Merge branch 'timers-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[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
68 #include <asm/pgtable.h>
69 #include <asm/pgalloc.h>
70 #include <asm/uaccess.h>
71 #include <asm/mmu_context.h>
72 #include <asm/cacheflush.h>
73 #include <asm/tlbflush.h>
74
75 #include <trace/events/sched.h>
76
77 /*
78  * Protected counters by write_lock_irq(&tasklist_lock)
79  */
80 unsigned long total_forks;      /* Handle normal Linux uptimes. */
81 int nr_threads;                 /* The idle threads do not count.. */
82
83 int max_threads;                /* tunable limit on nr_threads */
84
85 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
86
87 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
88
89 int nr_processes(void)
90 {
91         int cpu;
92         int total = 0;
93
94         for_each_online_cpu(cpu)
95                 total += per_cpu(process_counts, cpu);
96
97         return total;
98 }
99
100 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
101 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
102 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
103 static struct kmem_cache *task_struct_cachep;
104 #endif
105
106 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
107 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
108 {
109 #ifdef CONFIG_DEBUG_STACK_USAGE
110         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
111 #else
112         gfp_t mask = GFP_KERNEL;
113 #endif
114         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
115 }
116
117 static inline void free_thread_info(struct thread_info *ti)
118 {
119         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
120 }
121 #endif
122
123 /* SLAB cache for signal_struct structures (tsk->signal) */
124 static struct kmem_cache *signal_cachep;
125
126 /* SLAB cache for sighand_struct structures (tsk->sighand) */
127 struct kmem_cache *sighand_cachep;
128
129 /* SLAB cache for files_struct structures (tsk->files) */
130 struct kmem_cache *files_cachep;
131
132 /* SLAB cache for fs_struct structures (tsk->fs) */
133 struct kmem_cache *fs_cachep;
134
135 /* SLAB cache for vm_area_struct structures */
136 struct kmem_cache *vm_area_cachep;
137
138 /* SLAB cache for mm_struct structures (tsk->mm) */
139 static struct kmem_cache *mm_cachep;
140
141 static void account_kernel_stack(struct thread_info *ti, int account)
142 {
143         struct zone *zone = page_zone(virt_to_page(ti));
144
145         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
146 }
147
148 void free_task(struct task_struct *tsk)
149 {
150         prop_local_destroy_single(&tsk->dirties);
151         account_kernel_stack(tsk->stack, -1);
152         free_thread_info(tsk->stack);
153         rt_mutex_debug_task_free(tsk);
154         ftrace_graph_exit_task(tsk);
155         free_task_struct(tsk);
156 }
157 EXPORT_SYMBOL(free_task);
158
159 void __put_task_struct(struct task_struct *tsk)
160 {
161         WARN_ON(!tsk->exit_state);
162         WARN_ON(atomic_read(&tsk->usage));
163         WARN_ON(tsk == current);
164
165         exit_creds(tsk);
166         delayacct_tsk_free(tsk);
167
168         if (!profile_handoff_task(tsk))
169                 free_task(tsk);
170 }
171
172 /*
173  * macro override instead of weak attribute alias, to workaround
174  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
175  */
176 #ifndef arch_task_cache_init
177 #define arch_task_cache_init()
178 #endif
179
180 void __init fork_init(unsigned long mempages)
181 {
182 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
183 #ifndef ARCH_MIN_TASKALIGN
184 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
185 #endif
186         /* create a slab on which task_structs can be allocated */
187         task_struct_cachep =
188                 kmem_cache_create("task_struct", sizeof(struct task_struct),
189                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
190 #endif
191
192         /* do the arch specific task caches init */
193         arch_task_cache_init();
194
195         /*
196          * The default maximum number of threads is set to a safe
197          * value: the thread structures can take up at most half
198          * of memory.
199          */
200         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
201
202         /*
203          * we need to allow at least 20 threads to boot a system
204          */
205         if(max_threads < 20)
206                 max_threads = 20;
207
208         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
209         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
210         init_task.signal->rlim[RLIMIT_SIGPENDING] =
211                 init_task.signal->rlim[RLIMIT_NPROC];
212 }
213
214 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
215                                                struct task_struct *src)
216 {
217         *dst = *src;
218         return 0;
219 }
220
221 static struct task_struct *dup_task_struct(struct task_struct *orig)
222 {
223         struct task_struct *tsk;
224         struct thread_info *ti;
225         unsigned long *stackend;
226
227         int err;
228
229         prepare_to_copy(orig);
230
231         tsk = alloc_task_struct();
232         if (!tsk)
233                 return NULL;
234
235         ti = alloc_thread_info(tsk);
236         if (!ti) {
237                 free_task_struct(tsk);
238                 return NULL;
239         }
240
241         err = arch_dup_task_struct(tsk, orig);
242         if (err)
243                 goto out;
244
245         tsk->stack = ti;
246
247         err = prop_local_init_single(&tsk->dirties);
248         if (err)
249                 goto out;
250
251         setup_thread_stack(tsk, orig);
252         stackend = end_of_stack(tsk);
253         *stackend = STACK_END_MAGIC;    /* for overflow detection */
254
255 #ifdef CONFIG_CC_STACKPROTECTOR
256         tsk->stack_canary = get_random_int();
257 #endif
258
259         /* One for us, one for whoever does the "release_task()" (usually parent) */
260         atomic_set(&tsk->usage,2);
261         atomic_set(&tsk->fs_excl, 0);
262 #ifdef CONFIG_BLK_DEV_IO_TRACE
263         tsk->btrace_seq = 0;
264 #endif
265         tsk->splice_pipe = NULL;
266
267         account_kernel_stack(ti, 1);
268
269         return tsk;
270
271 out:
272         free_thread_info(ti);
273         free_task_struct(tsk);
274         return NULL;
275 }
276
277 #ifdef CONFIG_MMU
278 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
279 {
280         struct vm_area_struct *mpnt, *tmp, **pprev;
281         struct rb_node **rb_link, *rb_parent;
282         int retval;
283         unsigned long charge;
284         struct mempolicy *pol;
285
286         down_write(&oldmm->mmap_sem);
287         flush_cache_dup_mm(oldmm);
288         /*
289          * Not linked in yet - no deadlock potential:
290          */
291         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
292
293         mm->locked_vm = 0;
294         mm->mmap = NULL;
295         mm->mmap_cache = NULL;
296         mm->free_area_cache = oldmm->mmap_base;
297         mm->cached_hole_size = ~0UL;
298         mm->map_count = 0;
299         cpumask_clear(mm_cpumask(mm));
300         mm->mm_rb = RB_ROOT;
301         rb_link = &mm->mm_rb.rb_node;
302         rb_parent = NULL;
303         pprev = &mm->mmap;
304         retval = ksm_fork(mm, oldmm);
305         if (retval)
306                 goto out;
307
308         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
309                 struct file *file;
310
311                 if (mpnt->vm_flags & VM_DONTCOPY) {
312                         long pages = vma_pages(mpnt);
313                         mm->total_vm -= pages;
314                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
315                                                                 -pages);
316                         continue;
317                 }
318                 charge = 0;
319                 if (mpnt->vm_flags & VM_ACCOUNT) {
320                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
321                         if (security_vm_enough_memory(len))
322                                 goto fail_nomem;
323                         charge = len;
324                 }
325                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
326                 if (!tmp)
327                         goto fail_nomem;
328                 *tmp = *mpnt;
329                 pol = mpol_dup(vma_policy(mpnt));
330                 retval = PTR_ERR(pol);
331                 if (IS_ERR(pol))
332                         goto fail_nomem_policy;
333                 vma_set_policy(tmp, pol);
334                 tmp->vm_flags &= ~VM_LOCKED;
335                 tmp->vm_mm = mm;
336                 tmp->vm_next = NULL;
337                 anon_vma_link(tmp);
338                 file = tmp->vm_file;
339                 if (file) {
340                         struct inode *inode = file->f_path.dentry->d_inode;
341                         struct address_space *mapping = file->f_mapping;
342
343                         get_file(file);
344                         if (tmp->vm_flags & VM_DENYWRITE)
345                                 atomic_dec(&inode->i_writecount);
346                         spin_lock(&mapping->i_mmap_lock);
347                         if (tmp->vm_flags & VM_SHARED)
348                                 mapping->i_mmap_writable++;
349                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
350                         flush_dcache_mmap_lock(mapping);
351                         /* insert tmp into the share list, just after mpnt */
352                         vma_prio_tree_add(tmp, mpnt);
353                         flush_dcache_mmap_unlock(mapping);
354                         spin_unlock(&mapping->i_mmap_lock);
355                 }
356
357                 /*
358                  * Clear hugetlb-related page reserves for children. This only
359                  * affects MAP_PRIVATE mappings. Faults generated by the child
360                  * are not guaranteed to succeed, even if read-only
361                  */
362                 if (is_vm_hugetlb_page(tmp))
363                         reset_vma_resv_huge_pages(tmp);
364
365                 /*
366                  * Link in the new vma and copy the page table entries.
367                  */
368                 *pprev = tmp;
369                 pprev = &tmp->vm_next;
370
371                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
372                 rb_link = &tmp->vm_rb.rb_right;
373                 rb_parent = &tmp->vm_rb;
374
375                 mm->map_count++;
376                 retval = copy_page_range(mm, oldmm, mpnt);
377
378                 if (tmp->vm_ops && tmp->vm_ops->open)
379                         tmp->vm_ops->open(tmp);
380
381                 if (retval)
382                         goto out;
383         }
384         /* a new mm has just been created */
385         arch_dup_mmap(oldmm, mm);
386         retval = 0;
387 out:
388         up_write(&mm->mmap_sem);
389         flush_tlb_mm(oldmm);
390         up_write(&oldmm->mmap_sem);
391         return retval;
392 fail_nomem_policy:
393         kmem_cache_free(vm_area_cachep, tmp);
394 fail_nomem:
395         retval = -ENOMEM;
396         vm_unacct_memory(charge);
397         goto out;
398 }
399
400 static inline int mm_alloc_pgd(struct mm_struct * mm)
401 {
402         mm->pgd = pgd_alloc(mm);
403         if (unlikely(!mm->pgd))
404                 return -ENOMEM;
405         return 0;
406 }
407
408 static inline void mm_free_pgd(struct mm_struct * mm)
409 {
410         pgd_free(mm, mm->pgd);
411 }
412 #else
413 #define dup_mmap(mm, oldmm)     (0)
414 #define mm_alloc_pgd(mm)        (0)
415 #define mm_free_pgd(mm)
416 #endif /* CONFIG_MMU */
417
418 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
419
420 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
421 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
422
423 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
424
425 static int __init coredump_filter_setup(char *s)
426 {
427         default_dump_filter =
428                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
429                 MMF_DUMP_FILTER_MASK;
430         return 1;
431 }
432
433 __setup("coredump_filter=", coredump_filter_setup);
434
435 #include <linux/init_task.h>
436
437 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
438 {
439         atomic_set(&mm->mm_users, 1);
440         atomic_set(&mm->mm_count, 1);
441         init_rwsem(&mm->mmap_sem);
442         INIT_LIST_HEAD(&mm->mmlist);
443         mm->flags = (current->mm) ?
444                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
445         mm->core_state = NULL;
446         mm->nr_ptes = 0;
447         set_mm_counter(mm, file_rss, 0);
448         set_mm_counter(mm, anon_rss, 0);
449         spin_lock_init(&mm->page_table_lock);
450         spin_lock_init(&mm->ioctx_lock);
451         INIT_HLIST_HEAD(&mm->ioctx_list);
452         mm->free_area_cache = TASK_UNMAPPED_BASE;
453         mm->cached_hole_size = ~0UL;
454         mm_init_owner(mm, p);
455
456         if (likely(!mm_alloc_pgd(mm))) {
457                 mm->def_flags = 0;
458                 mmu_notifier_mm_init(mm);
459                 return mm;
460         }
461
462         free_mm(mm);
463         return NULL;
464 }
465
466 /*
467  * Allocate and initialize an mm_struct.
468  */
469 struct mm_struct * mm_alloc(void)
470 {
471         struct mm_struct * mm;
472
473         mm = allocate_mm();
474         if (mm) {
475                 memset(mm, 0, sizeof(*mm));
476                 mm = mm_init(mm, current);
477         }
478         return mm;
479 }
480
481 /*
482  * Called when the last reference to the mm
483  * is dropped: either by a lazy thread or by
484  * mmput. Free the page directory and the mm.
485  */
486 void __mmdrop(struct mm_struct *mm)
487 {
488         BUG_ON(mm == &init_mm);
489         mm_free_pgd(mm);
490         destroy_context(mm);
491         mmu_notifier_mm_destroy(mm);
492         free_mm(mm);
493 }
494 EXPORT_SYMBOL_GPL(__mmdrop);
495
496 /*
497  * Decrement the use count and release all resources for an mm.
498  */
499 void mmput(struct mm_struct *mm)
500 {
501         might_sleep();
502
503         if (atomic_dec_and_test(&mm->mm_users)) {
504                 exit_aio(mm);
505                 ksm_exit(mm);
506                 exit_mmap(mm);
507                 set_mm_exe_file(mm, NULL);
508                 if (!list_empty(&mm->mmlist)) {
509                         spin_lock(&mmlist_lock);
510                         list_del(&mm->mmlist);
511                         spin_unlock(&mmlist_lock);
512                 }
513                 put_swap_token(mm);
514                 mmdrop(mm);
515         }
516 }
517 EXPORT_SYMBOL_GPL(mmput);
518
519 /**
520  * get_task_mm - acquire a reference to the task's mm
521  *
522  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
523  * this kernel workthread has transiently adopted a user mm with use_mm,
524  * to do its AIO) is not set and if so returns a reference to it, after
525  * bumping up the use count.  User must release the mm via mmput()
526  * after use.  Typically used by /proc and ptrace.
527  */
528 struct mm_struct *get_task_mm(struct task_struct *task)
529 {
530         struct mm_struct *mm;
531
532         task_lock(task);
533         mm = task->mm;
534         if (mm) {
535                 if (task->flags & PF_KTHREAD)
536                         mm = NULL;
537                 else
538                         atomic_inc(&mm->mm_users);
539         }
540         task_unlock(task);
541         return mm;
542 }
543 EXPORT_SYMBOL_GPL(get_task_mm);
544
545 /* Please note the differences between mmput and mm_release.
546  * mmput is called whenever we stop holding onto a mm_struct,
547  * error success whatever.
548  *
549  * mm_release is called after a mm_struct has been removed
550  * from the current process.
551  *
552  * This difference is important for error handling, when we
553  * only half set up a mm_struct for a new process and need to restore
554  * the old one.  Because we mmput the new mm_struct before
555  * restoring the old one. . .
556  * Eric Biederman 10 January 1998
557  */
558 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
559 {
560         struct completion *vfork_done = tsk->vfork_done;
561
562         /* Get rid of any futexes when releasing the mm */
563 #ifdef CONFIG_FUTEX
564         if (unlikely(tsk->robust_list))
565                 exit_robust_list(tsk);
566 #ifdef CONFIG_COMPAT
567         if (unlikely(tsk->compat_robust_list))
568                 compat_exit_robust_list(tsk);
569 #endif
570 #endif
571
572         /* Get rid of any cached register state */
573         deactivate_mm(tsk, mm);
574
575         /* notify parent sleeping on vfork() */
576         if (vfork_done) {
577                 tsk->vfork_done = NULL;
578                 complete(vfork_done);
579         }
580
581         /*
582          * If we're exiting normally, clear a user-space tid field if
583          * requested.  We leave this alone when dying by signal, to leave
584          * the value intact in a core dump, and to save the unnecessary
585          * trouble otherwise.  Userland only wants this done for a sys_exit.
586          */
587         if (tsk->clear_child_tid) {
588                 if (!(tsk->flags & PF_SIGNALED) &&
589                     atomic_read(&mm->mm_users) > 1) {
590                         /*
591                          * We don't check the error code - if userspace has
592                          * not set up a proper pointer then tough luck.
593                          */
594                         put_user(0, tsk->clear_child_tid);
595                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
596                                         1, NULL, NULL, 0);
597                 }
598                 tsk->clear_child_tid = NULL;
599         }
600 }
601
602 /*
603  * Allocate a new mm structure and copy contents from the
604  * mm structure of the passed in task structure.
605  */
606 struct mm_struct *dup_mm(struct task_struct *tsk)
607 {
608         struct mm_struct *mm, *oldmm = current->mm;
609         int err;
610
611         if (!oldmm)
612                 return NULL;
613
614         mm = allocate_mm();
615         if (!mm)
616                 goto fail_nomem;
617
618         memcpy(mm, oldmm, sizeof(*mm));
619
620         /* Initializing for Swap token stuff */
621         mm->token_priority = 0;
622         mm->last_interval = 0;
623
624         if (!mm_init(mm, tsk))
625                 goto fail_nomem;
626
627         if (init_new_context(tsk, mm))
628                 goto fail_nocontext;
629
630         dup_mm_exe_file(oldmm, mm);
631
632         err = dup_mmap(mm, oldmm);
633         if (err)
634                 goto free_pt;
635
636         mm->hiwater_rss = get_mm_rss(mm);
637         mm->hiwater_vm = mm->total_vm;
638
639         return mm;
640
641 free_pt:
642         mmput(mm);
643
644 fail_nomem:
645         return NULL;
646
647 fail_nocontext:
648         /*
649          * If init_new_context() failed, we cannot use mmput() to free the mm
650          * because it calls destroy_context()
651          */
652         mm_free_pgd(mm);
653         free_mm(mm);
654         return NULL;
655 }
656
657 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
658 {
659         struct mm_struct * mm, *oldmm;
660         int retval;
661
662         tsk->min_flt = tsk->maj_flt = 0;
663         tsk->nvcsw = tsk->nivcsw = 0;
664 #ifdef CONFIG_DETECT_HUNG_TASK
665         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
666 #endif
667
668         tsk->mm = NULL;
669         tsk->active_mm = NULL;
670
671         /*
672          * Are we cloning a kernel thread?
673          *
674          * We need to steal a active VM for that..
675          */
676         oldmm = current->mm;
677         if (!oldmm)
678                 return 0;
679
680         if (clone_flags & CLONE_VM) {
681                 atomic_inc(&oldmm->mm_users);
682                 mm = oldmm;
683                 goto good_mm;
684         }
685
686         retval = -ENOMEM;
687         mm = dup_mm(tsk);
688         if (!mm)
689                 goto fail_nomem;
690
691 good_mm:
692         /* Initializing for Swap token stuff */
693         mm->token_priority = 0;
694         mm->last_interval = 0;
695
696         tsk->mm = mm;
697         tsk->active_mm = mm;
698         return 0;
699
700 fail_nomem:
701         return retval;
702 }
703
704 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
705 {
706         struct fs_struct *fs = current->fs;
707         if (clone_flags & CLONE_FS) {
708                 /* tsk->fs is already what we want */
709                 write_lock(&fs->lock);
710                 if (fs->in_exec) {
711                         write_unlock(&fs->lock);
712                         return -EAGAIN;
713                 }
714                 fs->users++;
715                 write_unlock(&fs->lock);
716                 return 0;
717         }
718         tsk->fs = copy_fs_struct(fs);
719         if (!tsk->fs)
720                 return -ENOMEM;
721         return 0;
722 }
723
724 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
725 {
726         struct files_struct *oldf, *newf;
727         int error = 0;
728
729         /*
730          * A background process may not have any files ...
731          */
732         oldf = current->files;
733         if (!oldf)
734                 goto out;
735
736         if (clone_flags & CLONE_FILES) {
737                 atomic_inc(&oldf->count);
738                 goto out;
739         }
740
741         newf = dup_fd(oldf, &error);
742         if (!newf)
743                 goto out;
744
745         tsk->files = newf;
746         error = 0;
747 out:
748         return error;
749 }
750
751 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
752 {
753 #ifdef CONFIG_BLOCK
754         struct io_context *ioc = current->io_context;
755
756         if (!ioc)
757                 return 0;
758         /*
759          * Share io context with parent, if CLONE_IO is set
760          */
761         if (clone_flags & CLONE_IO) {
762                 tsk->io_context = ioc_task_link(ioc);
763                 if (unlikely(!tsk->io_context))
764                         return -ENOMEM;
765         } else if (ioprio_valid(ioc->ioprio)) {
766                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
767                 if (unlikely(!tsk->io_context))
768                         return -ENOMEM;
769
770                 tsk->io_context->ioprio = ioc->ioprio;
771         }
772 #endif
773         return 0;
774 }
775
776 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
777 {
778         struct sighand_struct *sig;
779
780         if (clone_flags & CLONE_SIGHAND) {
781                 atomic_inc(&current->sighand->count);
782                 return 0;
783         }
784         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
785         rcu_assign_pointer(tsk->sighand, sig);
786         if (!sig)
787                 return -ENOMEM;
788         atomic_set(&sig->count, 1);
789         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
790         return 0;
791 }
792
793 void __cleanup_sighand(struct sighand_struct *sighand)
794 {
795         if (atomic_dec_and_test(&sighand->count))
796                 kmem_cache_free(sighand_cachep, sighand);
797 }
798
799
800 /*
801  * Initialize POSIX timer handling for a thread group.
802  */
803 static void posix_cpu_timers_init_group(struct signal_struct *sig)
804 {
805         /* Thread group counters. */
806         thread_group_cputime_init(sig);
807
808         /* Expiration times and increments. */
809         sig->it[CPUCLOCK_PROF].expires = cputime_zero;
810         sig->it[CPUCLOCK_PROF].incr = cputime_zero;
811         sig->it[CPUCLOCK_VIRT].expires = cputime_zero;
812         sig->it[CPUCLOCK_VIRT].incr = cputime_zero;
813
814         /* Cached expiration times. */
815         sig->cputime_expires.prof_exp = cputime_zero;
816         sig->cputime_expires.virt_exp = cputime_zero;
817         sig->cputime_expires.sched_exp = 0;
818
819         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
820                 sig->cputime_expires.prof_exp =
821                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
822                 sig->cputimer.running = 1;
823         }
824
825         /* The timer lists. */
826         INIT_LIST_HEAD(&sig->cpu_timers[0]);
827         INIT_LIST_HEAD(&sig->cpu_timers[1]);
828         INIT_LIST_HEAD(&sig->cpu_timers[2]);
829 }
830
831 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
832 {
833         struct signal_struct *sig;
834
835         if (clone_flags & CLONE_THREAD)
836                 return 0;
837
838         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
839         tsk->signal = sig;
840         if (!sig)
841                 return -ENOMEM;
842
843         atomic_set(&sig->count, 1);
844         atomic_set(&sig->live, 1);
845         init_waitqueue_head(&sig->wait_chldexit);
846         sig->flags = 0;
847         if (clone_flags & CLONE_NEWPID)
848                 sig->flags |= SIGNAL_UNKILLABLE;
849         sig->group_exit_code = 0;
850         sig->group_exit_task = NULL;
851         sig->group_stop_count = 0;
852         sig->curr_target = tsk;
853         init_sigpending(&sig->shared_pending);
854         INIT_LIST_HEAD(&sig->posix_timers);
855
856         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
857         sig->it_real_incr.tv64 = 0;
858         sig->real_timer.function = it_real_fn;
859
860         sig->leader = 0;        /* session leadership doesn't inherit */
861         sig->tty_old_pgrp = NULL;
862         sig->tty = NULL;
863
864         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
865         sig->gtime = cputime_zero;
866         sig->cgtime = cputime_zero;
867         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
868         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
869         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
870         sig->maxrss = sig->cmaxrss = 0;
871         task_io_accounting_init(&sig->ioac);
872         sig->sum_sched_runtime = 0;
873         taskstats_tgid_init(sig);
874
875         task_lock(current->group_leader);
876         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
877         task_unlock(current->group_leader);
878
879         posix_cpu_timers_init_group(sig);
880
881         acct_init_pacct(&sig->pacct);
882
883         tty_audit_fork(sig);
884
885         sig->oom_adj = current->signal->oom_adj;
886
887         return 0;
888 }
889
890 void __cleanup_signal(struct signal_struct *sig)
891 {
892         thread_group_cputime_free(sig);
893         tty_kref_put(sig->tty);
894         kmem_cache_free(signal_cachep, sig);
895 }
896
897 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
898 {
899         unsigned long new_flags = p->flags;
900
901         new_flags &= ~PF_SUPERPRIV;
902         new_flags |= PF_FORKNOEXEC;
903         new_flags |= PF_STARTING;
904         p->flags = new_flags;
905         clear_freeze_flag(p);
906 }
907
908 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
909 {
910         current->clear_child_tid = tidptr;
911
912         return task_pid_vnr(current);
913 }
914
915 static void rt_mutex_init_task(struct task_struct *p)
916 {
917         spin_lock_init(&p->pi_lock);
918 #ifdef CONFIG_RT_MUTEXES
919         plist_head_init(&p->pi_waiters, &p->pi_lock);
920         p->pi_blocked_on = NULL;
921 #endif
922 }
923
924 #ifdef CONFIG_MM_OWNER
925 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
926 {
927         mm->owner = p;
928 }
929 #endif /* CONFIG_MM_OWNER */
930
931 /*
932  * Initialize POSIX timer handling for a single task.
933  */
934 static void posix_cpu_timers_init(struct task_struct *tsk)
935 {
936         tsk->cputime_expires.prof_exp = cputime_zero;
937         tsk->cputime_expires.virt_exp = cputime_zero;
938         tsk->cputime_expires.sched_exp = 0;
939         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
940         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
941         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
942 }
943
944 /*
945  * This creates a new process as a copy of the old one,
946  * but does not actually start it yet.
947  *
948  * It copies the registers, and all the appropriate
949  * parts of the process environment (as per the clone
950  * flags). The actual kick-off is left to the caller.
951  */
952 static struct task_struct *copy_process(unsigned long clone_flags,
953                                         unsigned long stack_start,
954                                         struct pt_regs *regs,
955                                         unsigned long stack_size,
956                                         int __user *child_tidptr,
957                                         struct pid *pid,
958                                         int trace)
959 {
960         int retval;
961         struct task_struct *p;
962         int cgroup_callbacks_done = 0;
963
964         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
965                 return ERR_PTR(-EINVAL);
966
967         /*
968          * Thread groups must share signals as well, and detached threads
969          * can only be started up within the thread group.
970          */
971         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
972                 return ERR_PTR(-EINVAL);
973
974         /*
975          * Shared signal handlers imply shared VM. By way of the above,
976          * thread groups also imply shared VM. Blocking this case allows
977          * for various simplifications in other code.
978          */
979         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
980                 return ERR_PTR(-EINVAL);
981
982         retval = security_task_create(clone_flags);
983         if (retval)
984                 goto fork_out;
985
986         retval = -ENOMEM;
987         p = dup_task_struct(current);
988         if (!p)
989                 goto fork_out;
990
991         ftrace_graph_init_task(p);
992
993         rt_mutex_init_task(p);
994
995 #ifdef CONFIG_PROVE_LOCKING
996         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
997         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
998 #endif
999         retval = -EAGAIN;
1000         if (atomic_read(&p->real_cred->user->processes) >=
1001                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1002                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1003                     p->real_cred->user != INIT_USER)
1004                         goto bad_fork_free;
1005         }
1006
1007         retval = copy_creds(p, clone_flags);
1008         if (retval < 0)
1009                 goto bad_fork_free;
1010
1011         /*
1012          * If multiple threads are within copy_process(), then this check
1013          * triggers too late. This doesn't hurt, the check is only there
1014          * to stop root fork bombs.
1015          */
1016         retval = -EAGAIN;
1017         if (nr_threads >= max_threads)
1018                 goto bad_fork_cleanup_count;
1019
1020         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1021                 goto bad_fork_cleanup_count;
1022
1023         if (p->binfmt && !try_module_get(p->binfmt->module))
1024                 goto bad_fork_cleanup_put_domain;
1025
1026         p->did_exec = 0;
1027         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1028         copy_flags(clone_flags, p);
1029         INIT_LIST_HEAD(&p->children);
1030         INIT_LIST_HEAD(&p->sibling);
1031         rcu_copy_process(p);
1032         p->vfork_done = NULL;
1033         spin_lock_init(&p->alloc_lock);
1034
1035         init_sigpending(&p->pending);
1036
1037         p->utime = cputime_zero;
1038         p->stime = cputime_zero;
1039         p->gtime = cputime_zero;
1040         p->utimescaled = cputime_zero;
1041         p->stimescaled = cputime_zero;
1042         p->prev_utime = cputime_zero;
1043         p->prev_stime = cputime_zero;
1044
1045         p->default_timer_slack_ns = current->timer_slack_ns;
1046
1047         task_io_accounting_init(&p->ioac);
1048         acct_clear_integrals(p);
1049
1050         posix_cpu_timers_init(p);
1051
1052         p->lock_depth = -1;             /* -1 = no lock */
1053         do_posix_clock_monotonic_gettime(&p->start_time);
1054         p->real_start_time = p->start_time;
1055         monotonic_to_bootbased(&p->real_start_time);
1056         p->io_context = NULL;
1057         p->audit_context = NULL;
1058         cgroup_fork(p);
1059 #ifdef CONFIG_NUMA
1060         p->mempolicy = mpol_dup(p->mempolicy);
1061         if (IS_ERR(p->mempolicy)) {
1062                 retval = PTR_ERR(p->mempolicy);
1063                 p->mempolicy = NULL;
1064                 goto bad_fork_cleanup_cgroup;
1065         }
1066         mpol_fix_fork_child_flag(p);
1067 #endif
1068 #ifdef CONFIG_TRACE_IRQFLAGS
1069         p->irq_events = 0;
1070 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1071         p->hardirqs_enabled = 1;
1072 #else
1073         p->hardirqs_enabled = 0;
1074 #endif
1075         p->hardirq_enable_ip = 0;
1076         p->hardirq_enable_event = 0;
1077         p->hardirq_disable_ip = _THIS_IP_;
1078         p->hardirq_disable_event = 0;
1079         p->softirqs_enabled = 1;
1080         p->softirq_enable_ip = _THIS_IP_;
1081         p->softirq_enable_event = 0;
1082         p->softirq_disable_ip = 0;
1083         p->softirq_disable_event = 0;
1084         p->hardirq_context = 0;
1085         p->softirq_context = 0;
1086 #endif
1087 #ifdef CONFIG_LOCKDEP
1088         p->lockdep_depth = 0; /* no locks held yet */
1089         p->curr_chain_key = 0;
1090         p->lockdep_recursion = 0;
1091 #endif
1092
1093 #ifdef CONFIG_DEBUG_MUTEXES
1094         p->blocked_on = NULL; /* not blocked yet */
1095 #endif
1096
1097         p->bts = NULL;
1098
1099         p->stack_start = stack_start;
1100
1101         /* Perform scheduler related setup. Assign this task to a CPU. */
1102         sched_fork(p, clone_flags);
1103
1104         retval = perf_event_init_task(p);
1105         if (retval)
1106                 goto bad_fork_cleanup_policy;
1107
1108         if ((retval = audit_alloc(p)))
1109                 goto bad_fork_cleanup_policy;
1110         /* copy all the process information */
1111         if ((retval = copy_semundo(clone_flags, p)))
1112                 goto bad_fork_cleanup_audit;
1113         if ((retval = copy_files(clone_flags, p)))
1114                 goto bad_fork_cleanup_semundo;
1115         if ((retval = copy_fs(clone_flags, p)))
1116                 goto bad_fork_cleanup_files;
1117         if ((retval = copy_sighand(clone_flags, p)))
1118                 goto bad_fork_cleanup_fs;
1119         if ((retval = copy_signal(clone_flags, p)))
1120                 goto bad_fork_cleanup_sighand;
1121         if ((retval = copy_mm(clone_flags, p)))
1122                 goto bad_fork_cleanup_signal;
1123         if ((retval = copy_namespaces(clone_flags, p)))
1124                 goto bad_fork_cleanup_mm;
1125         if ((retval = copy_io(clone_flags, p)))
1126                 goto bad_fork_cleanup_namespaces;
1127         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1128         if (retval)
1129                 goto bad_fork_cleanup_io;
1130
1131         if (pid != &init_struct_pid) {
1132                 retval = -ENOMEM;
1133                 pid = alloc_pid(p->nsproxy->pid_ns);
1134                 if (!pid)
1135                         goto bad_fork_cleanup_io;
1136
1137                 if (clone_flags & CLONE_NEWPID) {
1138                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1139                         if (retval < 0)
1140                                 goto bad_fork_free_pid;
1141                 }
1142         }
1143
1144         p->pid = pid_nr(pid);
1145         p->tgid = p->pid;
1146         if (clone_flags & CLONE_THREAD)
1147                 p->tgid = current->tgid;
1148
1149         if (current->nsproxy != p->nsproxy) {
1150                 retval = ns_cgroup_clone(p, pid);
1151                 if (retval)
1152                         goto bad_fork_free_pid;
1153         }
1154
1155         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1156         /*
1157          * Clear TID on mm_release()?
1158          */
1159         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1160 #ifdef CONFIG_FUTEX
1161         p->robust_list = NULL;
1162 #ifdef CONFIG_COMPAT
1163         p->compat_robust_list = NULL;
1164 #endif
1165         INIT_LIST_HEAD(&p->pi_state_list);
1166         p->pi_state_cache = NULL;
1167 #endif
1168         /*
1169          * sigaltstack should be cleared when sharing the same VM
1170          */
1171         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1172                 p->sas_ss_sp = p->sas_ss_size = 0;
1173
1174         /*
1175          * Syscall tracing should be turned off in the child regardless
1176          * of CLONE_PTRACE.
1177          */
1178         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1179 #ifdef TIF_SYSCALL_EMU
1180         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1181 #endif
1182         clear_all_latency_tracing(p);
1183
1184         /* ok, now we should be set up.. */
1185         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1186         p->pdeath_signal = 0;
1187         p->exit_state = 0;
1188
1189         /*
1190          * Ok, make it visible to the rest of the system.
1191          * We dont wake it up yet.
1192          */
1193         p->group_leader = p;
1194         INIT_LIST_HEAD(&p->thread_group);
1195
1196         /* Now that the task is set up, run cgroup callbacks if
1197          * necessary. We need to run them before the task is visible
1198          * on the tasklist. */
1199         cgroup_fork_callbacks(p);
1200         cgroup_callbacks_done = 1;
1201
1202         /* Need tasklist lock for parent etc handling! */
1203         write_lock_irq(&tasklist_lock);
1204
1205         /*
1206          * The task hasn't been attached yet, so its cpus_allowed mask will
1207          * not be changed, nor will its assigned CPU.
1208          *
1209          * The cpus_allowed mask of the parent may have changed after it was
1210          * copied first time - so re-copy it here, then check the child's CPU
1211          * to ensure it is on a valid CPU (and if not, just force it back to
1212          * parent's CPU). This avoids alot of nasty races.
1213          */
1214         p->cpus_allowed = current->cpus_allowed;
1215         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1216         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1217                         !cpu_online(task_cpu(p))))
1218                 set_task_cpu(p, smp_processor_id());
1219
1220         /* CLONE_PARENT re-uses the old parent */
1221         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1222                 p->real_parent = current->real_parent;
1223                 p->parent_exec_id = current->parent_exec_id;
1224         } else {
1225                 p->real_parent = current;
1226                 p->parent_exec_id = current->self_exec_id;
1227         }
1228
1229         spin_lock(&current->sighand->siglock);
1230
1231         /*
1232          * Process group and session signals need to be delivered to just the
1233          * parent before the fork or both the parent and the child after the
1234          * fork. Restart if a signal comes in before we add the new process to
1235          * it's process group.
1236          * A fatal signal pending means that current will exit, so the new
1237          * thread can't slip out of an OOM kill (or normal SIGKILL).
1238          */
1239         recalc_sigpending();
1240         if (signal_pending(current)) {
1241                 spin_unlock(&current->sighand->siglock);
1242                 write_unlock_irq(&tasklist_lock);
1243                 retval = -ERESTARTNOINTR;
1244                 goto bad_fork_free_pid;
1245         }
1246
1247         if (clone_flags & CLONE_THREAD) {
1248                 atomic_inc(&current->signal->count);
1249                 atomic_inc(&current->signal->live);
1250                 p->group_leader = current->group_leader;
1251                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1252         }
1253
1254         if (likely(p->pid)) {
1255                 list_add_tail(&p->sibling, &p->real_parent->children);
1256                 tracehook_finish_clone(p, clone_flags, trace);
1257
1258                 if (thread_group_leader(p)) {
1259                         if (clone_flags & CLONE_NEWPID)
1260                                 p->nsproxy->pid_ns->child_reaper = p;
1261
1262                         p->signal->leader_pid = pid;
1263                         tty_kref_put(p->signal->tty);
1264                         p->signal->tty = tty_kref_get(current->signal->tty);
1265                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1266                         attach_pid(p, PIDTYPE_SID, task_session(current));
1267                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1268                         __get_cpu_var(process_counts)++;
1269                 }
1270                 attach_pid(p, PIDTYPE_PID, pid);
1271                 nr_threads++;
1272         }
1273
1274         total_forks++;
1275         spin_unlock(&current->sighand->siglock);
1276         write_unlock_irq(&tasklist_lock);
1277         proc_fork_connector(p);
1278         cgroup_post_fork(p);
1279         perf_event_fork(p);
1280         return p;
1281
1282 bad_fork_free_pid:
1283         if (pid != &init_struct_pid)
1284                 free_pid(pid);
1285 bad_fork_cleanup_io:
1286         put_io_context(p->io_context);
1287 bad_fork_cleanup_namespaces:
1288         exit_task_namespaces(p);
1289 bad_fork_cleanup_mm:
1290         if (p->mm)
1291                 mmput(p->mm);
1292 bad_fork_cleanup_signal:
1293         if (!(clone_flags & CLONE_THREAD))
1294                 __cleanup_signal(p->signal);
1295 bad_fork_cleanup_sighand:
1296         __cleanup_sighand(p->sighand);
1297 bad_fork_cleanup_fs:
1298         exit_fs(p); /* blocking */
1299 bad_fork_cleanup_files:
1300         exit_files(p); /* blocking */
1301 bad_fork_cleanup_semundo:
1302         exit_sem(p);
1303 bad_fork_cleanup_audit:
1304         audit_free(p);
1305 bad_fork_cleanup_policy:
1306         perf_event_free_task(p);
1307 #ifdef CONFIG_NUMA
1308         mpol_put(p->mempolicy);
1309 bad_fork_cleanup_cgroup:
1310 #endif
1311         cgroup_exit(p, cgroup_callbacks_done);
1312         delayacct_tsk_free(p);
1313         if (p->binfmt)
1314                 module_put(p->binfmt->module);
1315 bad_fork_cleanup_put_domain:
1316         module_put(task_thread_info(p)->exec_domain->module);
1317 bad_fork_cleanup_count:
1318         atomic_dec(&p->cred->user->processes);
1319         exit_creds(p);
1320 bad_fork_free:
1321         free_task(p);
1322 fork_out:
1323         return ERR_PTR(retval);
1324 }
1325
1326 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1327 {
1328         memset(regs, 0, sizeof(struct pt_regs));
1329         return regs;
1330 }
1331
1332 struct task_struct * __cpuinit fork_idle(int cpu)
1333 {
1334         struct task_struct *task;
1335         struct pt_regs regs;
1336
1337         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1338                             &init_struct_pid, 0);
1339         if (!IS_ERR(task))
1340                 init_idle(task, cpu);
1341
1342         return task;
1343 }
1344
1345 /*
1346  *  Ok, this is the main fork-routine.
1347  *
1348  * It copies the process, and if successful kick-starts
1349  * it and waits for it to finish using the VM if required.
1350  */
1351 long do_fork(unsigned long clone_flags,
1352               unsigned long stack_start,
1353               struct pt_regs *regs,
1354               unsigned long stack_size,
1355               int __user *parent_tidptr,
1356               int __user *child_tidptr)
1357 {
1358         struct task_struct *p;
1359         int trace = 0;
1360         long nr;
1361
1362         /*
1363          * Do some preliminary argument and permissions checking before we
1364          * actually start allocating stuff
1365          */
1366         if (clone_flags & CLONE_NEWUSER) {
1367                 if (clone_flags & CLONE_THREAD)
1368                         return -EINVAL;
1369                 /* hopefully this check will go away when userns support is
1370                  * complete
1371                  */
1372                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1373                                 !capable(CAP_SETGID))
1374                         return -EPERM;
1375         }
1376
1377         /*
1378          * We hope to recycle these flags after 2.6.26
1379          */
1380         if (unlikely(clone_flags & CLONE_STOPPED)) {
1381                 static int __read_mostly count = 100;
1382
1383                 if (count > 0 && printk_ratelimit()) {
1384                         char comm[TASK_COMM_LEN];
1385
1386                         count--;
1387                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1388                                         "clone flags 0x%lx\n",
1389                                 get_task_comm(comm, current),
1390                                 clone_flags & CLONE_STOPPED);
1391                 }
1392         }
1393
1394         /*
1395          * When called from kernel_thread, don't do user tracing stuff.
1396          */
1397         if (likely(user_mode(regs)))
1398                 trace = tracehook_prepare_clone(clone_flags);
1399
1400         p = copy_process(clone_flags, stack_start, regs, stack_size,
1401                          child_tidptr, NULL, trace);
1402         /*
1403          * Do this prior waking up the new thread - the thread pointer
1404          * might get invalid after that point, if the thread exits quickly.
1405          */
1406         if (!IS_ERR(p)) {
1407                 struct completion vfork;
1408
1409                 trace_sched_process_fork(current, p);
1410
1411                 nr = task_pid_vnr(p);
1412
1413                 if (clone_flags & CLONE_PARENT_SETTID)
1414                         put_user(nr, parent_tidptr);
1415
1416                 if (clone_flags & CLONE_VFORK) {
1417                         p->vfork_done = &vfork;
1418                         init_completion(&vfork);
1419                 }
1420
1421                 audit_finish_fork(p);
1422                 tracehook_report_clone(regs, clone_flags, nr, p);
1423
1424                 /*
1425                  * We set PF_STARTING at creation in case tracing wants to
1426                  * use this to distinguish a fully live task from one that
1427                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1428                  * clear it and set the child going.
1429                  */
1430                 p->flags &= ~PF_STARTING;
1431
1432                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1433                         /*
1434                          * We'll start up with an immediate SIGSTOP.
1435                          */
1436                         sigaddset(&p->pending.signal, SIGSTOP);
1437                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1438                         __set_task_state(p, TASK_STOPPED);
1439                 } else {
1440                         wake_up_new_task(p, clone_flags);
1441                 }
1442
1443                 tracehook_report_clone_complete(trace, regs,
1444                                                 clone_flags, nr, p);
1445
1446                 if (clone_flags & CLONE_VFORK) {
1447                         freezer_do_not_count();
1448                         wait_for_completion(&vfork);
1449                         freezer_count();
1450                         tracehook_report_vfork_done(p, nr);
1451                 }
1452         } else {
1453                 nr = PTR_ERR(p);
1454         }
1455         return nr;
1456 }
1457
1458 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1459 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1460 #endif
1461
1462 static void sighand_ctor(void *data)
1463 {
1464         struct sighand_struct *sighand = data;
1465
1466         spin_lock_init(&sighand->siglock);
1467         init_waitqueue_head(&sighand->signalfd_wqh);
1468 }
1469
1470 void __init proc_caches_init(void)
1471 {
1472         sighand_cachep = kmem_cache_create("sighand_cache",
1473                         sizeof(struct sighand_struct), 0,
1474                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1475                         SLAB_NOTRACK, sighand_ctor);
1476         signal_cachep = kmem_cache_create("signal_cache",
1477                         sizeof(struct signal_struct), 0,
1478                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1479         files_cachep = kmem_cache_create("files_cache",
1480                         sizeof(struct files_struct), 0,
1481                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1482         fs_cachep = kmem_cache_create("fs_cache",
1483                         sizeof(struct fs_struct), 0,
1484                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1485         mm_cachep = kmem_cache_create("mm_struct",
1486                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1487                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1488         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1489         mmap_init();
1490 }
1491
1492 /*
1493  * Check constraints on flags passed to the unshare system call and
1494  * force unsharing of additional process context as appropriate.
1495  */
1496 static void check_unshare_flags(unsigned long *flags_ptr)
1497 {
1498         /*
1499          * If unsharing a thread from a thread group, must also
1500          * unshare vm.
1501          */
1502         if (*flags_ptr & CLONE_THREAD)
1503                 *flags_ptr |= CLONE_VM;
1504
1505         /*
1506          * If unsharing vm, must also unshare signal handlers.
1507          */
1508         if (*flags_ptr & CLONE_VM)
1509                 *flags_ptr |= CLONE_SIGHAND;
1510
1511         /*
1512          * If unsharing signal handlers and the task was created
1513          * using CLONE_THREAD, then must unshare the thread
1514          */
1515         if ((*flags_ptr & CLONE_SIGHAND) &&
1516             (atomic_read(&current->signal->count) > 1))
1517                 *flags_ptr |= CLONE_THREAD;
1518
1519         /*
1520          * If unsharing namespace, must also unshare filesystem information.
1521          */
1522         if (*flags_ptr & CLONE_NEWNS)
1523                 *flags_ptr |= CLONE_FS;
1524 }
1525
1526 /*
1527  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1528  */
1529 static int unshare_thread(unsigned long unshare_flags)
1530 {
1531         if (unshare_flags & CLONE_THREAD)
1532                 return -EINVAL;
1533
1534         return 0;
1535 }
1536
1537 /*
1538  * Unshare the filesystem structure if it is being shared
1539  */
1540 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1541 {
1542         struct fs_struct *fs = current->fs;
1543
1544         if (!(unshare_flags & CLONE_FS) || !fs)
1545                 return 0;
1546
1547         /* don't need lock here; in the worst case we'll do useless copy */
1548         if (fs->users == 1)
1549                 return 0;
1550
1551         *new_fsp = copy_fs_struct(fs);
1552         if (!*new_fsp)
1553                 return -ENOMEM;
1554
1555         return 0;
1556 }
1557
1558 /*
1559  * Unsharing of sighand is not supported yet
1560  */
1561 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1562 {
1563         struct sighand_struct *sigh = current->sighand;
1564
1565         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1566                 return -EINVAL;
1567         else
1568                 return 0;
1569 }
1570
1571 /*
1572  * Unshare vm if it is being shared
1573  */
1574 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1575 {
1576         struct mm_struct *mm = current->mm;
1577
1578         if ((unshare_flags & CLONE_VM) &&
1579             (mm && atomic_read(&mm->mm_users) > 1)) {
1580                 return -EINVAL;
1581         }
1582
1583         return 0;
1584 }
1585
1586 /*
1587  * Unshare file descriptor table if it is being shared
1588  */
1589 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1590 {
1591         struct files_struct *fd = current->files;
1592         int error = 0;
1593
1594         if ((unshare_flags & CLONE_FILES) &&
1595             (fd && atomic_read(&fd->count) > 1)) {
1596                 *new_fdp = dup_fd(fd, &error);
1597                 if (!*new_fdp)
1598                         return error;
1599         }
1600
1601         return 0;
1602 }
1603
1604 /*
1605  * unshare allows a process to 'unshare' part of the process
1606  * context which was originally shared using clone.  copy_*
1607  * functions used by do_fork() cannot be used here directly
1608  * because they modify an inactive task_struct that is being
1609  * constructed. Here we are modifying the current, active,
1610  * task_struct.
1611  */
1612 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1613 {
1614         int err = 0;
1615         struct fs_struct *fs, *new_fs = NULL;
1616         struct sighand_struct *new_sigh = NULL;
1617         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1618         struct files_struct *fd, *new_fd = NULL;
1619         struct nsproxy *new_nsproxy = NULL;
1620         int do_sysvsem = 0;
1621
1622         check_unshare_flags(&unshare_flags);
1623
1624         /* Return -EINVAL for all unsupported flags */
1625         err = -EINVAL;
1626         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1627                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1628                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1629                 goto bad_unshare_out;
1630
1631         /*
1632          * CLONE_NEWIPC must also detach from the undolist: after switching
1633          * to a new ipc namespace, the semaphore arrays from the old
1634          * namespace are unreachable.
1635          */
1636         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1637                 do_sysvsem = 1;
1638         if ((err = unshare_thread(unshare_flags)))
1639                 goto bad_unshare_out;
1640         if ((err = unshare_fs(unshare_flags, &new_fs)))
1641                 goto bad_unshare_cleanup_thread;
1642         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1643                 goto bad_unshare_cleanup_fs;
1644         if ((err = unshare_vm(unshare_flags, &new_mm)))
1645                 goto bad_unshare_cleanup_sigh;
1646         if ((err = unshare_fd(unshare_flags, &new_fd)))
1647                 goto bad_unshare_cleanup_vm;
1648         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1649                         new_fs)))
1650                 goto bad_unshare_cleanup_fd;
1651
1652         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1653                 if (do_sysvsem) {
1654                         /*
1655                          * CLONE_SYSVSEM is equivalent to sys_exit().
1656                          */
1657                         exit_sem(current);
1658                 }
1659
1660                 if (new_nsproxy) {
1661                         switch_task_namespaces(current, new_nsproxy);
1662                         new_nsproxy = NULL;
1663                 }
1664
1665                 task_lock(current);
1666
1667                 if (new_fs) {
1668                         fs = current->fs;
1669                         write_lock(&fs->lock);
1670                         current->fs = new_fs;
1671                         if (--fs->users)
1672                                 new_fs = NULL;
1673                         else
1674                                 new_fs = fs;
1675                         write_unlock(&fs->lock);
1676                 }
1677
1678                 if (new_mm) {
1679                         mm = current->mm;
1680                         active_mm = current->active_mm;
1681                         current->mm = new_mm;
1682                         current->active_mm = new_mm;
1683                         activate_mm(active_mm, new_mm);
1684                         new_mm = mm;
1685                 }
1686
1687                 if (new_fd) {
1688                         fd = current->files;
1689                         current->files = new_fd;
1690                         new_fd = fd;
1691                 }
1692
1693                 task_unlock(current);
1694         }
1695
1696         if (new_nsproxy)
1697                 put_nsproxy(new_nsproxy);
1698
1699 bad_unshare_cleanup_fd:
1700         if (new_fd)
1701                 put_files_struct(new_fd);
1702
1703 bad_unshare_cleanup_vm:
1704         if (new_mm)
1705                 mmput(new_mm);
1706
1707 bad_unshare_cleanup_sigh:
1708         if (new_sigh)
1709                 if (atomic_dec_and_test(&new_sigh->count))
1710                         kmem_cache_free(sighand_cachep, new_sigh);
1711
1712 bad_unshare_cleanup_fs:
1713         if (new_fs)
1714                 free_fs_struct(new_fs);
1715
1716 bad_unshare_cleanup_thread:
1717 bad_unshare_out:
1718         return err;
1719 }
1720
1721 /*
1722  *      Helper to unshare the files of the current task.
1723  *      We don't want to expose copy_files internals to
1724  *      the exec layer of the kernel.
1725  */
1726
1727 int unshare_files(struct files_struct **displaced)
1728 {
1729         struct task_struct *task = current;
1730         struct files_struct *copy = NULL;
1731         int error;
1732
1733         error = unshare_fd(CLONE_FILES, &copy);
1734         if (error || !copy) {
1735                 *displaced = NULL;
1736                 return error;
1737         }
1738         *displaced = task->files;
1739         task_lock(task);
1740         task->files = copy;
1741         task_unlock(task);
1742         return 0;
1743 }