futex: Move exit_pi_state() call to release_mm()
[linux-2.6.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/tracehook.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/task_io_accounting_ops.h>
44 #include <linux/rcupdate.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/audit.h>
48 #include <linux/memcontrol.h>
49 #include <linux/ftrace.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/acct.h>
53 #include <linux/tsacct_kern.h>
54 #include <linux/cn_proc.h>
55 #include <linux/freezer.h>
56 #include <linux/delayacct.h>
57 #include <linux/taskstats_kern.h>
58 #include <linux/random.h>
59 #include <linux/tty.h>
60 #include <linux/proc_fs.h>
61 #include <linux/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_counter.h>
65
66 #include <asm/pgtable.h>
67 #include <asm/pgalloc.h>
68 #include <asm/uaccess.h>
69 #include <asm/mmu_context.h>
70 #include <asm/cacheflush.h>
71 #include <asm/tlbflush.h>
72
73 #include <trace/events/sched.h>
74
75 /*
76  * Protected counters by write_lock_irq(&tasklist_lock)
77  */
78 unsigned long total_forks;      /* Handle normal Linux uptimes. */
79 int nr_threads;                 /* The idle threads do not count.. */
80
81 int max_threads;                /* tunable limit on nr_threads */
82
83 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
84
85 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
86
87 int nr_processes(void)
88 {
89         int cpu;
90         int total = 0;
91
92         for_each_online_cpu(cpu)
93                 total += per_cpu(process_counts, cpu);
94
95         return total;
96 }
97
98 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
99 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
100 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
101 static struct kmem_cache *task_struct_cachep;
102 #endif
103
104 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
105 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
106 {
107 #ifdef CONFIG_DEBUG_STACK_USAGE
108         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
109 #else
110         gfp_t mask = GFP_KERNEL;
111 #endif
112         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
113 }
114
115 static inline void free_thread_info(struct thread_info *ti)
116 {
117         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
118 }
119 #endif
120
121 /* SLAB cache for signal_struct structures (tsk->signal) */
122 static struct kmem_cache *signal_cachep;
123
124 /* SLAB cache for sighand_struct structures (tsk->sighand) */
125 struct kmem_cache *sighand_cachep;
126
127 /* SLAB cache for files_struct structures (tsk->files) */
128 struct kmem_cache *files_cachep;
129
130 /* SLAB cache for fs_struct structures (tsk->fs) */
131 struct kmem_cache *fs_cachep;
132
133 /* SLAB cache for vm_area_struct structures */
134 struct kmem_cache *vm_area_cachep;
135
136 /* SLAB cache for mm_struct structures (tsk->mm) */
137 static struct kmem_cache *mm_cachep;
138
139 void free_task(struct task_struct *tsk)
140 {
141         prop_local_destroy_single(&tsk->dirties);
142         free_thread_info(tsk->stack);
143         rt_mutex_debug_task_free(tsk);
144         ftrace_graph_exit_task(tsk);
145         free_task_struct(tsk);
146 }
147 EXPORT_SYMBOL(free_task);
148
149 void __put_task_struct(struct task_struct *tsk)
150 {
151         WARN_ON(!tsk->exit_state);
152         WARN_ON(atomic_read(&tsk->usage));
153         WARN_ON(tsk == current);
154
155         exit_creds(tsk);
156         delayacct_tsk_free(tsk);
157
158         if (!profile_handoff_task(tsk))
159                 free_task(tsk);
160 }
161
162 /*
163  * macro override instead of weak attribute alias, to workaround
164  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
165  */
166 #ifndef arch_task_cache_init
167 #define arch_task_cache_init()
168 #endif
169
170 void __init fork_init(unsigned long mempages)
171 {
172 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
173 #ifndef ARCH_MIN_TASKALIGN
174 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
175 #endif
176         /* create a slab on which task_structs can be allocated */
177         task_struct_cachep =
178                 kmem_cache_create("task_struct", sizeof(struct task_struct),
179                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
180 #endif
181
182         /* do the arch specific task caches init */
183         arch_task_cache_init();
184
185         /*
186          * The default maximum number of threads is set to a safe
187          * value: the thread structures can take up at most half
188          * of memory.
189          */
190         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
191
192         /*
193          * we need to allow at least 20 threads to boot a system
194          */
195         if(max_threads < 20)
196                 max_threads = 20;
197
198         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
199         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
200         init_task.signal->rlim[RLIMIT_SIGPENDING] =
201                 init_task.signal->rlim[RLIMIT_NPROC];
202 }
203
204 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
205                                                struct task_struct *src)
206 {
207         *dst = *src;
208         return 0;
209 }
210
211 static struct task_struct *dup_task_struct(struct task_struct *orig)
212 {
213         struct task_struct *tsk;
214         struct thread_info *ti;
215         unsigned long *stackend;
216
217         int err;
218
219         prepare_to_copy(orig);
220
221         tsk = alloc_task_struct();
222         if (!tsk)
223                 return NULL;
224
225         ti = alloc_thread_info(tsk);
226         if (!ti) {
227                 free_task_struct(tsk);
228                 return NULL;
229         }
230
231         err = arch_dup_task_struct(tsk, orig);
232         if (err)
233                 goto out;
234
235         tsk->stack = ti;
236
237         err = prop_local_init_single(&tsk->dirties);
238         if (err)
239                 goto out;
240
241         setup_thread_stack(tsk, orig);
242         stackend = end_of_stack(tsk);
243         *stackend = STACK_END_MAGIC;    /* for overflow detection */
244
245 #ifdef CONFIG_CC_STACKPROTECTOR
246         tsk->stack_canary = get_random_int();
247 #endif
248
249         /* One for us, one for whoever does the "release_task()" (usually parent) */
250         atomic_set(&tsk->usage,2);
251         atomic_set(&tsk->fs_excl, 0);
252 #ifdef CONFIG_BLK_DEV_IO_TRACE
253         tsk->btrace_seq = 0;
254 #endif
255         tsk->splice_pipe = NULL;
256         return tsk;
257
258 out:
259         free_thread_info(ti);
260         free_task_struct(tsk);
261         return NULL;
262 }
263
264 #ifdef CONFIG_MMU
265 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
266 {
267         struct vm_area_struct *mpnt, *tmp, **pprev;
268         struct rb_node **rb_link, *rb_parent;
269         int retval;
270         unsigned long charge;
271         struct mempolicy *pol;
272
273         down_write(&oldmm->mmap_sem);
274         flush_cache_dup_mm(oldmm);
275         /*
276          * Not linked in yet - no deadlock potential:
277          */
278         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
279
280         mm->locked_vm = 0;
281         mm->mmap = NULL;
282         mm->mmap_cache = NULL;
283         mm->free_area_cache = oldmm->mmap_base;
284         mm->cached_hole_size = ~0UL;
285         mm->map_count = 0;
286         cpumask_clear(mm_cpumask(mm));
287         mm->mm_rb = RB_ROOT;
288         rb_link = &mm->mm_rb.rb_node;
289         rb_parent = NULL;
290         pprev = &mm->mmap;
291
292         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
293                 struct file *file;
294
295                 if (mpnt->vm_flags & VM_DONTCOPY) {
296                         long pages = vma_pages(mpnt);
297                         mm->total_vm -= pages;
298                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
299                                                                 -pages);
300                         continue;
301                 }
302                 charge = 0;
303                 if (mpnt->vm_flags & VM_ACCOUNT) {
304                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
305                         if (security_vm_enough_memory(len))
306                                 goto fail_nomem;
307                         charge = len;
308                 }
309                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
310                 if (!tmp)
311                         goto fail_nomem;
312                 *tmp = *mpnt;
313                 pol = mpol_dup(vma_policy(mpnt));
314                 retval = PTR_ERR(pol);
315                 if (IS_ERR(pol))
316                         goto fail_nomem_policy;
317                 vma_set_policy(tmp, pol);
318                 tmp->vm_flags &= ~VM_LOCKED;
319                 tmp->vm_mm = mm;
320                 tmp->vm_next = NULL;
321                 anon_vma_link(tmp);
322                 file = tmp->vm_file;
323                 if (file) {
324                         struct inode *inode = file->f_path.dentry->d_inode;
325                         struct address_space *mapping = file->f_mapping;
326
327                         get_file(file);
328                         if (tmp->vm_flags & VM_DENYWRITE)
329                                 atomic_dec(&inode->i_writecount);
330                         spin_lock(&mapping->i_mmap_lock);
331                         if (tmp->vm_flags & VM_SHARED)
332                                 mapping->i_mmap_writable++;
333                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
334                         flush_dcache_mmap_lock(mapping);
335                         /* insert tmp into the share list, just after mpnt */
336                         vma_prio_tree_add(tmp, mpnt);
337                         flush_dcache_mmap_unlock(mapping);
338                         spin_unlock(&mapping->i_mmap_lock);
339                 }
340
341                 /*
342                  * Clear hugetlb-related page reserves for children. This only
343                  * affects MAP_PRIVATE mappings. Faults generated by the child
344                  * are not guaranteed to succeed, even if read-only
345                  */
346                 if (is_vm_hugetlb_page(tmp))
347                         reset_vma_resv_huge_pages(tmp);
348
349                 /*
350                  * Link in the new vma and copy the page table entries.
351                  */
352                 *pprev = tmp;
353                 pprev = &tmp->vm_next;
354
355                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
356                 rb_link = &tmp->vm_rb.rb_right;
357                 rb_parent = &tmp->vm_rb;
358
359                 mm->map_count++;
360                 retval = copy_page_range(mm, oldmm, mpnt);
361
362                 if (tmp->vm_ops && tmp->vm_ops->open)
363                         tmp->vm_ops->open(tmp);
364
365                 if (retval)
366                         goto out;
367         }
368         /* a new mm has just been created */
369         arch_dup_mmap(oldmm, mm);
370         retval = 0;
371 out:
372         up_write(&mm->mmap_sem);
373         flush_tlb_mm(oldmm);
374         up_write(&oldmm->mmap_sem);
375         return retval;
376 fail_nomem_policy:
377         kmem_cache_free(vm_area_cachep, tmp);
378 fail_nomem:
379         retval = -ENOMEM;
380         vm_unacct_memory(charge);
381         goto out;
382 }
383
384 static inline int mm_alloc_pgd(struct mm_struct * mm)
385 {
386         mm->pgd = pgd_alloc(mm);
387         if (unlikely(!mm->pgd))
388                 return -ENOMEM;
389         return 0;
390 }
391
392 static inline void mm_free_pgd(struct mm_struct * mm)
393 {
394         pgd_free(mm, mm->pgd);
395 }
396 #else
397 #define dup_mmap(mm, oldmm)     (0)
398 #define mm_alloc_pgd(mm)        (0)
399 #define mm_free_pgd(mm)
400 #endif /* CONFIG_MMU */
401
402 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
403
404 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
405 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
406
407 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
408
409 static int __init coredump_filter_setup(char *s)
410 {
411         default_dump_filter =
412                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
413                 MMF_DUMP_FILTER_MASK;
414         return 1;
415 }
416
417 __setup("coredump_filter=", coredump_filter_setup);
418
419 #include <linux/init_task.h>
420
421 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
422 {
423         atomic_set(&mm->mm_users, 1);
424         atomic_set(&mm->mm_count, 1);
425         init_rwsem(&mm->mmap_sem);
426         INIT_LIST_HEAD(&mm->mmlist);
427         mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
428         mm->core_state = NULL;
429         mm->nr_ptes = 0;
430         set_mm_counter(mm, file_rss, 0);
431         set_mm_counter(mm, anon_rss, 0);
432         spin_lock_init(&mm->page_table_lock);
433         spin_lock_init(&mm->ioctx_lock);
434         INIT_HLIST_HEAD(&mm->ioctx_list);
435         mm->free_area_cache = TASK_UNMAPPED_BASE;
436         mm->cached_hole_size = ~0UL;
437         mm_init_owner(mm, p);
438
439         if (likely(!mm_alloc_pgd(mm))) {
440                 mm->def_flags = 0;
441                 mmu_notifier_mm_init(mm);
442                 return mm;
443         }
444
445         free_mm(mm);
446         return NULL;
447 }
448
449 /*
450  * Allocate and initialize an mm_struct.
451  */
452 struct mm_struct * mm_alloc(void)
453 {
454         struct mm_struct * mm;
455
456         mm = allocate_mm();
457         if (mm) {
458                 memset(mm, 0, sizeof(*mm));
459                 mm = mm_init(mm, current);
460         }
461         return mm;
462 }
463
464 /*
465  * Called when the last reference to the mm
466  * is dropped: either by a lazy thread or by
467  * mmput. Free the page directory and the mm.
468  */
469 void __mmdrop(struct mm_struct *mm)
470 {
471         BUG_ON(mm == &init_mm);
472         mm_free_pgd(mm);
473         destroy_context(mm);
474         mmu_notifier_mm_destroy(mm);
475         free_mm(mm);
476 }
477 EXPORT_SYMBOL_GPL(__mmdrop);
478
479 /*
480  * Decrement the use count and release all resources for an mm.
481  */
482 void mmput(struct mm_struct *mm)
483 {
484         might_sleep();
485
486         if (atomic_dec_and_test(&mm->mm_users)) {
487                 exit_aio(mm);
488                 exit_mmap(mm);
489                 set_mm_exe_file(mm, NULL);
490                 if (!list_empty(&mm->mmlist)) {
491                         spin_lock(&mmlist_lock);
492                         list_del(&mm->mmlist);
493                         spin_unlock(&mmlist_lock);
494                 }
495                 put_swap_token(mm);
496                 mmdrop(mm);
497         }
498 }
499 EXPORT_SYMBOL_GPL(mmput);
500
501 /**
502  * get_task_mm - acquire a reference to the task's mm
503  *
504  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
505  * this kernel workthread has transiently adopted a user mm with use_mm,
506  * to do its AIO) is not set and if so returns a reference to it, after
507  * bumping up the use count.  User must release the mm via mmput()
508  * after use.  Typically used by /proc and ptrace.
509  */
510 struct mm_struct *get_task_mm(struct task_struct *task)
511 {
512         struct mm_struct *mm;
513
514         task_lock(task);
515         mm = task->mm;
516         if (mm) {
517                 if (task->flags & PF_KTHREAD)
518                         mm = NULL;
519                 else
520                         atomic_inc(&mm->mm_users);
521         }
522         task_unlock(task);
523         return mm;
524 }
525 EXPORT_SYMBOL_GPL(get_task_mm);
526
527 /* Please note the differences between mmput and mm_release.
528  * mmput is called whenever we stop holding onto a mm_struct,
529  * error success whatever.
530  *
531  * mm_release is called after a mm_struct has been removed
532  * from the current process.
533  *
534  * This difference is important for error handling, when we
535  * only half set up a mm_struct for a new process and need to restore
536  * the old one.  Because we mmput the new mm_struct before
537  * restoring the old one. . .
538  * Eric Biederman 10 January 1998
539  */
540 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
541 {
542         struct completion *vfork_done = tsk->vfork_done;
543
544         /* Get rid of any futexes when releasing the mm */
545 #ifdef CONFIG_FUTEX
546         if (unlikely(tsk->robust_list)) {
547                 exit_robust_list(tsk);
548                 tsk->robust_list = NULL;
549         }
550 #ifdef CONFIG_COMPAT
551         if (unlikely(tsk->compat_robust_list)) {
552                 compat_exit_robust_list(tsk);
553                 tsk->compat_robust_list = NULL;
554         }
555 #endif
556         if (unlikely(!list_empty(&tsk->pi_state_list)))
557                 exit_pi_state_list(tsk);
558 #endif
559
560         /* Get rid of any cached register state */
561         deactivate_mm(tsk, mm);
562
563         /* notify parent sleeping on vfork() */
564         if (vfork_done) {
565                 tsk->vfork_done = NULL;
566                 complete(vfork_done);
567         }
568
569         /*
570          * If we're exiting normally, clear a user-space tid field if
571          * requested.  We leave this alone when dying by signal, to leave
572          * the value intact in a core dump, and to save the unnecessary
573          * trouble otherwise.  Userland only wants this done for a sys_exit.
574          */
575         if (tsk->clear_child_tid) {
576                 if (!(tsk->flags & PF_SIGNALED) &&
577                     atomic_read(&mm->mm_users) > 1) {
578                         /*
579                          * We don't check the error code - if userspace has
580                          * not set up a proper pointer then tough luck.
581                          */
582                         put_user(0, tsk->clear_child_tid);
583                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
584                                         1, NULL, NULL, 0);
585                 }
586                 tsk->clear_child_tid = NULL;
587         }
588 }
589
590 /*
591  * Allocate a new mm structure and copy contents from the
592  * mm structure of the passed in task structure.
593  */
594 struct mm_struct *dup_mm(struct task_struct *tsk)
595 {
596         struct mm_struct *mm, *oldmm = current->mm;
597         int err;
598
599         if (!oldmm)
600                 return NULL;
601
602         mm = allocate_mm();
603         if (!mm)
604                 goto fail_nomem;
605
606         memcpy(mm, oldmm, sizeof(*mm));
607
608         /* Initializing for Swap token stuff */
609         mm->token_priority = 0;
610         mm->last_interval = 0;
611
612         if (!mm_init(mm, tsk))
613                 goto fail_nomem;
614
615         if (init_new_context(tsk, mm))
616                 goto fail_nocontext;
617
618         dup_mm_exe_file(oldmm, mm);
619
620         err = dup_mmap(mm, oldmm);
621         if (err)
622                 goto free_pt;
623
624         mm->hiwater_rss = get_mm_rss(mm);
625         mm->hiwater_vm = mm->total_vm;
626
627         return mm;
628
629 free_pt:
630         mmput(mm);
631
632 fail_nomem:
633         return NULL;
634
635 fail_nocontext:
636         /*
637          * If init_new_context() failed, we cannot use mmput() to free the mm
638          * because it calls destroy_context()
639          */
640         mm_free_pgd(mm);
641         free_mm(mm);
642         return NULL;
643 }
644
645 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
646 {
647         struct mm_struct * mm, *oldmm;
648         int retval;
649
650         tsk->min_flt = tsk->maj_flt = 0;
651         tsk->nvcsw = tsk->nivcsw = 0;
652 #ifdef CONFIG_DETECT_HUNG_TASK
653         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
654 #endif
655
656         tsk->mm = NULL;
657         tsk->active_mm = NULL;
658
659         /*
660          * Are we cloning a kernel thread?
661          *
662          * We need to steal a active VM for that..
663          */
664         oldmm = current->mm;
665         if (!oldmm)
666                 return 0;
667
668         if (clone_flags & CLONE_VM) {
669                 atomic_inc(&oldmm->mm_users);
670                 mm = oldmm;
671                 goto good_mm;
672         }
673
674         retval = -ENOMEM;
675         mm = dup_mm(tsk);
676         if (!mm)
677                 goto fail_nomem;
678
679 good_mm:
680         /* Initializing for Swap token stuff */
681         mm->token_priority = 0;
682         mm->last_interval = 0;
683
684         tsk->mm = mm;
685         tsk->active_mm = mm;
686         return 0;
687
688 fail_nomem:
689         return retval;
690 }
691
692 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
693 {
694         struct fs_struct *fs = current->fs;
695         if (clone_flags & CLONE_FS) {
696                 /* tsk->fs is already what we want */
697                 write_lock(&fs->lock);
698                 if (fs->in_exec) {
699                         write_unlock(&fs->lock);
700                         return -EAGAIN;
701                 }
702                 fs->users++;
703                 write_unlock(&fs->lock);
704                 return 0;
705         }
706         tsk->fs = copy_fs_struct(fs);
707         if (!tsk->fs)
708                 return -ENOMEM;
709         return 0;
710 }
711
712 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
713 {
714         struct files_struct *oldf, *newf;
715         int error = 0;
716
717         /*
718          * A background process may not have any files ...
719          */
720         oldf = current->files;
721         if (!oldf)
722                 goto out;
723
724         if (clone_flags & CLONE_FILES) {
725                 atomic_inc(&oldf->count);
726                 goto out;
727         }
728
729         newf = dup_fd(oldf, &error);
730         if (!newf)
731                 goto out;
732
733         tsk->files = newf;
734         error = 0;
735 out:
736         return error;
737 }
738
739 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
740 {
741 #ifdef CONFIG_BLOCK
742         struct io_context *ioc = current->io_context;
743
744         if (!ioc)
745                 return 0;
746         /*
747          * Share io context with parent, if CLONE_IO is set
748          */
749         if (clone_flags & CLONE_IO) {
750                 tsk->io_context = ioc_task_link(ioc);
751                 if (unlikely(!tsk->io_context))
752                         return -ENOMEM;
753         } else if (ioprio_valid(ioc->ioprio)) {
754                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
755                 if (unlikely(!tsk->io_context))
756                         return -ENOMEM;
757
758                 tsk->io_context->ioprio = ioc->ioprio;
759         }
760 #endif
761         return 0;
762 }
763
764 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
765 {
766         struct sighand_struct *sig;
767
768         if (clone_flags & CLONE_SIGHAND) {
769                 atomic_inc(&current->sighand->count);
770                 return 0;
771         }
772         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
773         rcu_assign_pointer(tsk->sighand, sig);
774         if (!sig)
775                 return -ENOMEM;
776         atomic_set(&sig->count, 1);
777         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
778         return 0;
779 }
780
781 void __cleanup_sighand(struct sighand_struct *sighand)
782 {
783         if (atomic_dec_and_test(&sighand->count))
784                 kmem_cache_free(sighand_cachep, sighand);
785 }
786
787
788 /*
789  * Initialize POSIX timer handling for a thread group.
790  */
791 static void posix_cpu_timers_init_group(struct signal_struct *sig)
792 {
793         /* Thread group counters. */
794         thread_group_cputime_init(sig);
795
796         /* Expiration times and increments. */
797         sig->it_virt_expires = cputime_zero;
798         sig->it_virt_incr = cputime_zero;
799         sig->it_prof_expires = cputime_zero;
800         sig->it_prof_incr = cputime_zero;
801
802         /* Cached expiration times. */
803         sig->cputime_expires.prof_exp = cputime_zero;
804         sig->cputime_expires.virt_exp = cputime_zero;
805         sig->cputime_expires.sched_exp = 0;
806
807         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
808                 sig->cputime_expires.prof_exp =
809                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
810                 sig->cputimer.running = 1;
811         }
812
813         /* The timer lists. */
814         INIT_LIST_HEAD(&sig->cpu_timers[0]);
815         INIT_LIST_HEAD(&sig->cpu_timers[1]);
816         INIT_LIST_HEAD(&sig->cpu_timers[2]);
817 }
818
819 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
820 {
821         struct signal_struct *sig;
822
823         if (clone_flags & CLONE_THREAD)
824                 return 0;
825
826         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
827         tsk->signal = sig;
828         if (!sig)
829                 return -ENOMEM;
830
831         atomic_set(&sig->count, 1);
832         atomic_set(&sig->live, 1);
833         init_waitqueue_head(&sig->wait_chldexit);
834         sig->flags = 0;
835         if (clone_flags & CLONE_NEWPID)
836                 sig->flags |= SIGNAL_UNKILLABLE;
837         sig->group_exit_code = 0;
838         sig->group_exit_task = NULL;
839         sig->group_stop_count = 0;
840         sig->curr_target = tsk;
841         init_sigpending(&sig->shared_pending);
842         INIT_LIST_HEAD(&sig->posix_timers);
843
844         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
845         sig->it_real_incr.tv64 = 0;
846         sig->real_timer.function = it_real_fn;
847
848         sig->leader = 0;        /* session leadership doesn't inherit */
849         sig->tty_old_pgrp = NULL;
850         sig->tty = NULL;
851
852         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
853         sig->gtime = cputime_zero;
854         sig->cgtime = cputime_zero;
855         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
856         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
857         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
858         task_io_accounting_init(&sig->ioac);
859         sig->sum_sched_runtime = 0;
860         taskstats_tgid_init(sig);
861
862         task_lock(current->group_leader);
863         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
864         task_unlock(current->group_leader);
865
866         posix_cpu_timers_init_group(sig);
867
868         acct_init_pacct(&sig->pacct);
869
870         tty_audit_fork(sig);
871
872         return 0;
873 }
874
875 void __cleanup_signal(struct signal_struct *sig)
876 {
877         thread_group_cputime_free(sig);
878         tty_kref_put(sig->tty);
879         kmem_cache_free(signal_cachep, sig);
880 }
881
882 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
883 {
884         unsigned long new_flags = p->flags;
885
886         new_flags &= ~PF_SUPERPRIV;
887         new_flags |= PF_FORKNOEXEC;
888         new_flags |= PF_STARTING;
889         p->flags = new_flags;
890         clear_freeze_flag(p);
891 }
892
893 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
894 {
895         current->clear_child_tid = tidptr;
896
897         return task_pid_vnr(current);
898 }
899
900 static void rt_mutex_init_task(struct task_struct *p)
901 {
902         spin_lock_init(&p->pi_lock);
903 #ifdef CONFIG_RT_MUTEXES
904         plist_head_init(&p->pi_waiters, &p->pi_lock);
905         p->pi_blocked_on = NULL;
906 #endif
907 }
908
909 #ifdef CONFIG_MM_OWNER
910 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
911 {
912         mm->owner = p;
913 }
914 #endif /* CONFIG_MM_OWNER */
915
916 /*
917  * Initialize POSIX timer handling for a single task.
918  */
919 static void posix_cpu_timers_init(struct task_struct *tsk)
920 {
921         tsk->cputime_expires.prof_exp = cputime_zero;
922         tsk->cputime_expires.virt_exp = cputime_zero;
923         tsk->cputime_expires.sched_exp = 0;
924         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
925         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
926         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
927 }
928
929 /*
930  * This creates a new process as a copy of the old one,
931  * but does not actually start it yet.
932  *
933  * It copies the registers, and all the appropriate
934  * parts of the process environment (as per the clone
935  * flags). The actual kick-off is left to the caller.
936  */
937 static struct task_struct *copy_process(unsigned long clone_flags,
938                                         unsigned long stack_start,
939                                         struct pt_regs *regs,
940                                         unsigned long stack_size,
941                                         int __user *child_tidptr,
942                                         struct pid *pid,
943                                         int trace)
944 {
945         int retval;
946         struct task_struct *p;
947         int cgroup_callbacks_done = 0;
948
949         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
950                 return ERR_PTR(-EINVAL);
951
952         /*
953          * Thread groups must share signals as well, and detached threads
954          * can only be started up within the thread group.
955          */
956         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
957                 return ERR_PTR(-EINVAL);
958
959         /*
960          * Shared signal handlers imply shared VM. By way of the above,
961          * thread groups also imply shared VM. Blocking this case allows
962          * for various simplifications in other code.
963          */
964         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
965                 return ERR_PTR(-EINVAL);
966
967         retval = security_task_create(clone_flags);
968         if (retval)
969                 goto fork_out;
970
971         retval = -ENOMEM;
972         p = dup_task_struct(current);
973         if (!p)
974                 goto fork_out;
975
976         ftrace_graph_init_task(p);
977
978         rt_mutex_init_task(p);
979
980 #ifdef CONFIG_PROVE_LOCKING
981         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
982         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
983 #endif
984         retval = -EAGAIN;
985         if (atomic_read(&p->real_cred->user->processes) >=
986                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
987                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
988                     p->real_cred->user != INIT_USER)
989                         goto bad_fork_free;
990         }
991
992         retval = copy_creds(p, clone_flags);
993         if (retval < 0)
994                 goto bad_fork_free;
995
996         /*
997          * If multiple threads are within copy_process(), then this check
998          * triggers too late. This doesn't hurt, the check is only there
999          * to stop root fork bombs.
1000          */
1001         retval = -EAGAIN;
1002         if (nr_threads >= max_threads)
1003                 goto bad_fork_cleanup_count;
1004
1005         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1006                 goto bad_fork_cleanup_count;
1007
1008         if (p->binfmt && !try_module_get(p->binfmt->module))
1009                 goto bad_fork_cleanup_put_domain;
1010
1011         p->did_exec = 0;
1012         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1013         copy_flags(clone_flags, p);
1014         INIT_LIST_HEAD(&p->children);
1015         INIT_LIST_HEAD(&p->sibling);
1016         rcu_copy_process(p);
1017         p->vfork_done = NULL;
1018         spin_lock_init(&p->alloc_lock);
1019
1020         init_sigpending(&p->pending);
1021
1022         p->utime = cputime_zero;
1023         p->stime = cputime_zero;
1024         p->gtime = cputime_zero;
1025         p->utimescaled = cputime_zero;
1026         p->stimescaled = cputime_zero;
1027         p->prev_utime = cputime_zero;
1028         p->prev_stime = cputime_zero;
1029
1030         p->default_timer_slack_ns = current->timer_slack_ns;
1031
1032         task_io_accounting_init(&p->ioac);
1033         acct_clear_integrals(p);
1034
1035         posix_cpu_timers_init(p);
1036
1037         p->lock_depth = -1;             /* -1 = no lock */
1038         do_posix_clock_monotonic_gettime(&p->start_time);
1039         p->real_start_time = p->start_time;
1040         monotonic_to_bootbased(&p->real_start_time);
1041         p->io_context = NULL;
1042         p->audit_context = NULL;
1043         cgroup_fork(p);
1044 #ifdef CONFIG_NUMA
1045         p->mempolicy = mpol_dup(p->mempolicy);
1046         if (IS_ERR(p->mempolicy)) {
1047                 retval = PTR_ERR(p->mempolicy);
1048                 p->mempolicy = NULL;
1049                 goto bad_fork_cleanup_cgroup;
1050         }
1051         mpol_fix_fork_child_flag(p);
1052 #endif
1053 #ifdef CONFIG_TRACE_IRQFLAGS
1054         p->irq_events = 0;
1055 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1056         p->hardirqs_enabled = 1;
1057 #else
1058         p->hardirqs_enabled = 0;
1059 #endif
1060         p->hardirq_enable_ip = 0;
1061         p->hardirq_enable_event = 0;
1062         p->hardirq_disable_ip = _THIS_IP_;
1063         p->hardirq_disable_event = 0;
1064         p->softirqs_enabled = 1;
1065         p->softirq_enable_ip = _THIS_IP_;
1066         p->softirq_enable_event = 0;
1067         p->softirq_disable_ip = 0;
1068         p->softirq_disable_event = 0;
1069         p->hardirq_context = 0;
1070         p->softirq_context = 0;
1071 #endif
1072 #ifdef CONFIG_LOCKDEP
1073         p->lockdep_depth = 0; /* no locks held yet */
1074         p->curr_chain_key = 0;
1075         p->lockdep_recursion = 0;
1076 #endif
1077
1078 #ifdef CONFIG_DEBUG_MUTEXES
1079         p->blocked_on = NULL; /* not blocked yet */
1080 #endif
1081
1082         p->bts = NULL;
1083
1084         /* Perform scheduler related setup. Assign this task to a CPU. */
1085         sched_fork(p, clone_flags);
1086
1087         retval = perf_counter_init_task(p);
1088         if (retval)
1089                 goto bad_fork_cleanup_policy;
1090
1091         if ((retval = audit_alloc(p)))
1092                 goto bad_fork_cleanup_policy;
1093         /* copy all the process information */
1094         if ((retval = copy_semundo(clone_flags, p)))
1095                 goto bad_fork_cleanup_audit;
1096         if ((retval = copy_files(clone_flags, p)))
1097                 goto bad_fork_cleanup_semundo;
1098         if ((retval = copy_fs(clone_flags, p)))
1099                 goto bad_fork_cleanup_files;
1100         if ((retval = copy_sighand(clone_flags, p)))
1101                 goto bad_fork_cleanup_fs;
1102         if ((retval = copy_signal(clone_flags, p)))
1103                 goto bad_fork_cleanup_sighand;
1104         if ((retval = copy_mm(clone_flags, p)))
1105                 goto bad_fork_cleanup_signal;
1106         if ((retval = copy_namespaces(clone_flags, p)))
1107                 goto bad_fork_cleanup_mm;
1108         if ((retval = copy_io(clone_flags, p)))
1109                 goto bad_fork_cleanup_namespaces;
1110         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1111         if (retval)
1112                 goto bad_fork_cleanup_io;
1113
1114         if (pid != &init_struct_pid) {
1115                 retval = -ENOMEM;
1116                 pid = alloc_pid(p->nsproxy->pid_ns);
1117                 if (!pid)
1118                         goto bad_fork_cleanup_io;
1119
1120                 if (clone_flags & CLONE_NEWPID) {
1121                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1122                         if (retval < 0)
1123                                 goto bad_fork_free_pid;
1124                 }
1125         }
1126
1127         p->pid = pid_nr(pid);
1128         p->tgid = p->pid;
1129         if (clone_flags & CLONE_THREAD)
1130                 p->tgid = current->tgid;
1131
1132         if (current->nsproxy != p->nsproxy) {
1133                 retval = ns_cgroup_clone(p, pid);
1134                 if (retval)
1135                         goto bad_fork_free_pid;
1136         }
1137
1138         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1139         /*
1140          * Clear TID on mm_release()?
1141          */
1142         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1143 #ifdef CONFIG_FUTEX
1144         p->robust_list = NULL;
1145 #ifdef CONFIG_COMPAT
1146         p->compat_robust_list = NULL;
1147 #endif
1148         INIT_LIST_HEAD(&p->pi_state_list);
1149         p->pi_state_cache = NULL;
1150 #endif
1151         /*
1152          * sigaltstack should be cleared when sharing the same VM
1153          */
1154         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1155                 p->sas_ss_sp = p->sas_ss_size = 0;
1156
1157         /*
1158          * Syscall tracing should be turned off in the child regardless
1159          * of CLONE_PTRACE.
1160          */
1161         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1162 #ifdef TIF_SYSCALL_EMU
1163         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1164 #endif
1165         clear_all_latency_tracing(p);
1166
1167         /* ok, now we should be set up.. */
1168         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1169         p->pdeath_signal = 0;
1170         p->exit_state = 0;
1171
1172         /*
1173          * Ok, make it visible to the rest of the system.
1174          * We dont wake it up yet.
1175          */
1176         p->group_leader = p;
1177         INIT_LIST_HEAD(&p->thread_group);
1178
1179         /* Now that the task is set up, run cgroup callbacks if
1180          * necessary. We need to run them before the task is visible
1181          * on the tasklist. */
1182         cgroup_fork_callbacks(p);
1183         cgroup_callbacks_done = 1;
1184
1185         /* Need tasklist lock for parent etc handling! */
1186         write_lock_irq(&tasklist_lock);
1187
1188         /*
1189          * The task hasn't been attached yet, so its cpus_allowed mask will
1190          * not be changed, nor will its assigned CPU.
1191          *
1192          * The cpus_allowed mask of the parent may have changed after it was
1193          * copied first time - so re-copy it here, then check the child's CPU
1194          * to ensure it is on a valid CPU (and if not, just force it back to
1195          * parent's CPU). This avoids alot of nasty races.
1196          */
1197         p->cpus_allowed = current->cpus_allowed;
1198         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1199         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1200                         !cpu_online(task_cpu(p))))
1201                 set_task_cpu(p, smp_processor_id());
1202
1203         /* CLONE_PARENT re-uses the old parent */
1204         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1205                 p->real_parent = current->real_parent;
1206                 p->parent_exec_id = current->parent_exec_id;
1207         } else {
1208                 p->real_parent = current;
1209                 p->parent_exec_id = current->self_exec_id;
1210         }
1211
1212         spin_lock(&current->sighand->siglock);
1213
1214         /*
1215          * Process group and session signals need to be delivered to just the
1216          * parent before the fork or both the parent and the child after the
1217          * fork. Restart if a signal comes in before we add the new process to
1218          * it's process group.
1219          * A fatal signal pending means that current will exit, so the new
1220          * thread can't slip out of an OOM kill (or normal SIGKILL).
1221          */
1222         recalc_sigpending();
1223         if (signal_pending(current)) {
1224                 spin_unlock(&current->sighand->siglock);
1225                 write_unlock_irq(&tasklist_lock);
1226                 retval = -ERESTARTNOINTR;
1227                 goto bad_fork_free_pid;
1228         }
1229
1230         if (clone_flags & CLONE_THREAD) {
1231                 atomic_inc(&current->signal->count);
1232                 atomic_inc(&current->signal->live);
1233                 p->group_leader = current->group_leader;
1234                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1235         }
1236
1237         if (likely(p->pid)) {
1238                 list_add_tail(&p->sibling, &p->real_parent->children);
1239                 tracehook_finish_clone(p, clone_flags, trace);
1240
1241                 if (thread_group_leader(p)) {
1242                         if (clone_flags & CLONE_NEWPID)
1243                                 p->nsproxy->pid_ns->child_reaper = p;
1244
1245                         p->signal->leader_pid = pid;
1246                         tty_kref_put(p->signal->tty);
1247                         p->signal->tty = tty_kref_get(current->signal->tty);
1248                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1249                         attach_pid(p, PIDTYPE_SID, task_session(current));
1250                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1251                         __get_cpu_var(process_counts)++;
1252                 }
1253                 attach_pid(p, PIDTYPE_PID, pid);
1254                 nr_threads++;
1255         }
1256
1257         total_forks++;
1258         spin_unlock(&current->sighand->siglock);
1259         write_unlock_irq(&tasklist_lock);
1260         proc_fork_connector(p);
1261         cgroup_post_fork(p);
1262         perf_counter_fork(p);
1263         return p;
1264
1265 bad_fork_free_pid:
1266         if (pid != &init_struct_pid)
1267                 free_pid(pid);
1268 bad_fork_cleanup_io:
1269         put_io_context(p->io_context);
1270 bad_fork_cleanup_namespaces:
1271         exit_task_namespaces(p);
1272 bad_fork_cleanup_mm:
1273         if (p->mm)
1274                 mmput(p->mm);
1275 bad_fork_cleanup_signal:
1276         if (!(clone_flags & CLONE_THREAD))
1277                 __cleanup_signal(p->signal);
1278 bad_fork_cleanup_sighand:
1279         __cleanup_sighand(p->sighand);
1280 bad_fork_cleanup_fs:
1281         exit_fs(p); /* blocking */
1282 bad_fork_cleanup_files:
1283         exit_files(p); /* blocking */
1284 bad_fork_cleanup_semundo:
1285         exit_sem(p);
1286 bad_fork_cleanup_audit:
1287         audit_free(p);
1288 bad_fork_cleanup_policy:
1289         perf_counter_free_task(p);
1290 #ifdef CONFIG_NUMA
1291         mpol_put(p->mempolicy);
1292 bad_fork_cleanup_cgroup:
1293 #endif
1294         cgroup_exit(p, cgroup_callbacks_done);
1295         delayacct_tsk_free(p);
1296         if (p->binfmt)
1297                 module_put(p->binfmt->module);
1298 bad_fork_cleanup_put_domain:
1299         module_put(task_thread_info(p)->exec_domain->module);
1300 bad_fork_cleanup_count:
1301         atomic_dec(&p->cred->user->processes);
1302         exit_creds(p);
1303 bad_fork_free:
1304         free_task(p);
1305 fork_out:
1306         return ERR_PTR(retval);
1307 }
1308
1309 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1310 {
1311         memset(regs, 0, sizeof(struct pt_regs));
1312         return regs;
1313 }
1314
1315 struct task_struct * __cpuinit fork_idle(int cpu)
1316 {
1317         struct task_struct *task;
1318         struct pt_regs regs;
1319
1320         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1321                             &init_struct_pid, 0);
1322         if (!IS_ERR(task))
1323                 init_idle(task, cpu);
1324
1325         return task;
1326 }
1327
1328 /*
1329  *  Ok, this is the main fork-routine.
1330  *
1331  * It copies the process, and if successful kick-starts
1332  * it and waits for it to finish using the VM if required.
1333  */
1334 long do_fork(unsigned long clone_flags,
1335               unsigned long stack_start,
1336               struct pt_regs *regs,
1337               unsigned long stack_size,
1338               int __user *parent_tidptr,
1339               int __user *child_tidptr)
1340 {
1341         struct task_struct *p;
1342         int trace = 0;
1343         long nr;
1344
1345         /*
1346          * Do some preliminary argument and permissions checking before we
1347          * actually start allocating stuff
1348          */
1349         if (clone_flags & CLONE_NEWUSER) {
1350                 if (clone_flags & CLONE_THREAD)
1351                         return -EINVAL;
1352                 /* hopefully this check will go away when userns support is
1353                  * complete
1354                  */
1355                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1356                                 !capable(CAP_SETGID))
1357                         return -EPERM;
1358         }
1359
1360         /*
1361          * We hope to recycle these flags after 2.6.26
1362          */
1363         if (unlikely(clone_flags & CLONE_STOPPED)) {
1364                 static int __read_mostly count = 100;
1365
1366                 if (count > 0 && printk_ratelimit()) {
1367                         char comm[TASK_COMM_LEN];
1368
1369                         count--;
1370                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1371                                         "clone flags 0x%lx\n",
1372                                 get_task_comm(comm, current),
1373                                 clone_flags & CLONE_STOPPED);
1374                 }
1375         }
1376
1377         /*
1378          * When called from kernel_thread, don't do user tracing stuff.
1379          */
1380         if (likely(user_mode(regs)))
1381                 trace = tracehook_prepare_clone(clone_flags);
1382
1383         p = copy_process(clone_flags, stack_start, regs, stack_size,
1384                          child_tidptr, NULL, trace);
1385         /*
1386          * Do this prior waking up the new thread - the thread pointer
1387          * might get invalid after that point, if the thread exits quickly.
1388          */
1389         if (!IS_ERR(p)) {
1390                 struct completion vfork;
1391
1392                 trace_sched_process_fork(current, p);
1393
1394                 nr = task_pid_vnr(p);
1395
1396                 if (clone_flags & CLONE_PARENT_SETTID)
1397                         put_user(nr, parent_tidptr);
1398
1399                 if (clone_flags & CLONE_VFORK) {
1400                         p->vfork_done = &vfork;
1401                         init_completion(&vfork);
1402                 }
1403
1404                 audit_finish_fork(p);
1405                 tracehook_report_clone(regs, clone_flags, nr, p);
1406
1407                 /*
1408                  * We set PF_STARTING at creation in case tracing wants to
1409                  * use this to distinguish a fully live task from one that
1410                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1411                  * clear it and set the child going.
1412                  */
1413                 p->flags &= ~PF_STARTING;
1414
1415                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1416                         /*
1417                          * We'll start up with an immediate SIGSTOP.
1418                          */
1419                         sigaddset(&p->pending.signal, SIGSTOP);
1420                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1421                         __set_task_state(p, TASK_STOPPED);
1422                 } else {
1423                         wake_up_new_task(p, clone_flags);
1424                 }
1425
1426                 tracehook_report_clone_complete(trace, regs,
1427                                                 clone_flags, nr, p);
1428
1429                 if (clone_flags & CLONE_VFORK) {
1430                         freezer_do_not_count();
1431                         wait_for_completion(&vfork);
1432                         freezer_count();
1433                         tracehook_report_vfork_done(p, nr);
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(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                         SLAB_NOTRACK, sighand_ctor);
1459         signal_cachep = kmem_cache_create("signal_cache",
1460                         sizeof(struct signal_struct), 0,
1461                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1462         files_cachep = kmem_cache_create("files_cache",
1463                         sizeof(struct files_struct), 0,
1464                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1465         fs_cachep = kmem_cache_create("fs_cache",
1466                         sizeof(struct fs_struct), 0,
1467                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1468         mm_cachep = kmem_cache_create("mm_struct",
1469                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1470                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1471         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1472         mmap_init();
1473 }
1474
1475 /*
1476  * Check constraints on flags passed to the unshare system call and
1477  * force unsharing of additional process context as appropriate.
1478  */
1479 static void check_unshare_flags(unsigned long *flags_ptr)
1480 {
1481         /*
1482          * If unsharing a thread from a thread group, must also
1483          * unshare vm.
1484          */
1485         if (*flags_ptr & CLONE_THREAD)
1486                 *flags_ptr |= CLONE_VM;
1487
1488         /*
1489          * If unsharing vm, must also unshare signal handlers.
1490          */
1491         if (*flags_ptr & CLONE_VM)
1492                 *flags_ptr |= CLONE_SIGHAND;
1493
1494         /*
1495          * If unsharing signal handlers and the task was created
1496          * using CLONE_THREAD, then must unshare the thread
1497          */
1498         if ((*flags_ptr & CLONE_SIGHAND) &&
1499             (atomic_read(&current->signal->count) > 1))
1500                 *flags_ptr |= CLONE_THREAD;
1501
1502         /*
1503          * If unsharing namespace, must also unshare filesystem information.
1504          */
1505         if (*flags_ptr & CLONE_NEWNS)
1506                 *flags_ptr |= CLONE_FS;
1507 }
1508
1509 /*
1510  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1511  */
1512 static int unshare_thread(unsigned long unshare_flags)
1513 {
1514         if (unshare_flags & CLONE_THREAD)
1515                 return -EINVAL;
1516
1517         return 0;
1518 }
1519
1520 /*
1521  * Unshare the filesystem structure if it is being shared
1522  */
1523 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1524 {
1525         struct fs_struct *fs = current->fs;
1526
1527         if (!(unshare_flags & CLONE_FS) || !fs)
1528                 return 0;
1529
1530         /* don't need lock here; in the worst case we'll do useless copy */
1531         if (fs->users == 1)
1532                 return 0;
1533
1534         *new_fsp = copy_fs_struct(fs);
1535         if (!*new_fsp)
1536                 return -ENOMEM;
1537
1538         return 0;
1539 }
1540
1541 /*
1542  * Unsharing of sighand is not supported yet
1543  */
1544 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1545 {
1546         struct sighand_struct *sigh = current->sighand;
1547
1548         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1549                 return -EINVAL;
1550         else
1551                 return 0;
1552 }
1553
1554 /*
1555  * Unshare vm if it is being shared
1556  */
1557 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1558 {
1559         struct mm_struct *mm = current->mm;
1560
1561         if ((unshare_flags & CLONE_VM) &&
1562             (mm && atomic_read(&mm->mm_users) > 1)) {
1563                 return -EINVAL;
1564         }
1565
1566         return 0;
1567 }
1568
1569 /*
1570  * Unshare file descriptor table if it is being shared
1571  */
1572 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1573 {
1574         struct files_struct *fd = current->files;
1575         int error = 0;
1576
1577         if ((unshare_flags & CLONE_FILES) &&
1578             (fd && atomic_read(&fd->count) > 1)) {
1579                 *new_fdp = dup_fd(fd, &error);
1580                 if (!*new_fdp)
1581                         return error;
1582         }
1583
1584         return 0;
1585 }
1586
1587 /*
1588  * unshare allows a process to 'unshare' part of the process
1589  * context which was originally shared using clone.  copy_*
1590  * functions used by do_fork() cannot be used here directly
1591  * because they modify an inactive task_struct that is being
1592  * constructed. Here we are modifying the current, active,
1593  * task_struct.
1594  */
1595 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1596 {
1597         int err = 0;
1598         struct fs_struct *fs, *new_fs = NULL;
1599         struct sighand_struct *new_sigh = NULL;
1600         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1601         struct files_struct *fd, *new_fd = NULL;
1602         struct nsproxy *new_nsproxy = NULL;
1603         int do_sysvsem = 0;
1604
1605         check_unshare_flags(&unshare_flags);
1606
1607         /* Return -EINVAL for all unsupported flags */
1608         err = -EINVAL;
1609         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1610                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1611                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1612                 goto bad_unshare_out;
1613
1614         /*
1615          * CLONE_NEWIPC must also detach from the undolist: after switching
1616          * to a new ipc namespace, the semaphore arrays from the old
1617          * namespace are unreachable.
1618          */
1619         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1620                 do_sysvsem = 1;
1621         if ((err = unshare_thread(unshare_flags)))
1622                 goto bad_unshare_out;
1623         if ((err = unshare_fs(unshare_flags, &new_fs)))
1624                 goto bad_unshare_cleanup_thread;
1625         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1626                 goto bad_unshare_cleanup_fs;
1627         if ((err = unshare_vm(unshare_flags, &new_mm)))
1628                 goto bad_unshare_cleanup_sigh;
1629         if ((err = unshare_fd(unshare_flags, &new_fd)))
1630                 goto bad_unshare_cleanup_vm;
1631         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1632                         new_fs)))
1633                 goto bad_unshare_cleanup_fd;
1634
1635         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1636                 if (do_sysvsem) {
1637                         /*
1638                          * CLONE_SYSVSEM is equivalent to sys_exit().
1639                          */
1640                         exit_sem(current);
1641                 }
1642
1643                 if (new_nsproxy) {
1644                         switch_task_namespaces(current, new_nsproxy);
1645                         new_nsproxy = NULL;
1646                 }
1647
1648                 task_lock(current);
1649
1650                 if (new_fs) {
1651                         fs = current->fs;
1652                         write_lock(&fs->lock);
1653                         current->fs = new_fs;
1654                         if (--fs->users)
1655                                 new_fs = NULL;
1656                         else
1657                                 new_fs = fs;
1658                         write_unlock(&fs->lock);
1659                 }
1660
1661                 if (new_mm) {
1662                         mm = current->mm;
1663                         active_mm = current->active_mm;
1664                         current->mm = new_mm;
1665                         current->active_mm = new_mm;
1666                         activate_mm(active_mm, new_mm);
1667                         new_mm = mm;
1668                 }
1669
1670                 if (new_fd) {
1671                         fd = current->files;
1672                         current->files = new_fd;
1673                         new_fd = fd;
1674                 }
1675
1676                 task_unlock(current);
1677         }
1678
1679         if (new_nsproxy)
1680                 put_nsproxy(new_nsproxy);
1681
1682 bad_unshare_cleanup_fd:
1683         if (new_fd)
1684                 put_files_struct(new_fd);
1685
1686 bad_unshare_cleanup_vm:
1687         if (new_mm)
1688                 mmput(new_mm);
1689
1690 bad_unshare_cleanup_sigh:
1691         if (new_sigh)
1692                 if (atomic_dec_and_test(&new_sigh->count))
1693                         kmem_cache_free(sighand_cachep, new_sigh);
1694
1695 bad_unshare_cleanup_fs:
1696         if (new_fs)
1697                 free_fs_struct(new_fs);
1698
1699 bad_unshare_cleanup_thread:
1700 bad_unshare_out:
1701         return err;
1702 }
1703
1704 /*
1705  *      Helper to unshare the files of the current task.
1706  *      We don't want to expose copy_files internals to
1707  *      the exec layer of the kernel.
1708  */
1709
1710 int unshare_files(struct files_struct **displaced)
1711 {
1712         struct task_struct *task = current;
1713         struct files_struct *copy = NULL;
1714         int error;
1715
1716         error = unshare_fd(CLONE_FILES, &copy);
1717         if (error || !copy) {
1718                 *displaced = NULL;
1719                 return error;
1720         }
1721         *displaced = task->files;
1722         task_lock(task);
1723         task->files = copy;
1724         task_unlock(task);
1725         return 0;
1726 }