Merge branch 'linus' into x86/xen
[linux-2.6] / 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/mnt_namespace.h>
21 #include <linux/personality.h>
22 #include <linux/mempolicy.h>
23 #include <linux/sem.h>
24 #include <linux/file.h>
25 #include <linux/fdtable.h>
26 #include <linux/iocontext.h>
27 #include <linux/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/futex.h>
41 #include <linux/task_io_accounting_ops.h>
42 #include <linux/rcupdate.h>
43 #include <linux/ptrace.h>
44 #include <linux/mount.h>
45 #include <linux/audit.h>
46 #include <linux/memcontrol.h>
47 #include <linux/profile.h>
48 #include <linux/rmap.h>
49 #include <linux/acct.h>
50 #include <linux/tsacct_kern.h>
51 #include <linux/cn_proc.h>
52 #include <linux/freezer.h>
53 #include <linux/delayacct.h>
54 #include <linux/taskstats_kern.h>
55 #include <linux/random.h>
56 #include <linux/tty.h>
57 #include <linux/proc_fs.h>
58 #include <linux/blkdev.h>
59
60 #include <asm/pgtable.h>
61 #include <asm/pgalloc.h>
62 #include <asm/uaccess.h>
63 #include <asm/mmu_context.h>
64 #include <asm/cacheflush.h>
65 #include <asm/tlbflush.h>
66
67 /*
68  * Protected counters by write_lock_irq(&tasklist_lock)
69  */
70 unsigned long total_forks;      /* Handle normal Linux uptimes. */
71 int nr_threads;                 /* The idle threads do not count.. */
72
73 int max_threads;                /* tunable limit on nr_threads */
74
75 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
76
77 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
78
79 int nr_processes(void)
80 {
81         int cpu;
82         int total = 0;
83
84         for_each_online_cpu(cpu)
85                 total += per_cpu(process_counts, cpu);
86
87         return total;
88 }
89
90 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
91 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
92 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
93 static struct kmem_cache *task_struct_cachep;
94 #endif
95
96 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
97 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
98 {
99 #ifdef CONFIG_DEBUG_STACK_USAGE
100         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
101 #else
102         gfp_t mask = GFP_KERNEL;
103 #endif
104         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
105 }
106
107 static inline void free_thread_info(struct thread_info *ti)
108 {
109         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
110 }
111 #endif
112
113 /* SLAB cache for signal_struct structures (tsk->signal) */
114 static struct kmem_cache *signal_cachep;
115
116 /* SLAB cache for sighand_struct structures (tsk->sighand) */
117 struct kmem_cache *sighand_cachep;
118
119 /* SLAB cache for files_struct structures (tsk->files) */
120 struct kmem_cache *files_cachep;
121
122 /* SLAB cache for fs_struct structures (tsk->fs) */
123 struct kmem_cache *fs_cachep;
124
125 /* SLAB cache for vm_area_struct structures */
126 struct kmem_cache *vm_area_cachep;
127
128 /* SLAB cache for mm_struct structures (tsk->mm) */
129 static struct kmem_cache *mm_cachep;
130
131 void free_task(struct task_struct *tsk)
132 {
133         prop_local_destroy_single(&tsk->dirties);
134         free_thread_info(tsk->stack);
135         rt_mutex_debug_task_free(tsk);
136         free_task_struct(tsk);
137 }
138 EXPORT_SYMBOL(free_task);
139
140 void __put_task_struct(struct task_struct *tsk)
141 {
142         WARN_ON(!tsk->exit_state);
143         WARN_ON(atomic_read(&tsk->usage));
144         WARN_ON(tsk == current);
145
146         security_task_free(tsk);
147         free_uid(tsk->user);
148         put_group_info(tsk->group_info);
149         delayacct_tsk_free(tsk);
150
151         if (!profile_handoff_task(tsk))
152                 free_task(tsk);
153 }
154
155 /*
156  * macro override instead of weak attribute alias, to workaround
157  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
158  */
159 #ifndef arch_task_cache_init
160 #define arch_task_cache_init()
161 #endif
162
163 void __init fork_init(unsigned long mempages)
164 {
165 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
166 #ifndef ARCH_MIN_TASKALIGN
167 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
168 #endif
169         /* create a slab on which task_structs can be allocated */
170         task_struct_cachep =
171                 kmem_cache_create("task_struct", sizeof(struct task_struct),
172                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
173 #endif
174
175         /* do the arch specific task caches init */
176         arch_task_cache_init();
177
178         /*
179          * The default maximum number of threads is set to a safe
180          * value: the thread structures can take up at most half
181          * of memory.
182          */
183         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
184
185         /*
186          * we need to allow at least 20 threads to boot a system
187          */
188         if(max_threads < 20)
189                 max_threads = 20;
190
191         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
192         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
193         init_task.signal->rlim[RLIMIT_SIGPENDING] =
194                 init_task.signal->rlim[RLIMIT_NPROC];
195 }
196
197 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
198                                                struct task_struct *src)
199 {
200         *dst = *src;
201         return 0;
202 }
203
204 static struct task_struct *dup_task_struct(struct task_struct *orig)
205 {
206         struct task_struct *tsk;
207         struct thread_info *ti;
208         int err;
209
210         prepare_to_copy(orig);
211
212         tsk = alloc_task_struct();
213         if (!tsk)
214                 return NULL;
215
216         ti = alloc_thread_info(tsk);
217         if (!ti) {
218                 free_task_struct(tsk);
219                 return NULL;
220         }
221
222         err = arch_dup_task_struct(tsk, orig);
223         if (err)
224                 goto out;
225
226         tsk->stack = ti;
227
228         err = prop_local_init_single(&tsk->dirties);
229         if (err)
230                 goto out;
231
232         setup_thread_stack(tsk, orig);
233
234 #ifdef CONFIG_CC_STACKPROTECTOR
235         tsk->stack_canary = get_random_int();
236 #endif
237
238         /* One for us, one for whoever does the "release_task()" (usually parent) */
239         atomic_set(&tsk->usage,2);
240         atomic_set(&tsk->fs_excl, 0);
241 #ifdef CONFIG_BLK_DEV_IO_TRACE
242         tsk->btrace_seq = 0;
243 #endif
244         tsk->splice_pipe = NULL;
245         return tsk;
246
247 out:
248         free_thread_info(ti);
249         free_task_struct(tsk);
250         return NULL;
251 }
252
253 #ifdef CONFIG_MMU
254 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
255 {
256         struct vm_area_struct *mpnt, *tmp, **pprev;
257         struct rb_node **rb_link, *rb_parent;
258         int retval;
259         unsigned long charge;
260         struct mempolicy *pol;
261
262         down_write(&oldmm->mmap_sem);
263         flush_cache_dup_mm(oldmm);
264         /*
265          * Not linked in yet - no deadlock potential:
266          */
267         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
268
269         mm->locked_vm = 0;
270         mm->mmap = NULL;
271         mm->mmap_cache = NULL;
272         mm->free_area_cache = oldmm->mmap_base;
273         mm->cached_hole_size = ~0UL;
274         mm->map_count = 0;
275         cpus_clear(mm->cpu_vm_mask);
276         mm->mm_rb = RB_ROOT;
277         rb_link = &mm->mm_rb.rb_node;
278         rb_parent = NULL;
279         pprev = &mm->mmap;
280
281         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
282                 struct file *file;
283
284                 if (mpnt->vm_flags & VM_DONTCOPY) {
285                         long pages = vma_pages(mpnt);
286                         mm->total_vm -= pages;
287                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
288                                                                 -pages);
289                         continue;
290                 }
291                 charge = 0;
292                 if (mpnt->vm_flags & VM_ACCOUNT) {
293                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
294                         if (security_vm_enough_memory(len))
295                                 goto fail_nomem;
296                         charge = len;
297                 }
298                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
299                 if (!tmp)
300                         goto fail_nomem;
301                 *tmp = *mpnt;
302                 pol = mpol_dup(vma_policy(mpnt));
303                 retval = PTR_ERR(pol);
304                 if (IS_ERR(pol))
305                         goto fail_nomem_policy;
306                 vma_set_policy(tmp, pol);
307                 tmp->vm_flags &= ~VM_LOCKED;
308                 tmp->vm_mm = mm;
309                 tmp->vm_next = NULL;
310                 anon_vma_link(tmp);
311                 file = tmp->vm_file;
312                 if (file) {
313                         struct inode *inode = file->f_path.dentry->d_inode;
314                         get_file(file);
315                         if (tmp->vm_flags & VM_DENYWRITE)
316                                 atomic_dec(&inode->i_writecount);
317
318                         /* insert tmp into the share list, just after mpnt */
319                         spin_lock(&file->f_mapping->i_mmap_lock);
320                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
321                         flush_dcache_mmap_lock(file->f_mapping);
322                         vma_prio_tree_add(tmp, mpnt);
323                         flush_dcache_mmap_unlock(file->f_mapping);
324                         spin_unlock(&file->f_mapping->i_mmap_lock);
325                 }
326
327                 /*
328                  * Clear hugetlb-related page reserves for children. This only
329                  * affects MAP_PRIVATE mappings. Faults generated by the child
330                  * are not guaranteed to succeed, even if read-only
331                  */
332                 if (is_vm_hugetlb_page(tmp))
333                         reset_vma_resv_huge_pages(tmp);
334
335                 /*
336                  * Link in the new vma and copy the page table entries.
337                  */
338                 *pprev = tmp;
339                 pprev = &tmp->vm_next;
340
341                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
342                 rb_link = &tmp->vm_rb.rb_right;
343                 rb_parent = &tmp->vm_rb;
344
345                 mm->map_count++;
346                 retval = copy_page_range(mm, oldmm, mpnt);
347
348                 if (tmp->vm_ops && tmp->vm_ops->open)
349                         tmp->vm_ops->open(tmp);
350
351                 if (retval)
352                         goto out;
353         }
354         /* a new mm has just been created */
355         arch_dup_mmap(oldmm, mm);
356         retval = 0;
357 out:
358         up_write(&mm->mmap_sem);
359         flush_tlb_mm(oldmm);
360         up_write(&oldmm->mmap_sem);
361         return retval;
362 fail_nomem_policy:
363         kmem_cache_free(vm_area_cachep, tmp);
364 fail_nomem:
365         retval = -ENOMEM;
366         vm_unacct_memory(charge);
367         goto out;
368 }
369
370 static inline int mm_alloc_pgd(struct mm_struct * mm)
371 {
372         mm->pgd = pgd_alloc(mm);
373         if (unlikely(!mm->pgd))
374                 return -ENOMEM;
375         return 0;
376 }
377
378 static inline void mm_free_pgd(struct mm_struct * mm)
379 {
380         pgd_free(mm, mm->pgd);
381 }
382 #else
383 #define dup_mmap(mm, oldmm)     (0)
384 #define mm_alloc_pgd(mm)        (0)
385 #define mm_free_pgd(mm)
386 #endif /* CONFIG_MMU */
387
388 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
389
390 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
391 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
392
393 #include <linux/init_task.h>
394
395 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
396 {
397         atomic_set(&mm->mm_users, 1);
398         atomic_set(&mm->mm_count, 1);
399         init_rwsem(&mm->mmap_sem);
400         INIT_LIST_HEAD(&mm->mmlist);
401         mm->flags = (current->mm) ? current->mm->flags
402                                   : MMF_DUMP_FILTER_DEFAULT;
403         mm->core_state = NULL;
404         mm->nr_ptes = 0;
405         set_mm_counter(mm, file_rss, 0);
406         set_mm_counter(mm, anon_rss, 0);
407         spin_lock_init(&mm->page_table_lock);
408         rwlock_init(&mm->ioctx_list_lock);
409         mm->ioctx_list = NULL;
410         mm->free_area_cache = TASK_UNMAPPED_BASE;
411         mm->cached_hole_size = ~0UL;
412         mm_init_owner(mm, p);
413
414         if (likely(!mm_alloc_pgd(mm))) {
415                 mm->def_flags = 0;
416                 return mm;
417         }
418
419         free_mm(mm);
420         return NULL;
421 }
422
423 /*
424  * Allocate and initialize an mm_struct.
425  */
426 struct mm_struct * mm_alloc(void)
427 {
428         struct mm_struct * mm;
429
430         mm = allocate_mm();
431         if (mm) {
432                 memset(mm, 0, sizeof(*mm));
433                 mm = mm_init(mm, current);
434         }
435         return mm;
436 }
437
438 /*
439  * Called when the last reference to the mm
440  * is dropped: either by a lazy thread or by
441  * mmput. Free the page directory and the mm.
442  */
443 void __mmdrop(struct mm_struct *mm)
444 {
445         BUG_ON(mm == &init_mm);
446         mm_free_pgd(mm);
447         destroy_context(mm);
448         free_mm(mm);
449 }
450 EXPORT_SYMBOL_GPL(__mmdrop);
451
452 /*
453  * Decrement the use count and release all resources for an mm.
454  */
455 void mmput(struct mm_struct *mm)
456 {
457         might_sleep();
458
459         if (atomic_dec_and_test(&mm->mm_users)) {
460                 exit_aio(mm);
461                 exit_mmap(mm);
462                 set_mm_exe_file(mm, NULL);
463                 if (!list_empty(&mm->mmlist)) {
464                         spin_lock(&mmlist_lock);
465                         list_del(&mm->mmlist);
466                         spin_unlock(&mmlist_lock);
467                 }
468                 put_swap_token(mm);
469                 mmdrop(mm);
470         }
471 }
472 EXPORT_SYMBOL_GPL(mmput);
473
474 /**
475  * get_task_mm - acquire a reference to the task's mm
476  *
477  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
478  * this kernel workthread has transiently adopted a user mm with use_mm,
479  * to do its AIO) is not set and if so returns a reference to it, after
480  * bumping up the use count.  User must release the mm via mmput()
481  * after use.  Typically used by /proc and ptrace.
482  */
483 struct mm_struct *get_task_mm(struct task_struct *task)
484 {
485         struct mm_struct *mm;
486
487         task_lock(task);
488         mm = task->mm;
489         if (mm) {
490                 if (task->flags & PF_KTHREAD)
491                         mm = NULL;
492                 else
493                         atomic_inc(&mm->mm_users);
494         }
495         task_unlock(task);
496         return mm;
497 }
498 EXPORT_SYMBOL_GPL(get_task_mm);
499
500 /* Please note the differences between mmput and mm_release.
501  * mmput is called whenever we stop holding onto a mm_struct,
502  * error success whatever.
503  *
504  * mm_release is called after a mm_struct has been removed
505  * from the current process.
506  *
507  * This difference is important for error handling, when we
508  * only half set up a mm_struct for a new process and need to restore
509  * the old one.  Because we mmput the new mm_struct before
510  * restoring the old one. . .
511  * Eric Biederman 10 January 1998
512  */
513 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
514 {
515         struct completion *vfork_done = tsk->vfork_done;
516
517         /* Get rid of any cached register state */
518         deactivate_mm(tsk, mm);
519
520         /* notify parent sleeping on vfork() */
521         if (vfork_done) {
522                 tsk->vfork_done = NULL;
523                 complete(vfork_done);
524         }
525
526         /*
527          * If we're exiting normally, clear a user-space tid field if
528          * requested.  We leave this alone when dying by signal, to leave
529          * the value intact in a core dump, and to save the unnecessary
530          * trouble otherwise.  Userland only wants this done for a sys_exit.
531          */
532         if (tsk->clear_child_tid
533             && !(tsk->flags & PF_SIGNALED)
534             && atomic_read(&mm->mm_users) > 1) {
535                 u32 __user * tidptr = tsk->clear_child_tid;
536                 tsk->clear_child_tid = NULL;
537
538                 /*
539                  * We don't check the error code - if userspace has
540                  * not set up a proper pointer then tough luck.
541                  */
542                 put_user(0, tidptr);
543                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
544         }
545 }
546
547 /*
548  * Allocate a new mm structure and copy contents from the
549  * mm structure of the passed in task structure.
550  */
551 struct mm_struct *dup_mm(struct task_struct *tsk)
552 {
553         struct mm_struct *mm, *oldmm = current->mm;
554         int err;
555
556         if (!oldmm)
557                 return NULL;
558
559         mm = allocate_mm();
560         if (!mm)
561                 goto fail_nomem;
562
563         memcpy(mm, oldmm, sizeof(*mm));
564
565         /* Initializing for Swap token stuff */
566         mm->token_priority = 0;
567         mm->last_interval = 0;
568
569         if (!mm_init(mm, tsk))
570                 goto fail_nomem;
571
572         if (init_new_context(tsk, mm))
573                 goto fail_nocontext;
574
575         dup_mm_exe_file(oldmm, mm);
576
577         err = dup_mmap(mm, oldmm);
578         if (err)
579                 goto free_pt;
580
581         mm->hiwater_rss = get_mm_rss(mm);
582         mm->hiwater_vm = mm->total_vm;
583
584         return mm;
585
586 free_pt:
587         mmput(mm);
588
589 fail_nomem:
590         return NULL;
591
592 fail_nocontext:
593         /*
594          * If init_new_context() failed, we cannot use mmput() to free the mm
595          * because it calls destroy_context()
596          */
597         mm_free_pgd(mm);
598         free_mm(mm);
599         return NULL;
600 }
601
602 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
603 {
604         struct mm_struct * mm, *oldmm;
605         int retval;
606
607         tsk->min_flt = tsk->maj_flt = 0;
608         tsk->nvcsw = tsk->nivcsw = 0;
609
610         tsk->mm = NULL;
611         tsk->active_mm = NULL;
612
613         /*
614          * Are we cloning a kernel thread?
615          *
616          * We need to steal a active VM for that..
617          */
618         oldmm = current->mm;
619         if (!oldmm)
620                 return 0;
621
622         if (clone_flags & CLONE_VM) {
623                 atomic_inc(&oldmm->mm_users);
624                 mm = oldmm;
625                 goto good_mm;
626         }
627
628         retval = -ENOMEM;
629         mm = dup_mm(tsk);
630         if (!mm)
631                 goto fail_nomem;
632
633 good_mm:
634         /* Initializing for Swap token stuff */
635         mm->token_priority = 0;
636         mm->last_interval = 0;
637
638         tsk->mm = mm;
639         tsk->active_mm = mm;
640         return 0;
641
642 fail_nomem:
643         return retval;
644 }
645
646 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
647 {
648         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
649         /* We don't need to lock fs - think why ;-) */
650         if (fs) {
651                 atomic_set(&fs->count, 1);
652                 rwlock_init(&fs->lock);
653                 fs->umask = old->umask;
654                 read_lock(&old->lock);
655                 fs->root = old->root;
656                 path_get(&old->root);
657                 fs->pwd = old->pwd;
658                 path_get(&old->pwd);
659                 if (old->altroot.dentry) {
660                         fs->altroot = old->altroot;
661                         path_get(&old->altroot);
662                 } else {
663                         fs->altroot.mnt = NULL;
664                         fs->altroot.dentry = NULL;
665                 }
666                 read_unlock(&old->lock);
667         }
668         return fs;
669 }
670
671 struct fs_struct *copy_fs_struct(struct fs_struct *old)
672 {
673         return __copy_fs_struct(old);
674 }
675
676 EXPORT_SYMBOL_GPL(copy_fs_struct);
677
678 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
679 {
680         if (clone_flags & CLONE_FS) {
681                 atomic_inc(&current->fs->count);
682                 return 0;
683         }
684         tsk->fs = __copy_fs_struct(current->fs);
685         if (!tsk->fs)
686                 return -ENOMEM;
687         return 0;
688 }
689
690 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
691 {
692         struct files_struct *oldf, *newf;
693         int error = 0;
694
695         /*
696          * A background process may not have any files ...
697          */
698         oldf = current->files;
699         if (!oldf)
700                 goto out;
701
702         if (clone_flags & CLONE_FILES) {
703                 atomic_inc(&oldf->count);
704                 goto out;
705         }
706
707         newf = dup_fd(oldf, &error);
708         if (!newf)
709                 goto out;
710
711         tsk->files = newf;
712         error = 0;
713 out:
714         return error;
715 }
716
717 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
718 {
719 #ifdef CONFIG_BLOCK
720         struct io_context *ioc = current->io_context;
721
722         if (!ioc)
723                 return 0;
724         /*
725          * Share io context with parent, if CLONE_IO is set
726          */
727         if (clone_flags & CLONE_IO) {
728                 tsk->io_context = ioc_task_link(ioc);
729                 if (unlikely(!tsk->io_context))
730                         return -ENOMEM;
731         } else if (ioprio_valid(ioc->ioprio)) {
732                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
733                 if (unlikely(!tsk->io_context))
734                         return -ENOMEM;
735
736                 tsk->io_context->ioprio = ioc->ioprio;
737         }
738 #endif
739         return 0;
740 }
741
742 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
743 {
744         struct sighand_struct *sig;
745
746         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
747                 atomic_inc(&current->sighand->count);
748                 return 0;
749         }
750         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
751         rcu_assign_pointer(tsk->sighand, sig);
752         if (!sig)
753                 return -ENOMEM;
754         atomic_set(&sig->count, 1);
755         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
756         return 0;
757 }
758
759 void __cleanup_sighand(struct sighand_struct *sighand)
760 {
761         if (atomic_dec_and_test(&sighand->count))
762                 kmem_cache_free(sighand_cachep, sighand);
763 }
764
765 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
766 {
767         struct signal_struct *sig;
768         int ret;
769
770         if (clone_flags & CLONE_THREAD) {
771                 atomic_inc(&current->signal->count);
772                 atomic_inc(&current->signal->live);
773                 return 0;
774         }
775         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
776         tsk->signal = sig;
777         if (!sig)
778                 return -ENOMEM;
779
780         ret = copy_thread_group_keys(tsk);
781         if (ret < 0) {
782                 kmem_cache_free(signal_cachep, sig);
783                 return ret;
784         }
785
786         atomic_set(&sig->count, 1);
787         atomic_set(&sig->live, 1);
788         init_waitqueue_head(&sig->wait_chldexit);
789         sig->flags = 0;
790         sig->group_exit_code = 0;
791         sig->group_exit_task = NULL;
792         sig->group_stop_count = 0;
793         sig->curr_target = tsk;
794         init_sigpending(&sig->shared_pending);
795         INIT_LIST_HEAD(&sig->posix_timers);
796
797         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
798         sig->it_real_incr.tv64 = 0;
799         sig->real_timer.function = it_real_fn;
800
801         sig->it_virt_expires = cputime_zero;
802         sig->it_virt_incr = cputime_zero;
803         sig->it_prof_expires = cputime_zero;
804         sig->it_prof_incr = cputime_zero;
805
806         sig->leader = 0;        /* session leadership doesn't inherit */
807         sig->tty_old_pgrp = NULL;
808
809         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
810         sig->gtime = cputime_zero;
811         sig->cgtime = cputime_zero;
812         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
813         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
814         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
815 #ifdef CONFIG_TASK_XACCT
816         sig->rchar = sig->wchar = sig->syscr = sig->syscw = 0;
817 #endif
818 #ifdef CONFIG_TASK_IO_ACCOUNTING
819         memset(&sig->ioac, 0, sizeof(sig->ioac));
820 #endif
821         sig->sum_sched_runtime = 0;
822         INIT_LIST_HEAD(&sig->cpu_timers[0]);
823         INIT_LIST_HEAD(&sig->cpu_timers[1]);
824         INIT_LIST_HEAD(&sig->cpu_timers[2]);
825         taskstats_tgid_init(sig);
826
827         task_lock(current->group_leader);
828         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
829         task_unlock(current->group_leader);
830
831         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
832                 /*
833                  * New sole thread in the process gets an expiry time
834                  * of the whole CPU time limit.
835                  */
836                 tsk->it_prof_expires =
837                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
838         }
839         acct_init_pacct(&sig->pacct);
840
841         tty_audit_fork(sig);
842
843         return 0;
844 }
845
846 void __cleanup_signal(struct signal_struct *sig)
847 {
848         exit_thread_group_keys(sig);
849         kmem_cache_free(signal_cachep, sig);
850 }
851
852 static void cleanup_signal(struct task_struct *tsk)
853 {
854         struct signal_struct *sig = tsk->signal;
855
856         atomic_dec(&sig->live);
857
858         if (atomic_dec_and_test(&sig->count))
859                 __cleanup_signal(sig);
860 }
861
862 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
863 {
864         unsigned long new_flags = p->flags;
865
866         new_flags &= ~PF_SUPERPRIV;
867         new_flags |= PF_FORKNOEXEC;
868         if (!(clone_flags & CLONE_PTRACE))
869                 p->ptrace = 0;
870         p->flags = new_flags;
871         clear_freeze_flag(p);
872 }
873
874 asmlinkage long sys_set_tid_address(int __user *tidptr)
875 {
876         current->clear_child_tid = tidptr;
877
878         return task_pid_vnr(current);
879 }
880
881 static void rt_mutex_init_task(struct task_struct *p)
882 {
883         spin_lock_init(&p->pi_lock);
884 #ifdef CONFIG_RT_MUTEXES
885         plist_head_init(&p->pi_waiters, &p->pi_lock);
886         p->pi_blocked_on = NULL;
887 #endif
888 }
889
890 #ifdef CONFIG_MM_OWNER
891 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
892 {
893         mm->owner = p;
894 }
895 #endif /* CONFIG_MM_OWNER */
896
897 /*
898  * This creates a new process as a copy of the old one,
899  * but does not actually start it yet.
900  *
901  * It copies the registers, and all the appropriate
902  * parts of the process environment (as per the clone
903  * flags). The actual kick-off is left to the caller.
904  */
905 static struct task_struct *copy_process(unsigned long clone_flags,
906                                         unsigned long stack_start,
907                                         struct pt_regs *regs,
908                                         unsigned long stack_size,
909                                         int __user *child_tidptr,
910                                         struct pid *pid)
911 {
912         int retval;
913         struct task_struct *p;
914         int cgroup_callbacks_done = 0;
915
916         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
917                 return ERR_PTR(-EINVAL);
918
919         /*
920          * Thread groups must share signals as well, and detached threads
921          * can only be started up within the thread group.
922          */
923         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
924                 return ERR_PTR(-EINVAL);
925
926         /*
927          * Shared signal handlers imply shared VM. By way of the above,
928          * thread groups also imply shared VM. Blocking this case allows
929          * for various simplifications in other code.
930          */
931         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
932                 return ERR_PTR(-EINVAL);
933
934         retval = security_task_create(clone_flags);
935         if (retval)
936                 goto fork_out;
937
938         retval = -ENOMEM;
939         p = dup_task_struct(current);
940         if (!p)
941                 goto fork_out;
942
943         rt_mutex_init_task(p);
944
945 #ifdef CONFIG_PROVE_LOCKING
946         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
947         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
948 #endif
949         retval = -EAGAIN;
950         if (atomic_read(&p->user->processes) >=
951                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
952                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
953                     p->user != current->nsproxy->user_ns->root_user)
954                         goto bad_fork_free;
955         }
956
957         atomic_inc(&p->user->__count);
958         atomic_inc(&p->user->processes);
959         get_group_info(p->group_info);
960
961         /*
962          * If multiple threads are within copy_process(), then this check
963          * triggers too late. This doesn't hurt, the check is only there
964          * to stop root fork bombs.
965          */
966         if (nr_threads >= max_threads)
967                 goto bad_fork_cleanup_count;
968
969         if (!try_module_get(task_thread_info(p)->exec_domain->module))
970                 goto bad_fork_cleanup_count;
971
972         if (p->binfmt && !try_module_get(p->binfmt->module))
973                 goto bad_fork_cleanup_put_domain;
974
975         p->did_exec = 0;
976         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
977         copy_flags(clone_flags, p);
978         INIT_LIST_HEAD(&p->children);
979         INIT_LIST_HEAD(&p->sibling);
980 #ifdef CONFIG_PREEMPT_RCU
981         p->rcu_read_lock_nesting = 0;
982         p->rcu_flipctr_idx = 0;
983 #endif /* #ifdef CONFIG_PREEMPT_RCU */
984         p->vfork_done = NULL;
985         spin_lock_init(&p->alloc_lock);
986
987         clear_tsk_thread_flag(p, TIF_SIGPENDING);
988         init_sigpending(&p->pending);
989
990         p->utime = cputime_zero;
991         p->stime = cputime_zero;
992         p->gtime = cputime_zero;
993         p->utimescaled = cputime_zero;
994         p->stimescaled = cputime_zero;
995         p->prev_utime = cputime_zero;
996         p->prev_stime = cputime_zero;
997
998 #ifdef CONFIG_DETECT_SOFTLOCKUP
999         p->last_switch_count = 0;
1000         p->last_switch_timestamp = 0;
1001 #endif
1002
1003 #ifdef CONFIG_TASK_XACCT
1004         p->rchar = 0;           /* I/O counter: bytes read */
1005         p->wchar = 0;           /* I/O counter: bytes written */
1006         p->syscr = 0;           /* I/O counter: read syscalls */
1007         p->syscw = 0;           /* I/O counter: write syscalls */
1008 #endif
1009         task_io_accounting_init(p);
1010         acct_clear_integrals(p);
1011
1012         p->it_virt_expires = cputime_zero;
1013         p->it_prof_expires = cputime_zero;
1014         p->it_sched_expires = 0;
1015         INIT_LIST_HEAD(&p->cpu_timers[0]);
1016         INIT_LIST_HEAD(&p->cpu_timers[1]);
1017         INIT_LIST_HEAD(&p->cpu_timers[2]);
1018
1019         p->lock_depth = -1;             /* -1 = no lock */
1020         do_posix_clock_monotonic_gettime(&p->start_time);
1021         p->real_start_time = p->start_time;
1022         monotonic_to_bootbased(&p->real_start_time);
1023 #ifdef CONFIG_SECURITY
1024         p->security = NULL;
1025 #endif
1026         p->cap_bset = current->cap_bset;
1027         p->io_context = NULL;
1028         p->audit_context = NULL;
1029         cgroup_fork(p);
1030 #ifdef CONFIG_NUMA
1031         p->mempolicy = mpol_dup(p->mempolicy);
1032         if (IS_ERR(p->mempolicy)) {
1033                 retval = PTR_ERR(p->mempolicy);
1034                 p->mempolicy = NULL;
1035                 goto bad_fork_cleanup_cgroup;
1036         }
1037         mpol_fix_fork_child_flag(p);
1038 #endif
1039 #ifdef CONFIG_TRACE_IRQFLAGS
1040         p->irq_events = 0;
1041 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1042         p->hardirqs_enabled = 1;
1043 #else
1044         p->hardirqs_enabled = 0;
1045 #endif
1046         p->hardirq_enable_ip = 0;
1047         p->hardirq_enable_event = 0;
1048         p->hardirq_disable_ip = _THIS_IP_;
1049         p->hardirq_disable_event = 0;
1050         p->softirqs_enabled = 1;
1051         p->softirq_enable_ip = _THIS_IP_;
1052         p->softirq_enable_event = 0;
1053         p->softirq_disable_ip = 0;
1054         p->softirq_disable_event = 0;
1055         p->hardirq_context = 0;
1056         p->softirq_context = 0;
1057 #endif
1058 #ifdef CONFIG_LOCKDEP
1059         p->lockdep_depth = 0; /* no locks held yet */
1060         p->curr_chain_key = 0;
1061         p->lockdep_recursion = 0;
1062 #endif
1063
1064 #ifdef CONFIG_DEBUG_MUTEXES
1065         p->blocked_on = NULL; /* not blocked yet */
1066 #endif
1067
1068         /* Perform scheduler related setup. Assign this task to a CPU. */
1069         sched_fork(p, clone_flags);
1070
1071         if ((retval = security_task_alloc(p)))
1072                 goto bad_fork_cleanup_policy;
1073         if ((retval = audit_alloc(p)))
1074                 goto bad_fork_cleanup_security;
1075         /* copy all the process information */
1076         if ((retval = copy_semundo(clone_flags, p)))
1077                 goto bad_fork_cleanup_audit;
1078         if ((retval = copy_files(clone_flags, p)))
1079                 goto bad_fork_cleanup_semundo;
1080         if ((retval = copy_fs(clone_flags, p)))
1081                 goto bad_fork_cleanup_files;
1082         if ((retval = copy_sighand(clone_flags, p)))
1083                 goto bad_fork_cleanup_fs;
1084         if ((retval = copy_signal(clone_flags, p)))
1085                 goto bad_fork_cleanup_sighand;
1086         if ((retval = copy_mm(clone_flags, p)))
1087                 goto bad_fork_cleanup_signal;
1088         if ((retval = copy_keys(clone_flags, p)))
1089                 goto bad_fork_cleanup_mm;
1090         if ((retval = copy_namespaces(clone_flags, p)))
1091                 goto bad_fork_cleanup_keys;
1092         if ((retval = copy_io(clone_flags, p)))
1093                 goto bad_fork_cleanup_namespaces;
1094         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1095         if (retval)
1096                 goto bad_fork_cleanup_io;
1097
1098         if (pid != &init_struct_pid) {
1099                 retval = -ENOMEM;
1100                 pid = alloc_pid(task_active_pid_ns(p));
1101                 if (!pid)
1102                         goto bad_fork_cleanup_io;
1103
1104                 if (clone_flags & CLONE_NEWPID) {
1105                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1106                         if (retval < 0)
1107                                 goto bad_fork_free_pid;
1108                 }
1109         }
1110
1111         p->pid = pid_nr(pid);
1112         p->tgid = p->pid;
1113         if (clone_flags & CLONE_THREAD)
1114                 p->tgid = current->tgid;
1115
1116         if (current->nsproxy != p->nsproxy) {
1117                 retval = ns_cgroup_clone(p, pid);
1118                 if (retval)
1119                         goto bad_fork_free_pid;
1120         }
1121
1122         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1123         /*
1124          * Clear TID on mm_release()?
1125          */
1126         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1127 #ifdef CONFIG_FUTEX
1128         p->robust_list = NULL;
1129 #ifdef CONFIG_COMPAT
1130         p->compat_robust_list = NULL;
1131 #endif
1132         INIT_LIST_HEAD(&p->pi_state_list);
1133         p->pi_state_cache = NULL;
1134 #endif
1135         /*
1136          * sigaltstack should be cleared when sharing the same VM
1137          */
1138         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1139                 p->sas_ss_sp = p->sas_ss_size = 0;
1140
1141         /*
1142          * Syscall tracing should be turned off in the child regardless
1143          * of CLONE_PTRACE.
1144          */
1145         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1146 #ifdef TIF_SYSCALL_EMU
1147         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1148 #endif
1149         clear_all_latency_tracing(p);
1150
1151         /* Our parent execution domain becomes current domain
1152            These must match for thread signalling to apply */
1153         p->parent_exec_id = p->self_exec_id;
1154
1155         /* ok, now we should be set up.. */
1156         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1157         p->pdeath_signal = 0;
1158         p->exit_state = 0;
1159
1160         /*
1161          * Ok, make it visible to the rest of the system.
1162          * We dont wake it up yet.
1163          */
1164         p->group_leader = p;
1165         INIT_LIST_HEAD(&p->thread_group);
1166         INIT_LIST_HEAD(&p->ptrace_entry);
1167         INIT_LIST_HEAD(&p->ptraced);
1168
1169         /* Now that the task is set up, run cgroup callbacks if
1170          * necessary. We need to run them before the task is visible
1171          * on the tasklist. */
1172         cgroup_fork_callbacks(p);
1173         cgroup_callbacks_done = 1;
1174
1175         /* Need tasklist lock for parent etc handling! */
1176         write_lock_irq(&tasklist_lock);
1177
1178         /*
1179          * The task hasn't been attached yet, so its cpus_allowed mask will
1180          * not be changed, nor will its assigned CPU.
1181          *
1182          * The cpus_allowed mask of the parent may have changed after it was
1183          * copied first time - so re-copy it here, then check the child's CPU
1184          * to ensure it is on a valid CPU (and if not, just force it back to
1185          * parent's CPU). This avoids alot of nasty races.
1186          */
1187         p->cpus_allowed = current->cpus_allowed;
1188         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1189         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1190                         !cpu_online(task_cpu(p))))
1191                 set_task_cpu(p, smp_processor_id());
1192
1193         /* CLONE_PARENT re-uses the old parent */
1194         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1195                 p->real_parent = current->real_parent;
1196         else
1197                 p->real_parent = current;
1198         p->parent = p->real_parent;
1199
1200         spin_lock(&current->sighand->siglock);
1201
1202         /*
1203          * Process group and session signals need to be delivered to just the
1204          * parent before the fork or both the parent and the child after the
1205          * fork. Restart if a signal comes in before we add the new process to
1206          * it's process group.
1207          * A fatal signal pending means that current will exit, so the new
1208          * thread can't slip out of an OOM kill (or normal SIGKILL).
1209          */
1210         recalc_sigpending();
1211         if (signal_pending(current)) {
1212                 spin_unlock(&current->sighand->siglock);
1213                 write_unlock_irq(&tasklist_lock);
1214                 retval = -ERESTARTNOINTR;
1215                 goto bad_fork_free_pid;
1216         }
1217
1218         if (clone_flags & CLONE_THREAD) {
1219                 p->group_leader = current->group_leader;
1220                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1221
1222                 if (!cputime_eq(current->signal->it_virt_expires,
1223                                 cputime_zero) ||
1224                     !cputime_eq(current->signal->it_prof_expires,
1225                                 cputime_zero) ||
1226                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1227                     !list_empty(&current->signal->cpu_timers[0]) ||
1228                     !list_empty(&current->signal->cpu_timers[1]) ||
1229                     !list_empty(&current->signal->cpu_timers[2])) {
1230                         /*
1231                          * Have child wake up on its first tick to check
1232                          * for process CPU timers.
1233                          */
1234                         p->it_prof_expires = jiffies_to_cputime(1);
1235                 }
1236         }
1237
1238         if (likely(p->pid)) {
1239                 list_add_tail(&p->sibling, &p->real_parent->children);
1240                 if (unlikely(p->ptrace & PT_PTRACED))
1241                         __ptrace_link(p, current->parent);
1242
1243                 if (thread_group_leader(p)) {
1244                         if (clone_flags & CLONE_NEWPID)
1245                                 p->nsproxy->pid_ns->child_reaper = p;
1246
1247                         p->signal->leader_pid = pid;
1248                         p->signal->tty = current->signal->tty;
1249                         set_task_pgrp(p, task_pgrp_nr(current));
1250                         set_task_session(p, task_session_nr(current));
1251                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1252                         attach_pid(p, PIDTYPE_SID, task_session(current));
1253                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1254                         __get_cpu_var(process_counts)++;
1255                 }
1256                 attach_pid(p, PIDTYPE_PID, pid);
1257                 nr_threads++;
1258         }
1259
1260         total_forks++;
1261         spin_unlock(&current->sighand->siglock);
1262         write_unlock_irq(&tasklist_lock);
1263         proc_fork_connector(p);
1264         cgroup_post_fork(p);
1265         return p;
1266
1267 bad_fork_free_pid:
1268         if (pid != &init_struct_pid)
1269                 free_pid(pid);
1270 bad_fork_cleanup_io:
1271         put_io_context(p->io_context);
1272 bad_fork_cleanup_namespaces:
1273         exit_task_namespaces(p);
1274 bad_fork_cleanup_keys:
1275         exit_keys(p);
1276 bad_fork_cleanup_mm:
1277         if (p->mm)
1278                 mmput(p->mm);
1279 bad_fork_cleanup_signal:
1280         cleanup_signal(p);
1281 bad_fork_cleanup_sighand:
1282         __cleanup_sighand(p->sighand);
1283 bad_fork_cleanup_fs:
1284         exit_fs(p); /* blocking */
1285 bad_fork_cleanup_files:
1286         exit_files(p); /* blocking */
1287 bad_fork_cleanup_semundo:
1288         exit_sem(p);
1289 bad_fork_cleanup_audit:
1290         audit_free(p);
1291 bad_fork_cleanup_security:
1292         security_task_free(p);
1293 bad_fork_cleanup_policy:
1294 #ifdef CONFIG_NUMA
1295         mpol_put(p->mempolicy);
1296 bad_fork_cleanup_cgroup:
1297 #endif
1298         cgroup_exit(p, cgroup_callbacks_done);
1299         delayacct_tsk_free(p);
1300         if (p->binfmt)
1301                 module_put(p->binfmt->module);
1302 bad_fork_cleanup_put_domain:
1303         module_put(task_thread_info(p)->exec_domain->module);
1304 bad_fork_cleanup_count:
1305         put_group_info(p->group_info);
1306         atomic_dec(&p->user->processes);
1307         free_uid(p->user);
1308 bad_fork_free:
1309         free_task(p);
1310 fork_out:
1311         return ERR_PTR(retval);
1312 }
1313
1314 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1315 {
1316         memset(regs, 0, sizeof(struct pt_regs));
1317         return regs;
1318 }
1319
1320 struct task_struct * __cpuinit fork_idle(int cpu)
1321 {
1322         struct task_struct *task;
1323         struct pt_regs regs;
1324
1325         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1326                                 &init_struct_pid);
1327         if (!IS_ERR(task))
1328                 init_idle(task, cpu);
1329
1330         return task;
1331 }
1332
1333 static int fork_traceflag(unsigned clone_flags)
1334 {
1335         if (clone_flags & CLONE_UNTRACED)
1336                 return 0;
1337         else if (clone_flags & CLONE_VFORK) {
1338                 if (current->ptrace & PT_TRACE_VFORK)
1339                         return PTRACE_EVENT_VFORK;
1340         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1341                 if (current->ptrace & PT_TRACE_CLONE)
1342                         return PTRACE_EVENT_CLONE;
1343         } else if (current->ptrace & PT_TRACE_FORK)
1344                 return PTRACE_EVENT_FORK;
1345
1346         return 0;
1347 }
1348
1349 /*
1350  *  Ok, this is the main fork-routine.
1351  *
1352  * It copies the process, and if successful kick-starts
1353  * it and waits for it to finish using the VM if required.
1354  */
1355 long do_fork(unsigned long clone_flags,
1356               unsigned long stack_start,
1357               struct pt_regs *regs,
1358               unsigned long stack_size,
1359               int __user *parent_tidptr,
1360               int __user *child_tidptr)
1361 {
1362         struct task_struct *p;
1363         int trace = 0;
1364         long nr;
1365
1366         /*
1367          * We hope to recycle these flags after 2.6.26
1368          */
1369         if (unlikely(clone_flags & CLONE_STOPPED)) {
1370                 static int __read_mostly count = 100;
1371
1372                 if (count > 0 && printk_ratelimit()) {
1373                         char comm[TASK_COMM_LEN];
1374
1375                         count--;
1376                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1377                                         "clone flags 0x%lx\n",
1378                                 get_task_comm(comm, current),
1379                                 clone_flags & CLONE_STOPPED);
1380                 }
1381         }
1382
1383         if (unlikely(current->ptrace)) {
1384                 trace = fork_traceflag (clone_flags);
1385                 if (trace)
1386                         clone_flags |= CLONE_PTRACE;
1387         }
1388
1389         p = copy_process(clone_flags, stack_start, regs, stack_size,
1390                         child_tidptr, NULL);
1391         /*
1392          * Do this prior waking up the new thread - the thread pointer
1393          * might get invalid after that point, if the thread exits quickly.
1394          */
1395         if (!IS_ERR(p)) {
1396                 struct completion vfork;
1397
1398                 nr = task_pid_vnr(p);
1399
1400                 if (clone_flags & CLONE_PARENT_SETTID)
1401                         put_user(nr, parent_tidptr);
1402
1403                 if (clone_flags & CLONE_VFORK) {
1404                         p->vfork_done = &vfork;
1405                         init_completion(&vfork);
1406                 }
1407
1408                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1409                         /*
1410                          * We'll start up with an immediate SIGSTOP.
1411                          */
1412                         sigaddset(&p->pending.signal, SIGSTOP);
1413                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1414                 }
1415
1416                 if (!(clone_flags & CLONE_STOPPED))
1417                         wake_up_new_task(p, clone_flags);
1418                 else
1419                         __set_task_state(p, TASK_STOPPED);
1420
1421                 if (unlikely (trace)) {
1422                         current->ptrace_message = nr;
1423                         ptrace_notify ((trace << 8) | SIGTRAP);
1424                 }
1425
1426                 if (clone_flags & CLONE_VFORK) {
1427                         freezer_do_not_count();
1428                         wait_for_completion(&vfork);
1429                         freezer_count();
1430                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1431                                 current->ptrace_message = nr;
1432                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1433                         }
1434                 }
1435         } else {
1436                 nr = PTR_ERR(p);
1437         }
1438         return nr;
1439 }
1440
1441 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1442 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1443 #endif
1444
1445 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1446 {
1447         struct sighand_struct *sighand = data;
1448
1449         spin_lock_init(&sighand->siglock);
1450         init_waitqueue_head(&sighand->signalfd_wqh);
1451 }
1452
1453 void __init proc_caches_init(void)
1454 {
1455         sighand_cachep = kmem_cache_create("sighand_cache",
1456                         sizeof(struct sighand_struct), 0,
1457                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1458                         sighand_ctor);
1459         signal_cachep = kmem_cache_create("signal_cache",
1460                         sizeof(struct signal_struct), 0,
1461                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1462         files_cachep = kmem_cache_create("files_cache",
1463                         sizeof(struct files_struct), 0,
1464                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1465         fs_cachep = kmem_cache_create("fs_cache",
1466                         sizeof(struct fs_struct), 0,
1467                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1468         vm_area_cachep = kmem_cache_create("vm_area_struct",
1469                         sizeof(struct vm_area_struct), 0,
1470                         SLAB_PANIC, NULL);
1471         mm_cachep = kmem_cache_create("mm_struct",
1472                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1473                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1474 }
1475
1476 /*
1477  * Check constraints on flags passed to the unshare system call and
1478  * force unsharing of additional process context as appropriate.
1479  */
1480 static void check_unshare_flags(unsigned long *flags_ptr)
1481 {
1482         /*
1483          * If unsharing a thread from a thread group, must also
1484          * unshare vm.
1485          */
1486         if (*flags_ptr & CLONE_THREAD)
1487                 *flags_ptr |= CLONE_VM;
1488
1489         /*
1490          * If unsharing vm, must also unshare signal handlers.
1491          */
1492         if (*flags_ptr & CLONE_VM)
1493                 *flags_ptr |= CLONE_SIGHAND;
1494
1495         /*
1496          * If unsharing signal handlers and the task was created
1497          * using CLONE_THREAD, then must unshare the thread
1498          */
1499         if ((*flags_ptr & CLONE_SIGHAND) &&
1500             (atomic_read(&current->signal->count) > 1))
1501                 *flags_ptr |= CLONE_THREAD;
1502
1503         /*
1504          * If unsharing namespace, must also unshare filesystem information.
1505          */
1506         if (*flags_ptr & CLONE_NEWNS)
1507                 *flags_ptr |= CLONE_FS;
1508 }
1509
1510 /*
1511  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1512  */
1513 static int unshare_thread(unsigned long unshare_flags)
1514 {
1515         if (unshare_flags & CLONE_THREAD)
1516                 return -EINVAL;
1517
1518         return 0;
1519 }
1520
1521 /*
1522  * Unshare the filesystem structure if it is being shared
1523  */
1524 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1525 {
1526         struct fs_struct *fs = current->fs;
1527
1528         if ((unshare_flags & CLONE_FS) &&
1529             (fs && atomic_read(&fs->count) > 1)) {
1530                 *new_fsp = __copy_fs_struct(current->fs);
1531                 if (!*new_fsp)
1532                         return -ENOMEM;
1533         }
1534
1535         return 0;
1536 }
1537
1538 /*
1539  * Unsharing of sighand is not supported yet
1540  */
1541 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1542 {
1543         struct sighand_struct *sigh = current->sighand;
1544
1545         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1546                 return -EINVAL;
1547         else
1548                 return 0;
1549 }
1550
1551 /*
1552  * Unshare vm if it is being shared
1553  */
1554 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1555 {
1556         struct mm_struct *mm = current->mm;
1557
1558         if ((unshare_flags & CLONE_VM) &&
1559             (mm && atomic_read(&mm->mm_users) > 1)) {
1560                 return -EINVAL;
1561         }
1562
1563         return 0;
1564 }
1565
1566 /*
1567  * Unshare file descriptor table if it is being shared
1568  */
1569 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1570 {
1571         struct files_struct *fd = current->files;
1572         int error = 0;
1573
1574         if ((unshare_flags & CLONE_FILES) &&
1575             (fd && atomic_read(&fd->count) > 1)) {
1576                 *new_fdp = dup_fd(fd, &error);
1577                 if (!*new_fdp)
1578                         return error;
1579         }
1580
1581         return 0;
1582 }
1583
1584 /*
1585  * unshare allows a process to 'unshare' part of the process
1586  * context which was originally shared using clone.  copy_*
1587  * functions used by do_fork() cannot be used here directly
1588  * because they modify an inactive task_struct that is being
1589  * constructed. Here we are modifying the current, active,
1590  * task_struct.
1591  */
1592 asmlinkage long sys_unshare(unsigned long unshare_flags)
1593 {
1594         int err = 0;
1595         struct fs_struct *fs, *new_fs = NULL;
1596         struct sighand_struct *new_sigh = NULL;
1597         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1598         struct files_struct *fd, *new_fd = NULL;
1599         struct nsproxy *new_nsproxy = NULL;
1600         int do_sysvsem = 0;
1601
1602         check_unshare_flags(&unshare_flags);
1603
1604         /* Return -EINVAL for all unsupported flags */
1605         err = -EINVAL;
1606         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1607                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1608                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1609                                 CLONE_NEWNET))
1610                 goto bad_unshare_out;
1611
1612         /*
1613          * CLONE_NEWIPC must also detach from the undolist: after switching
1614          * to a new ipc namespace, the semaphore arrays from the old
1615          * namespace are unreachable.
1616          */
1617         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1618                 do_sysvsem = 1;
1619         if ((err = unshare_thread(unshare_flags)))
1620                 goto bad_unshare_out;
1621         if ((err = unshare_fs(unshare_flags, &new_fs)))
1622                 goto bad_unshare_cleanup_thread;
1623         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1624                 goto bad_unshare_cleanup_fs;
1625         if ((err = unshare_vm(unshare_flags, &new_mm)))
1626                 goto bad_unshare_cleanup_sigh;
1627         if ((err = unshare_fd(unshare_flags, &new_fd)))
1628                 goto bad_unshare_cleanup_vm;
1629         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1630                         new_fs)))
1631                 goto bad_unshare_cleanup_fd;
1632
1633         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1634                 if (do_sysvsem) {
1635                         /*
1636                          * CLONE_SYSVSEM is equivalent to sys_exit().
1637                          */
1638                         exit_sem(current);
1639                 }
1640
1641                 if (new_nsproxy) {
1642                         switch_task_namespaces(current, new_nsproxy);
1643                         new_nsproxy = NULL;
1644                 }
1645
1646                 task_lock(current);
1647
1648                 if (new_fs) {
1649                         fs = current->fs;
1650                         current->fs = new_fs;
1651                         new_fs = fs;
1652                 }
1653
1654                 if (new_mm) {
1655                         mm = current->mm;
1656                         active_mm = current->active_mm;
1657                         current->mm = new_mm;
1658                         current->active_mm = new_mm;
1659                         activate_mm(active_mm, new_mm);
1660                         new_mm = mm;
1661                 }
1662
1663                 if (new_fd) {
1664                         fd = current->files;
1665                         current->files = new_fd;
1666                         new_fd = fd;
1667                 }
1668
1669                 task_unlock(current);
1670         }
1671
1672         if (new_nsproxy)
1673                 put_nsproxy(new_nsproxy);
1674
1675 bad_unshare_cleanup_fd:
1676         if (new_fd)
1677                 put_files_struct(new_fd);
1678
1679 bad_unshare_cleanup_vm:
1680         if (new_mm)
1681                 mmput(new_mm);
1682
1683 bad_unshare_cleanup_sigh:
1684         if (new_sigh)
1685                 if (atomic_dec_and_test(&new_sigh->count))
1686                         kmem_cache_free(sighand_cachep, new_sigh);
1687
1688 bad_unshare_cleanup_fs:
1689         if (new_fs)
1690                 put_fs_struct(new_fs);
1691
1692 bad_unshare_cleanup_thread:
1693 bad_unshare_out:
1694         return err;
1695 }
1696
1697 /*
1698  *      Helper to unshare the files of the current task.
1699  *      We don't want to expose copy_files internals to
1700  *      the exec layer of the kernel.
1701  */
1702
1703 int unshare_files(struct files_struct **displaced)
1704 {
1705         struct task_struct *task = current;
1706         struct files_struct *copy = NULL;
1707         int error;
1708
1709         error = unshare_fd(CLONE_FILES, &copy);
1710         if (error || !copy) {
1711                 *displaced = NULL;
1712                 return error;
1713         }
1714         *displaced = task->files;
1715         task_lock(task);
1716         task->files = copy;
1717         task_unlock(task);
1718         return 0;
1719 }