Correct nr_processes() when CPUs have been unplugged
[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_possible_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 void mm_init_aio(struct mm_struct *mm)
438 {
439 #ifdef CONFIG_AIO
440         spin_lock_init(&mm->ioctx_lock);
441         INIT_HLIST_HEAD(&mm->ioctx_list);
442 #endif
443 }
444
445 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
446 {
447         atomic_set(&mm->mm_users, 1);
448         atomic_set(&mm->mm_count, 1);
449         init_rwsem(&mm->mmap_sem);
450         INIT_LIST_HEAD(&mm->mmlist);
451         mm->flags = (current->mm) ?
452                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
453         mm->core_state = NULL;
454         mm->nr_ptes = 0;
455         set_mm_counter(mm, file_rss, 0);
456         set_mm_counter(mm, anon_rss, 0);
457         spin_lock_init(&mm->page_table_lock);
458         mm->free_area_cache = TASK_UNMAPPED_BASE;
459         mm->cached_hole_size = ~0UL;
460         mm_init_aio(mm);
461         mm_init_owner(mm, p);
462
463         if (likely(!mm_alloc_pgd(mm))) {
464                 mm->def_flags = 0;
465                 mmu_notifier_mm_init(mm);
466                 return mm;
467         }
468
469         free_mm(mm);
470         return NULL;
471 }
472
473 /*
474  * Allocate and initialize an mm_struct.
475  */
476 struct mm_struct * mm_alloc(void)
477 {
478         struct mm_struct * mm;
479
480         mm = allocate_mm();
481         if (mm) {
482                 memset(mm, 0, sizeof(*mm));
483                 mm = mm_init(mm, current);
484         }
485         return mm;
486 }
487
488 /*
489  * Called when the last reference to the mm
490  * is dropped: either by a lazy thread or by
491  * mmput. Free the page directory and the mm.
492  */
493 void __mmdrop(struct mm_struct *mm)
494 {
495         BUG_ON(mm == &init_mm);
496         mm_free_pgd(mm);
497         destroy_context(mm);
498         mmu_notifier_mm_destroy(mm);
499         free_mm(mm);
500 }
501 EXPORT_SYMBOL_GPL(__mmdrop);
502
503 /*
504  * Decrement the use count and release all resources for an mm.
505  */
506 void mmput(struct mm_struct *mm)
507 {
508         might_sleep();
509
510         if (atomic_dec_and_test(&mm->mm_users)) {
511                 exit_aio(mm);
512                 ksm_exit(mm);
513                 exit_mmap(mm);
514                 set_mm_exe_file(mm, NULL);
515                 if (!list_empty(&mm->mmlist)) {
516                         spin_lock(&mmlist_lock);
517                         list_del(&mm->mmlist);
518                         spin_unlock(&mmlist_lock);
519                 }
520                 put_swap_token(mm);
521                 if (mm->binfmt)
522                         module_put(mm->binfmt->module);
523                 mmdrop(mm);
524         }
525 }
526 EXPORT_SYMBOL_GPL(mmput);
527
528 /**
529  * get_task_mm - acquire a reference to the task's mm
530  *
531  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
532  * this kernel workthread has transiently adopted a user mm with use_mm,
533  * to do its AIO) is not set and if so returns a reference to it, after
534  * bumping up the use count.  User must release the mm via mmput()
535  * after use.  Typically used by /proc and ptrace.
536  */
537 struct mm_struct *get_task_mm(struct task_struct *task)
538 {
539         struct mm_struct *mm;
540
541         task_lock(task);
542         mm = task->mm;
543         if (mm) {
544                 if (task->flags & PF_KTHREAD)
545                         mm = NULL;
546                 else
547                         atomic_inc(&mm->mm_users);
548         }
549         task_unlock(task);
550         return mm;
551 }
552 EXPORT_SYMBOL_GPL(get_task_mm);
553
554 /* Please note the differences between mmput and mm_release.
555  * mmput is called whenever we stop holding onto a mm_struct,
556  * error success whatever.
557  *
558  * mm_release is called after a mm_struct has been removed
559  * from the current process.
560  *
561  * This difference is important for error handling, when we
562  * only half set up a mm_struct for a new process and need to restore
563  * the old one.  Because we mmput the new mm_struct before
564  * restoring the old one. . .
565  * Eric Biederman 10 January 1998
566  */
567 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
568 {
569         struct completion *vfork_done = tsk->vfork_done;
570
571         /* Get rid of any futexes when releasing the mm */
572 #ifdef CONFIG_FUTEX
573         if (unlikely(tsk->robust_list)) {
574                 exit_robust_list(tsk);
575                 tsk->robust_list = NULL;
576         }
577 #ifdef CONFIG_COMPAT
578         if (unlikely(tsk->compat_robust_list)) {
579                 compat_exit_robust_list(tsk);
580                 tsk->compat_robust_list = NULL;
581         }
582 #endif
583         if (unlikely(!list_empty(&tsk->pi_state_list)))
584                 exit_pi_state_list(tsk);
585 #endif
586
587         /* Get rid of any cached register state */
588         deactivate_mm(tsk, mm);
589
590         /* notify parent sleeping on vfork() */
591         if (vfork_done) {
592                 tsk->vfork_done = NULL;
593                 complete(vfork_done);
594         }
595
596         /*
597          * If we're exiting normally, clear a user-space tid field if
598          * requested.  We leave this alone when dying by signal, to leave
599          * the value intact in a core dump, and to save the unnecessary
600          * trouble otherwise.  Userland only wants this done for a sys_exit.
601          */
602         if (tsk->clear_child_tid) {
603                 if (!(tsk->flags & PF_SIGNALED) &&
604                     atomic_read(&mm->mm_users) > 1) {
605                         /*
606                          * We don't check the error code - if userspace has
607                          * not set up a proper pointer then tough luck.
608                          */
609                         put_user(0, tsk->clear_child_tid);
610                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
611                                         1, NULL, NULL, 0);
612                 }
613                 tsk->clear_child_tid = NULL;
614         }
615 }
616
617 /*
618  * Allocate a new mm structure and copy contents from the
619  * mm structure of the passed in task structure.
620  */
621 struct mm_struct *dup_mm(struct task_struct *tsk)
622 {
623         struct mm_struct *mm, *oldmm = current->mm;
624         int err;
625
626         if (!oldmm)
627                 return NULL;
628
629         mm = allocate_mm();
630         if (!mm)
631                 goto fail_nomem;
632
633         memcpy(mm, oldmm, sizeof(*mm));
634
635         /* Initializing for Swap token stuff */
636         mm->token_priority = 0;
637         mm->last_interval = 0;
638
639         if (!mm_init(mm, tsk))
640                 goto fail_nomem;
641
642         if (init_new_context(tsk, mm))
643                 goto fail_nocontext;
644
645         dup_mm_exe_file(oldmm, mm);
646
647         err = dup_mmap(mm, oldmm);
648         if (err)
649                 goto free_pt;
650
651         mm->hiwater_rss = get_mm_rss(mm);
652         mm->hiwater_vm = mm->total_vm;
653
654         if (mm->binfmt && !try_module_get(mm->binfmt->module))
655                 goto free_pt;
656
657         return mm;
658
659 free_pt:
660         /* don't put binfmt in mmput, we haven't got module yet */
661         mm->binfmt = NULL;
662         mmput(mm);
663
664 fail_nomem:
665         return NULL;
666
667 fail_nocontext:
668         /*
669          * If init_new_context() failed, we cannot use mmput() to free the mm
670          * because it calls destroy_context()
671          */
672         mm_free_pgd(mm);
673         free_mm(mm);
674         return NULL;
675 }
676
677 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
678 {
679         struct mm_struct * mm, *oldmm;
680         int retval;
681
682         tsk->min_flt = tsk->maj_flt = 0;
683         tsk->nvcsw = tsk->nivcsw = 0;
684 #ifdef CONFIG_DETECT_HUNG_TASK
685         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
686 #endif
687
688         tsk->mm = NULL;
689         tsk->active_mm = NULL;
690
691         /*
692          * Are we cloning a kernel thread?
693          *
694          * We need to steal a active VM for that..
695          */
696         oldmm = current->mm;
697         if (!oldmm)
698                 return 0;
699
700         if (clone_flags & CLONE_VM) {
701                 atomic_inc(&oldmm->mm_users);
702                 mm = oldmm;
703                 goto good_mm;
704         }
705
706         retval = -ENOMEM;
707         mm = dup_mm(tsk);
708         if (!mm)
709                 goto fail_nomem;
710
711 good_mm:
712         /* Initializing for Swap token stuff */
713         mm->token_priority = 0;
714         mm->last_interval = 0;
715
716         tsk->mm = mm;
717         tsk->active_mm = mm;
718         return 0;
719
720 fail_nomem:
721         return retval;
722 }
723
724 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
725 {
726         struct fs_struct *fs = current->fs;
727         if (clone_flags & CLONE_FS) {
728                 /* tsk->fs is already what we want */
729                 write_lock(&fs->lock);
730                 if (fs->in_exec) {
731                         write_unlock(&fs->lock);
732                         return -EAGAIN;
733                 }
734                 fs->users++;
735                 write_unlock(&fs->lock);
736                 return 0;
737         }
738         tsk->fs = copy_fs_struct(fs);
739         if (!tsk->fs)
740                 return -ENOMEM;
741         return 0;
742 }
743
744 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
745 {
746         struct files_struct *oldf, *newf;
747         int error = 0;
748
749         /*
750          * A background process may not have any files ...
751          */
752         oldf = current->files;
753         if (!oldf)
754                 goto out;
755
756         if (clone_flags & CLONE_FILES) {
757                 atomic_inc(&oldf->count);
758                 goto out;
759         }
760
761         newf = dup_fd(oldf, &error);
762         if (!newf)
763                 goto out;
764
765         tsk->files = newf;
766         error = 0;
767 out:
768         return error;
769 }
770
771 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
772 {
773 #ifdef CONFIG_BLOCK
774         struct io_context *ioc = current->io_context;
775
776         if (!ioc)
777                 return 0;
778         /*
779          * Share io context with parent, if CLONE_IO is set
780          */
781         if (clone_flags & CLONE_IO) {
782                 tsk->io_context = ioc_task_link(ioc);
783                 if (unlikely(!tsk->io_context))
784                         return -ENOMEM;
785         } else if (ioprio_valid(ioc->ioprio)) {
786                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
787                 if (unlikely(!tsk->io_context))
788                         return -ENOMEM;
789
790                 tsk->io_context->ioprio = ioc->ioprio;
791         }
792 #endif
793         return 0;
794 }
795
796 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
797 {
798         struct sighand_struct *sig;
799
800         if (clone_flags & CLONE_SIGHAND) {
801                 atomic_inc(&current->sighand->count);
802                 return 0;
803         }
804         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
805         rcu_assign_pointer(tsk->sighand, sig);
806         if (!sig)
807                 return -ENOMEM;
808         atomic_set(&sig->count, 1);
809         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
810         return 0;
811 }
812
813 void __cleanup_sighand(struct sighand_struct *sighand)
814 {
815         if (atomic_dec_and_test(&sighand->count))
816                 kmem_cache_free(sighand_cachep, sighand);
817 }
818
819
820 /*
821  * Initialize POSIX timer handling for a thread group.
822  */
823 static void posix_cpu_timers_init_group(struct signal_struct *sig)
824 {
825         /* Thread group counters. */
826         thread_group_cputime_init(sig);
827
828         /* Expiration times and increments. */
829         sig->it[CPUCLOCK_PROF].expires = cputime_zero;
830         sig->it[CPUCLOCK_PROF].incr = cputime_zero;
831         sig->it[CPUCLOCK_VIRT].expires = cputime_zero;
832         sig->it[CPUCLOCK_VIRT].incr = cputime_zero;
833
834         /* Cached expiration times. */
835         sig->cputime_expires.prof_exp = cputime_zero;
836         sig->cputime_expires.virt_exp = cputime_zero;
837         sig->cputime_expires.sched_exp = 0;
838
839         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
840                 sig->cputime_expires.prof_exp =
841                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
842                 sig->cputimer.running = 1;
843         }
844
845         /* The timer lists. */
846         INIT_LIST_HEAD(&sig->cpu_timers[0]);
847         INIT_LIST_HEAD(&sig->cpu_timers[1]);
848         INIT_LIST_HEAD(&sig->cpu_timers[2]);
849 }
850
851 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
852 {
853         struct signal_struct *sig;
854
855         if (clone_flags & CLONE_THREAD)
856                 return 0;
857
858         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
859         tsk->signal = sig;
860         if (!sig)
861                 return -ENOMEM;
862
863         atomic_set(&sig->count, 1);
864         atomic_set(&sig->live, 1);
865         init_waitqueue_head(&sig->wait_chldexit);
866         sig->flags = 0;
867         if (clone_flags & CLONE_NEWPID)
868                 sig->flags |= SIGNAL_UNKILLABLE;
869         sig->group_exit_code = 0;
870         sig->group_exit_task = NULL;
871         sig->group_stop_count = 0;
872         sig->curr_target = tsk;
873         init_sigpending(&sig->shared_pending);
874         INIT_LIST_HEAD(&sig->posix_timers);
875
876         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
877         sig->it_real_incr.tv64 = 0;
878         sig->real_timer.function = it_real_fn;
879
880         sig->leader = 0;        /* session leadership doesn't inherit */
881         sig->tty_old_pgrp = NULL;
882         sig->tty = NULL;
883
884         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
885         sig->gtime = cputime_zero;
886         sig->cgtime = cputime_zero;
887         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
888         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
889         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
890         sig->maxrss = sig->cmaxrss = 0;
891         task_io_accounting_init(&sig->ioac);
892         sig->sum_sched_runtime = 0;
893         taskstats_tgid_init(sig);
894
895         task_lock(current->group_leader);
896         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
897         task_unlock(current->group_leader);
898
899         posix_cpu_timers_init_group(sig);
900
901         acct_init_pacct(&sig->pacct);
902
903         tty_audit_fork(sig);
904
905         sig->oom_adj = current->signal->oom_adj;
906
907         return 0;
908 }
909
910 void __cleanup_signal(struct signal_struct *sig)
911 {
912         thread_group_cputime_free(sig);
913         tty_kref_put(sig->tty);
914         kmem_cache_free(signal_cachep, sig);
915 }
916
917 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
918 {
919         unsigned long new_flags = p->flags;
920
921         new_flags &= ~PF_SUPERPRIV;
922         new_flags |= PF_FORKNOEXEC;
923         new_flags |= PF_STARTING;
924         p->flags = new_flags;
925         clear_freeze_flag(p);
926 }
927
928 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
929 {
930         current->clear_child_tid = tidptr;
931
932         return task_pid_vnr(current);
933 }
934
935 static void rt_mutex_init_task(struct task_struct *p)
936 {
937         spin_lock_init(&p->pi_lock);
938 #ifdef CONFIG_RT_MUTEXES
939         plist_head_init(&p->pi_waiters, &p->pi_lock);
940         p->pi_blocked_on = NULL;
941 #endif
942 }
943
944 #ifdef CONFIG_MM_OWNER
945 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
946 {
947         mm->owner = p;
948 }
949 #endif /* CONFIG_MM_OWNER */
950
951 /*
952  * Initialize POSIX timer handling for a single task.
953  */
954 static void posix_cpu_timers_init(struct task_struct *tsk)
955 {
956         tsk->cputime_expires.prof_exp = cputime_zero;
957         tsk->cputime_expires.virt_exp = cputime_zero;
958         tsk->cputime_expires.sched_exp = 0;
959         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
960         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
961         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
962 }
963
964 /*
965  * This creates a new process as a copy of the old one,
966  * but does not actually start it yet.
967  *
968  * It copies the registers, and all the appropriate
969  * parts of the process environment (as per the clone
970  * flags). The actual kick-off is left to the caller.
971  */
972 static struct task_struct *copy_process(unsigned long clone_flags,
973                                         unsigned long stack_start,
974                                         struct pt_regs *regs,
975                                         unsigned long stack_size,
976                                         int __user *child_tidptr,
977                                         struct pid *pid,
978                                         int trace)
979 {
980         int retval;
981         struct task_struct *p;
982         int cgroup_callbacks_done = 0;
983
984         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
985                 return ERR_PTR(-EINVAL);
986
987         /*
988          * Thread groups must share signals as well, and detached threads
989          * can only be started up within the thread group.
990          */
991         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
992                 return ERR_PTR(-EINVAL);
993
994         /*
995          * Shared signal handlers imply shared VM. By way of the above,
996          * thread groups also imply shared VM. Blocking this case allows
997          * for various simplifications in other code.
998          */
999         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1000                 return ERR_PTR(-EINVAL);
1001
1002         /*
1003          * Siblings of global init remain as zombies on exit since they are
1004          * not reaped by their parent (swapper). To solve this and to avoid
1005          * multi-rooted process trees, prevent global and container-inits
1006          * from creating siblings.
1007          */
1008         if ((clone_flags & CLONE_PARENT) &&
1009                                 current->signal->flags & SIGNAL_UNKILLABLE)
1010                 return ERR_PTR(-EINVAL);
1011
1012         retval = security_task_create(clone_flags);
1013         if (retval)
1014                 goto fork_out;
1015
1016         retval = -ENOMEM;
1017         p = dup_task_struct(current);
1018         if (!p)
1019                 goto fork_out;
1020
1021         ftrace_graph_init_task(p);
1022
1023         rt_mutex_init_task(p);
1024
1025 #ifdef CONFIG_PROVE_LOCKING
1026         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1027         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1028 #endif
1029         retval = -EAGAIN;
1030         if (atomic_read(&p->real_cred->user->processes) >=
1031                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1032                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1033                     p->real_cred->user != INIT_USER)
1034                         goto bad_fork_free;
1035         }
1036
1037         retval = copy_creds(p, clone_flags);
1038         if (retval < 0)
1039                 goto bad_fork_free;
1040
1041         /*
1042          * If multiple threads are within copy_process(), then this check
1043          * triggers too late. This doesn't hurt, the check is only there
1044          * to stop root fork bombs.
1045          */
1046         retval = -EAGAIN;
1047         if (nr_threads >= max_threads)
1048                 goto bad_fork_cleanup_count;
1049
1050         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1051                 goto bad_fork_cleanup_count;
1052
1053         p->did_exec = 0;
1054         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1055         copy_flags(clone_flags, p);
1056         INIT_LIST_HEAD(&p->children);
1057         INIT_LIST_HEAD(&p->sibling);
1058         rcu_copy_process(p);
1059         p->vfork_done = NULL;
1060         spin_lock_init(&p->alloc_lock);
1061
1062         init_sigpending(&p->pending);
1063
1064         p->utime = cputime_zero;
1065         p->stime = cputime_zero;
1066         p->gtime = cputime_zero;
1067         p->utimescaled = cputime_zero;
1068         p->stimescaled = cputime_zero;
1069         p->prev_utime = cputime_zero;
1070         p->prev_stime = cputime_zero;
1071
1072         p->default_timer_slack_ns = current->timer_slack_ns;
1073
1074         task_io_accounting_init(&p->ioac);
1075         acct_clear_integrals(p);
1076
1077         posix_cpu_timers_init(p);
1078
1079         p->lock_depth = -1;             /* -1 = no lock */
1080         do_posix_clock_monotonic_gettime(&p->start_time);
1081         p->real_start_time = p->start_time;
1082         monotonic_to_bootbased(&p->real_start_time);
1083         p->io_context = NULL;
1084         p->audit_context = NULL;
1085         cgroup_fork(p);
1086 #ifdef CONFIG_NUMA
1087         p->mempolicy = mpol_dup(p->mempolicy);
1088         if (IS_ERR(p->mempolicy)) {
1089                 retval = PTR_ERR(p->mempolicy);
1090                 p->mempolicy = NULL;
1091                 goto bad_fork_cleanup_cgroup;
1092         }
1093         mpol_fix_fork_child_flag(p);
1094 #endif
1095 #ifdef CONFIG_TRACE_IRQFLAGS
1096         p->irq_events = 0;
1097 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1098         p->hardirqs_enabled = 1;
1099 #else
1100         p->hardirqs_enabled = 0;
1101 #endif
1102         p->hardirq_enable_ip = 0;
1103         p->hardirq_enable_event = 0;
1104         p->hardirq_disable_ip = _THIS_IP_;
1105         p->hardirq_disable_event = 0;
1106         p->softirqs_enabled = 1;
1107         p->softirq_enable_ip = _THIS_IP_;
1108         p->softirq_enable_event = 0;
1109         p->softirq_disable_ip = 0;
1110         p->softirq_disable_event = 0;
1111         p->hardirq_context = 0;
1112         p->softirq_context = 0;
1113 #endif
1114 #ifdef CONFIG_LOCKDEP
1115         p->lockdep_depth = 0; /* no locks held yet */
1116         p->curr_chain_key = 0;
1117         p->lockdep_recursion = 0;
1118 #endif
1119
1120 #ifdef CONFIG_DEBUG_MUTEXES
1121         p->blocked_on = NULL; /* not blocked yet */
1122 #endif
1123
1124         p->bts = NULL;
1125
1126         p->stack_start = stack_start;
1127
1128         /* Perform scheduler related setup. Assign this task to a CPU. */
1129         sched_fork(p, clone_flags);
1130
1131         retval = perf_event_init_task(p);
1132         if (retval)
1133                 goto bad_fork_cleanup_policy;
1134
1135         if ((retval = audit_alloc(p)))
1136                 goto bad_fork_cleanup_policy;
1137         /* copy all the process information */
1138         if ((retval = copy_semundo(clone_flags, p)))
1139                 goto bad_fork_cleanup_audit;
1140         if ((retval = copy_files(clone_flags, p)))
1141                 goto bad_fork_cleanup_semundo;
1142         if ((retval = copy_fs(clone_flags, p)))
1143                 goto bad_fork_cleanup_files;
1144         if ((retval = copy_sighand(clone_flags, p)))
1145                 goto bad_fork_cleanup_fs;
1146         if ((retval = copy_signal(clone_flags, p)))
1147                 goto bad_fork_cleanup_sighand;
1148         if ((retval = copy_mm(clone_flags, p)))
1149                 goto bad_fork_cleanup_signal;
1150         if ((retval = copy_namespaces(clone_flags, p)))
1151                 goto bad_fork_cleanup_mm;
1152         if ((retval = copy_io(clone_flags, p)))
1153                 goto bad_fork_cleanup_namespaces;
1154         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1155         if (retval)
1156                 goto bad_fork_cleanup_io;
1157
1158         if (pid != &init_struct_pid) {
1159                 retval = -ENOMEM;
1160                 pid = alloc_pid(p->nsproxy->pid_ns);
1161                 if (!pid)
1162                         goto bad_fork_cleanup_io;
1163
1164                 if (clone_flags & CLONE_NEWPID) {
1165                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1166                         if (retval < 0)
1167                                 goto bad_fork_free_pid;
1168                 }
1169         }
1170
1171         p->pid = pid_nr(pid);
1172         p->tgid = p->pid;
1173         if (clone_flags & CLONE_THREAD)
1174                 p->tgid = current->tgid;
1175
1176         if (current->nsproxy != p->nsproxy) {
1177                 retval = ns_cgroup_clone(p, pid);
1178                 if (retval)
1179                         goto bad_fork_free_pid;
1180         }
1181
1182         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1183         /*
1184          * Clear TID on mm_release()?
1185          */
1186         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1187 #ifdef CONFIG_FUTEX
1188         p->robust_list = NULL;
1189 #ifdef CONFIG_COMPAT
1190         p->compat_robust_list = NULL;
1191 #endif
1192         INIT_LIST_HEAD(&p->pi_state_list);
1193         p->pi_state_cache = NULL;
1194 #endif
1195         /*
1196          * sigaltstack should be cleared when sharing the same VM
1197          */
1198         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1199                 p->sas_ss_sp = p->sas_ss_size = 0;
1200
1201         /*
1202          * Syscall tracing should be turned off in the child regardless
1203          * of CLONE_PTRACE.
1204          */
1205         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1206 #ifdef TIF_SYSCALL_EMU
1207         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1208 #endif
1209         clear_all_latency_tracing(p);
1210
1211         /* ok, now we should be set up.. */
1212         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1213         p->pdeath_signal = 0;
1214         p->exit_state = 0;
1215
1216         /*
1217          * Ok, make it visible to the rest of the system.
1218          * We dont wake it up yet.
1219          */
1220         p->group_leader = p;
1221         INIT_LIST_HEAD(&p->thread_group);
1222
1223         /* Now that the task is set up, run cgroup callbacks if
1224          * necessary. We need to run them before the task is visible
1225          * on the tasklist. */
1226         cgroup_fork_callbacks(p);
1227         cgroup_callbacks_done = 1;
1228
1229         /* Need tasklist lock for parent etc handling! */
1230         write_lock_irq(&tasklist_lock);
1231
1232         /*
1233          * The task hasn't been attached yet, so its cpus_allowed mask will
1234          * not be changed, nor will its assigned CPU.
1235          *
1236          * The cpus_allowed mask of the parent may have changed after it was
1237          * copied first time - so re-copy it here, then check the child's CPU
1238          * to ensure it is on a valid CPU (and if not, just force it back to
1239          * parent's CPU). This avoids alot of nasty races.
1240          */
1241         p->cpus_allowed = current->cpus_allowed;
1242         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1243         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1244                         !cpu_online(task_cpu(p))))
1245                 set_task_cpu(p, smp_processor_id());
1246
1247         /* CLONE_PARENT re-uses the old parent */
1248         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1249                 p->real_parent = current->real_parent;
1250                 p->parent_exec_id = current->parent_exec_id;
1251         } else {
1252                 p->real_parent = current;
1253                 p->parent_exec_id = current->self_exec_id;
1254         }
1255
1256         spin_lock(&current->sighand->siglock);
1257
1258         /*
1259          * Process group and session signals need to be delivered to just the
1260          * parent before the fork or both the parent and the child after the
1261          * fork. Restart if a signal comes in before we add the new process to
1262          * it's process group.
1263          * A fatal signal pending means that current will exit, so the new
1264          * thread can't slip out of an OOM kill (or normal SIGKILL).
1265          */
1266         recalc_sigpending();
1267         if (signal_pending(current)) {
1268                 spin_unlock(&current->sighand->siglock);
1269                 write_unlock_irq(&tasklist_lock);
1270                 retval = -ERESTARTNOINTR;
1271                 goto bad_fork_free_pid;
1272         }
1273
1274         if (clone_flags & CLONE_THREAD) {
1275                 atomic_inc(&current->signal->count);
1276                 atomic_inc(&current->signal->live);
1277                 p->group_leader = current->group_leader;
1278                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1279         }
1280
1281         if (likely(p->pid)) {
1282                 list_add_tail(&p->sibling, &p->real_parent->children);
1283                 tracehook_finish_clone(p, clone_flags, trace);
1284
1285                 if (thread_group_leader(p)) {
1286                         if (clone_flags & CLONE_NEWPID)
1287                                 p->nsproxy->pid_ns->child_reaper = p;
1288
1289                         p->signal->leader_pid = pid;
1290                         tty_kref_put(p->signal->tty);
1291                         p->signal->tty = tty_kref_get(current->signal->tty);
1292                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1293                         attach_pid(p, PIDTYPE_SID, task_session(current));
1294                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1295                         __get_cpu_var(process_counts)++;
1296                 }
1297                 attach_pid(p, PIDTYPE_PID, pid);
1298                 nr_threads++;
1299         }
1300
1301         total_forks++;
1302         spin_unlock(&current->sighand->siglock);
1303         write_unlock_irq(&tasklist_lock);
1304         proc_fork_connector(p);
1305         cgroup_post_fork(p);
1306         perf_event_fork(p);
1307         return p;
1308
1309 bad_fork_free_pid:
1310         if (pid != &init_struct_pid)
1311                 free_pid(pid);
1312 bad_fork_cleanup_io:
1313         put_io_context(p->io_context);
1314 bad_fork_cleanup_namespaces:
1315         exit_task_namespaces(p);
1316 bad_fork_cleanup_mm:
1317         if (p->mm)
1318                 mmput(p->mm);
1319 bad_fork_cleanup_signal:
1320         if (!(clone_flags & CLONE_THREAD))
1321                 __cleanup_signal(p->signal);
1322 bad_fork_cleanup_sighand:
1323         __cleanup_sighand(p->sighand);
1324 bad_fork_cleanup_fs:
1325         exit_fs(p); /* blocking */
1326 bad_fork_cleanup_files:
1327         exit_files(p); /* blocking */
1328 bad_fork_cleanup_semundo:
1329         exit_sem(p);
1330 bad_fork_cleanup_audit:
1331         audit_free(p);
1332 bad_fork_cleanup_policy:
1333         perf_event_free_task(p);
1334 #ifdef CONFIG_NUMA
1335         mpol_put(p->mempolicy);
1336 bad_fork_cleanup_cgroup:
1337 #endif
1338         cgroup_exit(p, cgroup_callbacks_done);
1339         delayacct_tsk_free(p);
1340         module_put(task_thread_info(p)->exec_domain->module);
1341 bad_fork_cleanup_count:
1342         atomic_dec(&p->cred->user->processes);
1343         exit_creds(p);
1344 bad_fork_free:
1345         free_task(p);
1346 fork_out:
1347         return ERR_PTR(retval);
1348 }
1349
1350 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1351 {
1352         memset(regs, 0, sizeof(struct pt_regs));
1353         return regs;
1354 }
1355
1356 struct task_struct * __cpuinit fork_idle(int cpu)
1357 {
1358         struct task_struct *task;
1359         struct pt_regs regs;
1360
1361         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1362                             &init_struct_pid, 0);
1363         if (!IS_ERR(task))
1364                 init_idle(task, cpu);
1365
1366         return task;
1367 }
1368
1369 /*
1370  *  Ok, this is the main fork-routine.
1371  *
1372  * It copies the process, and if successful kick-starts
1373  * it and waits for it to finish using the VM if required.
1374  */
1375 long do_fork(unsigned long clone_flags,
1376               unsigned long stack_start,
1377               struct pt_regs *regs,
1378               unsigned long stack_size,
1379               int __user *parent_tidptr,
1380               int __user *child_tidptr)
1381 {
1382         struct task_struct *p;
1383         int trace = 0;
1384         long nr;
1385
1386         /*
1387          * Do some preliminary argument and permissions checking before we
1388          * actually start allocating stuff
1389          */
1390         if (clone_flags & CLONE_NEWUSER) {
1391                 if (clone_flags & CLONE_THREAD)
1392                         return -EINVAL;
1393                 /* hopefully this check will go away when userns support is
1394                  * complete
1395                  */
1396                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1397                                 !capable(CAP_SETGID))
1398                         return -EPERM;
1399         }
1400
1401         /*
1402          * We hope to recycle these flags after 2.6.26
1403          */
1404         if (unlikely(clone_flags & CLONE_STOPPED)) {
1405                 static int __read_mostly count = 100;
1406
1407                 if (count > 0 && printk_ratelimit()) {
1408                         char comm[TASK_COMM_LEN];
1409
1410                         count--;
1411                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1412                                         "clone flags 0x%lx\n",
1413                                 get_task_comm(comm, current),
1414                                 clone_flags & CLONE_STOPPED);
1415                 }
1416         }
1417
1418         /*
1419          * When called from kernel_thread, don't do user tracing stuff.
1420          */
1421         if (likely(user_mode(regs)))
1422                 trace = tracehook_prepare_clone(clone_flags);
1423
1424         p = copy_process(clone_flags, stack_start, regs, stack_size,
1425                          child_tidptr, NULL, trace);
1426         /*
1427          * Do this prior waking up the new thread - the thread pointer
1428          * might get invalid after that point, if the thread exits quickly.
1429          */
1430         if (!IS_ERR(p)) {
1431                 struct completion vfork;
1432
1433                 trace_sched_process_fork(current, p);
1434
1435                 nr = task_pid_vnr(p);
1436
1437                 if (clone_flags & CLONE_PARENT_SETTID)
1438                         put_user(nr, parent_tidptr);
1439
1440                 if (clone_flags & CLONE_VFORK) {
1441                         p->vfork_done = &vfork;
1442                         init_completion(&vfork);
1443                 }
1444
1445                 audit_finish_fork(p);
1446                 tracehook_report_clone(regs, clone_flags, nr, p);
1447
1448                 /*
1449                  * We set PF_STARTING at creation in case tracing wants to
1450                  * use this to distinguish a fully live task from one that
1451                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1452                  * clear it and set the child going.
1453                  */
1454                 p->flags &= ~PF_STARTING;
1455
1456                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1457                         /*
1458                          * We'll start up with an immediate SIGSTOP.
1459                          */
1460                         sigaddset(&p->pending.signal, SIGSTOP);
1461                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1462                         __set_task_state(p, TASK_STOPPED);
1463                 } else {
1464                         wake_up_new_task(p, clone_flags);
1465                 }
1466
1467                 tracehook_report_clone_complete(trace, regs,
1468                                                 clone_flags, nr, p);
1469
1470                 if (clone_flags & CLONE_VFORK) {
1471                         freezer_do_not_count();
1472                         wait_for_completion(&vfork);
1473                         freezer_count();
1474                         tracehook_report_vfork_done(p, nr);
1475                 }
1476         } else {
1477                 nr = PTR_ERR(p);
1478         }
1479         return nr;
1480 }
1481
1482 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1483 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1484 #endif
1485
1486 static void sighand_ctor(void *data)
1487 {
1488         struct sighand_struct *sighand = data;
1489
1490         spin_lock_init(&sighand->siglock);
1491         init_waitqueue_head(&sighand->signalfd_wqh);
1492 }
1493
1494 void __init proc_caches_init(void)
1495 {
1496         sighand_cachep = kmem_cache_create("sighand_cache",
1497                         sizeof(struct sighand_struct), 0,
1498                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1499                         SLAB_NOTRACK, sighand_ctor);
1500         signal_cachep = kmem_cache_create("signal_cache",
1501                         sizeof(struct signal_struct), 0,
1502                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1503         files_cachep = kmem_cache_create("files_cache",
1504                         sizeof(struct files_struct), 0,
1505                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1506         fs_cachep = kmem_cache_create("fs_cache",
1507                         sizeof(struct fs_struct), 0,
1508                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1509         mm_cachep = kmem_cache_create("mm_struct",
1510                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1511                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1512         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1513         mmap_init();
1514 }
1515
1516 /*
1517  * Check constraints on flags passed to the unshare system call and
1518  * force unsharing of additional process context as appropriate.
1519  */
1520 static void check_unshare_flags(unsigned long *flags_ptr)
1521 {
1522         /*
1523          * If unsharing a thread from a thread group, must also
1524          * unshare vm.
1525          */
1526         if (*flags_ptr & CLONE_THREAD)
1527                 *flags_ptr |= CLONE_VM;
1528
1529         /*
1530          * If unsharing vm, must also unshare signal handlers.
1531          */
1532         if (*flags_ptr & CLONE_VM)
1533                 *flags_ptr |= CLONE_SIGHAND;
1534
1535         /*
1536          * If unsharing signal handlers and the task was created
1537          * using CLONE_THREAD, then must unshare the thread
1538          */
1539         if ((*flags_ptr & CLONE_SIGHAND) &&
1540             (atomic_read(&current->signal->count) > 1))
1541                 *flags_ptr |= CLONE_THREAD;
1542
1543         /*
1544          * If unsharing namespace, must also unshare filesystem information.
1545          */
1546         if (*flags_ptr & CLONE_NEWNS)
1547                 *flags_ptr |= CLONE_FS;
1548 }
1549
1550 /*
1551  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1552  */
1553 static int unshare_thread(unsigned long unshare_flags)
1554 {
1555         if (unshare_flags & CLONE_THREAD)
1556                 return -EINVAL;
1557
1558         return 0;
1559 }
1560
1561 /*
1562  * Unshare the filesystem structure if it is being shared
1563  */
1564 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1565 {
1566         struct fs_struct *fs = current->fs;
1567
1568         if (!(unshare_flags & CLONE_FS) || !fs)
1569                 return 0;
1570
1571         /* don't need lock here; in the worst case we'll do useless copy */
1572         if (fs->users == 1)
1573                 return 0;
1574
1575         *new_fsp = copy_fs_struct(fs);
1576         if (!*new_fsp)
1577                 return -ENOMEM;
1578
1579         return 0;
1580 }
1581
1582 /*
1583  * Unsharing of sighand is not supported yet
1584  */
1585 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1586 {
1587         struct sighand_struct *sigh = current->sighand;
1588
1589         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1590                 return -EINVAL;
1591         else
1592                 return 0;
1593 }
1594
1595 /*
1596  * Unshare vm if it is being shared
1597  */
1598 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1599 {
1600         struct mm_struct *mm = current->mm;
1601
1602         if ((unshare_flags & CLONE_VM) &&
1603             (mm && atomic_read(&mm->mm_users) > 1)) {
1604                 return -EINVAL;
1605         }
1606
1607         return 0;
1608 }
1609
1610 /*
1611  * Unshare file descriptor table if it is being shared
1612  */
1613 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1614 {
1615         struct files_struct *fd = current->files;
1616         int error = 0;
1617
1618         if ((unshare_flags & CLONE_FILES) &&
1619             (fd && atomic_read(&fd->count) > 1)) {
1620                 *new_fdp = dup_fd(fd, &error);
1621                 if (!*new_fdp)
1622                         return error;
1623         }
1624
1625         return 0;
1626 }
1627
1628 /*
1629  * unshare allows a process to 'unshare' part of the process
1630  * context which was originally shared using clone.  copy_*
1631  * functions used by do_fork() cannot be used here directly
1632  * because they modify an inactive task_struct that is being
1633  * constructed. Here we are modifying the current, active,
1634  * task_struct.
1635  */
1636 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1637 {
1638         int err = 0;
1639         struct fs_struct *fs, *new_fs = NULL;
1640         struct sighand_struct *new_sigh = NULL;
1641         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1642         struct files_struct *fd, *new_fd = NULL;
1643         struct nsproxy *new_nsproxy = NULL;
1644         int do_sysvsem = 0;
1645
1646         check_unshare_flags(&unshare_flags);
1647
1648         /* Return -EINVAL for all unsupported flags */
1649         err = -EINVAL;
1650         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1651                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1652                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1653                 goto bad_unshare_out;
1654
1655         /*
1656          * CLONE_NEWIPC must also detach from the undolist: after switching
1657          * to a new ipc namespace, the semaphore arrays from the old
1658          * namespace are unreachable.
1659          */
1660         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1661                 do_sysvsem = 1;
1662         if ((err = unshare_thread(unshare_flags)))
1663                 goto bad_unshare_out;
1664         if ((err = unshare_fs(unshare_flags, &new_fs)))
1665                 goto bad_unshare_cleanup_thread;
1666         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1667                 goto bad_unshare_cleanup_fs;
1668         if ((err = unshare_vm(unshare_flags, &new_mm)))
1669                 goto bad_unshare_cleanup_sigh;
1670         if ((err = unshare_fd(unshare_flags, &new_fd)))
1671                 goto bad_unshare_cleanup_vm;
1672         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1673                         new_fs)))
1674                 goto bad_unshare_cleanup_fd;
1675
1676         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1677                 if (do_sysvsem) {
1678                         /*
1679                          * CLONE_SYSVSEM is equivalent to sys_exit().
1680                          */
1681                         exit_sem(current);
1682                 }
1683
1684                 if (new_nsproxy) {
1685                         switch_task_namespaces(current, new_nsproxy);
1686                         new_nsproxy = NULL;
1687                 }
1688
1689                 task_lock(current);
1690
1691                 if (new_fs) {
1692                         fs = current->fs;
1693                         write_lock(&fs->lock);
1694                         current->fs = new_fs;
1695                         if (--fs->users)
1696                                 new_fs = NULL;
1697                         else
1698                                 new_fs = fs;
1699                         write_unlock(&fs->lock);
1700                 }
1701
1702                 if (new_mm) {
1703                         mm = current->mm;
1704                         active_mm = current->active_mm;
1705                         current->mm = new_mm;
1706                         current->active_mm = new_mm;
1707                         activate_mm(active_mm, new_mm);
1708                         new_mm = mm;
1709                 }
1710
1711                 if (new_fd) {
1712                         fd = current->files;
1713                         current->files = new_fd;
1714                         new_fd = fd;
1715                 }
1716
1717                 task_unlock(current);
1718         }
1719
1720         if (new_nsproxy)
1721                 put_nsproxy(new_nsproxy);
1722
1723 bad_unshare_cleanup_fd:
1724         if (new_fd)
1725                 put_files_struct(new_fd);
1726
1727 bad_unshare_cleanup_vm:
1728         if (new_mm)
1729                 mmput(new_mm);
1730
1731 bad_unshare_cleanup_sigh:
1732         if (new_sigh)
1733                 if (atomic_dec_and_test(&new_sigh->count))
1734                         kmem_cache_free(sighand_cachep, new_sigh);
1735
1736 bad_unshare_cleanup_fs:
1737         if (new_fs)
1738                 free_fs_struct(new_fs);
1739
1740 bad_unshare_cleanup_thread:
1741 bad_unshare_out:
1742         return err;
1743 }
1744
1745 /*
1746  *      Helper to unshare the files of the current task.
1747  *      We don't want to expose copy_files internals to
1748  *      the exec layer of the kernel.
1749  */
1750
1751 int unshare_files(struct files_struct **displaced)
1752 {
1753         struct task_struct *task = current;
1754         struct files_struct *copy = NULL;
1755         int error;
1756
1757         error = unshare_fd(CLONE_FILES, &copy);
1758         if (error || !copy) {
1759                 *displaced = NULL;
1760                 return error;
1761         }
1762         *displaced = task->files;
1763         task_lock(task);
1764         task->files = copy;
1765         task_unlock(task);
1766         return 0;
1767 }