mm: oom analysis: Show kernel stack usage in /proc/meminfo and OOM log output
[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/acct.h>
53 #include <linux/tsacct_kern.h>
54 #include <linux/cn_proc.h>
55 #include <linux/freezer.h>
56 #include <linux/delayacct.h>
57 #include <linux/taskstats_kern.h>
58 #include <linux/random.h>
59 #include <linux/tty.h>
60 #include <linux/proc_fs.h>
61 #include <linux/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_event.h>
65
66 #include <asm/pgtable.h>
67 #include <asm/pgalloc.h>
68 #include <asm/uaccess.h>
69 #include <asm/mmu_context.h>
70 #include <asm/cacheflush.h>
71 #include <asm/tlbflush.h>
72
73 #include <trace/events/sched.h>
74
75 /*
76  * Protected counters by write_lock_irq(&tasklist_lock)
77  */
78 unsigned long total_forks;      /* Handle normal Linux uptimes. */
79 int nr_threads;                 /* The idle threads do not count.. */
80
81 int max_threads;                /* tunable limit on nr_threads */
82
83 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
84
85 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
86
87 int nr_processes(void)
88 {
89         int cpu;
90         int total = 0;
91
92         for_each_online_cpu(cpu)
93                 total += per_cpu(process_counts, cpu);
94
95         return total;
96 }
97
98 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
99 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
100 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
101 static struct kmem_cache *task_struct_cachep;
102 #endif
103
104 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
105 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
106 {
107 #ifdef CONFIG_DEBUG_STACK_USAGE
108         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
109 #else
110         gfp_t mask = GFP_KERNEL;
111 #endif
112         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
113 }
114
115 static inline void free_thread_info(struct thread_info *ti)
116 {
117         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
118 }
119 #endif
120
121 /* SLAB cache for signal_struct structures (tsk->signal) */
122 static struct kmem_cache *signal_cachep;
123
124 /* SLAB cache for sighand_struct structures (tsk->sighand) */
125 struct kmem_cache *sighand_cachep;
126
127 /* SLAB cache for files_struct structures (tsk->files) */
128 struct kmem_cache *files_cachep;
129
130 /* SLAB cache for fs_struct structures (tsk->fs) */
131 struct kmem_cache *fs_cachep;
132
133 /* SLAB cache for vm_area_struct structures */
134 struct kmem_cache *vm_area_cachep;
135
136 /* SLAB cache for mm_struct structures (tsk->mm) */
137 static struct kmem_cache *mm_cachep;
138
139 static void account_kernel_stack(struct thread_info *ti, int account)
140 {
141         struct zone *zone = page_zone(virt_to_page(ti));
142
143         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
144 }
145
146 void free_task(struct task_struct *tsk)
147 {
148         prop_local_destroy_single(&tsk->dirties);
149         account_kernel_stack(tsk->stack, -1);
150         free_thread_info(tsk->stack);
151         rt_mutex_debug_task_free(tsk);
152         ftrace_graph_exit_task(tsk);
153         free_task_struct(tsk);
154 }
155 EXPORT_SYMBOL(free_task);
156
157 void __put_task_struct(struct task_struct *tsk)
158 {
159         WARN_ON(!tsk->exit_state);
160         WARN_ON(atomic_read(&tsk->usage));
161         WARN_ON(tsk == current);
162
163         exit_creds(tsk);
164         delayacct_tsk_free(tsk);
165
166         if (!profile_handoff_task(tsk))
167                 free_task(tsk);
168 }
169
170 /*
171  * macro override instead of weak attribute alias, to workaround
172  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
173  */
174 #ifndef arch_task_cache_init
175 #define arch_task_cache_init()
176 #endif
177
178 void __init fork_init(unsigned long mempages)
179 {
180 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
181 #ifndef ARCH_MIN_TASKALIGN
182 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
183 #endif
184         /* create a slab on which task_structs can be allocated */
185         task_struct_cachep =
186                 kmem_cache_create("task_struct", sizeof(struct task_struct),
187                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
188 #endif
189
190         /* do the arch specific task caches init */
191         arch_task_cache_init();
192
193         /*
194          * The default maximum number of threads is set to a safe
195          * value: the thread structures can take up at most half
196          * of memory.
197          */
198         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
199
200         /*
201          * we need to allow at least 20 threads to boot a system
202          */
203         if(max_threads < 20)
204                 max_threads = 20;
205
206         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
207         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
208         init_task.signal->rlim[RLIMIT_SIGPENDING] =
209                 init_task.signal->rlim[RLIMIT_NPROC];
210 }
211
212 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
213                                                struct task_struct *src)
214 {
215         *dst = *src;
216         return 0;
217 }
218
219 static struct task_struct *dup_task_struct(struct task_struct *orig)
220 {
221         struct task_struct *tsk;
222         struct thread_info *ti;
223         unsigned long *stackend;
224
225         int err;
226
227         prepare_to_copy(orig);
228
229         tsk = alloc_task_struct();
230         if (!tsk)
231                 return NULL;
232
233         ti = alloc_thread_info(tsk);
234         if (!ti) {
235                 free_task_struct(tsk);
236                 return NULL;
237         }
238
239         err = arch_dup_task_struct(tsk, orig);
240         if (err)
241                 goto out;
242
243         tsk->stack = ti;
244
245         err = prop_local_init_single(&tsk->dirties);
246         if (err)
247                 goto out;
248
249         setup_thread_stack(tsk, orig);
250         stackend = end_of_stack(tsk);
251         *stackend = STACK_END_MAGIC;    /* for overflow detection */
252
253 #ifdef CONFIG_CC_STACKPROTECTOR
254         tsk->stack_canary = get_random_int();
255 #endif
256
257         /* One for us, one for whoever does the "release_task()" (usually parent) */
258         atomic_set(&tsk->usage,2);
259         atomic_set(&tsk->fs_excl, 0);
260 #ifdef CONFIG_BLK_DEV_IO_TRACE
261         tsk->btrace_seq = 0;
262 #endif
263         tsk->splice_pipe = NULL;
264
265         account_kernel_stack(ti, 1);
266
267         return tsk;
268
269 out:
270         free_thread_info(ti);
271         free_task_struct(tsk);
272         return NULL;
273 }
274
275 #ifdef CONFIG_MMU
276 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
277 {
278         struct vm_area_struct *mpnt, *tmp, **pprev;
279         struct rb_node **rb_link, *rb_parent;
280         int retval;
281         unsigned long charge;
282         struct mempolicy *pol;
283
284         down_write(&oldmm->mmap_sem);
285         flush_cache_dup_mm(oldmm);
286         /*
287          * Not linked in yet - no deadlock potential:
288          */
289         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
290
291         mm->locked_vm = 0;
292         mm->mmap = NULL;
293         mm->mmap_cache = NULL;
294         mm->free_area_cache = oldmm->mmap_base;
295         mm->cached_hole_size = ~0UL;
296         mm->map_count = 0;
297         cpumask_clear(mm_cpumask(mm));
298         mm->mm_rb = RB_ROOT;
299         rb_link = &mm->mm_rb.rb_node;
300         rb_parent = NULL;
301         pprev = &mm->mmap;
302
303         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
304                 struct file *file;
305
306                 if (mpnt->vm_flags & VM_DONTCOPY) {
307                         long pages = vma_pages(mpnt);
308                         mm->total_vm -= pages;
309                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
310                                                                 -pages);
311                         continue;
312                 }
313                 charge = 0;
314                 if (mpnt->vm_flags & VM_ACCOUNT) {
315                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
316                         if (security_vm_enough_memory(len))
317                                 goto fail_nomem;
318                         charge = len;
319                 }
320                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
321                 if (!tmp)
322                         goto fail_nomem;
323                 *tmp = *mpnt;
324                 pol = mpol_dup(vma_policy(mpnt));
325                 retval = PTR_ERR(pol);
326                 if (IS_ERR(pol))
327                         goto fail_nomem_policy;
328                 vma_set_policy(tmp, pol);
329                 tmp->vm_flags &= ~VM_LOCKED;
330                 tmp->vm_mm = mm;
331                 tmp->vm_next = NULL;
332                 anon_vma_link(tmp);
333                 file = tmp->vm_file;
334                 if (file) {
335                         struct inode *inode = file->f_path.dentry->d_inode;
336                         struct address_space *mapping = file->f_mapping;
337
338                         get_file(file);
339                         if (tmp->vm_flags & VM_DENYWRITE)
340                                 atomic_dec(&inode->i_writecount);
341                         spin_lock(&mapping->i_mmap_lock);
342                         if (tmp->vm_flags & VM_SHARED)
343                                 mapping->i_mmap_writable++;
344                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
345                         flush_dcache_mmap_lock(mapping);
346                         /* insert tmp into the share list, just after mpnt */
347                         vma_prio_tree_add(tmp, mpnt);
348                         flush_dcache_mmap_unlock(mapping);
349                         spin_unlock(&mapping->i_mmap_lock);
350                 }
351
352                 /*
353                  * Clear hugetlb-related page reserves for children. This only
354                  * affects MAP_PRIVATE mappings. Faults generated by the child
355                  * are not guaranteed to succeed, even if read-only
356                  */
357                 if (is_vm_hugetlb_page(tmp))
358                         reset_vma_resv_huge_pages(tmp);
359
360                 /*
361                  * Link in the new vma and copy the page table entries.
362                  */
363                 *pprev = tmp;
364                 pprev = &tmp->vm_next;
365
366                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
367                 rb_link = &tmp->vm_rb.rb_right;
368                 rb_parent = &tmp->vm_rb;
369
370                 mm->map_count++;
371                 retval = copy_page_range(mm, oldmm, mpnt);
372
373                 if (tmp->vm_ops && tmp->vm_ops->open)
374                         tmp->vm_ops->open(tmp);
375
376                 if (retval)
377                         goto out;
378         }
379         /* a new mm has just been created */
380         arch_dup_mmap(oldmm, mm);
381         retval = 0;
382 out:
383         up_write(&mm->mmap_sem);
384         flush_tlb_mm(oldmm);
385         up_write(&oldmm->mmap_sem);
386         return retval;
387 fail_nomem_policy:
388         kmem_cache_free(vm_area_cachep, tmp);
389 fail_nomem:
390         retval = -ENOMEM;
391         vm_unacct_memory(charge);
392         goto out;
393 }
394
395 static inline int mm_alloc_pgd(struct mm_struct * mm)
396 {
397         mm->pgd = pgd_alloc(mm);
398         if (unlikely(!mm->pgd))
399                 return -ENOMEM;
400         return 0;
401 }
402
403 static inline void mm_free_pgd(struct mm_struct * mm)
404 {
405         pgd_free(mm, mm->pgd);
406 }
407 #else
408 #define dup_mmap(mm, oldmm)     (0)
409 #define mm_alloc_pgd(mm)        (0)
410 #define mm_free_pgd(mm)
411 #endif /* CONFIG_MMU */
412
413 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
414
415 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
416 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
417
418 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
419
420 static int __init coredump_filter_setup(char *s)
421 {
422         default_dump_filter =
423                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
424                 MMF_DUMP_FILTER_MASK;
425         return 1;
426 }
427
428 __setup("coredump_filter=", coredump_filter_setup);
429
430 #include <linux/init_task.h>
431
432 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
433 {
434         atomic_set(&mm->mm_users, 1);
435         atomic_set(&mm->mm_count, 1);
436         init_rwsem(&mm->mmap_sem);
437         INIT_LIST_HEAD(&mm->mmlist);
438         mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
439         mm->core_state = NULL;
440         mm->nr_ptes = 0;
441         set_mm_counter(mm, file_rss, 0);
442         set_mm_counter(mm, anon_rss, 0);
443         spin_lock_init(&mm->page_table_lock);
444         spin_lock_init(&mm->ioctx_lock);
445         INIT_HLIST_HEAD(&mm->ioctx_list);
446         mm->free_area_cache = TASK_UNMAPPED_BASE;
447         mm->cached_hole_size = ~0UL;
448         mm_init_owner(mm, p);
449
450         if (likely(!mm_alloc_pgd(mm))) {
451                 mm->def_flags = 0;
452                 mmu_notifier_mm_init(mm);
453                 return mm;
454         }
455
456         free_mm(mm);
457         return NULL;
458 }
459
460 /*
461  * Allocate and initialize an mm_struct.
462  */
463 struct mm_struct * mm_alloc(void)
464 {
465         struct mm_struct * mm;
466
467         mm = allocate_mm();
468         if (mm) {
469                 memset(mm, 0, sizeof(*mm));
470                 mm = mm_init(mm, current);
471         }
472         return mm;
473 }
474
475 /*
476  * Called when the last reference to the mm
477  * is dropped: either by a lazy thread or by
478  * mmput. Free the page directory and the mm.
479  */
480 void __mmdrop(struct mm_struct *mm)
481 {
482         BUG_ON(mm == &init_mm);
483         mm_free_pgd(mm);
484         destroy_context(mm);
485         mmu_notifier_mm_destroy(mm);
486         free_mm(mm);
487 }
488 EXPORT_SYMBOL_GPL(__mmdrop);
489
490 /*
491  * Decrement the use count and release all resources for an mm.
492  */
493 void mmput(struct mm_struct *mm)
494 {
495         might_sleep();
496
497         if (atomic_dec_and_test(&mm->mm_users)) {
498                 exit_aio(mm);
499                 exit_mmap(mm);
500                 set_mm_exe_file(mm, NULL);
501                 if (!list_empty(&mm->mmlist)) {
502                         spin_lock(&mmlist_lock);
503                         list_del(&mm->mmlist);
504                         spin_unlock(&mmlist_lock);
505                 }
506                 put_swap_token(mm);
507                 mmdrop(mm);
508         }
509 }
510 EXPORT_SYMBOL_GPL(mmput);
511
512 /**
513  * get_task_mm - acquire a reference to the task's mm
514  *
515  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
516  * this kernel workthread has transiently adopted a user mm with use_mm,
517  * to do its AIO) is not set and if so returns a reference to it, after
518  * bumping up the use count.  User must release the mm via mmput()
519  * after use.  Typically used by /proc and ptrace.
520  */
521 struct mm_struct *get_task_mm(struct task_struct *task)
522 {
523         struct mm_struct *mm;
524
525         task_lock(task);
526         mm = task->mm;
527         if (mm) {
528                 if (task->flags & PF_KTHREAD)
529                         mm = NULL;
530                 else
531                         atomic_inc(&mm->mm_users);
532         }
533         task_unlock(task);
534         return mm;
535 }
536 EXPORT_SYMBOL_GPL(get_task_mm);
537
538 /* Please note the differences between mmput and mm_release.
539  * mmput is called whenever we stop holding onto a mm_struct,
540  * error success whatever.
541  *
542  * mm_release is called after a mm_struct has been removed
543  * from the current process.
544  *
545  * This difference is important for error handling, when we
546  * only half set up a mm_struct for a new process and need to restore
547  * the old one.  Because we mmput the new mm_struct before
548  * restoring the old one. . .
549  * Eric Biederman 10 January 1998
550  */
551 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
552 {
553         struct completion *vfork_done = tsk->vfork_done;
554
555         /* Get rid of any futexes when releasing the mm */
556 #ifdef CONFIG_FUTEX
557         if (unlikely(tsk->robust_list))
558                 exit_robust_list(tsk);
559 #ifdef CONFIG_COMPAT
560         if (unlikely(tsk->compat_robust_list))
561                 compat_exit_robust_list(tsk);
562 #endif
563 #endif
564
565         /* Get rid of any cached register state */
566         deactivate_mm(tsk, mm);
567
568         /* notify parent sleeping on vfork() */
569         if (vfork_done) {
570                 tsk->vfork_done = NULL;
571                 complete(vfork_done);
572         }
573
574         /*
575          * If we're exiting normally, clear a user-space tid field if
576          * requested.  We leave this alone when dying by signal, to leave
577          * the value intact in a core dump, and to save the unnecessary
578          * trouble otherwise.  Userland only wants this done for a sys_exit.
579          */
580         if (tsk->clear_child_tid) {
581                 if (!(tsk->flags & PF_SIGNALED) &&
582                     atomic_read(&mm->mm_users) > 1) {
583                         /*
584                          * We don't check the error code - if userspace has
585                          * not set up a proper pointer then tough luck.
586                          */
587                         put_user(0, tsk->clear_child_tid);
588                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
589                                         1, NULL, NULL, 0);
590                 }
591                 tsk->clear_child_tid = NULL;
592         }
593 }
594
595 /*
596  * Allocate a new mm structure and copy contents from the
597  * mm structure of the passed in task structure.
598  */
599 struct mm_struct *dup_mm(struct task_struct *tsk)
600 {
601         struct mm_struct *mm, *oldmm = current->mm;
602         int err;
603
604         if (!oldmm)
605                 return NULL;
606
607         mm = allocate_mm();
608         if (!mm)
609                 goto fail_nomem;
610
611         memcpy(mm, oldmm, sizeof(*mm));
612
613         /* Initializing for Swap token stuff */
614         mm->token_priority = 0;
615         mm->last_interval = 0;
616
617         if (!mm_init(mm, tsk))
618                 goto fail_nomem;
619
620         if (init_new_context(tsk, mm))
621                 goto fail_nocontext;
622
623         dup_mm_exe_file(oldmm, mm);
624
625         err = dup_mmap(mm, oldmm);
626         if (err)
627                 goto free_pt;
628
629         mm->hiwater_rss = get_mm_rss(mm);
630         mm->hiwater_vm = mm->total_vm;
631
632         return mm;
633
634 free_pt:
635         mmput(mm);
636
637 fail_nomem:
638         return NULL;
639
640 fail_nocontext:
641         /*
642          * If init_new_context() failed, we cannot use mmput() to free the mm
643          * because it calls destroy_context()
644          */
645         mm_free_pgd(mm);
646         free_mm(mm);
647         return NULL;
648 }
649
650 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
651 {
652         struct mm_struct * mm, *oldmm;
653         int retval;
654
655         tsk->min_flt = tsk->maj_flt = 0;
656         tsk->nvcsw = tsk->nivcsw = 0;
657 #ifdef CONFIG_DETECT_HUNG_TASK
658         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
659 #endif
660
661         tsk->mm = NULL;
662         tsk->active_mm = NULL;
663
664         /*
665          * Are we cloning a kernel thread?
666          *
667          * We need to steal a active VM for that..
668          */
669         oldmm = current->mm;
670         if (!oldmm)
671                 return 0;
672
673         if (clone_flags & CLONE_VM) {
674                 atomic_inc(&oldmm->mm_users);
675                 mm = oldmm;
676                 goto good_mm;
677         }
678
679         retval = -ENOMEM;
680         mm = dup_mm(tsk);
681         if (!mm)
682                 goto fail_nomem;
683
684 good_mm:
685         /* Initializing for Swap token stuff */
686         mm->token_priority = 0;
687         mm->last_interval = 0;
688
689         tsk->mm = mm;
690         tsk->active_mm = mm;
691         return 0;
692
693 fail_nomem:
694         return retval;
695 }
696
697 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
698 {
699         struct fs_struct *fs = current->fs;
700         if (clone_flags & CLONE_FS) {
701                 /* tsk->fs is already what we want */
702                 write_lock(&fs->lock);
703                 if (fs->in_exec) {
704                         write_unlock(&fs->lock);
705                         return -EAGAIN;
706                 }
707                 fs->users++;
708                 write_unlock(&fs->lock);
709                 return 0;
710         }
711         tsk->fs = copy_fs_struct(fs);
712         if (!tsk->fs)
713                 return -ENOMEM;
714         return 0;
715 }
716
717 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
718 {
719         struct files_struct *oldf, *newf;
720         int error = 0;
721
722         /*
723          * A background process may not have any files ...
724          */
725         oldf = current->files;
726         if (!oldf)
727                 goto out;
728
729         if (clone_flags & CLONE_FILES) {
730                 atomic_inc(&oldf->count);
731                 goto out;
732         }
733
734         newf = dup_fd(oldf, &error);
735         if (!newf)
736                 goto out;
737
738         tsk->files = newf;
739         error = 0;
740 out:
741         return error;
742 }
743
744 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
745 {
746 #ifdef CONFIG_BLOCK
747         struct io_context *ioc = current->io_context;
748
749         if (!ioc)
750                 return 0;
751         /*
752          * Share io context with parent, if CLONE_IO is set
753          */
754         if (clone_flags & CLONE_IO) {
755                 tsk->io_context = ioc_task_link(ioc);
756                 if (unlikely(!tsk->io_context))
757                         return -ENOMEM;
758         } else if (ioprio_valid(ioc->ioprio)) {
759                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
760                 if (unlikely(!tsk->io_context))
761                         return -ENOMEM;
762
763                 tsk->io_context->ioprio = ioc->ioprio;
764         }
765 #endif
766         return 0;
767 }
768
769 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
770 {
771         struct sighand_struct *sig;
772
773         if (clone_flags & CLONE_SIGHAND) {
774                 atomic_inc(&current->sighand->count);
775                 return 0;
776         }
777         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
778         rcu_assign_pointer(tsk->sighand, sig);
779         if (!sig)
780                 return -ENOMEM;
781         atomic_set(&sig->count, 1);
782         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
783         return 0;
784 }
785
786 void __cleanup_sighand(struct sighand_struct *sighand)
787 {
788         if (atomic_dec_and_test(&sighand->count))
789                 kmem_cache_free(sighand_cachep, sighand);
790 }
791
792
793 /*
794  * Initialize POSIX timer handling for a thread group.
795  */
796 static void posix_cpu_timers_init_group(struct signal_struct *sig)
797 {
798         /* Thread group counters. */
799         thread_group_cputime_init(sig);
800
801         /* Expiration times and increments. */
802         sig->it_virt_expires = cputime_zero;
803         sig->it_virt_incr = cputime_zero;
804         sig->it_prof_expires = cputime_zero;
805         sig->it_prof_incr = cputime_zero;
806
807         /* Cached expiration times. */
808         sig->cputime_expires.prof_exp = cputime_zero;
809         sig->cputime_expires.virt_exp = cputime_zero;
810         sig->cputime_expires.sched_exp = 0;
811
812         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
813                 sig->cputime_expires.prof_exp =
814                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
815                 sig->cputimer.running = 1;
816         }
817
818         /* The timer lists. */
819         INIT_LIST_HEAD(&sig->cpu_timers[0]);
820         INIT_LIST_HEAD(&sig->cpu_timers[1]);
821         INIT_LIST_HEAD(&sig->cpu_timers[2]);
822 }
823
824 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
825 {
826         struct signal_struct *sig;
827
828         if (clone_flags & CLONE_THREAD)
829                 return 0;
830
831         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
832         tsk->signal = sig;
833         if (!sig)
834                 return -ENOMEM;
835
836         atomic_set(&sig->count, 1);
837         atomic_set(&sig->live, 1);
838         init_waitqueue_head(&sig->wait_chldexit);
839         sig->flags = 0;
840         if (clone_flags & CLONE_NEWPID)
841                 sig->flags |= SIGNAL_UNKILLABLE;
842         sig->group_exit_code = 0;
843         sig->group_exit_task = NULL;
844         sig->group_stop_count = 0;
845         sig->curr_target = tsk;
846         init_sigpending(&sig->shared_pending);
847         INIT_LIST_HEAD(&sig->posix_timers);
848
849         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
850         sig->it_real_incr.tv64 = 0;
851         sig->real_timer.function = it_real_fn;
852
853         sig->leader = 0;        /* session leadership doesn't inherit */
854         sig->tty_old_pgrp = NULL;
855         sig->tty = NULL;
856
857         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
858         sig->gtime = cputime_zero;
859         sig->cgtime = cputime_zero;
860         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
861         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
862         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
863         task_io_accounting_init(&sig->ioac);
864         sig->sum_sched_runtime = 0;
865         taskstats_tgid_init(sig);
866
867         task_lock(current->group_leader);
868         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
869         task_unlock(current->group_leader);
870
871         posix_cpu_timers_init_group(sig);
872
873         acct_init_pacct(&sig->pacct);
874
875         tty_audit_fork(sig);
876
877         return 0;
878 }
879
880 void __cleanup_signal(struct signal_struct *sig)
881 {
882         thread_group_cputime_free(sig);
883         tty_kref_put(sig->tty);
884         kmem_cache_free(signal_cachep, sig);
885 }
886
887 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
888 {
889         unsigned long new_flags = p->flags;
890
891         new_flags &= ~PF_SUPERPRIV;
892         new_flags |= PF_FORKNOEXEC;
893         new_flags |= PF_STARTING;
894         p->flags = new_flags;
895         clear_freeze_flag(p);
896 }
897
898 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
899 {
900         current->clear_child_tid = tidptr;
901
902         return task_pid_vnr(current);
903 }
904
905 static void rt_mutex_init_task(struct task_struct *p)
906 {
907         spin_lock_init(&p->pi_lock);
908 #ifdef CONFIG_RT_MUTEXES
909         plist_head_init(&p->pi_waiters, &p->pi_lock);
910         p->pi_blocked_on = NULL;
911 #endif
912 }
913
914 #ifdef CONFIG_MM_OWNER
915 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
916 {
917         mm->owner = p;
918 }
919 #endif /* CONFIG_MM_OWNER */
920
921 /*
922  * Initialize POSIX timer handling for a single task.
923  */
924 static void posix_cpu_timers_init(struct task_struct *tsk)
925 {
926         tsk->cputime_expires.prof_exp = cputime_zero;
927         tsk->cputime_expires.virt_exp = cputime_zero;
928         tsk->cputime_expires.sched_exp = 0;
929         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
930         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
931         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
932 }
933
934 /*
935  * This creates a new process as a copy of the old one,
936  * but does not actually start it yet.
937  *
938  * It copies the registers, and all the appropriate
939  * parts of the process environment (as per the clone
940  * flags). The actual kick-off is left to the caller.
941  */
942 static struct task_struct *copy_process(unsigned long clone_flags,
943                                         unsigned long stack_start,
944                                         struct pt_regs *regs,
945                                         unsigned long stack_size,
946                                         int __user *child_tidptr,
947                                         struct pid *pid,
948                                         int trace)
949 {
950         int retval;
951         struct task_struct *p;
952         int cgroup_callbacks_done = 0;
953
954         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
955                 return ERR_PTR(-EINVAL);
956
957         /*
958          * Thread groups must share signals as well, and detached threads
959          * can only be started up within the thread group.
960          */
961         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
962                 return ERR_PTR(-EINVAL);
963
964         /*
965          * Shared signal handlers imply shared VM. By way of the above,
966          * thread groups also imply shared VM. Blocking this case allows
967          * for various simplifications in other code.
968          */
969         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
970                 return ERR_PTR(-EINVAL);
971
972         retval = security_task_create(clone_flags);
973         if (retval)
974                 goto fork_out;
975
976         retval = -ENOMEM;
977         p = dup_task_struct(current);
978         if (!p)
979                 goto fork_out;
980
981         ftrace_graph_init_task(p);
982
983         rt_mutex_init_task(p);
984
985 #ifdef CONFIG_PROVE_LOCKING
986         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
987         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
988 #endif
989         retval = -EAGAIN;
990         if (atomic_read(&p->real_cred->user->processes) >=
991                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
992                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
993                     p->real_cred->user != INIT_USER)
994                         goto bad_fork_free;
995         }
996
997         retval = copy_creds(p, clone_flags);
998         if (retval < 0)
999                 goto bad_fork_free;
1000
1001         /*
1002          * If multiple threads are within copy_process(), then this check
1003          * triggers too late. This doesn't hurt, the check is only there
1004          * to stop root fork bombs.
1005          */
1006         retval = -EAGAIN;
1007         if (nr_threads >= max_threads)
1008                 goto bad_fork_cleanup_count;
1009
1010         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1011                 goto bad_fork_cleanup_count;
1012
1013         if (p->binfmt && !try_module_get(p->binfmt->module))
1014                 goto bad_fork_cleanup_put_domain;
1015
1016         p->did_exec = 0;
1017         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1018         copy_flags(clone_flags, p);
1019         INIT_LIST_HEAD(&p->children);
1020         INIT_LIST_HEAD(&p->sibling);
1021         rcu_copy_process(p);
1022         p->vfork_done = NULL;
1023         spin_lock_init(&p->alloc_lock);
1024
1025         init_sigpending(&p->pending);
1026
1027         p->utime = cputime_zero;
1028         p->stime = cputime_zero;
1029         p->gtime = cputime_zero;
1030         p->utimescaled = cputime_zero;
1031         p->stimescaled = cputime_zero;
1032         p->prev_utime = cputime_zero;
1033         p->prev_stime = cputime_zero;
1034
1035         p->default_timer_slack_ns = current->timer_slack_ns;
1036
1037         task_io_accounting_init(&p->ioac);
1038         acct_clear_integrals(p);
1039
1040         posix_cpu_timers_init(p);
1041
1042         p->lock_depth = -1;             /* -1 = no lock */
1043         do_posix_clock_monotonic_gettime(&p->start_time);
1044         p->real_start_time = p->start_time;
1045         monotonic_to_bootbased(&p->real_start_time);
1046         p->io_context = NULL;
1047         p->audit_context = NULL;
1048         cgroup_fork(p);
1049 #ifdef CONFIG_NUMA
1050         p->mempolicy = mpol_dup(p->mempolicy);
1051         if (IS_ERR(p->mempolicy)) {
1052                 retval = PTR_ERR(p->mempolicy);
1053                 p->mempolicy = NULL;
1054                 goto bad_fork_cleanup_cgroup;
1055         }
1056         mpol_fix_fork_child_flag(p);
1057 #endif
1058 #ifdef CONFIG_TRACE_IRQFLAGS
1059         p->irq_events = 0;
1060 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1061         p->hardirqs_enabled = 1;
1062 #else
1063         p->hardirqs_enabled = 0;
1064 #endif
1065         p->hardirq_enable_ip = 0;
1066         p->hardirq_enable_event = 0;
1067         p->hardirq_disable_ip = _THIS_IP_;
1068         p->hardirq_disable_event = 0;
1069         p->softirqs_enabled = 1;
1070         p->softirq_enable_ip = _THIS_IP_;
1071         p->softirq_enable_event = 0;
1072         p->softirq_disable_ip = 0;
1073         p->softirq_disable_event = 0;
1074         p->hardirq_context = 0;
1075         p->softirq_context = 0;
1076 #endif
1077 #ifdef CONFIG_LOCKDEP
1078         p->lockdep_depth = 0; /* no locks held yet */
1079         p->curr_chain_key = 0;
1080         p->lockdep_recursion = 0;
1081 #endif
1082
1083 #ifdef CONFIG_DEBUG_MUTEXES
1084         p->blocked_on = NULL; /* not blocked yet */
1085 #endif
1086
1087         p->bts = NULL;
1088
1089         /* Perform scheduler related setup. Assign this task to a CPU. */
1090         sched_fork(p, clone_flags);
1091
1092         retval = perf_event_init_task(p);
1093         if (retval)
1094                 goto bad_fork_cleanup_policy;
1095
1096         if ((retval = audit_alloc(p)))
1097                 goto bad_fork_cleanup_policy;
1098         /* copy all the process information */
1099         if ((retval = copy_semundo(clone_flags, p)))
1100                 goto bad_fork_cleanup_audit;
1101         if ((retval = copy_files(clone_flags, p)))
1102                 goto bad_fork_cleanup_semundo;
1103         if ((retval = copy_fs(clone_flags, p)))
1104                 goto bad_fork_cleanup_files;
1105         if ((retval = copy_sighand(clone_flags, p)))
1106                 goto bad_fork_cleanup_fs;
1107         if ((retval = copy_signal(clone_flags, p)))
1108                 goto bad_fork_cleanup_sighand;
1109         if ((retval = copy_mm(clone_flags, p)))
1110                 goto bad_fork_cleanup_signal;
1111         if ((retval = copy_namespaces(clone_flags, p)))
1112                 goto bad_fork_cleanup_mm;
1113         if ((retval = copy_io(clone_flags, p)))
1114                 goto bad_fork_cleanup_namespaces;
1115         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1116         if (retval)
1117                 goto bad_fork_cleanup_io;
1118
1119         if (pid != &init_struct_pid) {
1120                 retval = -ENOMEM;
1121                 pid = alloc_pid(p->nsproxy->pid_ns);
1122                 if (!pid)
1123                         goto bad_fork_cleanup_io;
1124
1125                 if (clone_flags & CLONE_NEWPID) {
1126                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1127                         if (retval < 0)
1128                                 goto bad_fork_free_pid;
1129                 }
1130         }
1131
1132         p->pid = pid_nr(pid);
1133         p->tgid = p->pid;
1134         if (clone_flags & CLONE_THREAD)
1135                 p->tgid = current->tgid;
1136
1137         if (current->nsproxy != p->nsproxy) {
1138                 retval = ns_cgroup_clone(p, pid);
1139                 if (retval)
1140                         goto bad_fork_free_pid;
1141         }
1142
1143         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1144         /*
1145          * Clear TID on mm_release()?
1146          */
1147         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1148 #ifdef CONFIG_FUTEX
1149         p->robust_list = NULL;
1150 #ifdef CONFIG_COMPAT
1151         p->compat_robust_list = NULL;
1152 #endif
1153         INIT_LIST_HEAD(&p->pi_state_list);
1154         p->pi_state_cache = NULL;
1155 #endif
1156         /*
1157          * sigaltstack should be cleared when sharing the same VM
1158          */
1159         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1160                 p->sas_ss_sp = p->sas_ss_size = 0;
1161
1162         /*
1163          * Syscall tracing should be turned off in the child regardless
1164          * of CLONE_PTRACE.
1165          */
1166         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1167 #ifdef TIF_SYSCALL_EMU
1168         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1169 #endif
1170         clear_all_latency_tracing(p);
1171
1172         /* ok, now we should be set up.. */
1173         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1174         p->pdeath_signal = 0;
1175         p->exit_state = 0;
1176
1177         /*
1178          * Ok, make it visible to the rest of the system.
1179          * We dont wake it up yet.
1180          */
1181         p->group_leader = p;
1182         INIT_LIST_HEAD(&p->thread_group);
1183
1184         /* Now that the task is set up, run cgroup callbacks if
1185          * necessary. We need to run them before the task is visible
1186          * on the tasklist. */
1187         cgroup_fork_callbacks(p);
1188         cgroup_callbacks_done = 1;
1189
1190         /* Need tasklist lock for parent etc handling! */
1191         write_lock_irq(&tasklist_lock);
1192
1193         /*
1194          * The task hasn't been attached yet, so its cpus_allowed mask will
1195          * not be changed, nor will its assigned CPU.
1196          *
1197          * The cpus_allowed mask of the parent may have changed after it was
1198          * copied first time - so re-copy it here, then check the child's CPU
1199          * to ensure it is on a valid CPU (and if not, just force it back to
1200          * parent's CPU). This avoids alot of nasty races.
1201          */
1202         p->cpus_allowed = current->cpus_allowed;
1203         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1204         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1205                         !cpu_online(task_cpu(p))))
1206                 set_task_cpu(p, smp_processor_id());
1207
1208         /* CLONE_PARENT re-uses the old parent */
1209         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1210                 p->real_parent = current->real_parent;
1211                 p->parent_exec_id = current->parent_exec_id;
1212         } else {
1213                 p->real_parent = current;
1214                 p->parent_exec_id = current->self_exec_id;
1215         }
1216
1217         spin_lock(&current->sighand->siglock);
1218
1219         /*
1220          * Process group and session signals need to be delivered to just the
1221          * parent before the fork or both the parent and the child after the
1222          * fork. Restart if a signal comes in before we add the new process to
1223          * it's process group.
1224          * A fatal signal pending means that current will exit, so the new
1225          * thread can't slip out of an OOM kill (or normal SIGKILL).
1226          */
1227         recalc_sigpending();
1228         if (signal_pending(current)) {
1229                 spin_unlock(&current->sighand->siglock);
1230                 write_unlock_irq(&tasklist_lock);
1231                 retval = -ERESTARTNOINTR;
1232                 goto bad_fork_free_pid;
1233         }
1234
1235         if (clone_flags & CLONE_THREAD) {
1236                 atomic_inc(&current->signal->count);
1237                 atomic_inc(&current->signal->live);
1238                 p->group_leader = current->group_leader;
1239                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1240         }
1241
1242         if (likely(p->pid)) {
1243                 list_add_tail(&p->sibling, &p->real_parent->children);
1244                 tracehook_finish_clone(p, clone_flags, trace);
1245
1246                 if (thread_group_leader(p)) {
1247                         if (clone_flags & CLONE_NEWPID)
1248                                 p->nsproxy->pid_ns->child_reaper = p;
1249
1250                         p->signal->leader_pid = pid;
1251                         tty_kref_put(p->signal->tty);
1252                         p->signal->tty = tty_kref_get(current->signal->tty);
1253                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1254                         attach_pid(p, PIDTYPE_SID, task_session(current));
1255                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1256                         __get_cpu_var(process_counts)++;
1257                 }
1258                 attach_pid(p, PIDTYPE_PID, pid);
1259                 nr_threads++;
1260         }
1261
1262         total_forks++;
1263         spin_unlock(&current->sighand->siglock);
1264         write_unlock_irq(&tasklist_lock);
1265         proc_fork_connector(p);
1266         cgroup_post_fork(p);
1267         perf_event_fork(p);
1268         return p;
1269
1270 bad_fork_free_pid:
1271         if (pid != &init_struct_pid)
1272                 free_pid(pid);
1273 bad_fork_cleanup_io:
1274         put_io_context(p->io_context);
1275 bad_fork_cleanup_namespaces:
1276         exit_task_namespaces(p);
1277 bad_fork_cleanup_mm:
1278         if (p->mm)
1279                 mmput(p->mm);
1280 bad_fork_cleanup_signal:
1281         if (!(clone_flags & CLONE_THREAD))
1282                 __cleanup_signal(p->signal);
1283 bad_fork_cleanup_sighand:
1284         __cleanup_sighand(p->sighand);
1285 bad_fork_cleanup_fs:
1286         exit_fs(p); /* blocking */
1287 bad_fork_cleanup_files:
1288         exit_files(p); /* blocking */
1289 bad_fork_cleanup_semundo:
1290         exit_sem(p);
1291 bad_fork_cleanup_audit:
1292         audit_free(p);
1293 bad_fork_cleanup_policy:
1294         perf_event_free_task(p);
1295 #ifdef CONFIG_NUMA
1296         mpol_put(p->mempolicy);
1297 bad_fork_cleanup_cgroup:
1298 #endif
1299         cgroup_exit(p, cgroup_callbacks_done);
1300         delayacct_tsk_free(p);
1301         if (p->binfmt)
1302                 module_put(p->binfmt->module);
1303 bad_fork_cleanup_put_domain:
1304         module_put(task_thread_info(p)->exec_domain->module);
1305 bad_fork_cleanup_count:
1306         atomic_dec(&p->cred->user->processes);
1307         exit_creds(p);
1308 bad_fork_free:
1309         free_task(p);
1310 fork_out:
1311         return ERR_PTR(retval);
1312 }
1313
1314 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1315 {
1316         memset(regs, 0, sizeof(struct pt_regs));
1317         return regs;
1318 }
1319
1320 struct task_struct * __cpuinit fork_idle(int cpu)
1321 {
1322         struct task_struct *task;
1323         struct pt_regs regs;
1324
1325         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1326                             &init_struct_pid, 0);
1327         if (!IS_ERR(task))
1328                 init_idle(task, cpu);
1329
1330         return task;
1331 }
1332
1333 /*
1334  *  Ok, this is the main fork-routine.
1335  *
1336  * It copies the process, and if successful kick-starts
1337  * it and waits for it to finish using the VM if required.
1338  */
1339 long do_fork(unsigned long clone_flags,
1340               unsigned long stack_start,
1341               struct pt_regs *regs,
1342               unsigned long stack_size,
1343               int __user *parent_tidptr,
1344               int __user *child_tidptr)
1345 {
1346         struct task_struct *p;
1347         int trace = 0;
1348         long nr;
1349
1350         /*
1351          * Do some preliminary argument and permissions checking before we
1352          * actually start allocating stuff
1353          */
1354         if (clone_flags & CLONE_NEWUSER) {
1355                 if (clone_flags & CLONE_THREAD)
1356                         return -EINVAL;
1357                 /* hopefully this check will go away when userns support is
1358                  * complete
1359                  */
1360                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1361                                 !capable(CAP_SETGID))
1362                         return -EPERM;
1363         }
1364
1365         /*
1366          * We hope to recycle these flags after 2.6.26
1367          */
1368         if (unlikely(clone_flags & CLONE_STOPPED)) {
1369                 static int __read_mostly count = 100;
1370
1371                 if (count > 0 && printk_ratelimit()) {
1372                         char comm[TASK_COMM_LEN];
1373
1374                         count--;
1375                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1376                                         "clone flags 0x%lx\n",
1377                                 get_task_comm(comm, current),
1378                                 clone_flags & CLONE_STOPPED);
1379                 }
1380         }
1381
1382         /*
1383          * When called from kernel_thread, don't do user tracing stuff.
1384          */
1385         if (likely(user_mode(regs)))
1386                 trace = tracehook_prepare_clone(clone_flags);
1387
1388         p = copy_process(clone_flags, stack_start, regs, stack_size,
1389                          child_tidptr, NULL, trace);
1390         /*
1391          * Do this prior waking up the new thread - the thread pointer
1392          * might get invalid after that point, if the thread exits quickly.
1393          */
1394         if (!IS_ERR(p)) {
1395                 struct completion vfork;
1396
1397                 trace_sched_process_fork(current, p);
1398
1399                 nr = task_pid_vnr(p);
1400
1401                 if (clone_flags & CLONE_PARENT_SETTID)
1402                         put_user(nr, parent_tidptr);
1403
1404                 if (clone_flags & CLONE_VFORK) {
1405                         p->vfork_done = &vfork;
1406                         init_completion(&vfork);
1407                 }
1408
1409                 audit_finish_fork(p);
1410                 tracehook_report_clone(regs, clone_flags, nr, p);
1411
1412                 /*
1413                  * We set PF_STARTING at creation in case tracing wants to
1414                  * use this to distinguish a fully live task from one that
1415                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1416                  * clear it and set the child going.
1417                  */
1418                 p->flags &= ~PF_STARTING;
1419
1420                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1421                         /*
1422                          * We'll start up with an immediate SIGSTOP.
1423                          */
1424                         sigaddset(&p->pending.signal, SIGSTOP);
1425                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1426                         __set_task_state(p, TASK_STOPPED);
1427                 } else {
1428                         wake_up_new_task(p, clone_flags);
1429                 }
1430
1431                 tracehook_report_clone_complete(trace, regs,
1432                                                 clone_flags, nr, p);
1433
1434                 if (clone_flags & CLONE_VFORK) {
1435                         freezer_do_not_count();
1436                         wait_for_completion(&vfork);
1437                         freezer_count();
1438                         tracehook_report_vfork_done(p, nr);
1439                 }
1440         } else {
1441                 nr = PTR_ERR(p);
1442         }
1443         return nr;
1444 }
1445
1446 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1447 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1448 #endif
1449
1450 static void sighand_ctor(void *data)
1451 {
1452         struct sighand_struct *sighand = data;
1453
1454         spin_lock_init(&sighand->siglock);
1455         init_waitqueue_head(&sighand->signalfd_wqh);
1456 }
1457
1458 void __init proc_caches_init(void)
1459 {
1460         sighand_cachep = kmem_cache_create("sighand_cache",
1461                         sizeof(struct sighand_struct), 0,
1462                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1463                         SLAB_NOTRACK, sighand_ctor);
1464         signal_cachep = kmem_cache_create("signal_cache",
1465                         sizeof(struct signal_struct), 0,
1466                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1467         files_cachep = kmem_cache_create("files_cache",
1468                         sizeof(struct files_struct), 0,
1469                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1470         fs_cachep = kmem_cache_create("fs_cache",
1471                         sizeof(struct fs_struct), 0,
1472                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1473         mm_cachep = kmem_cache_create("mm_struct",
1474                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1475                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1476         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1477         mmap_init();
1478 }
1479
1480 /*
1481  * Check constraints on flags passed to the unshare system call and
1482  * force unsharing of additional process context as appropriate.
1483  */
1484 static void check_unshare_flags(unsigned long *flags_ptr)
1485 {
1486         /*
1487          * If unsharing a thread from a thread group, must also
1488          * unshare vm.
1489          */
1490         if (*flags_ptr & CLONE_THREAD)
1491                 *flags_ptr |= CLONE_VM;
1492
1493         /*
1494          * If unsharing vm, must also unshare signal handlers.
1495          */
1496         if (*flags_ptr & CLONE_VM)
1497                 *flags_ptr |= CLONE_SIGHAND;
1498
1499         /*
1500          * If unsharing signal handlers and the task was created
1501          * using CLONE_THREAD, then must unshare the thread
1502          */
1503         if ((*flags_ptr & CLONE_SIGHAND) &&
1504             (atomic_read(&current->signal->count) > 1))
1505                 *flags_ptr |= CLONE_THREAD;
1506
1507         /*
1508          * If unsharing namespace, must also unshare filesystem information.
1509          */
1510         if (*flags_ptr & CLONE_NEWNS)
1511                 *flags_ptr |= CLONE_FS;
1512 }
1513
1514 /*
1515  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1516  */
1517 static int unshare_thread(unsigned long unshare_flags)
1518 {
1519         if (unshare_flags & CLONE_THREAD)
1520                 return -EINVAL;
1521
1522         return 0;
1523 }
1524
1525 /*
1526  * Unshare the filesystem structure if it is being shared
1527  */
1528 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1529 {
1530         struct fs_struct *fs = current->fs;
1531
1532         if (!(unshare_flags & CLONE_FS) || !fs)
1533                 return 0;
1534
1535         /* don't need lock here; in the worst case we'll do useless copy */
1536         if (fs->users == 1)
1537                 return 0;
1538
1539         *new_fsp = copy_fs_struct(fs);
1540         if (!*new_fsp)
1541                 return -ENOMEM;
1542
1543         return 0;
1544 }
1545
1546 /*
1547  * Unsharing of sighand is not supported yet
1548  */
1549 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1550 {
1551         struct sighand_struct *sigh = current->sighand;
1552
1553         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1554                 return -EINVAL;
1555         else
1556                 return 0;
1557 }
1558
1559 /*
1560  * Unshare vm if it is being shared
1561  */
1562 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1563 {
1564         struct mm_struct *mm = current->mm;
1565
1566         if ((unshare_flags & CLONE_VM) &&
1567             (mm && atomic_read(&mm->mm_users) > 1)) {
1568                 return -EINVAL;
1569         }
1570
1571         return 0;
1572 }
1573
1574 /*
1575  * Unshare file descriptor table if it is being shared
1576  */
1577 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1578 {
1579         struct files_struct *fd = current->files;
1580         int error = 0;
1581
1582         if ((unshare_flags & CLONE_FILES) &&
1583             (fd && atomic_read(&fd->count) > 1)) {
1584                 *new_fdp = dup_fd(fd, &error);
1585                 if (!*new_fdp)
1586                         return error;
1587         }
1588
1589         return 0;
1590 }
1591
1592 /*
1593  * unshare allows a process to 'unshare' part of the process
1594  * context which was originally shared using clone.  copy_*
1595  * functions used by do_fork() cannot be used here directly
1596  * because they modify an inactive task_struct that is being
1597  * constructed. Here we are modifying the current, active,
1598  * task_struct.
1599  */
1600 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1601 {
1602         int err = 0;
1603         struct fs_struct *fs, *new_fs = NULL;
1604         struct sighand_struct *new_sigh = NULL;
1605         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1606         struct files_struct *fd, *new_fd = NULL;
1607         struct nsproxy *new_nsproxy = NULL;
1608         int do_sysvsem = 0;
1609
1610         check_unshare_flags(&unshare_flags);
1611
1612         /* Return -EINVAL for all unsupported flags */
1613         err = -EINVAL;
1614         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1615                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1616                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1617                 goto bad_unshare_out;
1618
1619         /*
1620          * CLONE_NEWIPC must also detach from the undolist: after switching
1621          * to a new ipc namespace, the semaphore arrays from the old
1622          * namespace are unreachable.
1623          */
1624         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1625                 do_sysvsem = 1;
1626         if ((err = unshare_thread(unshare_flags)))
1627                 goto bad_unshare_out;
1628         if ((err = unshare_fs(unshare_flags, &new_fs)))
1629                 goto bad_unshare_cleanup_thread;
1630         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1631                 goto bad_unshare_cleanup_fs;
1632         if ((err = unshare_vm(unshare_flags, &new_mm)))
1633                 goto bad_unshare_cleanup_sigh;
1634         if ((err = unshare_fd(unshare_flags, &new_fd)))
1635                 goto bad_unshare_cleanup_vm;
1636         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1637                         new_fs)))
1638                 goto bad_unshare_cleanup_fd;
1639
1640         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1641                 if (do_sysvsem) {
1642                         /*
1643                          * CLONE_SYSVSEM is equivalent to sys_exit().
1644                          */
1645                         exit_sem(current);
1646                 }
1647
1648                 if (new_nsproxy) {
1649                         switch_task_namespaces(current, new_nsproxy);
1650                         new_nsproxy = NULL;
1651                 }
1652
1653                 task_lock(current);
1654
1655                 if (new_fs) {
1656                         fs = current->fs;
1657                         write_lock(&fs->lock);
1658                         current->fs = new_fs;
1659                         if (--fs->users)
1660                                 new_fs = NULL;
1661                         else
1662                                 new_fs = fs;
1663                         write_unlock(&fs->lock);
1664                 }
1665
1666                 if (new_mm) {
1667                         mm = current->mm;
1668                         active_mm = current->active_mm;
1669                         current->mm = new_mm;
1670                         current->active_mm = new_mm;
1671                         activate_mm(active_mm, new_mm);
1672                         new_mm = mm;
1673                 }
1674
1675                 if (new_fd) {
1676                         fd = current->files;
1677                         current->files = new_fd;
1678                         new_fd = fd;
1679                 }
1680
1681                 task_unlock(current);
1682         }
1683
1684         if (new_nsproxy)
1685                 put_nsproxy(new_nsproxy);
1686
1687 bad_unshare_cleanup_fd:
1688         if (new_fd)
1689                 put_files_struct(new_fd);
1690
1691 bad_unshare_cleanup_vm:
1692         if (new_mm)
1693                 mmput(new_mm);
1694
1695 bad_unshare_cleanup_sigh:
1696         if (new_sigh)
1697                 if (atomic_dec_and_test(&new_sigh->count))
1698                         kmem_cache_free(sighand_cachep, new_sigh);
1699
1700 bad_unshare_cleanup_fs:
1701         if (new_fs)
1702                 free_fs_struct(new_fs);
1703
1704 bad_unshare_cleanup_thread:
1705 bad_unshare_out:
1706         return err;
1707 }
1708
1709 /*
1710  *      Helper to unshare the files of the current task.
1711  *      We don't want to expose copy_files internals to
1712  *      the exec layer of the kernel.
1713  */
1714
1715 int unshare_files(struct files_struct **displaced)
1716 {
1717         struct task_struct *task = current;
1718         struct files_struct *copy = NULL;
1719         int error;
1720
1721         error = unshare_fd(CLONE_FILES, &copy);
1722         if (error || !copy) {
1723                 *displaced = NULL;
1724                 return error;
1725         }
1726         *displaced = task->files;
1727         task_lock(task);
1728         task->files = copy;
1729         task_unlock(task);
1730         return 0;
1731 }