043b5d88049b0ef7bd83ba2e8b7fc8ce4058e672
[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 #ifdef CONFIG_COMPAT
549         if (unlikely(tsk->compat_robust_list))
550                 compat_exit_robust_list(tsk);
551 #endif
552 #endif
553
554         /* Get rid of any cached register state */
555         deactivate_mm(tsk, mm);
556
557         /* notify parent sleeping on vfork() */
558         if (vfork_done) {
559                 tsk->vfork_done = NULL;
560                 complete(vfork_done);
561         }
562
563         /*
564          * If we're exiting normally, clear a user-space tid field if
565          * requested.  We leave this alone when dying by signal, to leave
566          * the value intact in a core dump, and to save the unnecessary
567          * trouble otherwise.  Userland only wants this done for a sys_exit.
568          */
569         if (tsk->clear_child_tid) {
570                 if (!(tsk->flags & PF_SIGNALED) &&
571                     atomic_read(&mm->mm_users) > 1) {
572                         /*
573                          * We don't check the error code - if userspace has
574                          * not set up a proper pointer then tough luck.
575                          */
576                         put_user(0, tsk->clear_child_tid);
577                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
578                                         1, NULL, NULL, 0);
579                 }
580                 tsk->clear_child_tid = NULL;
581         }
582 }
583
584 /*
585  * Allocate a new mm structure and copy contents from the
586  * mm structure of the passed in task structure.
587  */
588 struct mm_struct *dup_mm(struct task_struct *tsk)
589 {
590         struct mm_struct *mm, *oldmm = current->mm;
591         int err;
592
593         if (!oldmm)
594                 return NULL;
595
596         mm = allocate_mm();
597         if (!mm)
598                 goto fail_nomem;
599
600         memcpy(mm, oldmm, sizeof(*mm));
601
602         /* Initializing for Swap token stuff */
603         mm->token_priority = 0;
604         mm->last_interval = 0;
605
606         if (!mm_init(mm, tsk))
607                 goto fail_nomem;
608
609         if (init_new_context(tsk, mm))
610                 goto fail_nocontext;
611
612         dup_mm_exe_file(oldmm, mm);
613
614         err = dup_mmap(mm, oldmm);
615         if (err)
616                 goto free_pt;
617
618         mm->hiwater_rss = get_mm_rss(mm);
619         mm->hiwater_vm = mm->total_vm;
620
621         return mm;
622
623 free_pt:
624         mmput(mm);
625
626 fail_nomem:
627         return NULL;
628
629 fail_nocontext:
630         /*
631          * If init_new_context() failed, we cannot use mmput() to free the mm
632          * because it calls destroy_context()
633          */
634         mm_free_pgd(mm);
635         free_mm(mm);
636         return NULL;
637 }
638
639 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
640 {
641         struct mm_struct * mm, *oldmm;
642         int retval;
643
644         tsk->min_flt = tsk->maj_flt = 0;
645         tsk->nvcsw = tsk->nivcsw = 0;
646 #ifdef CONFIG_DETECT_HUNG_TASK
647         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
648 #endif
649
650         tsk->mm = NULL;
651         tsk->active_mm = NULL;
652
653         /*
654          * Are we cloning a kernel thread?
655          *
656          * We need to steal a active VM for that..
657          */
658         oldmm = current->mm;
659         if (!oldmm)
660                 return 0;
661
662         if (clone_flags & CLONE_VM) {
663                 atomic_inc(&oldmm->mm_users);
664                 mm = oldmm;
665                 goto good_mm;
666         }
667
668         retval = -ENOMEM;
669         mm = dup_mm(tsk);
670         if (!mm)
671                 goto fail_nomem;
672
673 good_mm:
674         /* Initializing for Swap token stuff */
675         mm->token_priority = 0;
676         mm->last_interval = 0;
677
678         tsk->mm = mm;
679         tsk->active_mm = mm;
680         return 0;
681
682 fail_nomem:
683         return retval;
684 }
685
686 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
687 {
688         struct fs_struct *fs = current->fs;
689         if (clone_flags & CLONE_FS) {
690                 /* tsk->fs is already what we want */
691                 write_lock(&fs->lock);
692                 if (fs->in_exec) {
693                         write_unlock(&fs->lock);
694                         return -EAGAIN;
695                 }
696                 fs->users++;
697                 write_unlock(&fs->lock);
698                 return 0;
699         }
700         tsk->fs = copy_fs_struct(fs);
701         if (!tsk->fs)
702                 return -ENOMEM;
703         return 0;
704 }
705
706 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
707 {
708         struct files_struct *oldf, *newf;
709         int error = 0;
710
711         /*
712          * A background process may not have any files ...
713          */
714         oldf = current->files;
715         if (!oldf)
716                 goto out;
717
718         if (clone_flags & CLONE_FILES) {
719                 atomic_inc(&oldf->count);
720                 goto out;
721         }
722
723         newf = dup_fd(oldf, &error);
724         if (!newf)
725                 goto out;
726
727         tsk->files = newf;
728         error = 0;
729 out:
730         return error;
731 }
732
733 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
734 {
735 #ifdef CONFIG_BLOCK
736         struct io_context *ioc = current->io_context;
737
738         if (!ioc)
739                 return 0;
740         /*
741          * Share io context with parent, if CLONE_IO is set
742          */
743         if (clone_flags & CLONE_IO) {
744                 tsk->io_context = ioc_task_link(ioc);
745                 if (unlikely(!tsk->io_context))
746                         return -ENOMEM;
747         } else if (ioprio_valid(ioc->ioprio)) {
748                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
749                 if (unlikely(!tsk->io_context))
750                         return -ENOMEM;
751
752                 tsk->io_context->ioprio = ioc->ioprio;
753         }
754 #endif
755         return 0;
756 }
757
758 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
759 {
760         struct sighand_struct *sig;
761
762         if (clone_flags & CLONE_SIGHAND) {
763                 atomic_inc(&current->sighand->count);
764                 return 0;
765         }
766         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
767         rcu_assign_pointer(tsk->sighand, sig);
768         if (!sig)
769                 return -ENOMEM;
770         atomic_set(&sig->count, 1);
771         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
772         return 0;
773 }
774
775 void __cleanup_sighand(struct sighand_struct *sighand)
776 {
777         if (atomic_dec_and_test(&sighand->count))
778                 kmem_cache_free(sighand_cachep, sighand);
779 }
780
781
782 /*
783  * Initialize POSIX timer handling for a thread group.
784  */
785 static void posix_cpu_timers_init_group(struct signal_struct *sig)
786 {
787         /* Thread group counters. */
788         thread_group_cputime_init(sig);
789
790         /* Expiration times and increments. */
791         sig->it_virt_expires = cputime_zero;
792         sig->it_virt_incr = cputime_zero;
793         sig->it_prof_expires = cputime_zero;
794         sig->it_prof_incr = cputime_zero;
795
796         /* Cached expiration times. */
797         sig->cputime_expires.prof_exp = cputime_zero;
798         sig->cputime_expires.virt_exp = cputime_zero;
799         sig->cputime_expires.sched_exp = 0;
800
801         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
802                 sig->cputime_expires.prof_exp =
803                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
804                 sig->cputimer.running = 1;
805         }
806
807         /* The timer lists. */
808         INIT_LIST_HEAD(&sig->cpu_timers[0]);
809         INIT_LIST_HEAD(&sig->cpu_timers[1]);
810         INIT_LIST_HEAD(&sig->cpu_timers[2]);
811 }
812
813 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
814 {
815         struct signal_struct *sig;
816
817         if (clone_flags & CLONE_THREAD) {
818                 atomic_inc(&current->signal->count);
819                 atomic_inc(&current->signal->live);
820                 return 0;
821         }
822
823         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
824         tsk->signal = sig;
825         if (!sig)
826                 return -ENOMEM;
827
828         atomic_set(&sig->count, 1);
829         atomic_set(&sig->live, 1);
830         init_waitqueue_head(&sig->wait_chldexit);
831         sig->flags = 0;
832         if (clone_flags & CLONE_NEWPID)
833                 sig->flags |= SIGNAL_UNKILLABLE;
834         sig->group_exit_code = 0;
835         sig->group_exit_task = NULL;
836         sig->group_stop_count = 0;
837         sig->curr_target = tsk;
838         init_sigpending(&sig->shared_pending);
839         INIT_LIST_HEAD(&sig->posix_timers);
840
841         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
842         sig->it_real_incr.tv64 = 0;
843         sig->real_timer.function = it_real_fn;
844
845         sig->leader = 0;        /* session leadership doesn't inherit */
846         sig->tty_old_pgrp = NULL;
847         sig->tty = NULL;
848
849         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
850         sig->gtime = cputime_zero;
851         sig->cgtime = cputime_zero;
852         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
853         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
854         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
855         task_io_accounting_init(&sig->ioac);
856         sig->sum_sched_runtime = 0;
857         taskstats_tgid_init(sig);
858
859         task_lock(current->group_leader);
860         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
861         task_unlock(current->group_leader);
862
863         posix_cpu_timers_init_group(sig);
864
865         acct_init_pacct(&sig->pacct);
866
867         tty_audit_fork(sig);
868
869         return 0;
870 }
871
872 void __cleanup_signal(struct signal_struct *sig)
873 {
874         thread_group_cputime_free(sig);
875         tty_kref_put(sig->tty);
876         kmem_cache_free(signal_cachep, sig);
877 }
878
879 static void cleanup_signal(struct task_struct *tsk)
880 {
881         struct signal_struct *sig = tsk->signal;
882
883         atomic_dec(&sig->live);
884
885         if (atomic_dec_and_test(&sig->count))
886                 __cleanup_signal(sig);
887 }
888
889 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
890 {
891         unsigned long new_flags = p->flags;
892
893         new_flags &= ~PF_SUPERPRIV;
894         new_flags |= PF_FORKNOEXEC;
895         new_flags |= PF_STARTING;
896         p->flags = new_flags;
897         clear_freeze_flag(p);
898 }
899
900 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
901 {
902         current->clear_child_tid = tidptr;
903
904         return task_pid_vnr(current);
905 }
906
907 static void rt_mutex_init_task(struct task_struct *p)
908 {
909         spin_lock_init(&p->pi_lock);
910 #ifdef CONFIG_RT_MUTEXES
911         plist_head_init(&p->pi_waiters, &p->pi_lock);
912         p->pi_blocked_on = NULL;
913 #endif
914 }
915
916 #ifdef CONFIG_MM_OWNER
917 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
918 {
919         mm->owner = p;
920 }
921 #endif /* CONFIG_MM_OWNER */
922
923 /*
924  * Initialize POSIX timer handling for a single task.
925  */
926 static void posix_cpu_timers_init(struct task_struct *tsk)
927 {
928         tsk->cputime_expires.prof_exp = cputime_zero;
929         tsk->cputime_expires.virt_exp = cputime_zero;
930         tsk->cputime_expires.sched_exp = 0;
931         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
932         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
933         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
934 }
935
936 /*
937  * This creates a new process as a copy of the old one,
938  * but does not actually start it yet.
939  *
940  * It copies the registers, and all the appropriate
941  * parts of the process environment (as per the clone
942  * flags). The actual kick-off is left to the caller.
943  */
944 static struct task_struct *copy_process(unsigned long clone_flags,
945                                         unsigned long stack_start,
946                                         struct pt_regs *regs,
947                                         unsigned long stack_size,
948                                         int __user *child_tidptr,
949                                         struct pid *pid,
950                                         int trace)
951 {
952         int retval;
953         struct task_struct *p;
954         int cgroup_callbacks_done = 0;
955
956         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
957                 return ERR_PTR(-EINVAL);
958
959         /*
960          * Thread groups must share signals as well, and detached threads
961          * can only be started up within the thread group.
962          */
963         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
964                 return ERR_PTR(-EINVAL);
965
966         /*
967          * Shared signal handlers imply shared VM. By way of the above,
968          * thread groups also imply shared VM. Blocking this case allows
969          * for various simplifications in other code.
970          */
971         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
972                 return ERR_PTR(-EINVAL);
973
974         retval = security_task_create(clone_flags);
975         if (retval)
976                 goto fork_out;
977
978         retval = -ENOMEM;
979         p = dup_task_struct(current);
980         if (!p)
981                 goto fork_out;
982
983         ftrace_graph_init_task(p);
984
985         rt_mutex_init_task(p);
986
987 #ifdef CONFIG_PROVE_LOCKING
988         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
989         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
990 #endif
991         retval = -EAGAIN;
992         if (atomic_read(&p->real_cred->user->processes) >=
993                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
994                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
995                     p->real_cred->user != INIT_USER)
996                         goto bad_fork_free;
997         }
998
999         retval = copy_creds(p, clone_flags);
1000         if (retval < 0)
1001                 goto bad_fork_free;
1002
1003         /*
1004          * If multiple threads are within copy_process(), then this check
1005          * triggers too late. This doesn't hurt, the check is only there
1006          * to stop root fork bombs.
1007          */
1008         retval = -EAGAIN;
1009         if (nr_threads >= max_threads)
1010                 goto bad_fork_cleanup_count;
1011
1012         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1013                 goto bad_fork_cleanup_count;
1014
1015         if (p->binfmt && !try_module_get(p->binfmt->module))
1016                 goto bad_fork_cleanup_put_domain;
1017
1018         p->did_exec = 0;
1019         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1020         copy_flags(clone_flags, p);
1021         INIT_LIST_HEAD(&p->children);
1022         INIT_LIST_HEAD(&p->sibling);
1023 #ifdef CONFIG_PREEMPT_RCU
1024         p->rcu_read_lock_nesting = 0;
1025         p->rcu_flipctr_idx = 0;
1026 #endif /* #ifdef CONFIG_PREEMPT_RCU */
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_counter_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                 p->group_leader = current->group_leader;
1242                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1243         }
1244
1245         if (likely(p->pid)) {
1246                 list_add_tail(&p->sibling, &p->real_parent->children);
1247                 tracehook_finish_clone(p, clone_flags, trace);
1248
1249                 if (thread_group_leader(p)) {
1250                         if (clone_flags & CLONE_NEWPID)
1251                                 p->nsproxy->pid_ns->child_reaper = p;
1252
1253                         p->signal->leader_pid = pid;
1254                         tty_kref_put(p->signal->tty);
1255                         p->signal->tty = tty_kref_get(current->signal->tty);
1256                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1257                         attach_pid(p, PIDTYPE_SID, task_session(current));
1258                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1259                         __get_cpu_var(process_counts)++;
1260                 }
1261                 attach_pid(p, PIDTYPE_PID, pid);
1262                 nr_threads++;
1263         }
1264
1265         total_forks++;
1266         spin_unlock(&current->sighand->siglock);
1267         write_unlock_irq(&tasklist_lock);
1268         proc_fork_connector(p);
1269         cgroup_post_fork(p);
1270         perf_counter_fork(p);
1271         return p;
1272
1273 bad_fork_free_pid:
1274         if (pid != &init_struct_pid)
1275                 free_pid(pid);
1276 bad_fork_cleanup_io:
1277         put_io_context(p->io_context);
1278 bad_fork_cleanup_namespaces:
1279         exit_task_namespaces(p);
1280 bad_fork_cleanup_mm:
1281         if (p->mm)
1282                 mmput(p->mm);
1283 bad_fork_cleanup_signal:
1284         cleanup_signal(p);
1285 bad_fork_cleanup_sighand:
1286         __cleanup_sighand(p->sighand);
1287 bad_fork_cleanup_fs:
1288         exit_fs(p); /* blocking */
1289 bad_fork_cleanup_files:
1290         exit_files(p); /* blocking */
1291 bad_fork_cleanup_semundo:
1292         exit_sem(p);
1293 bad_fork_cleanup_audit:
1294         audit_free(p);
1295 bad_fork_cleanup_policy:
1296         perf_counter_free_task(p);
1297 #ifdef CONFIG_NUMA
1298         mpol_put(p->mempolicy);
1299 bad_fork_cleanup_cgroup:
1300 #endif
1301         cgroup_exit(p, cgroup_callbacks_done);
1302         delayacct_tsk_free(p);
1303         if (p->binfmt)
1304                 module_put(p->binfmt->module);
1305 bad_fork_cleanup_put_domain:
1306         module_put(task_thread_info(p)->exec_domain->module);
1307 bad_fork_cleanup_count:
1308         atomic_dec(&p->cred->user->processes);
1309         exit_creds(p);
1310 bad_fork_free:
1311         free_task(p);
1312 fork_out:
1313         return ERR_PTR(retval);
1314 }
1315
1316 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1317 {
1318         memset(regs, 0, sizeof(struct pt_regs));
1319         return regs;
1320 }
1321
1322 struct task_struct * __cpuinit fork_idle(int cpu)
1323 {
1324         struct task_struct *task;
1325         struct pt_regs regs;
1326
1327         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1328                             &init_struct_pid, 0);
1329         if (!IS_ERR(task))
1330                 init_idle(task, cpu);
1331
1332         return task;
1333 }
1334
1335 /*
1336  *  Ok, this is the main fork-routine.
1337  *
1338  * It copies the process, and if successful kick-starts
1339  * it and waits for it to finish using the VM if required.
1340  */
1341 long do_fork(unsigned long clone_flags,
1342               unsigned long stack_start,
1343               struct pt_regs *regs,
1344               unsigned long stack_size,
1345               int __user *parent_tidptr,
1346               int __user *child_tidptr)
1347 {
1348         struct task_struct *p;
1349         int trace = 0;
1350         long nr;
1351
1352         /*
1353          * Do some preliminary argument and permissions checking before we
1354          * actually start allocating stuff
1355          */
1356         if (clone_flags & CLONE_NEWUSER) {
1357                 if (clone_flags & CLONE_THREAD)
1358                         return -EINVAL;
1359                 /* hopefully this check will go away when userns support is
1360                  * complete
1361                  */
1362                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1363                                 !capable(CAP_SETGID))
1364                         return -EPERM;
1365         }
1366
1367         /*
1368          * We hope to recycle these flags after 2.6.26
1369          */
1370         if (unlikely(clone_flags & CLONE_STOPPED)) {
1371                 static int __read_mostly count = 100;
1372
1373                 if (count > 0 && printk_ratelimit()) {
1374                         char comm[TASK_COMM_LEN];
1375
1376                         count--;
1377                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1378                                         "clone flags 0x%lx\n",
1379                                 get_task_comm(comm, current),
1380                                 clone_flags & CLONE_STOPPED);
1381                 }
1382         }
1383
1384         /*
1385          * When called from kernel_thread, don't do user tracing stuff.
1386          */
1387         if (likely(user_mode(regs)))
1388                 trace = tracehook_prepare_clone(clone_flags);
1389
1390         p = copy_process(clone_flags, stack_start, regs, stack_size,
1391                          child_tidptr, NULL, trace);
1392         /*
1393          * Do this prior waking up the new thread - the thread pointer
1394          * might get invalid after that point, if the thread exits quickly.
1395          */
1396         if (!IS_ERR(p)) {
1397                 struct completion vfork;
1398
1399                 trace_sched_process_fork(current, p);
1400
1401                 nr = task_pid_vnr(p);
1402
1403                 if (clone_flags & CLONE_PARENT_SETTID)
1404                         put_user(nr, parent_tidptr);
1405
1406                 if (clone_flags & CLONE_VFORK) {
1407                         p->vfork_done = &vfork;
1408                         init_completion(&vfork);
1409                 }
1410
1411                 audit_finish_fork(p);
1412                 tracehook_report_clone(regs, clone_flags, nr, p);
1413
1414                 /*
1415                  * We set PF_STARTING at creation in case tracing wants to
1416                  * use this to distinguish a fully live task from one that
1417                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1418                  * clear it and set the child going.
1419                  */
1420                 p->flags &= ~PF_STARTING;
1421
1422                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1423                         /*
1424                          * We'll start up with an immediate SIGSTOP.
1425                          */
1426                         sigaddset(&p->pending.signal, SIGSTOP);
1427                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1428                         __set_task_state(p, TASK_STOPPED);
1429                 } else {
1430                         wake_up_new_task(p, clone_flags);
1431                 }
1432
1433                 tracehook_report_clone_complete(trace, regs,
1434                                                 clone_flags, nr, p);
1435
1436                 if (clone_flags & CLONE_VFORK) {
1437                         freezer_do_not_count();
1438                         wait_for_completion(&vfork);
1439                         freezer_count();
1440                         tracehook_report_vfork_done(p, nr);
1441                 }
1442         } else {
1443                 nr = PTR_ERR(p);
1444         }
1445         return nr;
1446 }
1447
1448 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1449 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1450 #endif
1451
1452 static void sighand_ctor(void *data)
1453 {
1454         struct sighand_struct *sighand = data;
1455
1456         spin_lock_init(&sighand->siglock);
1457         init_waitqueue_head(&sighand->signalfd_wqh);
1458 }
1459
1460 void __init proc_caches_init(void)
1461 {
1462         sighand_cachep = kmem_cache_create("sighand_cache",
1463                         sizeof(struct sighand_struct), 0,
1464                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1465                         SLAB_NOTRACK, sighand_ctor);
1466         signal_cachep = kmem_cache_create("signal_cache",
1467                         sizeof(struct signal_struct), 0,
1468                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1469         files_cachep = kmem_cache_create("files_cache",
1470                         sizeof(struct files_struct), 0,
1471                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1472         fs_cachep = kmem_cache_create("fs_cache",
1473                         sizeof(struct fs_struct), 0,
1474                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1475         mm_cachep = kmem_cache_create("mm_struct",
1476                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1477                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1478         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1479         mmap_init();
1480 }
1481
1482 /*
1483  * Check constraints on flags passed to the unshare system call and
1484  * force unsharing of additional process context as appropriate.
1485  */
1486 static void check_unshare_flags(unsigned long *flags_ptr)
1487 {
1488         /*
1489          * If unsharing a thread from a thread group, must also
1490          * unshare vm.
1491          */
1492         if (*flags_ptr & CLONE_THREAD)
1493                 *flags_ptr |= CLONE_VM;
1494
1495         /*
1496          * If unsharing vm, must also unshare signal handlers.
1497          */
1498         if (*flags_ptr & CLONE_VM)
1499                 *flags_ptr |= CLONE_SIGHAND;
1500
1501         /*
1502          * If unsharing signal handlers and the task was created
1503          * using CLONE_THREAD, then must unshare the thread
1504          */
1505         if ((*flags_ptr & CLONE_SIGHAND) &&
1506             (atomic_read(&current->signal->count) > 1))
1507                 *flags_ptr |= CLONE_THREAD;
1508
1509         /*
1510          * If unsharing namespace, must also unshare filesystem information.
1511          */
1512         if (*flags_ptr & CLONE_NEWNS)
1513                 *flags_ptr |= CLONE_FS;
1514 }
1515
1516 /*
1517  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1518  */
1519 static int unshare_thread(unsigned long unshare_flags)
1520 {
1521         if (unshare_flags & CLONE_THREAD)
1522                 return -EINVAL;
1523
1524         return 0;
1525 }
1526
1527 /*
1528  * Unshare the filesystem structure if it is being shared
1529  */
1530 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1531 {
1532         struct fs_struct *fs = current->fs;
1533
1534         if (!(unshare_flags & CLONE_FS) || !fs)
1535                 return 0;
1536
1537         /* don't need lock here; in the worst case we'll do useless copy */
1538         if (fs->users == 1)
1539                 return 0;
1540
1541         *new_fsp = copy_fs_struct(fs);
1542         if (!*new_fsp)
1543                 return -ENOMEM;
1544
1545         return 0;
1546 }
1547
1548 /*
1549  * Unsharing of sighand is not supported yet
1550  */
1551 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1552 {
1553         struct sighand_struct *sigh = current->sighand;
1554
1555         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1556                 return -EINVAL;
1557         else
1558                 return 0;
1559 }
1560
1561 /*
1562  * Unshare vm if it is being shared
1563  */
1564 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1565 {
1566         struct mm_struct *mm = current->mm;
1567
1568         if ((unshare_flags & CLONE_VM) &&
1569             (mm && atomic_read(&mm->mm_users) > 1)) {
1570                 return -EINVAL;
1571         }
1572
1573         return 0;
1574 }
1575
1576 /*
1577  * Unshare file descriptor table if it is being shared
1578  */
1579 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1580 {
1581         struct files_struct *fd = current->files;
1582         int error = 0;
1583
1584         if ((unshare_flags & CLONE_FILES) &&
1585             (fd && atomic_read(&fd->count) > 1)) {
1586                 *new_fdp = dup_fd(fd, &error);
1587                 if (!*new_fdp)
1588                         return error;
1589         }
1590
1591         return 0;
1592 }
1593
1594 /*
1595  * unshare allows a process to 'unshare' part of the process
1596  * context which was originally shared using clone.  copy_*
1597  * functions used by do_fork() cannot be used here directly
1598  * because they modify an inactive task_struct that is being
1599  * constructed. Here we are modifying the current, active,
1600  * task_struct.
1601  */
1602 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1603 {
1604         int err = 0;
1605         struct fs_struct *fs, *new_fs = NULL;
1606         struct sighand_struct *new_sigh = NULL;
1607         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1608         struct files_struct *fd, *new_fd = NULL;
1609         struct nsproxy *new_nsproxy = NULL;
1610         int do_sysvsem = 0;
1611
1612         check_unshare_flags(&unshare_flags);
1613
1614         /* Return -EINVAL for all unsupported flags */
1615         err = -EINVAL;
1616         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1617                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1618                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1619                 goto bad_unshare_out;
1620
1621         /*
1622          * CLONE_NEWIPC must also detach from the undolist: after switching
1623          * to a new ipc namespace, the semaphore arrays from the old
1624          * namespace are unreachable.
1625          */
1626         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1627                 do_sysvsem = 1;
1628         if ((err = unshare_thread(unshare_flags)))
1629                 goto bad_unshare_out;
1630         if ((err = unshare_fs(unshare_flags, &new_fs)))
1631                 goto bad_unshare_cleanup_thread;
1632         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1633                 goto bad_unshare_cleanup_fs;
1634         if ((err = unshare_vm(unshare_flags, &new_mm)))
1635                 goto bad_unshare_cleanup_sigh;
1636         if ((err = unshare_fd(unshare_flags, &new_fd)))
1637                 goto bad_unshare_cleanup_vm;
1638         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1639                         new_fs)))
1640                 goto bad_unshare_cleanup_fd;
1641
1642         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1643                 if (do_sysvsem) {
1644                         /*
1645                          * CLONE_SYSVSEM is equivalent to sys_exit().
1646                          */
1647                         exit_sem(current);
1648                 }
1649
1650                 if (new_nsproxy) {
1651                         switch_task_namespaces(current, new_nsproxy);
1652                         new_nsproxy = NULL;
1653                 }
1654
1655                 task_lock(current);
1656
1657                 if (new_fs) {
1658                         fs = current->fs;
1659                         write_lock(&fs->lock);
1660                         current->fs = new_fs;
1661                         if (--fs->users)
1662                                 new_fs = NULL;
1663                         else
1664                                 new_fs = fs;
1665                         write_unlock(&fs->lock);
1666                 }
1667
1668                 if (new_mm) {
1669                         mm = current->mm;
1670                         active_mm = current->active_mm;
1671                         current->mm = new_mm;
1672                         current->active_mm = new_mm;
1673                         activate_mm(active_mm, new_mm);
1674                         new_mm = mm;
1675                 }
1676
1677                 if (new_fd) {
1678                         fd = current->files;
1679                         current->files = new_fd;
1680                         new_fd = fd;
1681                 }
1682
1683                 task_unlock(current);
1684         }
1685
1686         if (new_nsproxy)
1687                 put_nsproxy(new_nsproxy);
1688
1689 bad_unshare_cleanup_fd:
1690         if (new_fd)
1691                 put_files_struct(new_fd);
1692
1693 bad_unshare_cleanup_vm:
1694         if (new_mm)
1695                 mmput(new_mm);
1696
1697 bad_unshare_cleanup_sigh:
1698         if (new_sigh)
1699                 if (atomic_dec_and_test(&new_sigh->count))
1700                         kmem_cache_free(sighand_cachep, new_sigh);
1701
1702 bad_unshare_cleanup_fs:
1703         if (new_fs)
1704                 free_fs_struct(new_fs);
1705
1706 bad_unshare_cleanup_thread:
1707 bad_unshare_out:
1708         return err;
1709 }
1710
1711 /*
1712  *      Helper to unshare the files of the current task.
1713  *      We don't want to expose copy_files internals to
1714  *      the exec layer of the kernel.
1715  */
1716
1717 int unshare_files(struct files_struct **displaced)
1718 {
1719         struct task_struct *task = current;
1720         struct files_struct *copy = NULL;
1721         int error;
1722
1723         error = unshare_fd(CLONE_FILES, &copy);
1724         if (error || !copy) {
1725                 *displaced = NULL;
1726                 return error;
1727         }
1728         *displaced = task->files;
1729         task_lock(task);
1730         task->files = copy;
1731         task_unlock(task);
1732         return 0;
1733 }