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