task_struct cleanup: move binfmt field to 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                 if (mm->binfmt)
522                         module_put(mm->binfmt->module);
523                 mmdrop(mm);
524         }
525 }
526 EXPORT_SYMBOL_GPL(mmput);
527
528 /**
529  * get_task_mm - acquire a reference to the task's mm
530  *
531  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
532  * this kernel workthread has transiently adopted a user mm with use_mm,
533  * to do its AIO) is not set and if so returns a reference to it, after
534  * bumping up the use count.  User must release the mm via mmput()
535  * after use.  Typically used by /proc and ptrace.
536  */
537 struct mm_struct *get_task_mm(struct task_struct *task)
538 {
539         struct mm_struct *mm;
540
541         task_lock(task);
542         mm = task->mm;
543         if (mm) {
544                 if (task->flags & PF_KTHREAD)
545                         mm = NULL;
546                 else
547                         atomic_inc(&mm->mm_users);
548         }
549         task_unlock(task);
550         return mm;
551 }
552 EXPORT_SYMBOL_GPL(get_task_mm);
553
554 /* Please note the differences between mmput and mm_release.
555  * mmput is called whenever we stop holding onto a mm_struct,
556  * error success whatever.
557  *
558  * mm_release is called after a mm_struct has been removed
559  * from the current process.
560  *
561  * This difference is important for error handling, when we
562  * only half set up a mm_struct for a new process and need to restore
563  * the old one.  Because we mmput the new mm_struct before
564  * restoring the old one. . .
565  * Eric Biederman 10 January 1998
566  */
567 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
568 {
569         struct completion *vfork_done = tsk->vfork_done;
570
571         /* Get rid of any futexes when releasing the mm */
572 #ifdef CONFIG_FUTEX
573         if (unlikely(tsk->robust_list))
574                 exit_robust_list(tsk);
575 #ifdef CONFIG_COMPAT
576         if (unlikely(tsk->compat_robust_list))
577                 compat_exit_robust_list(tsk);
578 #endif
579 #endif
580
581         /* Get rid of any cached register state */
582         deactivate_mm(tsk, mm);
583
584         /* notify parent sleeping on vfork() */
585         if (vfork_done) {
586                 tsk->vfork_done = NULL;
587                 complete(vfork_done);
588         }
589
590         /*
591          * If we're exiting normally, clear a user-space tid field if
592          * requested.  We leave this alone when dying by signal, to leave
593          * the value intact in a core dump, and to save the unnecessary
594          * trouble otherwise.  Userland only wants this done for a sys_exit.
595          */
596         if (tsk->clear_child_tid) {
597                 if (!(tsk->flags & PF_SIGNALED) &&
598                     atomic_read(&mm->mm_users) > 1) {
599                         /*
600                          * We don't check the error code - if userspace has
601                          * not set up a proper pointer then tough luck.
602                          */
603                         put_user(0, tsk->clear_child_tid);
604                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
605                                         1, NULL, NULL, 0);
606                 }
607                 tsk->clear_child_tid = NULL;
608         }
609 }
610
611 /*
612  * Allocate a new mm structure and copy contents from the
613  * mm structure of the passed in task structure.
614  */
615 struct mm_struct *dup_mm(struct task_struct *tsk)
616 {
617         struct mm_struct *mm, *oldmm = current->mm;
618         int err;
619
620         if (!oldmm)
621                 return NULL;
622
623         mm = allocate_mm();
624         if (!mm)
625                 goto fail_nomem;
626
627         memcpy(mm, oldmm, sizeof(*mm));
628
629         /* Initializing for Swap token stuff */
630         mm->token_priority = 0;
631         mm->last_interval = 0;
632
633         if (!mm_init(mm, tsk))
634                 goto fail_nomem;
635
636         if (init_new_context(tsk, mm))
637                 goto fail_nocontext;
638
639         dup_mm_exe_file(oldmm, mm);
640
641         err = dup_mmap(mm, oldmm);
642         if (err)
643                 goto free_pt;
644
645         mm->hiwater_rss = get_mm_rss(mm);
646         mm->hiwater_vm = mm->total_vm;
647
648         if (mm->binfmt && !try_module_get(mm->binfmt->module))
649                 goto free_pt;
650
651         return mm;
652
653 free_pt:
654         /* don't put binfmt in mmput, we haven't got module yet */
655         mm->binfmt = NULL;
656         mmput(mm);
657
658 fail_nomem:
659         return NULL;
660
661 fail_nocontext:
662         /*
663          * If init_new_context() failed, we cannot use mmput() to free the mm
664          * because it calls destroy_context()
665          */
666         mm_free_pgd(mm);
667         free_mm(mm);
668         return NULL;
669 }
670
671 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
672 {
673         struct mm_struct * mm, *oldmm;
674         int retval;
675
676         tsk->min_flt = tsk->maj_flt = 0;
677         tsk->nvcsw = tsk->nivcsw = 0;
678 #ifdef CONFIG_DETECT_HUNG_TASK
679         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
680 #endif
681
682         tsk->mm = NULL;
683         tsk->active_mm = NULL;
684
685         /*
686          * Are we cloning a kernel thread?
687          *
688          * We need to steal a active VM for that..
689          */
690         oldmm = current->mm;
691         if (!oldmm)
692                 return 0;
693
694         if (clone_flags & CLONE_VM) {
695                 atomic_inc(&oldmm->mm_users);
696                 mm = oldmm;
697                 goto good_mm;
698         }
699
700         retval = -ENOMEM;
701         mm = dup_mm(tsk);
702         if (!mm)
703                 goto fail_nomem;
704
705 good_mm:
706         /* Initializing for Swap token stuff */
707         mm->token_priority = 0;
708         mm->last_interval = 0;
709
710         tsk->mm = mm;
711         tsk->active_mm = mm;
712         return 0;
713
714 fail_nomem:
715         return retval;
716 }
717
718 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
719 {
720         struct fs_struct *fs = current->fs;
721         if (clone_flags & CLONE_FS) {
722                 /* tsk->fs is already what we want */
723                 write_lock(&fs->lock);
724                 if (fs->in_exec) {
725                         write_unlock(&fs->lock);
726                         return -EAGAIN;
727                 }
728                 fs->users++;
729                 write_unlock(&fs->lock);
730                 return 0;
731         }
732         tsk->fs = copy_fs_struct(fs);
733         if (!tsk->fs)
734                 return -ENOMEM;
735         return 0;
736 }
737
738 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
739 {
740         struct files_struct *oldf, *newf;
741         int error = 0;
742
743         /*
744          * A background process may not have any files ...
745          */
746         oldf = current->files;
747         if (!oldf)
748                 goto out;
749
750         if (clone_flags & CLONE_FILES) {
751                 atomic_inc(&oldf->count);
752                 goto out;
753         }
754
755         newf = dup_fd(oldf, &error);
756         if (!newf)
757                 goto out;
758
759         tsk->files = newf;
760         error = 0;
761 out:
762         return error;
763 }
764
765 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
766 {
767 #ifdef CONFIG_BLOCK
768         struct io_context *ioc = current->io_context;
769
770         if (!ioc)
771                 return 0;
772         /*
773          * Share io context with parent, if CLONE_IO is set
774          */
775         if (clone_flags & CLONE_IO) {
776                 tsk->io_context = ioc_task_link(ioc);
777                 if (unlikely(!tsk->io_context))
778                         return -ENOMEM;
779         } else if (ioprio_valid(ioc->ioprio)) {
780                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
781                 if (unlikely(!tsk->io_context))
782                         return -ENOMEM;
783
784                 tsk->io_context->ioprio = ioc->ioprio;
785         }
786 #endif
787         return 0;
788 }
789
790 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
791 {
792         struct sighand_struct *sig;
793
794         if (clone_flags & CLONE_SIGHAND) {
795                 atomic_inc(&current->sighand->count);
796                 return 0;
797         }
798         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
799         rcu_assign_pointer(tsk->sighand, sig);
800         if (!sig)
801                 return -ENOMEM;
802         atomic_set(&sig->count, 1);
803         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
804         return 0;
805 }
806
807 void __cleanup_sighand(struct sighand_struct *sighand)
808 {
809         if (atomic_dec_and_test(&sighand->count))
810                 kmem_cache_free(sighand_cachep, sighand);
811 }
812
813
814 /*
815  * Initialize POSIX timer handling for a thread group.
816  */
817 static void posix_cpu_timers_init_group(struct signal_struct *sig)
818 {
819         /* Thread group counters. */
820         thread_group_cputime_init(sig);
821
822         /* Expiration times and increments. */
823         sig->it[CPUCLOCK_PROF].expires = cputime_zero;
824         sig->it[CPUCLOCK_PROF].incr = cputime_zero;
825         sig->it[CPUCLOCK_VIRT].expires = cputime_zero;
826         sig->it[CPUCLOCK_VIRT].incr = cputime_zero;
827
828         /* Cached expiration times. */
829         sig->cputime_expires.prof_exp = cputime_zero;
830         sig->cputime_expires.virt_exp = cputime_zero;
831         sig->cputime_expires.sched_exp = 0;
832
833         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
834                 sig->cputime_expires.prof_exp =
835                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
836                 sig->cputimer.running = 1;
837         }
838
839         /* The timer lists. */
840         INIT_LIST_HEAD(&sig->cpu_timers[0]);
841         INIT_LIST_HEAD(&sig->cpu_timers[1]);
842         INIT_LIST_HEAD(&sig->cpu_timers[2]);
843 }
844
845 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
846 {
847         struct signal_struct *sig;
848
849         if (clone_flags & CLONE_THREAD)
850                 return 0;
851
852         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
853         tsk->signal = sig;
854         if (!sig)
855                 return -ENOMEM;
856
857         atomic_set(&sig->count, 1);
858         atomic_set(&sig->live, 1);
859         init_waitqueue_head(&sig->wait_chldexit);
860         sig->flags = 0;
861         if (clone_flags & CLONE_NEWPID)
862                 sig->flags |= SIGNAL_UNKILLABLE;
863         sig->group_exit_code = 0;
864         sig->group_exit_task = NULL;
865         sig->group_stop_count = 0;
866         sig->curr_target = tsk;
867         init_sigpending(&sig->shared_pending);
868         INIT_LIST_HEAD(&sig->posix_timers);
869
870         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
871         sig->it_real_incr.tv64 = 0;
872         sig->real_timer.function = it_real_fn;
873
874         sig->leader = 0;        /* session leadership doesn't inherit */
875         sig->tty_old_pgrp = NULL;
876         sig->tty = NULL;
877
878         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
879         sig->gtime = cputime_zero;
880         sig->cgtime = cputime_zero;
881         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
882         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
883         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
884         sig->maxrss = sig->cmaxrss = 0;
885         task_io_accounting_init(&sig->ioac);
886         sig->sum_sched_runtime = 0;
887         taskstats_tgid_init(sig);
888
889         task_lock(current->group_leader);
890         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
891         task_unlock(current->group_leader);
892
893         posix_cpu_timers_init_group(sig);
894
895         acct_init_pacct(&sig->pacct);
896
897         tty_audit_fork(sig);
898
899         sig->oom_adj = current->signal->oom_adj;
900
901         return 0;
902 }
903
904 void __cleanup_signal(struct signal_struct *sig)
905 {
906         thread_group_cputime_free(sig);
907         tty_kref_put(sig->tty);
908         kmem_cache_free(signal_cachep, sig);
909 }
910
911 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
912 {
913         unsigned long new_flags = p->flags;
914
915         new_flags &= ~PF_SUPERPRIV;
916         new_flags |= PF_FORKNOEXEC;
917         new_flags |= PF_STARTING;
918         p->flags = new_flags;
919         clear_freeze_flag(p);
920 }
921
922 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
923 {
924         current->clear_child_tid = tidptr;
925
926         return task_pid_vnr(current);
927 }
928
929 static void rt_mutex_init_task(struct task_struct *p)
930 {
931         spin_lock_init(&p->pi_lock);
932 #ifdef CONFIG_RT_MUTEXES
933         plist_head_init(&p->pi_waiters, &p->pi_lock);
934         p->pi_blocked_on = NULL;
935 #endif
936 }
937
938 #ifdef CONFIG_MM_OWNER
939 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
940 {
941         mm->owner = p;
942 }
943 #endif /* CONFIG_MM_OWNER */
944
945 /*
946  * Initialize POSIX timer handling for a single task.
947  */
948 static void posix_cpu_timers_init(struct task_struct *tsk)
949 {
950         tsk->cputime_expires.prof_exp = cputime_zero;
951         tsk->cputime_expires.virt_exp = cputime_zero;
952         tsk->cputime_expires.sched_exp = 0;
953         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
954         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
955         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
956 }
957
958 /*
959  * This creates a new process as a copy of the old one,
960  * but does not actually start it yet.
961  *
962  * It copies the registers, and all the appropriate
963  * parts of the process environment (as per the clone
964  * flags). The actual kick-off is left to the caller.
965  */
966 static struct task_struct *copy_process(unsigned long clone_flags,
967                                         unsigned long stack_start,
968                                         struct pt_regs *regs,
969                                         unsigned long stack_size,
970                                         int __user *child_tidptr,
971                                         struct pid *pid,
972                                         int trace)
973 {
974         int retval;
975         struct task_struct *p;
976         int cgroup_callbacks_done = 0;
977
978         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
979                 return ERR_PTR(-EINVAL);
980
981         /*
982          * Thread groups must share signals as well, and detached threads
983          * can only be started up within the thread group.
984          */
985         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
986                 return ERR_PTR(-EINVAL);
987
988         /*
989          * Shared signal handlers imply shared VM. By way of the above,
990          * thread groups also imply shared VM. Blocking this case allows
991          * for various simplifications in other code.
992          */
993         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
994                 return ERR_PTR(-EINVAL);
995
996         /*
997          * Siblings of global init remain as zombies on exit since they are
998          * not reaped by their parent (swapper). To solve this and to avoid
999          * multi-rooted process trees, prevent global and container-inits
1000          * from creating siblings.
1001          */
1002         if ((clone_flags & CLONE_PARENT) &&
1003                                 current->signal->flags & SIGNAL_UNKILLABLE)
1004                 return ERR_PTR(-EINVAL);
1005
1006         retval = security_task_create(clone_flags);
1007         if (retval)
1008                 goto fork_out;
1009
1010         retval = -ENOMEM;
1011         p = dup_task_struct(current);
1012         if (!p)
1013                 goto fork_out;
1014
1015         ftrace_graph_init_task(p);
1016
1017         rt_mutex_init_task(p);
1018
1019 #ifdef CONFIG_PROVE_LOCKING
1020         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1021         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1022 #endif
1023         retval = -EAGAIN;
1024         if (atomic_read(&p->real_cred->user->processes) >=
1025                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1026                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1027                     p->real_cred->user != INIT_USER)
1028                         goto bad_fork_free;
1029         }
1030
1031         retval = copy_creds(p, clone_flags);
1032         if (retval < 0)
1033                 goto bad_fork_free;
1034
1035         /*
1036          * If multiple threads are within copy_process(), then this check
1037          * triggers too late. This doesn't hurt, the check is only there
1038          * to stop root fork bombs.
1039          */
1040         retval = -EAGAIN;
1041         if (nr_threads >= max_threads)
1042                 goto bad_fork_cleanup_count;
1043
1044         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1045                 goto bad_fork_cleanup_count;
1046
1047         p->did_exec = 0;
1048         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1049         copy_flags(clone_flags, p);
1050         INIT_LIST_HEAD(&p->children);
1051         INIT_LIST_HEAD(&p->sibling);
1052         rcu_copy_process(p);
1053         p->vfork_done = NULL;
1054         spin_lock_init(&p->alloc_lock);
1055
1056         init_sigpending(&p->pending);
1057
1058         p->utime = cputime_zero;
1059         p->stime = cputime_zero;
1060         p->gtime = cputime_zero;
1061         p->utimescaled = cputime_zero;
1062         p->stimescaled = cputime_zero;
1063         p->prev_utime = cputime_zero;
1064         p->prev_stime = cputime_zero;
1065
1066         p->default_timer_slack_ns = current->timer_slack_ns;
1067
1068         task_io_accounting_init(&p->ioac);
1069         acct_clear_integrals(p);
1070
1071         posix_cpu_timers_init(p);
1072
1073         p->lock_depth = -1;             /* -1 = no lock */
1074         do_posix_clock_monotonic_gettime(&p->start_time);
1075         p->real_start_time = p->start_time;
1076         monotonic_to_bootbased(&p->real_start_time);
1077         p->io_context = NULL;
1078         p->audit_context = NULL;
1079         cgroup_fork(p);
1080 #ifdef CONFIG_NUMA
1081         p->mempolicy = mpol_dup(p->mempolicy);
1082         if (IS_ERR(p->mempolicy)) {
1083                 retval = PTR_ERR(p->mempolicy);
1084                 p->mempolicy = NULL;
1085                 goto bad_fork_cleanup_cgroup;
1086         }
1087         mpol_fix_fork_child_flag(p);
1088 #endif
1089 #ifdef CONFIG_TRACE_IRQFLAGS
1090         p->irq_events = 0;
1091 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1092         p->hardirqs_enabled = 1;
1093 #else
1094         p->hardirqs_enabled = 0;
1095 #endif
1096         p->hardirq_enable_ip = 0;
1097         p->hardirq_enable_event = 0;
1098         p->hardirq_disable_ip = _THIS_IP_;
1099         p->hardirq_disable_event = 0;
1100         p->softirqs_enabled = 1;
1101         p->softirq_enable_ip = _THIS_IP_;
1102         p->softirq_enable_event = 0;
1103         p->softirq_disable_ip = 0;
1104         p->softirq_disable_event = 0;
1105         p->hardirq_context = 0;
1106         p->softirq_context = 0;
1107 #endif
1108 #ifdef CONFIG_LOCKDEP
1109         p->lockdep_depth = 0; /* no locks held yet */
1110         p->curr_chain_key = 0;
1111         p->lockdep_recursion = 0;
1112 #endif
1113
1114 #ifdef CONFIG_DEBUG_MUTEXES
1115         p->blocked_on = NULL; /* not blocked yet */
1116 #endif
1117
1118         p->bts = NULL;
1119
1120         p->stack_start = stack_start;
1121
1122         /* Perform scheduler related setup. Assign this task to a CPU. */
1123         sched_fork(p, clone_flags);
1124
1125         retval = perf_event_init_task(p);
1126         if (retval)
1127                 goto bad_fork_cleanup_policy;
1128
1129         if ((retval = audit_alloc(p)))
1130                 goto bad_fork_cleanup_policy;
1131         /* copy all the process information */
1132         if ((retval = copy_semundo(clone_flags, p)))
1133                 goto bad_fork_cleanup_audit;
1134         if ((retval = copy_files(clone_flags, p)))
1135                 goto bad_fork_cleanup_semundo;
1136         if ((retval = copy_fs(clone_flags, p)))
1137                 goto bad_fork_cleanup_files;
1138         if ((retval = copy_sighand(clone_flags, p)))
1139                 goto bad_fork_cleanup_fs;
1140         if ((retval = copy_signal(clone_flags, p)))
1141                 goto bad_fork_cleanup_sighand;
1142         if ((retval = copy_mm(clone_flags, p)))
1143                 goto bad_fork_cleanup_signal;
1144         if ((retval = copy_namespaces(clone_flags, p)))
1145                 goto bad_fork_cleanup_mm;
1146         if ((retval = copy_io(clone_flags, p)))
1147                 goto bad_fork_cleanup_namespaces;
1148         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1149         if (retval)
1150                 goto bad_fork_cleanup_io;
1151
1152         if (pid != &init_struct_pid) {
1153                 retval = -ENOMEM;
1154                 pid = alloc_pid(p->nsproxy->pid_ns);
1155                 if (!pid)
1156                         goto bad_fork_cleanup_io;
1157
1158                 if (clone_flags & CLONE_NEWPID) {
1159                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1160                         if (retval < 0)
1161                                 goto bad_fork_free_pid;
1162                 }
1163         }
1164
1165         p->pid = pid_nr(pid);
1166         p->tgid = p->pid;
1167         if (clone_flags & CLONE_THREAD)
1168                 p->tgid = current->tgid;
1169
1170         if (current->nsproxy != p->nsproxy) {
1171                 retval = ns_cgroup_clone(p, pid);
1172                 if (retval)
1173                         goto bad_fork_free_pid;
1174         }
1175
1176         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1177         /*
1178          * Clear TID on mm_release()?
1179          */
1180         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1181 #ifdef CONFIG_FUTEX
1182         p->robust_list = NULL;
1183 #ifdef CONFIG_COMPAT
1184         p->compat_robust_list = NULL;
1185 #endif
1186         INIT_LIST_HEAD(&p->pi_state_list);
1187         p->pi_state_cache = NULL;
1188 #endif
1189         /*
1190          * sigaltstack should be cleared when sharing the same VM
1191          */
1192         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1193                 p->sas_ss_sp = p->sas_ss_size = 0;
1194
1195         /*
1196          * Syscall tracing should be turned off in the child regardless
1197          * of CLONE_PTRACE.
1198          */
1199         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1200 #ifdef TIF_SYSCALL_EMU
1201         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1202 #endif
1203         clear_all_latency_tracing(p);
1204
1205         /* ok, now we should be set up.. */
1206         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1207         p->pdeath_signal = 0;
1208         p->exit_state = 0;
1209
1210         /*
1211          * Ok, make it visible to the rest of the system.
1212          * We dont wake it up yet.
1213          */
1214         p->group_leader = p;
1215         INIT_LIST_HEAD(&p->thread_group);
1216
1217         /* Now that the task is set up, run cgroup callbacks if
1218          * necessary. We need to run them before the task is visible
1219          * on the tasklist. */
1220         cgroup_fork_callbacks(p);
1221         cgroup_callbacks_done = 1;
1222
1223         /* Need tasklist lock for parent etc handling! */
1224         write_lock_irq(&tasklist_lock);
1225
1226         /*
1227          * The task hasn't been attached yet, so its cpus_allowed mask will
1228          * not be changed, nor will its assigned CPU.
1229          *
1230          * The cpus_allowed mask of the parent may have changed after it was
1231          * copied first time - so re-copy it here, then check the child's CPU
1232          * to ensure it is on a valid CPU (and if not, just force it back to
1233          * parent's CPU). This avoids alot of nasty races.
1234          */
1235         p->cpus_allowed = current->cpus_allowed;
1236         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1237         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1238                         !cpu_online(task_cpu(p))))
1239                 set_task_cpu(p, smp_processor_id());
1240
1241         /* CLONE_PARENT re-uses the old parent */
1242         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1243                 p->real_parent = current->real_parent;
1244                 p->parent_exec_id = current->parent_exec_id;
1245         } else {
1246                 p->real_parent = current;
1247                 p->parent_exec_id = current->self_exec_id;
1248         }
1249
1250         spin_lock(&current->sighand->siglock);
1251
1252         /*
1253          * Process group and session signals need to be delivered to just the
1254          * parent before the fork or both the parent and the child after the
1255          * fork. Restart if a signal comes in before we add the new process to
1256          * it's process group.
1257          * A fatal signal pending means that current will exit, so the new
1258          * thread can't slip out of an OOM kill (or normal SIGKILL).
1259          */
1260         recalc_sigpending();
1261         if (signal_pending(current)) {
1262                 spin_unlock(&current->sighand->siglock);
1263                 write_unlock_irq(&tasklist_lock);
1264                 retval = -ERESTARTNOINTR;
1265                 goto bad_fork_free_pid;
1266         }
1267
1268         if (clone_flags & CLONE_THREAD) {
1269                 atomic_inc(&current->signal->count);
1270                 atomic_inc(&current->signal->live);
1271                 p->group_leader = current->group_leader;
1272                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1273         }
1274
1275         if (likely(p->pid)) {
1276                 list_add_tail(&p->sibling, &p->real_parent->children);
1277                 tracehook_finish_clone(p, clone_flags, trace);
1278
1279                 if (thread_group_leader(p)) {
1280                         if (clone_flags & CLONE_NEWPID)
1281                                 p->nsproxy->pid_ns->child_reaper = p;
1282
1283                         p->signal->leader_pid = pid;
1284                         tty_kref_put(p->signal->tty);
1285                         p->signal->tty = tty_kref_get(current->signal->tty);
1286                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1287                         attach_pid(p, PIDTYPE_SID, task_session(current));
1288                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1289                         __get_cpu_var(process_counts)++;
1290                 }
1291                 attach_pid(p, PIDTYPE_PID, pid);
1292                 nr_threads++;
1293         }
1294
1295         total_forks++;
1296         spin_unlock(&current->sighand->siglock);
1297         write_unlock_irq(&tasklist_lock);
1298         proc_fork_connector(p);
1299         cgroup_post_fork(p);
1300         perf_event_fork(p);
1301         return p;
1302
1303 bad_fork_free_pid:
1304         if (pid != &init_struct_pid)
1305                 free_pid(pid);
1306 bad_fork_cleanup_io:
1307         put_io_context(p->io_context);
1308 bad_fork_cleanup_namespaces:
1309         exit_task_namespaces(p);
1310 bad_fork_cleanup_mm:
1311         if (p->mm)
1312                 mmput(p->mm);
1313 bad_fork_cleanup_signal:
1314         if (!(clone_flags & CLONE_THREAD))
1315                 __cleanup_signal(p->signal);
1316 bad_fork_cleanup_sighand:
1317         __cleanup_sighand(p->sighand);
1318 bad_fork_cleanup_fs:
1319         exit_fs(p); /* blocking */
1320 bad_fork_cleanup_files:
1321         exit_files(p); /* blocking */
1322 bad_fork_cleanup_semundo:
1323         exit_sem(p);
1324 bad_fork_cleanup_audit:
1325         audit_free(p);
1326 bad_fork_cleanup_policy:
1327         perf_event_free_task(p);
1328 #ifdef CONFIG_NUMA
1329         mpol_put(p->mempolicy);
1330 bad_fork_cleanup_cgroup:
1331 #endif
1332         cgroup_exit(p, cgroup_callbacks_done);
1333         delayacct_tsk_free(p);
1334         module_put(task_thread_info(p)->exec_domain->module);
1335 bad_fork_cleanup_count:
1336         atomic_dec(&p->cred->user->processes);
1337         exit_creds(p);
1338 bad_fork_free:
1339         free_task(p);
1340 fork_out:
1341         return ERR_PTR(retval);
1342 }
1343
1344 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1345 {
1346         memset(regs, 0, sizeof(struct pt_regs));
1347         return regs;
1348 }
1349
1350 struct task_struct * __cpuinit fork_idle(int cpu)
1351 {
1352         struct task_struct *task;
1353         struct pt_regs regs;
1354
1355         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1356                             &init_struct_pid, 0);
1357         if (!IS_ERR(task))
1358                 init_idle(task, cpu);
1359
1360         return task;
1361 }
1362
1363 /*
1364  *  Ok, this is the main fork-routine.
1365  *
1366  * It copies the process, and if successful kick-starts
1367  * it and waits for it to finish using the VM if required.
1368  */
1369 long do_fork(unsigned long clone_flags,
1370               unsigned long stack_start,
1371               struct pt_regs *regs,
1372               unsigned long stack_size,
1373               int __user *parent_tidptr,
1374               int __user *child_tidptr)
1375 {
1376         struct task_struct *p;
1377         int trace = 0;
1378         long nr;
1379
1380         /*
1381          * Do some preliminary argument and permissions checking before we
1382          * actually start allocating stuff
1383          */
1384         if (clone_flags & CLONE_NEWUSER) {
1385                 if (clone_flags & CLONE_THREAD)
1386                         return -EINVAL;
1387                 /* hopefully this check will go away when userns support is
1388                  * complete
1389                  */
1390                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1391                                 !capable(CAP_SETGID))
1392                         return -EPERM;
1393         }
1394
1395         /*
1396          * We hope to recycle these flags after 2.6.26
1397          */
1398         if (unlikely(clone_flags & CLONE_STOPPED)) {
1399                 static int __read_mostly count = 100;
1400
1401                 if (count > 0 && printk_ratelimit()) {
1402                         char comm[TASK_COMM_LEN];
1403
1404                         count--;
1405                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1406                                         "clone flags 0x%lx\n",
1407                                 get_task_comm(comm, current),
1408                                 clone_flags & CLONE_STOPPED);
1409                 }
1410         }
1411
1412         /*
1413          * When called from kernel_thread, don't do user tracing stuff.
1414          */
1415         if (likely(user_mode(regs)))
1416                 trace = tracehook_prepare_clone(clone_flags);
1417
1418         p = copy_process(clone_flags, stack_start, regs, stack_size,
1419                          child_tidptr, NULL, trace);
1420         /*
1421          * Do this prior waking up the new thread - the thread pointer
1422          * might get invalid after that point, if the thread exits quickly.
1423          */
1424         if (!IS_ERR(p)) {
1425                 struct completion vfork;
1426
1427                 trace_sched_process_fork(current, p);
1428
1429                 nr = task_pid_vnr(p);
1430
1431                 if (clone_flags & CLONE_PARENT_SETTID)
1432                         put_user(nr, parent_tidptr);
1433
1434                 if (clone_flags & CLONE_VFORK) {
1435                         p->vfork_done = &vfork;
1436                         init_completion(&vfork);
1437                 }
1438
1439                 audit_finish_fork(p);
1440                 tracehook_report_clone(regs, clone_flags, nr, p);
1441
1442                 /*
1443                  * We set PF_STARTING at creation in case tracing wants to
1444                  * use this to distinguish a fully live task from one that
1445                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1446                  * clear it and set the child going.
1447                  */
1448                 p->flags &= ~PF_STARTING;
1449
1450                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1451                         /*
1452                          * We'll start up with an immediate SIGSTOP.
1453                          */
1454                         sigaddset(&p->pending.signal, SIGSTOP);
1455                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1456                         __set_task_state(p, TASK_STOPPED);
1457                 } else {
1458                         wake_up_new_task(p, clone_flags);
1459                 }
1460
1461                 tracehook_report_clone_complete(trace, regs,
1462                                                 clone_flags, nr, p);
1463
1464                 if (clone_flags & CLONE_VFORK) {
1465                         freezer_do_not_count();
1466                         wait_for_completion(&vfork);
1467                         freezer_count();
1468                         tracehook_report_vfork_done(p, nr);
1469                 }
1470         } else {
1471                 nr = PTR_ERR(p);
1472         }
1473         return nr;
1474 }
1475
1476 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1477 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1478 #endif
1479
1480 static void sighand_ctor(void *data)
1481 {
1482         struct sighand_struct *sighand = data;
1483
1484         spin_lock_init(&sighand->siglock);
1485         init_waitqueue_head(&sighand->signalfd_wqh);
1486 }
1487
1488 void __init proc_caches_init(void)
1489 {
1490         sighand_cachep = kmem_cache_create("sighand_cache",
1491                         sizeof(struct sighand_struct), 0,
1492                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1493                         SLAB_NOTRACK, sighand_ctor);
1494         signal_cachep = kmem_cache_create("signal_cache",
1495                         sizeof(struct signal_struct), 0,
1496                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1497         files_cachep = kmem_cache_create("files_cache",
1498                         sizeof(struct files_struct), 0,
1499                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1500         fs_cachep = kmem_cache_create("fs_cache",
1501                         sizeof(struct fs_struct), 0,
1502                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1503         mm_cachep = kmem_cache_create("mm_struct",
1504                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1505                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1506         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1507         mmap_init();
1508 }
1509
1510 /*
1511  * Check constraints on flags passed to the unshare system call and
1512  * force unsharing of additional process context as appropriate.
1513  */
1514 static void check_unshare_flags(unsigned long *flags_ptr)
1515 {
1516         /*
1517          * If unsharing a thread from a thread group, must also
1518          * unshare vm.
1519          */
1520         if (*flags_ptr & CLONE_THREAD)
1521                 *flags_ptr |= CLONE_VM;
1522
1523         /*
1524          * If unsharing vm, must also unshare signal handlers.
1525          */
1526         if (*flags_ptr & CLONE_VM)
1527                 *flags_ptr |= CLONE_SIGHAND;
1528
1529         /*
1530          * If unsharing signal handlers and the task was created
1531          * using CLONE_THREAD, then must unshare the thread
1532          */
1533         if ((*flags_ptr & CLONE_SIGHAND) &&
1534             (atomic_read(&current->signal->count) > 1))
1535                 *flags_ptr |= CLONE_THREAD;
1536
1537         /*
1538          * If unsharing namespace, must also unshare filesystem information.
1539          */
1540         if (*flags_ptr & CLONE_NEWNS)
1541                 *flags_ptr |= CLONE_FS;
1542 }
1543
1544 /*
1545  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1546  */
1547 static int unshare_thread(unsigned long unshare_flags)
1548 {
1549         if (unshare_flags & CLONE_THREAD)
1550                 return -EINVAL;
1551
1552         return 0;
1553 }
1554
1555 /*
1556  * Unshare the filesystem structure if it is being shared
1557  */
1558 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1559 {
1560         struct fs_struct *fs = current->fs;
1561
1562         if (!(unshare_flags & CLONE_FS) || !fs)
1563                 return 0;
1564
1565         /* don't need lock here; in the worst case we'll do useless copy */
1566         if (fs->users == 1)
1567                 return 0;
1568
1569         *new_fsp = copy_fs_struct(fs);
1570         if (!*new_fsp)
1571                 return -ENOMEM;
1572
1573         return 0;
1574 }
1575
1576 /*
1577  * Unsharing of sighand is not supported yet
1578  */
1579 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1580 {
1581         struct sighand_struct *sigh = current->sighand;
1582
1583         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1584                 return -EINVAL;
1585         else
1586                 return 0;
1587 }
1588
1589 /*
1590  * Unshare vm if it is being shared
1591  */
1592 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1593 {
1594         struct mm_struct *mm = current->mm;
1595
1596         if ((unshare_flags & CLONE_VM) &&
1597             (mm && atomic_read(&mm->mm_users) > 1)) {
1598                 return -EINVAL;
1599         }
1600
1601         return 0;
1602 }
1603
1604 /*
1605  * Unshare file descriptor table if it is being shared
1606  */
1607 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1608 {
1609         struct files_struct *fd = current->files;
1610         int error = 0;
1611
1612         if ((unshare_flags & CLONE_FILES) &&
1613             (fd && atomic_read(&fd->count) > 1)) {
1614                 *new_fdp = dup_fd(fd, &error);
1615                 if (!*new_fdp)
1616                         return error;
1617         }
1618
1619         return 0;
1620 }
1621
1622 /*
1623  * unshare allows a process to 'unshare' part of the process
1624  * context which was originally shared using clone.  copy_*
1625  * functions used by do_fork() cannot be used here directly
1626  * because they modify an inactive task_struct that is being
1627  * constructed. Here we are modifying the current, active,
1628  * task_struct.
1629  */
1630 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1631 {
1632         int err = 0;
1633         struct fs_struct *fs, *new_fs = NULL;
1634         struct sighand_struct *new_sigh = NULL;
1635         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1636         struct files_struct *fd, *new_fd = NULL;
1637         struct nsproxy *new_nsproxy = NULL;
1638         int do_sysvsem = 0;
1639
1640         check_unshare_flags(&unshare_flags);
1641
1642         /* Return -EINVAL for all unsupported flags */
1643         err = -EINVAL;
1644         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1645                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1646                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1647                 goto bad_unshare_out;
1648
1649         /*
1650          * CLONE_NEWIPC must also detach from the undolist: after switching
1651          * to a new ipc namespace, the semaphore arrays from the old
1652          * namespace are unreachable.
1653          */
1654         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1655                 do_sysvsem = 1;
1656         if ((err = unshare_thread(unshare_flags)))
1657                 goto bad_unshare_out;
1658         if ((err = unshare_fs(unshare_flags, &new_fs)))
1659                 goto bad_unshare_cleanup_thread;
1660         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1661                 goto bad_unshare_cleanup_fs;
1662         if ((err = unshare_vm(unshare_flags, &new_mm)))
1663                 goto bad_unshare_cleanup_sigh;
1664         if ((err = unshare_fd(unshare_flags, &new_fd)))
1665                 goto bad_unshare_cleanup_vm;
1666         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1667                         new_fs)))
1668                 goto bad_unshare_cleanup_fd;
1669
1670         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1671                 if (do_sysvsem) {
1672                         /*
1673                          * CLONE_SYSVSEM is equivalent to sys_exit().
1674                          */
1675                         exit_sem(current);
1676                 }
1677
1678                 if (new_nsproxy) {
1679                         switch_task_namespaces(current, new_nsproxy);
1680                         new_nsproxy = NULL;
1681                 }
1682
1683                 task_lock(current);
1684
1685                 if (new_fs) {
1686                         fs = current->fs;
1687                         write_lock(&fs->lock);
1688                         current->fs = new_fs;
1689                         if (--fs->users)
1690                                 new_fs = NULL;
1691                         else
1692                                 new_fs = fs;
1693                         write_unlock(&fs->lock);
1694                 }
1695
1696                 if (new_mm) {
1697                         mm = current->mm;
1698                         active_mm = current->active_mm;
1699                         current->mm = new_mm;
1700                         current->active_mm = new_mm;
1701                         activate_mm(active_mm, new_mm);
1702                         new_mm = mm;
1703                 }
1704
1705                 if (new_fd) {
1706                         fd = current->files;
1707                         current->files = new_fd;
1708                         new_fd = fd;
1709                 }
1710
1711                 task_unlock(current);
1712         }
1713
1714         if (new_nsproxy)
1715                 put_nsproxy(new_nsproxy);
1716
1717 bad_unshare_cleanup_fd:
1718         if (new_fd)
1719                 put_files_struct(new_fd);
1720
1721 bad_unshare_cleanup_vm:
1722         if (new_mm)
1723                 mmput(new_mm);
1724
1725 bad_unshare_cleanup_sigh:
1726         if (new_sigh)
1727                 if (atomic_dec_and_test(&new_sigh->count))
1728                         kmem_cache_free(sighand_cachep, new_sigh);
1729
1730 bad_unshare_cleanup_fs:
1731         if (new_fs)
1732                 free_fs_struct(new_fs);
1733
1734 bad_unshare_cleanup_thread:
1735 bad_unshare_out:
1736         return err;
1737 }
1738
1739 /*
1740  *      Helper to unshare the files of the current task.
1741  *      We don't want to expose copy_files internals to
1742  *      the exec layer of the kernel.
1743  */
1744
1745 int unshare_files(struct files_struct **displaced)
1746 {
1747         struct task_struct *task = current;
1748         struct files_struct *copy = NULL;
1749         int error;
1750
1751         error = unshare_fd(CLONE_FILES, &copy);
1752         if (error || !copy) {
1753                 *displaced = NULL;
1754                 return error;
1755         }
1756         *displaced = task->files;
1757         task_lock(task);
1758         task->files = copy;
1759         task_unlock(task);
1760         return 0;
1761 }