aio: ifdef fields in mm_struct
[linux-2.6.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/tracehook.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/task_io_accounting_ops.h>
44 #include <linux/rcupdate.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/audit.h>
48 #include <linux/memcontrol.h>
49 #include <linux/ftrace.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/ksm.h>
53 #include <linux/acct.h>
54 #include <linux/tsacct_kern.h>
55 #include <linux/cn_proc.h>
56 #include <linux/freezer.h>
57 #include <linux/delayacct.h>
58 #include <linux/taskstats_kern.h>
59 #include <linux/random.h>
60 #include <linux/tty.h>
61 #include <linux/proc_fs.h>
62 #include <linux/blkdev.h>
63 #include <linux/fs_struct.h>
64 #include <linux/magic.h>
65 #include <linux/perf_event.h>
66 #include <linux/posix-timers.h>
67
68 #include <asm/pgtable.h>
69 #include <asm/pgalloc.h>
70 #include <asm/uaccess.h>
71 #include <asm/mmu_context.h>
72 #include <asm/cacheflush.h>
73 #include <asm/tlbflush.h>
74
75 #include <trace/events/sched.h>
76
77 /*
78  * Protected counters by write_lock_irq(&tasklist_lock)
79  */
80 unsigned long total_forks;      /* Handle normal Linux uptimes. */
81 int nr_threads;                 /* The idle threads do not count.. */
82
83 int max_threads;                /* tunable limit on nr_threads */
84
85 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
86
87 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
88
89 int nr_processes(void)
90 {
91         int cpu;
92         int total = 0;
93
94         for_each_online_cpu(cpu)
95                 total += per_cpu(process_counts, cpu);
96
97         return total;
98 }
99
100 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
101 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
102 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
103 static struct kmem_cache *task_struct_cachep;
104 #endif
105
106 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
107 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
108 {
109 #ifdef CONFIG_DEBUG_STACK_USAGE
110         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
111 #else
112         gfp_t mask = GFP_KERNEL;
113 #endif
114         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
115 }
116
117 static inline void free_thread_info(struct thread_info *ti)
118 {
119         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
120 }
121 #endif
122
123 /* SLAB cache for signal_struct structures (tsk->signal) */
124 static struct kmem_cache *signal_cachep;
125
126 /* SLAB cache for sighand_struct structures (tsk->sighand) */
127 struct kmem_cache *sighand_cachep;
128
129 /* SLAB cache for files_struct structures (tsk->files) */
130 struct kmem_cache *files_cachep;
131
132 /* SLAB cache for fs_struct structures (tsk->fs) */
133 struct kmem_cache *fs_cachep;
134
135 /* SLAB cache for vm_area_struct structures */
136 struct kmem_cache *vm_area_cachep;
137
138 /* SLAB cache for mm_struct structures (tsk->mm) */
139 static struct kmem_cache *mm_cachep;
140
141 static void account_kernel_stack(struct thread_info *ti, int account)
142 {
143         struct zone *zone = page_zone(virt_to_page(ti));
144
145         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
146 }
147
148 void free_task(struct task_struct *tsk)
149 {
150         prop_local_destroy_single(&tsk->dirties);
151         account_kernel_stack(tsk->stack, -1);
152         free_thread_info(tsk->stack);
153         rt_mutex_debug_task_free(tsk);
154         ftrace_graph_exit_task(tsk);
155         free_task_struct(tsk);
156 }
157 EXPORT_SYMBOL(free_task);
158
159 void __put_task_struct(struct task_struct *tsk)
160 {
161         WARN_ON(!tsk->exit_state);
162         WARN_ON(atomic_read(&tsk->usage));
163         WARN_ON(tsk == current);
164
165         exit_creds(tsk);
166         delayacct_tsk_free(tsk);
167
168         if (!profile_handoff_task(tsk))
169                 free_task(tsk);
170 }
171
172 /*
173  * macro override instead of weak attribute alias, to workaround
174  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
175  */
176 #ifndef arch_task_cache_init
177 #define arch_task_cache_init()
178 #endif
179
180 void __init fork_init(unsigned long mempages)
181 {
182 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
183 #ifndef ARCH_MIN_TASKALIGN
184 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
185 #endif
186         /* create a slab on which task_structs can be allocated */
187         task_struct_cachep =
188                 kmem_cache_create("task_struct", sizeof(struct task_struct),
189                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
190 #endif
191
192         /* do the arch specific task caches init */
193         arch_task_cache_init();
194
195         /*
196          * The default maximum number of threads is set to a safe
197          * value: the thread structures can take up at most half
198          * of memory.
199          */
200         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
201
202         /*
203          * we need to allow at least 20 threads to boot a system
204          */
205         if(max_threads < 20)
206                 max_threads = 20;
207
208         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
209         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
210         init_task.signal->rlim[RLIMIT_SIGPENDING] =
211                 init_task.signal->rlim[RLIMIT_NPROC];
212 }
213
214 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
215                                                struct task_struct *src)
216 {
217         *dst = *src;
218         return 0;
219 }
220
221 static struct task_struct *dup_task_struct(struct task_struct *orig)
222 {
223         struct task_struct *tsk;
224         struct thread_info *ti;
225         unsigned long *stackend;
226
227         int err;
228
229         prepare_to_copy(orig);
230
231         tsk = alloc_task_struct();
232         if (!tsk)
233                 return NULL;
234
235         ti = alloc_thread_info(tsk);
236         if (!ti) {
237                 free_task_struct(tsk);
238                 return NULL;
239         }
240
241         err = arch_dup_task_struct(tsk, orig);
242         if (err)
243                 goto out;
244
245         tsk->stack = ti;
246
247         err = prop_local_init_single(&tsk->dirties);
248         if (err)
249                 goto out;
250
251         setup_thread_stack(tsk, orig);
252         stackend = end_of_stack(tsk);
253         *stackend = STACK_END_MAGIC;    /* for overflow detection */
254
255 #ifdef CONFIG_CC_STACKPROTECTOR
256         tsk->stack_canary = get_random_int();
257 #endif
258
259         /* One for us, one for whoever does the "release_task()" (usually parent) */
260         atomic_set(&tsk->usage,2);
261         atomic_set(&tsk->fs_excl, 0);
262 #ifdef CONFIG_BLK_DEV_IO_TRACE
263         tsk->btrace_seq = 0;
264 #endif
265         tsk->splice_pipe = NULL;
266
267         account_kernel_stack(ti, 1);
268
269         return tsk;
270
271 out:
272         free_thread_info(ti);
273         free_task_struct(tsk);
274         return NULL;
275 }
276
277 #ifdef CONFIG_MMU
278 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
279 {
280         struct vm_area_struct *mpnt, *tmp, **pprev;
281         struct rb_node **rb_link, *rb_parent;
282         int retval;
283         unsigned long charge;
284         struct mempolicy *pol;
285
286         down_write(&oldmm->mmap_sem);
287         flush_cache_dup_mm(oldmm);
288         /*
289          * Not linked in yet - no deadlock potential:
290          */
291         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
292
293         mm->locked_vm = 0;
294         mm->mmap = NULL;
295         mm->mmap_cache = NULL;
296         mm->free_area_cache = oldmm->mmap_base;
297         mm->cached_hole_size = ~0UL;
298         mm->map_count = 0;
299         cpumask_clear(mm_cpumask(mm));
300         mm->mm_rb = RB_ROOT;
301         rb_link = &mm->mm_rb.rb_node;
302         rb_parent = NULL;
303         pprev = &mm->mmap;
304         retval = ksm_fork(mm, oldmm);
305         if (retval)
306                 goto out;
307
308         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
309                 struct file *file;
310
311                 if (mpnt->vm_flags & VM_DONTCOPY) {
312                         long pages = vma_pages(mpnt);
313                         mm->total_vm -= pages;
314                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
315                                                                 -pages);
316                         continue;
317                 }
318                 charge = 0;
319                 if (mpnt->vm_flags & VM_ACCOUNT) {
320                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
321                         if (security_vm_enough_memory(len))
322                                 goto fail_nomem;
323                         charge = len;
324                 }
325                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
326                 if (!tmp)
327                         goto fail_nomem;
328                 *tmp = *mpnt;
329                 pol = mpol_dup(vma_policy(mpnt));
330                 retval = PTR_ERR(pol);
331                 if (IS_ERR(pol))
332                         goto fail_nomem_policy;
333                 vma_set_policy(tmp, pol);
334                 tmp->vm_flags &= ~VM_LOCKED;
335                 tmp->vm_mm = mm;
336                 tmp->vm_next = NULL;
337                 anon_vma_link(tmp);
338                 file = tmp->vm_file;
339                 if (file) {
340                         struct inode *inode = file->f_path.dentry->d_inode;
341                         struct address_space *mapping = file->f_mapping;
342
343                         get_file(file);
344                         if (tmp->vm_flags & VM_DENYWRITE)
345                                 atomic_dec(&inode->i_writecount);
346                         spin_lock(&mapping->i_mmap_lock);
347                         if (tmp->vm_flags & VM_SHARED)
348                                 mapping->i_mmap_writable++;
349                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
350                         flush_dcache_mmap_lock(mapping);
351                         /* insert tmp into the share list, just after mpnt */
352                         vma_prio_tree_add(tmp, mpnt);
353                         flush_dcache_mmap_unlock(mapping);
354                         spin_unlock(&mapping->i_mmap_lock);
355                 }
356
357                 /*
358                  * Clear hugetlb-related page reserves for children. This only
359                  * affects MAP_PRIVATE mappings. Faults generated by the child
360                  * are not guaranteed to succeed, even if read-only
361                  */
362                 if (is_vm_hugetlb_page(tmp))
363                         reset_vma_resv_huge_pages(tmp);
364
365                 /*
366                  * Link in the new vma and copy the page table entries.
367                  */
368                 *pprev = tmp;
369                 pprev = &tmp->vm_next;
370
371                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
372                 rb_link = &tmp->vm_rb.rb_right;
373                 rb_parent = &tmp->vm_rb;
374
375                 mm->map_count++;
376                 retval = copy_page_range(mm, oldmm, mpnt);
377
378                 if (tmp->vm_ops && tmp->vm_ops->open)
379                         tmp->vm_ops->open(tmp);
380
381                 if (retval)
382                         goto out;
383         }
384         /* a new mm has just been created */
385         arch_dup_mmap(oldmm, mm);
386         retval = 0;
387 out:
388         up_write(&mm->mmap_sem);
389         flush_tlb_mm(oldmm);
390         up_write(&oldmm->mmap_sem);
391         return retval;
392 fail_nomem_policy:
393         kmem_cache_free(vm_area_cachep, tmp);
394 fail_nomem:
395         retval = -ENOMEM;
396         vm_unacct_memory(charge);
397         goto out;
398 }
399
400 static inline int mm_alloc_pgd(struct mm_struct * mm)
401 {
402         mm->pgd = pgd_alloc(mm);
403         if (unlikely(!mm->pgd))
404                 return -ENOMEM;
405         return 0;
406 }
407
408 static inline void mm_free_pgd(struct mm_struct * mm)
409 {
410         pgd_free(mm, mm->pgd);
411 }
412 #else
413 #define dup_mmap(mm, oldmm)     (0)
414 #define mm_alloc_pgd(mm)        (0)
415 #define mm_free_pgd(mm)
416 #endif /* CONFIG_MMU */
417
418 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
419
420 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
421 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
422
423 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
424
425 static int __init coredump_filter_setup(char *s)
426 {
427         default_dump_filter =
428                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
429                 MMF_DUMP_FILTER_MASK;
430         return 1;
431 }
432
433 __setup("coredump_filter=", coredump_filter_setup);
434
435 #include <linux/init_task.h>
436
437 static void mm_init_aio(struct mm_struct *mm)
438 {
439 #ifdef CONFIG_AIO
440         spin_lock_init(&mm->ioctx_lock);
441         INIT_HLIST_HEAD(&mm->ioctx_list);
442 #endif
443 }
444
445 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
446 {
447         atomic_set(&mm->mm_users, 1);
448         atomic_set(&mm->mm_count, 1);
449         init_rwsem(&mm->mmap_sem);
450         INIT_LIST_HEAD(&mm->mmlist);
451         mm->flags = (current->mm) ?
452                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
453         mm->core_state = NULL;
454         mm->nr_ptes = 0;
455         set_mm_counter(mm, file_rss, 0);
456         set_mm_counter(mm, anon_rss, 0);
457         spin_lock_init(&mm->page_table_lock);
458         mm->free_area_cache = TASK_UNMAPPED_BASE;
459         mm->cached_hole_size = ~0UL;
460         mm_init_aio(mm);
461         mm_init_owner(mm, p);
462
463         if (likely(!mm_alloc_pgd(mm))) {
464                 mm->def_flags = 0;
465                 mmu_notifier_mm_init(mm);
466                 return mm;
467         }
468
469         free_mm(mm);
470         return NULL;
471 }
472
473 /*
474  * Allocate and initialize an mm_struct.
475  */
476 struct mm_struct * mm_alloc(void)
477 {
478         struct mm_struct * mm;
479
480         mm = allocate_mm();
481         if (mm) {
482                 memset(mm, 0, sizeof(*mm));
483                 mm = mm_init(mm, current);
484         }
485         return mm;
486 }
487
488 /*
489  * Called when the last reference to the mm
490  * is dropped: either by a lazy thread or by
491  * mmput. Free the page directory and the mm.
492  */
493 void __mmdrop(struct mm_struct *mm)
494 {
495         BUG_ON(mm == &init_mm);
496         mm_free_pgd(mm);
497         destroy_context(mm);
498         mmu_notifier_mm_destroy(mm);
499         free_mm(mm);
500 }
501 EXPORT_SYMBOL_GPL(__mmdrop);
502
503 /*
504  * Decrement the use count and release all resources for an mm.
505  */
506 void mmput(struct mm_struct *mm)
507 {
508         might_sleep();
509
510         if (atomic_dec_and_test(&mm->mm_users)) {
511                 exit_aio(mm);
512                 ksm_exit(mm);
513                 exit_mmap(mm);
514                 set_mm_exe_file(mm, NULL);
515                 if (!list_empty(&mm->mmlist)) {
516                         spin_lock(&mmlist_lock);
517                         list_del(&mm->mmlist);
518                         spin_unlock(&mmlist_lock);
519                 }
520                 put_swap_token(mm);
521                 mmdrop(mm);
522         }
523 }
524 EXPORT_SYMBOL_GPL(mmput);
525
526 /**
527  * get_task_mm - acquire a reference to the task's mm
528  *
529  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
530  * this kernel workthread has transiently adopted a user mm with use_mm,
531  * to do its AIO) is not set and if so returns a reference to it, after
532  * bumping up the use count.  User must release the mm via mmput()
533  * after use.  Typically used by /proc and ptrace.
534  */
535 struct mm_struct *get_task_mm(struct task_struct *task)
536 {
537         struct mm_struct *mm;
538
539         task_lock(task);
540         mm = task->mm;
541         if (mm) {
542                 if (task->flags & PF_KTHREAD)
543                         mm = NULL;
544                 else
545                         atomic_inc(&mm->mm_users);
546         }
547         task_unlock(task);
548         return mm;
549 }
550 EXPORT_SYMBOL_GPL(get_task_mm);
551
552 /* Please note the differences between mmput and mm_release.
553  * mmput is called whenever we stop holding onto a mm_struct,
554  * error success whatever.
555  *
556  * mm_release is called after a mm_struct has been removed
557  * from the current process.
558  *
559  * This difference is important for error handling, when we
560  * only half set up a mm_struct for a new process and need to restore
561  * the old one.  Because we mmput the new mm_struct before
562  * restoring the old one. . .
563  * Eric Biederman 10 January 1998
564  */
565 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
566 {
567         struct completion *vfork_done = tsk->vfork_done;
568
569         /* Get rid of any futexes when releasing the mm */
570 #ifdef CONFIG_FUTEX
571         if (unlikely(tsk->robust_list))
572                 exit_robust_list(tsk);
573 #ifdef CONFIG_COMPAT
574         if (unlikely(tsk->compat_robust_list))
575                 compat_exit_robust_list(tsk);
576 #endif
577 #endif
578
579         /* Get rid of any cached register state */
580         deactivate_mm(tsk, mm);
581
582         /* notify parent sleeping on vfork() */
583         if (vfork_done) {
584                 tsk->vfork_done = NULL;
585                 complete(vfork_done);
586         }
587
588         /*
589          * If we're exiting normally, clear a user-space tid field if
590          * requested.  We leave this alone when dying by signal, to leave
591          * the value intact in a core dump, and to save the unnecessary
592          * trouble otherwise.  Userland only wants this done for a sys_exit.
593          */
594         if (tsk->clear_child_tid) {
595                 if (!(tsk->flags & PF_SIGNALED) &&
596                     atomic_read(&mm->mm_users) > 1) {
597                         /*
598                          * We don't check the error code - if userspace has
599                          * not set up a proper pointer then tough luck.
600                          */
601                         put_user(0, tsk->clear_child_tid);
602                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
603                                         1, NULL, NULL, 0);
604                 }
605                 tsk->clear_child_tid = NULL;
606         }
607 }
608
609 /*
610  * Allocate a new mm structure and copy contents from the
611  * mm structure of the passed in task structure.
612  */
613 struct mm_struct *dup_mm(struct task_struct *tsk)
614 {
615         struct mm_struct *mm, *oldmm = current->mm;
616         int err;
617
618         if (!oldmm)
619                 return NULL;
620
621         mm = allocate_mm();
622         if (!mm)
623                 goto fail_nomem;
624
625         memcpy(mm, oldmm, sizeof(*mm));
626
627         /* Initializing for Swap token stuff */
628         mm->token_priority = 0;
629         mm->last_interval = 0;
630
631         if (!mm_init(mm, tsk))
632                 goto fail_nomem;
633
634         if (init_new_context(tsk, mm))
635                 goto fail_nocontext;
636
637         dup_mm_exe_file(oldmm, mm);
638
639         err = dup_mmap(mm, oldmm);
640         if (err)
641                 goto free_pt;
642
643         mm->hiwater_rss = get_mm_rss(mm);
644         mm->hiwater_vm = mm->total_vm;
645
646         return mm;
647
648 free_pt:
649         mmput(mm);
650
651 fail_nomem:
652         return NULL;
653
654 fail_nocontext:
655         /*
656          * If init_new_context() failed, we cannot use mmput() to free the mm
657          * because it calls destroy_context()
658          */
659         mm_free_pgd(mm);
660         free_mm(mm);
661         return NULL;
662 }
663
664 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
665 {
666         struct mm_struct * mm, *oldmm;
667         int retval;
668
669         tsk->min_flt = tsk->maj_flt = 0;
670         tsk->nvcsw = tsk->nivcsw = 0;
671 #ifdef CONFIG_DETECT_HUNG_TASK
672         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
673 #endif
674
675         tsk->mm = NULL;
676         tsk->active_mm = NULL;
677
678         /*
679          * Are we cloning a kernel thread?
680          *
681          * We need to steal a active VM for that..
682          */
683         oldmm = current->mm;
684         if (!oldmm)
685                 return 0;
686
687         if (clone_flags & CLONE_VM) {
688                 atomic_inc(&oldmm->mm_users);
689                 mm = oldmm;
690                 goto good_mm;
691         }
692
693         retval = -ENOMEM;
694         mm = dup_mm(tsk);
695         if (!mm)
696                 goto fail_nomem;
697
698 good_mm:
699         /* Initializing for Swap token stuff */
700         mm->token_priority = 0;
701         mm->last_interval = 0;
702
703         tsk->mm = mm;
704         tsk->active_mm = mm;
705         return 0;
706
707 fail_nomem:
708         return retval;
709 }
710
711 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
712 {
713         struct fs_struct *fs = current->fs;
714         if (clone_flags & CLONE_FS) {
715                 /* tsk->fs is already what we want */
716                 write_lock(&fs->lock);
717                 if (fs->in_exec) {
718                         write_unlock(&fs->lock);
719                         return -EAGAIN;
720                 }
721                 fs->users++;
722                 write_unlock(&fs->lock);
723                 return 0;
724         }
725         tsk->fs = copy_fs_struct(fs);
726         if (!tsk->fs)
727                 return -ENOMEM;
728         return 0;
729 }
730
731 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
732 {
733         struct files_struct *oldf, *newf;
734         int error = 0;
735
736         /*
737          * A background process may not have any files ...
738          */
739         oldf = current->files;
740         if (!oldf)
741                 goto out;
742
743         if (clone_flags & CLONE_FILES) {
744                 atomic_inc(&oldf->count);
745                 goto out;
746         }
747
748         newf = dup_fd(oldf, &error);
749         if (!newf)
750                 goto out;
751
752         tsk->files = newf;
753         error = 0;
754 out:
755         return error;
756 }
757
758 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
759 {
760 #ifdef CONFIG_BLOCK
761         struct io_context *ioc = current->io_context;
762
763         if (!ioc)
764                 return 0;
765         /*
766          * Share io context with parent, if CLONE_IO is set
767          */
768         if (clone_flags & CLONE_IO) {
769                 tsk->io_context = ioc_task_link(ioc);
770                 if (unlikely(!tsk->io_context))
771                         return -ENOMEM;
772         } else if (ioprio_valid(ioc->ioprio)) {
773                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
774                 if (unlikely(!tsk->io_context))
775                         return -ENOMEM;
776
777                 tsk->io_context->ioprio = ioc->ioprio;
778         }
779 #endif
780         return 0;
781 }
782
783 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
784 {
785         struct sighand_struct *sig;
786
787         if (clone_flags & CLONE_SIGHAND) {
788                 atomic_inc(&current->sighand->count);
789                 return 0;
790         }
791         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
792         rcu_assign_pointer(tsk->sighand, sig);
793         if (!sig)
794                 return -ENOMEM;
795         atomic_set(&sig->count, 1);
796         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
797         return 0;
798 }
799
800 void __cleanup_sighand(struct sighand_struct *sighand)
801 {
802         if (atomic_dec_and_test(&sighand->count))
803                 kmem_cache_free(sighand_cachep, sighand);
804 }
805
806
807 /*
808  * Initialize POSIX timer handling for a thread group.
809  */
810 static void posix_cpu_timers_init_group(struct signal_struct *sig)
811 {
812         /* Thread group counters. */
813         thread_group_cputime_init(sig);
814
815         /* Expiration times and increments. */
816         sig->it[CPUCLOCK_PROF].expires = cputime_zero;
817         sig->it[CPUCLOCK_PROF].incr = cputime_zero;
818         sig->it[CPUCLOCK_VIRT].expires = cputime_zero;
819         sig->it[CPUCLOCK_VIRT].incr = cputime_zero;
820
821         /* Cached expiration times. */
822         sig->cputime_expires.prof_exp = cputime_zero;
823         sig->cputime_expires.virt_exp = cputime_zero;
824         sig->cputime_expires.sched_exp = 0;
825
826         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
827                 sig->cputime_expires.prof_exp =
828                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
829                 sig->cputimer.running = 1;
830         }
831
832         /* The timer lists. */
833         INIT_LIST_HEAD(&sig->cpu_timers[0]);
834         INIT_LIST_HEAD(&sig->cpu_timers[1]);
835         INIT_LIST_HEAD(&sig->cpu_timers[2]);
836 }
837
838 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
839 {
840         struct signal_struct *sig;
841
842         if (clone_flags & CLONE_THREAD)
843                 return 0;
844
845         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
846         tsk->signal = sig;
847         if (!sig)
848                 return -ENOMEM;
849
850         atomic_set(&sig->count, 1);
851         atomic_set(&sig->live, 1);
852         init_waitqueue_head(&sig->wait_chldexit);
853         sig->flags = 0;
854         if (clone_flags & CLONE_NEWPID)
855                 sig->flags |= SIGNAL_UNKILLABLE;
856         sig->group_exit_code = 0;
857         sig->group_exit_task = NULL;
858         sig->group_stop_count = 0;
859         sig->curr_target = tsk;
860         init_sigpending(&sig->shared_pending);
861         INIT_LIST_HEAD(&sig->posix_timers);
862
863         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
864         sig->it_real_incr.tv64 = 0;
865         sig->real_timer.function = it_real_fn;
866
867         sig->leader = 0;        /* session leadership doesn't inherit */
868         sig->tty_old_pgrp = NULL;
869         sig->tty = NULL;
870
871         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
872         sig->gtime = cputime_zero;
873         sig->cgtime = cputime_zero;
874         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
875         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
876         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
877         sig->maxrss = sig->cmaxrss = 0;
878         task_io_accounting_init(&sig->ioac);
879         sig->sum_sched_runtime = 0;
880         taskstats_tgid_init(sig);
881
882         task_lock(current->group_leader);
883         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
884         task_unlock(current->group_leader);
885
886         posix_cpu_timers_init_group(sig);
887
888         acct_init_pacct(&sig->pacct);
889
890         tty_audit_fork(sig);
891
892         sig->oom_adj = current->signal->oom_adj;
893
894         return 0;
895 }
896
897 void __cleanup_signal(struct signal_struct *sig)
898 {
899         thread_group_cputime_free(sig);
900         tty_kref_put(sig->tty);
901         kmem_cache_free(signal_cachep, sig);
902 }
903
904 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
905 {
906         unsigned long new_flags = p->flags;
907
908         new_flags &= ~PF_SUPERPRIV;
909         new_flags |= PF_FORKNOEXEC;
910         new_flags |= PF_STARTING;
911         p->flags = new_flags;
912         clear_freeze_flag(p);
913 }
914
915 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
916 {
917         current->clear_child_tid = tidptr;
918
919         return task_pid_vnr(current);
920 }
921
922 static void rt_mutex_init_task(struct task_struct *p)
923 {
924         spin_lock_init(&p->pi_lock);
925 #ifdef CONFIG_RT_MUTEXES
926         plist_head_init(&p->pi_waiters, &p->pi_lock);
927         p->pi_blocked_on = NULL;
928 #endif
929 }
930
931 #ifdef CONFIG_MM_OWNER
932 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
933 {
934         mm->owner = p;
935 }
936 #endif /* CONFIG_MM_OWNER */
937
938 /*
939  * Initialize POSIX timer handling for a single task.
940  */
941 static void posix_cpu_timers_init(struct task_struct *tsk)
942 {
943         tsk->cputime_expires.prof_exp = cputime_zero;
944         tsk->cputime_expires.virt_exp = cputime_zero;
945         tsk->cputime_expires.sched_exp = 0;
946         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
947         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
948         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
949 }
950
951 /*
952  * This creates a new process as a copy of the old one,
953  * but does not actually start it yet.
954  *
955  * It copies the registers, and all the appropriate
956  * parts of the process environment (as per the clone
957  * flags). The actual kick-off is left to the caller.
958  */
959 static struct task_struct *copy_process(unsigned long clone_flags,
960                                         unsigned long stack_start,
961                                         struct pt_regs *regs,
962                                         unsigned long stack_size,
963                                         int __user *child_tidptr,
964                                         struct pid *pid,
965                                         int trace)
966 {
967         int retval;
968         struct task_struct *p;
969         int cgroup_callbacks_done = 0;
970
971         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
972                 return ERR_PTR(-EINVAL);
973
974         /*
975          * Thread groups must share signals as well, and detached threads
976          * can only be started up within the thread group.
977          */
978         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
979                 return ERR_PTR(-EINVAL);
980
981         /*
982          * Shared signal handlers imply shared VM. By way of the above,
983          * thread groups also imply shared VM. Blocking this case allows
984          * for various simplifications in other code.
985          */
986         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
987                 return ERR_PTR(-EINVAL);
988
989         /*
990          * Siblings of global init remain as zombies on exit since they are
991          * not reaped by their parent (swapper). To solve this and to avoid
992          * multi-rooted process trees, prevent global and container-inits
993          * from creating siblings.
994          */
995         if ((clone_flags & CLONE_PARENT) &&
996                                 current->signal->flags & SIGNAL_UNKILLABLE)
997                 return ERR_PTR(-EINVAL);
998
999         retval = security_task_create(clone_flags);
1000         if (retval)
1001                 goto fork_out;
1002
1003         retval = -ENOMEM;
1004         p = dup_task_struct(current);
1005         if (!p)
1006                 goto fork_out;
1007
1008         ftrace_graph_init_task(p);
1009
1010         rt_mutex_init_task(p);
1011
1012 #ifdef CONFIG_PROVE_LOCKING
1013         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1014         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1015 #endif
1016         retval = -EAGAIN;
1017         if (atomic_read(&p->real_cred->user->processes) >=
1018                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1019                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1020                     p->real_cred->user != INIT_USER)
1021                         goto bad_fork_free;
1022         }
1023
1024         retval = copy_creds(p, clone_flags);
1025         if (retval < 0)
1026                 goto bad_fork_free;
1027
1028         /*
1029          * If multiple threads are within copy_process(), then this check
1030          * triggers too late. This doesn't hurt, the check is only there
1031          * to stop root fork bombs.
1032          */
1033         retval = -EAGAIN;
1034         if (nr_threads >= max_threads)
1035                 goto bad_fork_cleanup_count;
1036
1037         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1038                 goto bad_fork_cleanup_count;
1039
1040         if (p->binfmt && !try_module_get(p->binfmt->module))
1041                 goto bad_fork_cleanup_put_domain;
1042
1043         p->did_exec = 0;
1044         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1045         copy_flags(clone_flags, p);
1046         INIT_LIST_HEAD(&p->children);
1047         INIT_LIST_HEAD(&p->sibling);
1048         rcu_copy_process(p);
1049         p->vfork_done = NULL;
1050         spin_lock_init(&p->alloc_lock);
1051
1052         init_sigpending(&p->pending);
1053
1054         p->utime = cputime_zero;
1055         p->stime = cputime_zero;
1056         p->gtime = cputime_zero;
1057         p->utimescaled = cputime_zero;
1058         p->stimescaled = cputime_zero;
1059         p->prev_utime = cputime_zero;
1060         p->prev_stime = cputime_zero;
1061
1062         p->default_timer_slack_ns = current->timer_slack_ns;
1063
1064         task_io_accounting_init(&p->ioac);
1065         acct_clear_integrals(p);
1066
1067         posix_cpu_timers_init(p);
1068
1069         p->lock_depth = -1;             /* -1 = no lock */
1070         do_posix_clock_monotonic_gettime(&p->start_time);
1071         p->real_start_time = p->start_time;
1072         monotonic_to_bootbased(&p->real_start_time);
1073         p->io_context = NULL;
1074         p->audit_context = NULL;
1075         cgroup_fork(p);
1076 #ifdef CONFIG_NUMA
1077         p->mempolicy = mpol_dup(p->mempolicy);
1078         if (IS_ERR(p->mempolicy)) {
1079                 retval = PTR_ERR(p->mempolicy);
1080                 p->mempolicy = NULL;
1081                 goto bad_fork_cleanup_cgroup;
1082         }
1083         mpol_fix_fork_child_flag(p);
1084 #endif
1085 #ifdef CONFIG_TRACE_IRQFLAGS
1086         p->irq_events = 0;
1087 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1088         p->hardirqs_enabled = 1;
1089 #else
1090         p->hardirqs_enabled = 0;
1091 #endif
1092         p->hardirq_enable_ip = 0;
1093         p->hardirq_enable_event = 0;
1094         p->hardirq_disable_ip = _THIS_IP_;
1095         p->hardirq_disable_event = 0;
1096         p->softirqs_enabled = 1;
1097         p->softirq_enable_ip = _THIS_IP_;
1098         p->softirq_enable_event = 0;
1099         p->softirq_disable_ip = 0;
1100         p->softirq_disable_event = 0;
1101         p->hardirq_context = 0;
1102         p->softirq_context = 0;
1103 #endif
1104 #ifdef CONFIG_LOCKDEP
1105         p->lockdep_depth = 0; /* no locks held yet */
1106         p->curr_chain_key = 0;
1107         p->lockdep_recursion = 0;
1108 #endif
1109
1110 #ifdef CONFIG_DEBUG_MUTEXES
1111         p->blocked_on = NULL; /* not blocked yet */
1112 #endif
1113
1114         p->bts = NULL;
1115
1116         p->stack_start = stack_start;
1117
1118         /* Perform scheduler related setup. Assign this task to a CPU. */
1119         sched_fork(p, clone_flags);
1120
1121         retval = perf_event_init_task(p);
1122         if (retval)
1123                 goto bad_fork_cleanup_policy;
1124
1125         if ((retval = audit_alloc(p)))
1126                 goto bad_fork_cleanup_policy;
1127         /* copy all the process information */
1128         if ((retval = copy_semundo(clone_flags, p)))
1129                 goto bad_fork_cleanup_audit;
1130         if ((retval = copy_files(clone_flags, p)))
1131                 goto bad_fork_cleanup_semundo;
1132         if ((retval = copy_fs(clone_flags, p)))
1133                 goto bad_fork_cleanup_files;
1134         if ((retval = copy_sighand(clone_flags, p)))
1135                 goto bad_fork_cleanup_fs;
1136         if ((retval = copy_signal(clone_flags, p)))
1137                 goto bad_fork_cleanup_sighand;
1138         if ((retval = copy_mm(clone_flags, p)))
1139                 goto bad_fork_cleanup_signal;
1140         if ((retval = copy_namespaces(clone_flags, p)))
1141                 goto bad_fork_cleanup_mm;
1142         if ((retval = copy_io(clone_flags, p)))
1143                 goto bad_fork_cleanup_namespaces;
1144         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1145         if (retval)
1146                 goto bad_fork_cleanup_io;
1147
1148         if (pid != &init_struct_pid) {
1149                 retval = -ENOMEM;
1150                 pid = alloc_pid(p->nsproxy->pid_ns);
1151                 if (!pid)
1152                         goto bad_fork_cleanup_io;
1153
1154                 if (clone_flags & CLONE_NEWPID) {
1155                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1156                         if (retval < 0)
1157                                 goto bad_fork_free_pid;
1158                 }
1159         }
1160
1161         p->pid = pid_nr(pid);
1162         p->tgid = p->pid;
1163         if (clone_flags & CLONE_THREAD)
1164                 p->tgid = current->tgid;
1165
1166         if (current->nsproxy != p->nsproxy) {
1167                 retval = ns_cgroup_clone(p, pid);
1168                 if (retval)
1169                         goto bad_fork_free_pid;
1170         }
1171
1172         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1173         /*
1174          * Clear TID on mm_release()?
1175          */
1176         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1177 #ifdef CONFIG_FUTEX
1178         p->robust_list = NULL;
1179 #ifdef CONFIG_COMPAT
1180         p->compat_robust_list = NULL;
1181 #endif
1182         INIT_LIST_HEAD(&p->pi_state_list);
1183         p->pi_state_cache = NULL;
1184 #endif
1185         /*
1186          * sigaltstack should be cleared when sharing the same VM
1187          */
1188         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1189                 p->sas_ss_sp = p->sas_ss_size = 0;
1190
1191         /*
1192          * Syscall tracing should be turned off in the child regardless
1193          * of CLONE_PTRACE.
1194          */
1195         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1196 #ifdef TIF_SYSCALL_EMU
1197         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1198 #endif
1199         clear_all_latency_tracing(p);
1200
1201         /* ok, now we should be set up.. */
1202         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1203         p->pdeath_signal = 0;
1204         p->exit_state = 0;
1205
1206         /*
1207          * Ok, make it visible to the rest of the system.
1208          * We dont wake it up yet.
1209          */
1210         p->group_leader = p;
1211         INIT_LIST_HEAD(&p->thread_group);
1212
1213         /* Now that the task is set up, run cgroup callbacks if
1214          * necessary. We need to run them before the task is visible
1215          * on the tasklist. */
1216         cgroup_fork_callbacks(p);
1217         cgroup_callbacks_done = 1;
1218
1219         /* Need tasklist lock for parent etc handling! */
1220         write_lock_irq(&tasklist_lock);
1221
1222         /*
1223          * The task hasn't been attached yet, so its cpus_allowed mask will
1224          * not be changed, nor will its assigned CPU.
1225          *
1226          * The cpus_allowed mask of the parent may have changed after it was
1227          * copied first time - so re-copy it here, then check the child's CPU
1228          * to ensure it is on a valid CPU (and if not, just force it back to
1229          * parent's CPU). This avoids alot of nasty races.
1230          */
1231         p->cpus_allowed = current->cpus_allowed;
1232         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1233         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1234                         !cpu_online(task_cpu(p))))
1235                 set_task_cpu(p, smp_processor_id());
1236
1237         /* CLONE_PARENT re-uses the old parent */
1238         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1239                 p->real_parent = current->real_parent;
1240                 p->parent_exec_id = current->parent_exec_id;
1241         } else {
1242                 p->real_parent = current;
1243                 p->parent_exec_id = current->self_exec_id;
1244         }
1245
1246         spin_lock(&current->sighand->siglock);
1247
1248         /*
1249          * Process group and session signals need to be delivered to just the
1250          * parent before the fork or both the parent and the child after the
1251          * fork. Restart if a signal comes in before we add the new process to
1252          * it's process group.
1253          * A fatal signal pending means that current will exit, so the new
1254          * thread can't slip out of an OOM kill (or normal SIGKILL).
1255          */
1256         recalc_sigpending();
1257         if (signal_pending(current)) {
1258                 spin_unlock(&current->sighand->siglock);
1259                 write_unlock_irq(&tasklist_lock);
1260                 retval = -ERESTARTNOINTR;
1261                 goto bad_fork_free_pid;
1262         }
1263
1264         if (clone_flags & CLONE_THREAD) {
1265                 atomic_inc(&current->signal->count);
1266                 atomic_inc(&current->signal->live);
1267                 p->group_leader = current->group_leader;
1268                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1269         }
1270
1271         if (likely(p->pid)) {
1272                 list_add_tail(&p->sibling, &p->real_parent->children);
1273                 tracehook_finish_clone(p, clone_flags, trace);
1274
1275                 if (thread_group_leader(p)) {
1276                         if (clone_flags & CLONE_NEWPID)
1277                                 p->nsproxy->pid_ns->child_reaper = p;
1278
1279                         p->signal->leader_pid = pid;
1280                         tty_kref_put(p->signal->tty);
1281                         p->signal->tty = tty_kref_get(current->signal->tty);
1282                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1283                         attach_pid(p, PIDTYPE_SID, task_session(current));
1284                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1285                         __get_cpu_var(process_counts)++;
1286                 }
1287                 attach_pid(p, PIDTYPE_PID, pid);
1288                 nr_threads++;
1289         }
1290
1291         total_forks++;
1292         spin_unlock(&current->sighand->siglock);
1293         write_unlock_irq(&tasklist_lock);
1294         proc_fork_connector(p);
1295         cgroup_post_fork(p);
1296         perf_event_fork(p);
1297         return p;
1298
1299 bad_fork_free_pid:
1300         if (pid != &init_struct_pid)
1301                 free_pid(pid);
1302 bad_fork_cleanup_io:
1303         put_io_context(p->io_context);
1304 bad_fork_cleanup_namespaces:
1305         exit_task_namespaces(p);
1306 bad_fork_cleanup_mm:
1307         if (p->mm)
1308                 mmput(p->mm);
1309 bad_fork_cleanup_signal:
1310         if (!(clone_flags & CLONE_THREAD))
1311                 __cleanup_signal(p->signal);
1312 bad_fork_cleanup_sighand:
1313         __cleanup_sighand(p->sighand);
1314 bad_fork_cleanup_fs:
1315         exit_fs(p); /* blocking */
1316 bad_fork_cleanup_files:
1317         exit_files(p); /* blocking */
1318 bad_fork_cleanup_semundo:
1319         exit_sem(p);
1320 bad_fork_cleanup_audit:
1321         audit_free(p);
1322 bad_fork_cleanup_policy:
1323         perf_event_free_task(p);
1324 #ifdef CONFIG_NUMA
1325         mpol_put(p->mempolicy);
1326 bad_fork_cleanup_cgroup:
1327 #endif
1328         cgroup_exit(p, cgroup_callbacks_done);
1329         delayacct_tsk_free(p);
1330         if (p->binfmt)
1331                 module_put(p->binfmt->module);
1332 bad_fork_cleanup_put_domain:
1333         module_put(task_thread_info(p)->exec_domain->module);
1334 bad_fork_cleanup_count:
1335         atomic_dec(&p->cred->user->processes);
1336         exit_creds(p);
1337 bad_fork_free:
1338         free_task(p);
1339 fork_out:
1340         return ERR_PTR(retval);
1341 }
1342
1343 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1344 {
1345         memset(regs, 0, sizeof(struct pt_regs));
1346         return regs;
1347 }
1348
1349 struct task_struct * __cpuinit fork_idle(int cpu)
1350 {
1351         struct task_struct *task;
1352         struct pt_regs regs;
1353
1354         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1355                             &init_struct_pid, 0);
1356         if (!IS_ERR(task))
1357                 init_idle(task, cpu);
1358
1359         return task;
1360 }
1361
1362 /*
1363  *  Ok, this is the main fork-routine.
1364  *
1365  * It copies the process, and if successful kick-starts
1366  * it and waits for it to finish using the VM if required.
1367  */
1368 long do_fork(unsigned long clone_flags,
1369               unsigned long stack_start,
1370               struct pt_regs *regs,
1371               unsigned long stack_size,
1372               int __user *parent_tidptr,
1373               int __user *child_tidptr)
1374 {
1375         struct task_struct *p;
1376         int trace = 0;
1377         long nr;
1378
1379         /*
1380          * Do some preliminary argument and permissions checking before we
1381          * actually start allocating stuff
1382          */
1383         if (clone_flags & CLONE_NEWUSER) {
1384                 if (clone_flags & CLONE_THREAD)
1385                         return -EINVAL;
1386                 /* hopefully this check will go away when userns support is
1387                  * complete
1388                  */
1389                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1390                                 !capable(CAP_SETGID))
1391                         return -EPERM;
1392         }
1393
1394         /*
1395          * We hope to recycle these flags after 2.6.26
1396          */
1397         if (unlikely(clone_flags & CLONE_STOPPED)) {
1398                 static int __read_mostly count = 100;
1399
1400                 if (count > 0 && printk_ratelimit()) {
1401                         char comm[TASK_COMM_LEN];
1402
1403                         count--;
1404                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1405                                         "clone flags 0x%lx\n",
1406                                 get_task_comm(comm, current),
1407                                 clone_flags & CLONE_STOPPED);
1408                 }
1409         }
1410
1411         /*
1412          * When called from kernel_thread, don't do user tracing stuff.
1413          */
1414         if (likely(user_mode(regs)))
1415                 trace = tracehook_prepare_clone(clone_flags);
1416
1417         p = copy_process(clone_flags, stack_start, regs, stack_size,
1418                          child_tidptr, NULL, trace);
1419         /*
1420          * Do this prior waking up the new thread - the thread pointer
1421          * might get invalid after that point, if the thread exits quickly.
1422          */
1423         if (!IS_ERR(p)) {
1424                 struct completion vfork;
1425
1426                 trace_sched_process_fork(current, p);
1427
1428                 nr = task_pid_vnr(p);
1429
1430                 if (clone_flags & CLONE_PARENT_SETTID)
1431                         put_user(nr, parent_tidptr);
1432
1433                 if (clone_flags & CLONE_VFORK) {
1434                         p->vfork_done = &vfork;
1435                         init_completion(&vfork);
1436                 }
1437
1438                 audit_finish_fork(p);
1439                 tracehook_report_clone(regs, clone_flags, nr, p);
1440
1441                 /*
1442                  * We set PF_STARTING at creation in case tracing wants to
1443                  * use this to distinguish a fully live task from one that
1444                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1445                  * clear it and set the child going.
1446                  */
1447                 p->flags &= ~PF_STARTING;
1448
1449                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1450                         /*
1451                          * We'll start up with an immediate SIGSTOP.
1452                          */
1453                         sigaddset(&p->pending.signal, SIGSTOP);
1454                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1455                         __set_task_state(p, TASK_STOPPED);
1456                 } else {
1457                         wake_up_new_task(p, clone_flags);
1458                 }
1459
1460                 tracehook_report_clone_complete(trace, regs,
1461                                                 clone_flags, nr, p);
1462
1463                 if (clone_flags & CLONE_VFORK) {
1464                         freezer_do_not_count();
1465                         wait_for_completion(&vfork);
1466                         freezer_count();
1467                         tracehook_report_vfork_done(p, nr);
1468                 }
1469         } else {
1470                 nr = PTR_ERR(p);
1471         }
1472         return nr;
1473 }
1474
1475 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1476 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1477 #endif
1478
1479 static void sighand_ctor(void *data)
1480 {
1481         struct sighand_struct *sighand = data;
1482
1483         spin_lock_init(&sighand->siglock);
1484         init_waitqueue_head(&sighand->signalfd_wqh);
1485 }
1486
1487 void __init proc_caches_init(void)
1488 {
1489         sighand_cachep = kmem_cache_create("sighand_cache",
1490                         sizeof(struct sighand_struct), 0,
1491                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1492                         SLAB_NOTRACK, sighand_ctor);
1493         signal_cachep = kmem_cache_create("signal_cache",
1494                         sizeof(struct signal_struct), 0,
1495                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1496         files_cachep = kmem_cache_create("files_cache",
1497                         sizeof(struct files_struct), 0,
1498                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1499         fs_cachep = kmem_cache_create("fs_cache",
1500                         sizeof(struct fs_struct), 0,
1501                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1502         mm_cachep = kmem_cache_create("mm_struct",
1503                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1504                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1505         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1506         mmap_init();
1507 }
1508
1509 /*
1510  * Check constraints on flags passed to the unshare system call and
1511  * force unsharing of additional process context as appropriate.
1512  */
1513 static void check_unshare_flags(unsigned long *flags_ptr)
1514 {
1515         /*
1516          * If unsharing a thread from a thread group, must also
1517          * unshare vm.
1518          */
1519         if (*flags_ptr & CLONE_THREAD)
1520                 *flags_ptr |= CLONE_VM;
1521
1522         /*
1523          * If unsharing vm, must also unshare signal handlers.
1524          */
1525         if (*flags_ptr & CLONE_VM)
1526                 *flags_ptr |= CLONE_SIGHAND;
1527
1528         /*
1529          * If unsharing signal handlers and the task was created
1530          * using CLONE_THREAD, then must unshare the thread
1531          */
1532         if ((*flags_ptr & CLONE_SIGHAND) &&
1533             (atomic_read(&current->signal->count) > 1))
1534                 *flags_ptr |= CLONE_THREAD;
1535
1536         /*
1537          * If unsharing namespace, must also unshare filesystem information.
1538          */
1539         if (*flags_ptr & CLONE_NEWNS)
1540                 *flags_ptr |= CLONE_FS;
1541 }
1542
1543 /*
1544  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1545  */
1546 static int unshare_thread(unsigned long unshare_flags)
1547 {
1548         if (unshare_flags & CLONE_THREAD)
1549                 return -EINVAL;
1550
1551         return 0;
1552 }
1553
1554 /*
1555  * Unshare the filesystem structure if it is being shared
1556  */
1557 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1558 {
1559         struct fs_struct *fs = current->fs;
1560
1561         if (!(unshare_flags & CLONE_FS) || !fs)
1562                 return 0;
1563
1564         /* don't need lock here; in the worst case we'll do useless copy */
1565         if (fs->users == 1)
1566                 return 0;
1567
1568         *new_fsp = copy_fs_struct(fs);
1569         if (!*new_fsp)
1570                 return -ENOMEM;
1571
1572         return 0;
1573 }
1574
1575 /*
1576  * Unsharing of sighand is not supported yet
1577  */
1578 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1579 {
1580         struct sighand_struct *sigh = current->sighand;
1581
1582         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1583                 return -EINVAL;
1584         else
1585                 return 0;
1586 }
1587
1588 /*
1589  * Unshare vm if it is being shared
1590  */
1591 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1592 {
1593         struct mm_struct *mm = current->mm;
1594
1595         if ((unshare_flags & CLONE_VM) &&
1596             (mm && atomic_read(&mm->mm_users) > 1)) {
1597                 return -EINVAL;
1598         }
1599
1600         return 0;
1601 }
1602
1603 /*
1604  * Unshare file descriptor table if it is being shared
1605  */
1606 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1607 {
1608         struct files_struct *fd = current->files;
1609         int error = 0;
1610
1611         if ((unshare_flags & CLONE_FILES) &&
1612             (fd && atomic_read(&fd->count) > 1)) {
1613                 *new_fdp = dup_fd(fd, &error);
1614                 if (!*new_fdp)
1615                         return error;
1616         }
1617
1618         return 0;
1619 }
1620
1621 /*
1622  * unshare allows a process to 'unshare' part of the process
1623  * context which was originally shared using clone.  copy_*
1624  * functions used by do_fork() cannot be used here directly
1625  * because they modify an inactive task_struct that is being
1626  * constructed. Here we are modifying the current, active,
1627  * task_struct.
1628  */
1629 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1630 {
1631         int err = 0;
1632         struct fs_struct *fs, *new_fs = NULL;
1633         struct sighand_struct *new_sigh = NULL;
1634         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1635         struct files_struct *fd, *new_fd = NULL;
1636         struct nsproxy *new_nsproxy = NULL;
1637         int do_sysvsem = 0;
1638
1639         check_unshare_flags(&unshare_flags);
1640
1641         /* Return -EINVAL for all unsupported flags */
1642         err = -EINVAL;
1643         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1644                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1645                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1646                 goto bad_unshare_out;
1647
1648         /*
1649          * CLONE_NEWIPC must also detach from the undolist: after switching
1650          * to a new ipc namespace, the semaphore arrays from the old
1651          * namespace are unreachable.
1652          */
1653         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1654                 do_sysvsem = 1;
1655         if ((err = unshare_thread(unshare_flags)))
1656                 goto bad_unshare_out;
1657         if ((err = unshare_fs(unshare_flags, &new_fs)))
1658                 goto bad_unshare_cleanup_thread;
1659         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1660                 goto bad_unshare_cleanup_fs;
1661         if ((err = unshare_vm(unshare_flags, &new_mm)))
1662                 goto bad_unshare_cleanup_sigh;
1663         if ((err = unshare_fd(unshare_flags, &new_fd)))
1664                 goto bad_unshare_cleanup_vm;
1665         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1666                         new_fs)))
1667                 goto bad_unshare_cleanup_fd;
1668
1669         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1670                 if (do_sysvsem) {
1671                         /*
1672                          * CLONE_SYSVSEM is equivalent to sys_exit().
1673                          */
1674                         exit_sem(current);
1675                 }
1676
1677                 if (new_nsproxy) {
1678                         switch_task_namespaces(current, new_nsproxy);
1679                         new_nsproxy = NULL;
1680                 }
1681
1682                 task_lock(current);
1683
1684                 if (new_fs) {
1685                         fs = current->fs;
1686                         write_lock(&fs->lock);
1687                         current->fs = new_fs;
1688                         if (--fs->users)
1689                                 new_fs = NULL;
1690                         else
1691                                 new_fs = fs;
1692                         write_unlock(&fs->lock);
1693                 }
1694
1695                 if (new_mm) {
1696                         mm = current->mm;
1697                         active_mm = current->active_mm;
1698                         current->mm = new_mm;
1699                         current->active_mm = new_mm;
1700                         activate_mm(active_mm, new_mm);
1701                         new_mm = mm;
1702                 }
1703
1704                 if (new_fd) {
1705                         fd = current->files;
1706                         current->files = new_fd;
1707                         new_fd = fd;
1708                 }
1709
1710                 task_unlock(current);
1711         }
1712
1713         if (new_nsproxy)
1714                 put_nsproxy(new_nsproxy);
1715
1716 bad_unshare_cleanup_fd:
1717         if (new_fd)
1718                 put_files_struct(new_fd);
1719
1720 bad_unshare_cleanup_vm:
1721         if (new_mm)
1722                 mmput(new_mm);
1723
1724 bad_unshare_cleanup_sigh:
1725         if (new_sigh)
1726                 if (atomic_dec_and_test(&new_sigh->count))
1727                         kmem_cache_free(sighand_cachep, new_sigh);
1728
1729 bad_unshare_cleanup_fs:
1730         if (new_fs)
1731                 free_fs_struct(new_fs);
1732
1733 bad_unshare_cleanup_thread:
1734 bad_unshare_out:
1735         return err;
1736 }
1737
1738 /*
1739  *      Helper to unshare the files of the current task.
1740  *      We don't want to expose copy_files internals to
1741  *      the exec layer of the kernel.
1742  */
1743
1744 int unshare_files(struct files_struct **displaced)
1745 {
1746         struct task_struct *task = current;
1747         struct files_struct *copy = NULL;
1748         int error;
1749
1750         error = unshare_fd(CLONE_FILES, &copy);
1751         if (error || !copy) {
1752                 *displaced = NULL;
1753                 return error;
1754         }
1755         *displaced = task->files;
1756         task_lock(task);
1757         task->files = copy;
1758         task_unlock(task);
1759         return 0;
1760 }