1435                 trace = tracehook_prepare_clone(clone_flags);
1436
1437         p = copy_process(clone_flags, stack_start, regs, stack_size,
1438                          child_tidptr, NULL, trace);
1439         /*
1440          * Do this prior waking up the new thread - the thread pointer
1441          * might get invalid after that point, if the thread exits quickly.
1442          */
1443         if (!IS_ERR(p)) {
1444                 struct completion vfork;
1445
1446                 trace_sched_process_fork(current, p);
1447
1448                 nr = task_pid_vnr(p);
1449
1450                 if (clone_flags & CLONE_PARENT_SETTID)
1451                         put_user(nr, parent_tidptr);
1452
1453                 if (clone_flags & CLONE_VFORK) {
1454                         p->vfork_done = &vfork;
1455                         init_completion(&vfork);
1456                 }
1457
1458                 audit_finish_fork(p);
1459                 tracehook_report_clone(regs, clone_flags, nr, p);
1460
1461                 /*
1462                  * We set PF_STARTING at creation in case tracing wants to
1463                  * use this to distinguish a fully live task from one that
1464                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1465                  * clear it and set the child going.
1466                  */
1467                 p->flags &= ~PF_STARTING;
1468
1469                 wake_up_new_task(p, clone_flags);
1470
1471                 tracehook_report_clone_complete(trace, regs,
1472                                                 clone_flags, nr, p);
1473
1474                 if (clone_flags & CLONE_VFORK) {
1475                         freezer_do_not_count();
1476                         wait_for_completion(&vfork);
1477                         freezer_count();
1478                         tracehook_report_vfork_done(p, nr);
1479                 }
1480         } else {
1481                 nr = PTR_ERR(p);
1482         }
1483         return nr;
1484 }
1485
1486 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1487 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1488 #endif
1489
1490 static void sighand_ctor(void *data)
1491 {
1492         struct sighand_struct *sighand = data;
1493
1494         spin_lock_init(&sighand->siglock);
1495         init_waitqueue_head(&sighand->signalfd_wqh);
1496 }
1497
1498 void __init proc_caches_init(void)
1499 {
1500         sighand_cachep = kmem_cache_create("sighand_cache",
1501                         sizeof(struct sighand_struct), 0,
1502                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1503                         SLAB_NOTRACK, sighand_ctor);
1504         signal_cachep = kmem_cache_create("signal_cache",
1505                         sizeof(struct signal_struct), 0,
1506                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1507         files_cachep = kmem_cache_create("files_cache",
1508                         sizeof(struct files_struct), 0,
1509                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1510         fs_cachep = kmem_cache_create("fs_cache",
1511                         sizeof(struct fs_struct), 0,
1512                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1513         mm_cachep = kmem_cache_create("mm_struct",
1514                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1515                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1516         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1517         mmap_init();
1518 }
1519
1520 /*
1521  * Check constraints on flags passed to the unshare system call and
1522  * force unsharing of additional process context as appropriate.
1523  */
1524 static void check_unshare_flags(unsigned long *flags_ptr)
1525 {
1526         /*
1527          * If unsharing a thread from a thread group, must also
1528          * unshare vm.
1529          */
1530         if (*flags_ptr & CLONE_THREAD)
1531                 *flags_ptr |= CLONE_VM;
1532
1533         /*
1534          * If unsharing vm, must also unshare signal handlers.
1535          */
1536         if (*flags_ptr & CLONE_VM)
1537                 *flags_ptr |= CLONE_SIGHAND;
1538
1539         /*
1540          * If unsharing namespace, must also unshare filesystem information.
1541          */
1542         if (*flags_ptr & CLONE_NEWNS)
1543                 *flags_ptr |= CLONE_FS;
1544 }
1545
1546 /*
1547  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1548  */
1549 static int unshare_thread(unsigned long unshare_flags)
1550 {
1551         if (unshare_flags & CLONE_THREAD)
1552                 return -EINVAL;
1553
1554         return 0;
1555 }
1556
1557 /*
1558  * Unshare the filesystem structure if it is being shared
1559  */
1560 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1561 {
1562         struct fs_struct *fs = current->fs;
1563
1564         if (!(unshare_flags & CLONE_FS) || !fs)
1565                 return 0;
1566
1567         /* don't need lock here; in the worst case we'll do useless copy */
1568         if (fs->users == 1)
1569                 return 0;
1570
1571         *new_fsp = copy_fs_struct(fs);
1572         if (!*new_fsp)
1573                 return -ENOMEM;
1574
1575         return 0;
1576 }
1577
1578 /*
1579  * Unsharing of sighand is not supported yet
1580  */
1581 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1582 {
1583         struct sighand_struct *sigh = current->sighand;
1584
1585         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1586                 return -EINVAL;
1587         else
1588                 return 0;
1589 }
1590
1591 /*
1592  * Unshare vm if it is being shared
1593  */
1594 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1595 {
1596         struct mm_struct *mm = current->mm;
1597
1598         if ((unshare_flags & CLONE_VM) &&
1599             (mm && atomic_read(&mm->mm_users) > 1)) {
1600                 return -EINVAL;
1601         }
1602
1603         return 0;
1604 }
1605
1606 /*
1607  * Unshare file descriptor table if it is being shared
1608  */
1609 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1610 {
1611         struct files_struct *fd = current->files;
1612         int error = 0;
1613
1614         if ((unshare_flags & CLONE_FILES) &&
1615             (fd && atomic_read(&fd->count) > 1)) {
1616                 *new_fdp = dup_fd(fd, &error);
1617                 if (!*new_fdp)
1618                         return error;
1619         }
1620
1621         return 0;
1622 }
1623
1624 /*
1625  * unshare allows a process to 'unshare' part of the process
1626  * context which was originally shared using clone.  copy_*
1627  * functions used by do_fork() cannot be used here directly
1628  * because they modify an inactive task_struct that is being
1629  * constructed. Here we are modifying the current, active,
1630  * task_struct.
1631  */
1632 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1633 {
1634         int err = 0;
1635         struct fs_struct *fs, *new_fs = NULL;
1636         struct sighand_struct *new_sigh = NULL;
1637         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1638         struct files_struct *fd, *new_fd = NULL;
1639         struct nsproxy *new_nsproxy = NULL;
1640         int do_sysvsem = 0;
1641
1642         check_unshare_flags(&unshare_flags);
1643
1644         /* Return -EINVAL for all unsupported flags */
1645         err = -EINVAL;
1646         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1647                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1648                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1649                 goto bad_unshare_out;
1650
1651         /*
1652          * CLONE_NEWIPC must also detach from the undolist: after switching
1653          * to a new ipc namespace, the semaphore arrays from the old
1654          * namespace are unreachable.
1655          */
1656         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1657                 do_sysvsem = 1;
1658         if ((err = unshare_thread(unshare_flags)))
1659                 goto bad_unshare_out;
1660         if ((err = unshare_fs(unshare_flags, &new_fs)))
1661                 goto bad_unshare_cleanup_thread;
1662         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1663                 goto bad_unshare_cleanup_fs;
1664         if ((err = unshare_vm(unshare_flags, &new_mm)))
1665                 goto bad_unshare_cleanup_sigh;
1666         if ((err = unshare_fd(unshare_flags, &new_fd)))
1667                 goto bad_unshare_cleanup_vm;
1668         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1669                         new_fs)))
1670                 goto bad_unshare_cleanup_fd;
1671
1672         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1673                 if (do_sysvsem) {
1674                         /*
1675                          * CLONE_SYSVSEM is equivalent to sys_exit().
1676                          */
1677                         exit_sem(current);
1678                 }
1679
1680                 if (new_nsproxy) {
1681                         switch_task_namespaces(current, new_nsproxy);
1682                         new_nsproxy = NULL;
1683                 }
1684
1685                 task_lock(current);
1686
1687                 if (new_fs) {
1688                         fs = current->fs;
1689                         spin_lock(&fs->lock);
1690                         current->fs = new_fs;
1691                         if (--fs->users)
1692                                 new_fs = NULL;
1693                         else
1694                                 new_fs = fs;
1695                         spin_unlock(&fs->lock);
1696                 }
1697
1698                 if (new_mm) {
1699                         mm = current->mm;
1700                         active_mm = current->active_mm;
1701                         current->mm = new_mm;
1702                         current->active_mm = new_mm;
1703                         if (current->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
1704                                 atomic_dec(&mm->oom_disable_count);
1705                                 atomic_inc(&new_mm->oom_disable_count);
1706                         }
1707                         activate_mm(active_mm, new_mm);
1708                         new_mm = mm;
1709                 }
1710
1711                 if (new_fd) {
1712                         fd = current->files;
1713                         current->files = new_fd;
1714                         new_fd = fd;
1715                 }
1716
1717                 task_unlock(current);
1718         }
1719
1720         if (new_nsproxy)
1721                 put_nsproxy(new_nsproxy);
1722
1723 bad_unshare_cleanup_fd:
1724         if (new_fd)
1725                 put_files_struct(new_fd);
1726
1727 bad_unshare_cleanup_vm:
1728         if (new_mm)
1729                 mmput(new_mm);
1730
1731 bad_unshare_cleanup_sigh:
1732         if (new_sigh)
1733                 if (atomic_dec_and_test(&new_sigh->count))
1734                         kmem_cache_free(sighand_cachep, new_sigh);
1735
1736 bad_unshare_cleanup_fs:
1737         if (new_fs)
1738                 free_fs_struct(new_fs);
1739
1740 bad_unshare_cleanup_thread:
1741 bad_unshare_out:
1742         return err;
1743 }
1744
1745 /*
1746  *      Helper to unshare the files of the current task.
1747  *      We don't want to expose copy_files internals to
1748  *      the exec layer of the kernel.
1749  */
1750
1751 int unshare_files(struct files_struct **displaced)
1752 {
1753         struct task_struct *task = current;
1754         struct files_struct *copy = NULL;
1755         int error;
1756
1757         error = unshare_fd(CLONE_FILES, &copy);
1758         if (error || !copy) {
1759                 *displaced = NULL;
1760                 return error;
1761         }
1762         *displaced = task->files;
1763         task_lock(task);
1764         task->files = copy;
1765         task_unlock(task);
1766         return 0;
1767 }