Merge commit 'v2.6.37' into sched/core
[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 #include <linux/user-return-notifier.h>
68 #include <linux/oom.h>
69
70 #include <asm/pgtable.h>
71 #include <asm/pgalloc.h>
72 #include <asm/uaccess.h>
73 #include <asm/mmu_context.h>
74 #include <asm/cacheflush.h>
75 #include <asm/tlbflush.h>
76
77 #include <trace/events/sched.h>
78
79 /*
80  * Protected counters by write_lock_irq(&tasklist_lock)
81  */
82 unsigned long total_forks;      /* Handle normal Linux uptimes. */
83 int nr_threads;                 /* The idle threads do not count.. */
84
85 int max_threads;                /* tunable limit on nr_threads */
86
87 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
88
89 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
90
91 #ifdef CONFIG_PROVE_RCU
92 int lockdep_tasklist_lock_is_held(void)
93 {
94         return lockdep_is_held(&tasklist_lock);
95 }
96 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
97 #endif /* #ifdef CONFIG_PROVE_RCU */
98
99 int nr_processes(void)
100 {
101         int cpu;
102         int total = 0;
103
104         for_each_possible_cpu(cpu)
105                 total += per_cpu(process_counts, cpu);
106
107         return total;
108 }
109
110 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
111 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
112 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
113 static struct kmem_cache *task_struct_cachep;
114 #endif
115
116 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
117 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
118 {
119 #ifdef CONFIG_DEBUG_STACK_USAGE
120         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
121 #else
122         gfp_t mask = GFP_KERNEL;
123 #endif
124         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
125 }
126
127 static inline void free_thread_info(struct thread_info *ti)
128 {
129         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
130 }
131 #endif
132
133 /* SLAB cache for signal_struct structures (tsk->signal) */
134 static struct kmem_cache *signal_cachep;
135
136 /* SLAB cache for sighand_struct structures (tsk->sighand) */
137 struct kmem_cache *sighand_cachep;
138
139 /* SLAB cache for files_struct structures (tsk->files) */
140 struct kmem_cache *files_cachep;
141
142 /* SLAB cache for fs_struct structures (tsk->fs) */
143 struct kmem_cache *fs_cachep;
144
145 /* SLAB cache for vm_area_struct structures */
146 struct kmem_cache *vm_area_cachep;
147
148 /* SLAB cache for mm_struct structures (tsk->mm) */
149 static struct kmem_cache *mm_cachep;
150
151 static void account_kernel_stack(struct thread_info *ti, int account)
152 {
153         struct zone *zone = page_zone(virt_to_page(ti));
154
155         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
156 }
157
158 void free_task(struct task_struct *tsk)
159 {
160         prop_local_destroy_single(&tsk->dirties);
161         account_kernel_stack(tsk->stack, -1);
162         free_thread_info(tsk->stack);
163         rt_mutex_debug_task_free(tsk);
164         ftrace_graph_exit_task(tsk);
165         free_task_struct(tsk);
166 }
167 EXPORT_SYMBOL(free_task);
168
169 static inline void free_signal_struct(struct signal_struct *sig)
170 {
171         taskstats_tgid_free(sig);
172         kmem_cache_free(signal_cachep, sig);
173 }
174
175 static inline void put_signal_struct(struct signal_struct *sig)
176 {
177         if (atomic_dec_and_test(&sig->sigcnt)) {
178                 sched_autogroup_exit(sig);
179                 free_signal_struct(sig);
180         }
181 }
182
183 void __put_task_struct(struct task_struct *tsk)
184 {
185         WARN_ON(!tsk->exit_state);
186         WARN_ON(atomic_read(&tsk->usage));
187         WARN_ON(tsk == current);
188
189         exit_creds(tsk);
190         delayacct_tsk_free(tsk);
191         put_signal_struct(tsk->signal);
192
193         if (!profile_handoff_task(tsk))
194                 free_task(tsk);
195 }
196
197 /*
198  * macro override instead of weak attribute alias, to workaround
199  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
200  */
201 #ifndef arch_task_cache_init
202 #define arch_task_cache_init()
203 #endif
204
205 void __init fork_init(unsigned long mempages)
206 {
207 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
208 #ifndef ARCH_MIN_TASKALIGN
209 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
210 #endif
211         /* create a slab on which task_structs can be allocated */
212         task_struct_cachep =
213                 kmem_cache_create("task_struct", sizeof(struct task_struct),
214                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
215 #endif
216
217         /* do the arch specific task caches init */
218         arch_task_cache_init();
219
220         /*
221          * The default maximum number of threads is set to a safe
222          * value: the thread structures can take up at most half
223          * of memory.
224          */
225         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
226
227         /*
228          * we need to allow at least 20 threads to boot a system
229          */
230         if(max_threads < 20)
231                 max_threads = 20;
232
233         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
234         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
235         init_task.signal->rlim[RLIMIT_SIGPENDING] =
236                 init_task.signal->rlim[RLIMIT_NPROC];
237 }
238
239 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
240                                                struct task_struct *src)
241 {
242         *dst = *src;
243         return 0;
244 }
245
246 static struct task_struct *dup_task_struct(struct task_struct *orig)
247 {
248         struct task_struct *tsk;
249         struct thread_info *ti;
250         unsigned long *stackend;
251
252         int err;
253
254         prepare_to_copy(orig);
255
256         tsk = alloc_task_struct();
257         if (!tsk)
258                 return NULL;
259
260         ti = alloc_thread_info(tsk);
261         if (!ti) {
262                 free_task_struct(tsk);
263                 return NULL;
264         }
265
266         err = arch_dup_task_struct(tsk, orig);
267         if (err)
268                 goto out;
269
270         tsk->stack = ti;
271
272         err = prop_local_init_single(&tsk->dirties);
273         if (err)
274                 goto out;
275
276         setup_thread_stack(tsk, orig);
277         clear_user_return_notifier(tsk);
278         clear_tsk_need_resched(tsk);
279         stackend = end_of_stack(tsk);
280         *stackend = STACK_END_MAGIC;    /* for overflow detection */
281
282 #ifdef CONFIG_CC_STACKPROTECTOR
283         tsk->stack_canary = get_random_int();
284 #endif
285
286         /* One for us, one for whoever does the "release_task()" (usually parent) */
287         atomic_set(&tsk->usage,2);
288         atomic_set(&tsk->fs_excl, 0);
289 #ifdef CONFIG_BLK_DEV_IO_TRACE
290         tsk->btrace_seq = 0;
291 #endif
292         tsk->splice_pipe = NULL;
293
294         account_kernel_stack(ti, 1);
295
296         return tsk;
297
298 out:
299         free_thread_info(ti);
300         free_task_struct(tsk);
301         return NULL;
302 }
303
304 #ifdef CONFIG_MMU
305 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
306 {
307         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
308         struct rb_node **rb_link, *rb_parent;
309         int retval;
310         unsigned long charge;
311         struct mempolicy *pol;
312
313         down_write(&oldmm->mmap_sem);
314         flush_cache_dup_mm(oldmm);
315         /*
316          * Not linked in yet - no deadlock potential:
317          */
318         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
319
320         mm->locked_vm = 0;
321         mm->mmap = NULL;
322         mm->mmap_cache = NULL;
323         mm->free_area_cache = oldmm->mmap_base;
324         mm->cached_hole_size = ~0UL;
325         mm->map_count = 0;
326         cpumask_clear(mm_cpumask(mm));
327         mm->mm_rb = RB_ROOT;
328         rb_link = &mm->mm_rb.rb_node;
329         rb_parent = NULL;
330         pprev = &mm->mmap;
331         retval = ksm_fork(mm, oldmm);
332         if (retval)
333                 goto out;
334
335         prev = NULL;
336         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
337                 struct file *file;
338
339                 if (mpnt->vm_flags & VM_DONTCOPY) {
340                         long pages = vma_pages(mpnt);
341                         mm->total_vm -= pages;
342                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
343                                                                 -pages);
344                         continue;
345                 }
346                 charge = 0;
347                 if (mpnt->vm_flags & VM_ACCOUNT) {
348                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
349                         if (security_vm_enough_memory(len))
350                                 goto fail_nomem;
351                         charge = len;
352                 }
353                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
354                 if (!tmp)
355                         goto fail_nomem;
356                 *tmp = *mpnt;
357                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
358                 pol = mpol_dup(vma_policy(mpnt));
359                 retval = PTR_ERR(pol);
360                 if (IS_ERR(pol))
361                         goto fail_nomem_policy;
362                 vma_set_policy(tmp, pol);
363                 tmp->vm_mm = mm;
364                 if (anon_vma_fork(tmp, mpnt))
365                         goto fail_nomem_anon_vma_fork;
366                 tmp->vm_flags &= ~VM_LOCKED;
367                 tmp->vm_next = tmp->vm_prev = NULL;
368                 file = tmp->vm_file;
369                 if (file) {
370                         struct inode *inode = file->f_path.dentry->d_inode;
371                         struct address_space *mapping = file->f_mapping;
372
373                         get_file(file);
374                         if (tmp->vm_flags & VM_DENYWRITE)
375                                 atomic_dec(&inode->i_writecount);
376                         spin_lock(&mapping->i_mmap_lock);
377                         if (tmp->vm_flags & VM_SHARED)
378                                 mapping->i_mmap_writable++;
379                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
380                         flush_dcache_mmap_lock(mapping);
381                         /* insert tmp into the share list, just after mpnt */
382                         vma_prio_tree_add(tmp, mpnt);
383                         flush_dcache_mmap_unlock(mapping);
384                         spin_unlock(&mapping->i_mmap_lock);
385                 }
386
387                 /*
388                  * Clear hugetlb-related page reserves for children. This only
389                  * affects MAP_PRIVATE mappings. Faults generated by the child
390                  * are not guaranteed to succeed, even if read-only
391                  */
392                 if (is_vm_hugetlb_page(tmp))
393                         reset_vma_resv_huge_pages(tmp);
394
395                 /*
396                  * Link in the new vma and copy the page table entries.
397                  */
398                 *pprev = tmp;
399                 pprev = &tmp->vm_next;
400                 tmp->vm_prev = prev;
401                 prev = tmp;
402
403                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
404                 rb_link = &tmp->vm_rb.rb_right;
405                 rb_parent = &tmp->vm_rb;
406
407                 mm->map_count++;
408                 retval = copy_page_range(mm, oldmm, mpnt);
409
410                 if (tmp->vm_ops && tmp->vm_ops->open)
411                         tmp->vm_ops->open(tmp);
412
413                 if (retval)
414                         goto out;
415         }
416         /* a new mm has just been created */
417         arch_dup_mmap(oldmm, mm);
418         retval = 0;
419 out:
420         up_write(&mm->mmap_sem);
421         flush_tlb_mm(oldmm);
422         up_write(&oldmm->mmap_sem);
423         return retval;
424 fail_nomem_anon_vma_fork:
425         mpol_put(pol);
426 fail_nomem_policy:
427         kmem_cache_free(vm_area_cachep, tmp);
428 fail_nomem:
429         retval = -ENOMEM;
430         vm_unacct_memory(charge);
431         goto out;
432 }
433
434 static inline int mm_alloc_pgd(struct mm_struct * mm)
435 {
436         mm->pgd = pgd_alloc(mm);
437         if (unlikely(!mm->pgd))
438                 return -ENOMEM;
439         return 0;
440 }
441
442 static inline void mm_free_pgd(struct mm_struct * mm)
443 {
444         pgd_free(mm, mm->pgd);
445 }
446 #else
447 #define dup_mmap(mm, oldmm)     (0)
448 #define mm_alloc_pgd(mm)        (0)
449 #define mm_free_pgd(mm)
450 #endif /* CONFIG_MMU */
451
452 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
453
454 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
455 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
456
457 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
458
459 static int __init coredump_filter_setup(char *s)
460 {
461         default_dump_filter =
462                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
463                 MMF_DUMP_FILTER_MASK;
464         return 1;
465 }
466
467 __setup("coredump_filter=", coredump_filter_setup);
468
469 #include <linux/init_task.h>
470
471 static void mm_init_aio(struct mm_struct *mm)
472 {
473 #ifdef CONFIG_AIO
474         spin_lock_init(&mm->ioctx_lock);
475         INIT_HLIST_HEAD(&mm->ioctx_list);
476 #endif
477 }
478
479 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
480 {
481         atomic_set(&mm->mm_users, 1);
482         atomic_set(&mm->mm_count, 1);
483         init_rwsem(&mm->mmap_sem);
484         INIT_LIST_HEAD(&mm->mmlist);
485         mm->flags = (current->mm) ?
486                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
487         mm->core_state = NULL;
488         mm->nr_ptes = 0;
489         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
490         spin_lock_init(&mm->page_table_lock);
491         mm->free_area_cache = TASK_UNMAPPED_BASE;
492         mm->cached_hole_size = ~0UL;
493         mm_init_aio(mm);
494         mm_init_owner(mm, p);
495         atomic_set(&mm->oom_disable_count, 0);
496
497         if (likely(!mm_alloc_pgd(mm))) {
498                 mm->def_flags = 0;
499                 mmu_notifier_mm_init(mm);
500                 return mm;
501         }
502
503         free_mm(mm);
504         return NULL;
505 }
506
507 /*
508  * Allocate and initialize an mm_struct.
509  */
510 struct mm_struct * mm_alloc(void)
511 {
512         struct mm_struct * mm;
513
514         mm = allocate_mm();
515         if (mm) {
516                 memset(mm, 0, sizeof(*mm));
517                 mm = mm_init(mm, current);
518         }
519         return mm;
520 }
521
522 /*
523  * Called when the last reference to the mm
524  * is dropped: either by a lazy thread or by
525  * mmput. Free the page directory and the mm.
526  */
527 void __mmdrop(struct mm_struct *mm)
528 {
529         BUG_ON(mm == &init_mm);
530         mm_free_pgd(mm);
531         destroy_context(mm);
532         mmu_notifier_mm_destroy(mm);
533         free_mm(mm);
534 }
535 EXPORT_SYMBOL_GPL(__mmdrop);
536
537 /*
538  * Decrement the use count and release all resources for an mm.
539  */
540 void mmput(struct mm_struct *mm)
541 {
542         might_sleep();
543
544         if (atomic_dec_and_test(&mm->mm_users)) {
545                 exit_aio(mm);
546                 ksm_exit(mm);
547                 exit_mmap(mm);
548                 set_mm_exe_file(mm, NULL);
549                 if (!list_empty(&mm->mmlist)) {
550                         spin_lock(&mmlist_lock);
551                         list_del(&mm->mmlist);
552                         spin_unlock(&mmlist_lock);
553                 }
554                 put_swap_token(mm);
555                 if (mm->binfmt)
556                         module_put(mm->binfmt->module);
557                 mmdrop(mm);
558         }
559 }
560 EXPORT_SYMBOL_GPL(mmput);
561
562 /**
563  * get_task_mm - acquire a reference to the task's mm
564  *
565  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
566  * this kernel workthread has transiently adopted a user mm with use_mm,
567  * to do its AIO) is not set and if so returns a reference to it, after
568  * bumping up the use count.  User must release the mm via mmput()
569  * after use.  Typically used by /proc and ptrace.
570  */
571 struct mm_struct *get_task_mm(struct task_struct *task)
572 {
573         struct mm_struct *mm;
574
575         task_lock(task);
576         mm = task->mm;
577         if (mm) {
578                 if (task->flags & PF_KTHREAD)
579                         mm = NULL;
580                 else
581                         atomic_inc(&mm->mm_users);
582         }
583         task_unlock(task);
584         return mm;
585 }
586 EXPORT_SYMBOL_GPL(get_task_mm);
587
588 /* Please note the differences between mmput and mm_release.
589  * mmput is called whenever we stop holding onto a mm_struct,
590  * error success whatever.
591  *
592  * mm_release is called after a mm_struct has been removed
593  * from the current process.
594  *
595  * This difference is important for error handling, when we
596  * only half set up a mm_struct for a new process and need to restore
597  * the old one.  Because we mmput the new mm_struct before
598  * restoring the old one. . .
599  * Eric Biederman 10 January 1998
600  */
601 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
602 {
603         struct completion *vfork_done = tsk->vfork_done;
604
605         /* Get rid of any futexes when releasing the mm */
606 #ifdef CONFIG_FUTEX
607         if (unlikely(tsk->robust_list)) {
608                 exit_robust_list(tsk);
609                 tsk->robust_list = NULL;
610         }
611 #ifdef CONFIG_COMPAT
612         if (unlikely(tsk->compat_robust_list)) {
613                 compat_exit_robust_list(tsk);
614                 tsk->compat_robust_list = NULL;
615         }
616 #endif
617         if (unlikely(!list_empty(&tsk->pi_state_list)))
618                 exit_pi_state_list(tsk);
619 #endif
620
621         /* Get rid of any cached register state */
622         deactivate_mm(tsk, mm);
623
624         /* notify parent sleeping on vfork() */
625         if (vfork_done) {
626                 tsk->vfork_done = NULL;
627                 complete(vfork_done);
628         }
629
630         /*
631          * If we're exiting normally, clear a user-space tid field if
632          * requested.  We leave this alone when dying by signal, to leave
633          * the value intact in a core dump, and to save the unnecessary
634          * trouble otherwise.  Userland only wants this done for a sys_exit.
635          */
636         if (tsk->clear_child_tid) {
637                 if (!(tsk->flags & PF_SIGNALED) &&
638                     atomic_read(&mm->mm_users) > 1) {
639                         /*
640                          * We don't check the error code - if userspace has
641                          * not set up a proper pointer then tough luck.
642                          */
643                         put_user(0, tsk->clear_child_tid);
644                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
645                                         1, NULL, NULL, 0);
646                 }
647                 tsk->clear_child_tid = NULL;
648         }
649 }
650
651 /*
652  * Allocate a new mm structure and copy contents from the
653  * mm structure of the passed in task structure.
654  */
655 struct mm_struct *dup_mm(struct task_struct *tsk)
656 {
657         struct mm_struct *mm, *oldmm = current->mm;
658         int err;
659
660         if (!oldmm)
661                 return NULL;
662
663         mm = allocate_mm();
664         if (!mm)
665                 goto fail_nomem;
666
667         memcpy(mm, oldmm, sizeof(*mm));
668
669         /* Initializing for Swap token stuff */
670         mm->token_priority = 0;
671         mm->last_interval = 0;
672
673         if (!mm_init(mm, tsk))
674                 goto fail_nomem;
675
676         if (init_new_context(tsk, mm))
677                 goto fail_nocontext;
678
679         dup_mm_exe_file(oldmm, mm);
680
681         err = dup_mmap(mm, oldmm);
682         if (err)
683                 goto free_pt;
684
685         mm->hiwater_rss = get_mm_rss(mm);
686         mm->hiwater_vm = mm->total_vm;
687
688         if (mm->binfmt && !try_module_get(mm->binfmt->module))
689                 goto free_pt;
690
691         return mm;
692
693 free_pt:
694         /* don't put binfmt in mmput, we haven't got module yet */
695         mm->binfmt = NULL;
696         mmput(mm);
697
698 fail_nomem:
699         return NULL;
700
701 fail_nocontext:
702         /*
703          * If init_new_context() failed, we cannot use mmput() to free the mm
704          * because it calls destroy_context()
705          */
706         mm_free_pgd(mm);
707         free_mm(mm);
708         return NULL;
709 }
710
711 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
712 {
713         struct mm_struct * mm, *oldmm;
714         int retval;
715
716         tsk->min_flt = tsk->maj_flt = 0;
717         tsk->nvcsw = tsk->nivcsw = 0;
718 #ifdef CONFIG_DETECT_HUNG_TASK
719         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
720 #endif
721
722         tsk->mm = NULL;
723         tsk->active_mm = NULL;
724
725         /*
726          * Are we cloning a kernel thread?
727          *
728          * We need to steal a active VM for that..
729          */
730         oldmm = current->mm;
731         if (!oldmm)
732                 return 0;
733
734         if (clone_flags & CLONE_VM) {
735                 atomic_inc(&oldmm->mm_users);
736                 mm = oldmm;
737                 goto good_mm;
738         }
739
740         retval = -ENOMEM;
741         mm = dup_mm(tsk);
742         if (!mm)
743                 goto fail_nomem;
744
745 good_mm:
746         /* Initializing for Swap token stuff */
747         mm->token_priority = 0;
748         mm->last_interval = 0;
749         if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
750                 atomic_inc(&mm->oom_disable_count);
751
752         tsk->mm = mm;
753         tsk->active_mm = mm;
754         return 0;
755
756 fail_nomem:
757         return retval;
758 }
759
760 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
761 {
762         struct fs_struct *fs = current->fs;
763         if (clone_flags & CLONE_FS) {
764                 /* tsk->fs is already what we want */
765                 spin_lock(&fs->lock);
766                 if (fs->in_exec) {
767                         spin_unlock(&fs->lock);
768                         return -EAGAIN;
769                 }
770                 fs->users++;
771                 spin_unlock(&fs->lock);
772                 return 0;
773         }
774         tsk->fs = copy_fs_struct(fs);
775         if (!tsk->fs)
776                 return -ENOMEM;
777         return 0;
778 }
779
780 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
781 {
782         struct files_struct *oldf, *newf;
783         int error = 0;
784
785         /*
786          * A background process may not have any files ...
787          */
788         oldf = current->files;
789         if (!oldf)
790                 goto out;
791
792         if (clone_flags & CLONE_FILES) {
793                 atomic_inc(&oldf->count);
794                 goto out;
795         }
796
797         newf = dup_fd(oldf, &error);
798         if (!newf)
799                 goto out;
800
801         tsk->files = newf;
802         error = 0;
803 out:
804         return error;
805 }
806
807 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
808 {
809 #ifdef CONFIG_BLOCK
810         struct io_context *ioc = current->io_context;
811
812         if (!ioc)
813                 return 0;
814         /*
815          * Share io context with parent, if CLONE_IO is set
816          */
817         if (clone_flags & CLONE_IO) {
818                 tsk->io_context = ioc_task_link(ioc);
819                 if (unlikely(!tsk->io_context))
820                         return -ENOMEM;
821         } else if (ioprio_valid(ioc->ioprio)) {
822                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
823                 if (unlikely(!tsk->io_context))
824                         return -ENOMEM;
825
826                 tsk->io_context->ioprio = ioc->ioprio;
827         }
828 #endif
829         return 0;
830 }
831
832 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
833 {
834         struct sighand_struct *sig;
835
836         if (clone_flags & CLONE_SIGHAND) {
837                 atomic_inc(&current->sighand->count);
838                 return 0;
839         }
840         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
841         rcu_assign_pointer(tsk->sighand, sig);
842         if (!sig)
843                 return -ENOMEM;
844         atomic_set(&sig->count, 1);
845         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
846         return 0;
847 }
848
849 void __cleanup_sighand(struct sighand_struct *sighand)
850 {
851         if (atomic_dec_and_test(&sighand->count))
852                 kmem_cache_free(sighand_cachep, sighand);
853 }
854
855
856 /*
857  * Initialize POSIX timer handling for a thread group.
858  */
859 static void posix_cpu_timers_init_group(struct signal_struct *sig)
860 {
861         unsigned long cpu_limit;
862
863         /* Thread group counters. */
864         thread_group_cputime_init(sig);
865
866         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
867         if (cpu_limit != RLIM_INFINITY) {
868                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
869                 sig->cputimer.running = 1;
870         }
871
872         /* The timer lists. */
873         INIT_LIST_HEAD(&sig->cpu_timers[0]);
874         INIT_LIST_HEAD(&sig->cpu_timers[1]);
875         INIT_LIST_HEAD(&sig->cpu_timers[2]);
876 }
877
878 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
879 {
880         struct signal_struct *sig;
881
882         if (clone_flags & CLONE_THREAD)
883                 return 0;
884
885         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
886         tsk->signal = sig;
887         if (!sig)
888                 return -ENOMEM;
889
890         sig->nr_threads = 1;
891         atomic_set(&sig->live, 1);
892         atomic_set(&sig->sigcnt, 1);
893         init_waitqueue_head(&sig->wait_chldexit);
894         if (clone_flags & CLONE_NEWPID)
895                 sig->flags |= SIGNAL_UNKILLABLE;
896         sig->curr_target = tsk;
897         init_sigpending(&sig->shared_pending);
898         INIT_LIST_HEAD(&sig->posix_timers);
899
900         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
901         sig->real_timer.function = it_real_fn;
902
903         task_lock(current->group_leader);
904         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
905         task_unlock(current->group_leader);
906
907         posix_cpu_timers_init_group(sig);
908
909         tty_audit_fork(sig);
910         sched_autogroup_fork(sig);
911
912         sig->oom_adj = current->signal->oom_adj;
913         sig->oom_score_adj = current->signal->oom_score_adj;
914
915         mutex_init(&sig->cred_guard_mutex);
916
917         return 0;
918 }
919
920 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
921 {
922         unsigned long new_flags = p->flags;
923
924         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
925         new_flags |= PF_FORKNOEXEC;
926         new_flags |= PF_STARTING;
927         p->flags = new_flags;
928         clear_freeze_flag(p);
929 }
930
931 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
932 {
933         current->clear_child_tid = tidptr;
934
935         return task_pid_vnr(current);
936 }
937
938 static void rt_mutex_init_task(struct task_struct *p)
939 {
940         raw_spin_lock_init(&p->pi_lock);
941 #ifdef CONFIG_RT_MUTEXES
942         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
943         p->pi_blocked_on = NULL;
944 #endif
945 }
946
947 #ifdef CONFIG_MM_OWNER
948 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
949 {
950         mm->owner = p;
951 }
952 #endif /* CONFIG_MM_OWNER */
953
954 /*
955  * Initialize POSIX timer handling for a single task.
956  */
957 static void posix_cpu_timers_init(struct task_struct *tsk)
958 {
959         tsk->cputime_expires.prof_exp = cputime_zero;
960         tsk->cputime_expires.virt_exp = cputime_zero;
961         tsk->cputime_expires.sched_exp = 0;
962         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
963         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
964         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
965 }
966
967 /*
968  * This creates a new process as a copy of the old one,
969  * but does not actually start it yet.
970  *
971  * It copies the registers, and all the appropriate
972  * parts of the process environment (as per the clone
973  * flags). The actual kick-off is left to the caller.
974  */
975 static struct task_struct *copy_process(unsigned long clone_flags,
976                                         unsigned long stack_start,
977                                         struct pt_regs *regs,
978                                         unsigned long stack_size,
979                                         int __user *child_tidptr,
980                                         struct pid *pid,
981                                         int trace)
982 {
983         int retval;
984         struct task_struct *p;
985         int cgroup_callbacks_done = 0;
986
987         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
988                 return ERR_PTR(-EINVAL);
989
990         /*
991          * Thread groups must share signals as well, and detached threads
992          * can only be started up within the thread group.
993          */
994         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
995                 return ERR_PTR(-EINVAL);
996
997         /*
998          * Shared signal handlers imply shared VM. By way of the above,
999          * thread groups also imply shared VM. Blocking this case allows
1000          * for various simplifications in other code.
1001          */
1002         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1003                 return ERR_PTR(-EINVAL);
1004
1005         /*
1006          * Siblings of global init remain as zombies on exit since they are
1007          * not reaped by their parent (swapper). To solve this and to avoid
1008          * multi-rooted process trees, prevent global and container-inits
1009          * from creating siblings.
1010          */
1011         if ((clone_flags & CLONE_PARENT) &&
1012                                 current->signal->flags & SIGNAL_UNKILLABLE)
1013                 return ERR_PTR(-EINVAL);
1014
1015         retval = security_task_create(clone_flags);
1016         if (retval)
1017                 goto fork_out;
1018
1019         retval = -ENOMEM;
1020         p = dup_task_struct(current);
1021         if (!p)
1022                 goto fork_out;
1023
1024         ftrace_graph_init_task(p);
1025
1026         rt_mutex_init_task(p);
1027
1028 #ifdef CONFIG_PROVE_LOCKING
1029         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1030         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1031 #endif
1032         retval = -EAGAIN;
1033         if (atomic_read(&p->real_cred->user->processes) >=
1034                         task_rlimit(p, RLIMIT_NPROC)) {
1035                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1036                     p->real_cred->user != INIT_USER)
1037                         goto bad_fork_free;
1038         }
1039
1040         retval = copy_creds(p, clone_flags);
1041         if (retval < 0)
1042                 goto bad_fork_free;
1043
1044         /*
1045          * If multiple threads are within copy_process(), then this check
1046          * triggers too late. This doesn't hurt, the check is only there
1047          * to stop root fork bombs.
1048          */
1049         retval = -EAGAIN;
1050         if (nr_threads >= max_threads)
1051                 goto bad_fork_cleanup_count;
1052
1053         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1054                 goto bad_fork_cleanup_count;
1055
1056         p->did_exec = 0;
1057         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1058         copy_flags(clone_flags, p);
1059         INIT_LIST_HEAD(&p->children);
1060         INIT_LIST_HEAD(&p->sibling);
1061         rcu_copy_process(p);
1062         p->vfork_done = NULL;
1063         spin_lock_init(&p->alloc_lock);
1064
1065         init_sigpending(&p->pending);
1066
1067         p->utime = cputime_zero;
1068         p->stime = cputime_zero;
1069         p->gtime = cputime_zero;
1070         p->utimescaled = cputime_zero;
1071         p->stimescaled = cputime_zero;
1072 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1073         p->prev_utime = cputime_zero;
1074         p->prev_stime = cputime_zero;
1075 #endif
1076 #if defined(SPLIT_RSS_COUNTING)
1077         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1078 #endif
1079
1080         p->default_timer_slack_ns = current->timer_slack_ns;
1081
1082         task_io_accounting_init(&p->ioac);
1083         acct_clear_integrals(p);
1084
1085         posix_cpu_timers_init(p);
1086
1087         p->lock_depth = -1;             /* -1 = no lock */
1088         do_posix_clock_monotonic_gettime(&p->start_time);
1089         p->real_start_time = p->start_time;
1090         monotonic_to_bootbased(&p->real_start_time);
1091         p->io_context = NULL;
1092         p->audit_context = NULL;
1093         cgroup_fork(p);
1094 #ifdef CONFIG_NUMA
1095         p->mempolicy = mpol_dup(p->mempolicy);
1096         if (IS_ERR(p->mempolicy)) {
1097                 retval = PTR_ERR(p->mempolicy);
1098                 p->mempolicy = NULL;
1099                 goto bad_fork_cleanup_cgroup;
1100         }
1101         mpol_fix_fork_child_flag(p);
1102 #endif
1103 #ifdef CONFIG_TRACE_IRQFLAGS
1104         p->irq_events = 0;
1105 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1106         p->hardirqs_enabled = 1;
1107 #else
1108         p->hardirqs_enabled = 0;
1109 #endif
1110         p->hardirq_enable_ip = 0;
1111         p->hardirq_enable_event = 0;
1112         p->hardirq_disable_ip = _THIS_IP_;
1113         p->hardirq_disable_event = 0;
1114         p->softirqs_enabled = 1;
1115         p->softirq_enable_ip = _THIS_IP_;
1116         p->softirq_enable_event = 0;
1117         p->softirq_disable_ip = 0;
1118         p->softirq_disable_event = 0;
1119         p->hardirq_context = 0;
1120         p->softirq_context = 0;
1121 #endif
1122 #ifdef CONFIG_LOCKDEP
1123         p->lockdep_depth = 0; /* no locks held yet */
1124         p->curr_chain_key = 0;
1125         p->lockdep_recursion = 0;
1126 #endif
1127
1128 #ifdef CONFIG_DEBUG_MUTEXES
1129         p->blocked_on = NULL; /* not blocked yet */
1130 #endif
1131 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1132         p->memcg_batch.do_batch = 0;
1133         p->memcg_batch.memcg = NULL;
1134 #endif
1135
1136         /* Perform scheduler related setup. Assign this task to a CPU. */
1137         sched_fork(p, clone_flags);
1138
1139         retval = perf_event_init_task(p);
1140         if (retval)
1141                 goto bad_fork_cleanup_policy;
1142
1143         if ((retval = audit_alloc(p)))
1144                 goto bad_fork_cleanup_policy;
1145         /* copy all the process information */
1146         if ((retval = copy_semundo(clone_flags, p)))
1147                 goto bad_fork_cleanup_audit;
1148         if ((retval = copy_files(clone_flags, p)))
1149                 goto bad_fork_cleanup_semundo;
1150         if ((retval = copy_fs(clone_flags, p)))
1151                 goto bad_fork_cleanup_files;
1152         if ((retval = copy_sighand(clone_flags, p)))
1153                 goto bad_fork_cleanup_fs;
1154         if ((retval = copy_signal(clone_flags, p)))
1155                 goto bad_fork_cleanup_sighand;
1156         if ((retval = copy_mm(clone_flags, p)))
1157                 goto bad_fork_cleanup_signal;
1158         if ((retval = copy_namespaces(clone_flags, p)))
1159                 goto bad_fork_cleanup_mm;
1160         if ((retval = copy_io(clone_flags, p)))
1161                 goto bad_fork_cleanup_namespaces;
1162         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1163         if (retval)
1164                 goto bad_fork_cleanup_io;
1165
1166         if (pid != &init_struct_pid) {
1167                 retval = -ENOMEM;
1168                 pid = alloc_pid(p->nsproxy->pid_ns);
1169                 if (!pid)
1170                         goto bad_fork_cleanup_io;
1171
1172                 if (clone_flags & CLONE_NEWPID) {
1173                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1174                         if (retval < 0)
1175                                 goto bad_fork_free_pid;
1176                 }
1177         }
1178
1179         p->pid = pid_nr(pid);
1180         p->tgid = p->pid;
1181         if (clone_flags & CLONE_THREAD)
1182                 p->tgid = current->tgid;
1183
1184         if (current->nsproxy != p->nsproxy) {
1185                 retval = ns_cgroup_clone(p, pid);
1186                 if (retval)
1187                         goto bad_fork_free_pid;
1188         }
1189
1190         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1191         /*
1192          * Clear TID on mm_release()?
1193          */
1194         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1195 #ifdef CONFIG_FUTEX
1196         p->robust_list = NULL;
1197 #ifdef CONFIG_COMPAT
1198         p->compat_robust_list = NULL;
1199 #endif
1200         INIT_LIST_HEAD(&p->pi_state_list);
1201         p->pi_state_cache = NULL;
1202 #endif
1203         /*
1204          * sigaltstack should be cleared when sharing the same VM
1205          */
1206         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1207                 p->sas_ss_sp = p->sas_ss_size = 0;
1208
1209         /*
1210          * Syscall tracing and stepping should be turned off in the
1211          * child regardless of CLONE_PTRACE.
1212          */
1213         user_disable_single_step(p);
1214         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1215 #ifdef TIF_SYSCALL_EMU
1216         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1217 #endif
1218         clear_all_latency_tracing(p);
1219
1220         /* ok, now we should be set up.. */
1221         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1222         p->pdeath_signal = 0;
1223         p->exit_state = 0;
1224
1225         /*
1226          * Ok, make it visible to the rest of the system.
1227          * We dont wake it up yet.
1228          */
1229         p->group_leader = p;
1230         INIT_LIST_HEAD(&p->thread_group);
1231
1232         /* Now that the task is set up, run cgroup callbacks if
1233          * necessary. We need to run them before the task is visible
1234          * on the tasklist. */
1235         cgroup_fork_callbacks(p);
1236         cgroup_callbacks_done = 1;
1237
1238         /* Need tasklist lock for parent etc handling! */
1239         write_lock_irq(&tasklist_lock);
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                 current->signal->nr_threads++;
1270                 atomic_inc(&current->signal->live);
1271                 atomic_inc(&current->signal->sigcnt);
1272                 p->group_leader = current->group_leader;
1273                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1274         }
1275
1276         if (likely(p->pid)) {
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                         p->signal->tty = tty_kref_get(current->signal->tty);
1285                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1286                         attach_pid(p, PIDTYPE_SID, task_session(current));
1287                         list_add_tail(&p->sibling, &p->real_parent->children);
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         if (p->io_context)
1308                 exit_io_context(p);
1309 bad_fork_cleanup_namespaces:
1310         exit_task_namespaces(p);
1311 bad_fork_cleanup_mm:
1312         if (p->mm) {
1313                 task_lock(p);
1314                 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1315                         atomic_dec(&p->mm->oom_disable_count);
1316                 task_unlock(p);
1317                 mmput(p->mm);
1318         }
1319 bad_fork_cleanup_signal:
1320         if (!(clone_flags & CLONE_THREAD))
1321                 put_signal_struct(p->signal);
1322 bad_fork_cleanup_sighand:
1323         __cleanup_sighand(p->sighand);
1324 bad_fork_cleanup_fs:
1325         exit_fs(p); /* blocking */
1326 bad_fork_cleanup_files:
1327         exit_files(p); /* blocking */
1328 bad_fork_cleanup_semundo:
1329         exit_sem(p);
1330 bad_fork_cleanup_audit:
1331         audit_free(p);
1332 bad_fork_cleanup_policy:
1333         perf_event_free_task(p);
1334 #ifdef CONFIG_NUMA
1335         mpol_put(p->mempolicy);
1336 bad_fork_cleanup_cgroup:
1337 #endif
1338         cgroup_exit(p, cgroup_callbacks_done);
1339         delayacct_tsk_free(p);
1340         module_put(task_thread_info(p)->exec_domain->module);
1341 bad_fork_cleanup_count:
1342         atomic_dec(&p->cred->user->processes);
1343         exit_creds(p);
1344 bad_fork_free:
1345         free_task(p);
1346 fork_out:
1347         return ERR_PTR(retval);
1348 }
1349
1350 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1351 {
1352         memset(regs, 0, sizeof(struct pt_regs));
1353         return regs;
1354 }
1355
1356 static inline void init_idle_pids(struct pid_link *links)
1357 {
1358         enum pid_type type;
1359
1360         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1361                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1362                 links[type].pid = &init_struct_pid;
1363         }
1364 }
1365
1366 struct task_struct * __cpuinit fork_idle(int cpu)
1367 {
1368         struct task_struct *task;
1369         struct pt_regs regs;
1370
1371         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1372                             &init_struct_pid, 0);
1373         if (!IS_ERR(task)) {
1374                 init_idle_pids(task->pids);
1375                 init_idle(task, cpu);
1376         }
1377
1378         return task;
1379 }
1380
1381 /*
1382  *  Ok, this is the main fork-routine.
1383  *
1384  * It copies the process, and if successful kick-starts
1385  * it and waits for it to finish using the VM if required.
1386  */
1387 long do_fork(unsigned long clone_flags,
1388               unsigned long stack_start,
1389               struct pt_regs *regs,
1390               unsigned long stack_size,
1391               int __user *parent_tidptr,
1392               int __user *child_tidptr)
1393 {
1394         struct task_struct *p;
1395         int trace = 0;
1396         long nr;
1397
1398         /*
1399          * Do some preliminary argument and permissions checking before we
1400          * actually start allocating stuff
1401          */
1402         if (clone_flags & CLONE_NEWUSER) {
1403                 if (clone_flags & CLONE_THREAD)
1404                         return -EINVAL;
1405                 /* hopefully this check will go away when userns support is
1406                  * complete
1407                  */
1408                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1409                                 !capable(CAP_SETGID))
1410                         return -EPERM;
1411         }
1412
1413         /*
1414          * We hope to recycle these flags after 2.6.26
1415          */
1416         if (unlikely(clone_flags & CLONE_STOPPED)) {
1417                 static int __read_mostly count = 100;
1418
1419                 if (count > 0 && printk_ratelimit()) {
1420                         char comm[TASK_COMM_LEN];
1421
1422                         count--;
1423                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1424                                         "clone flags 0x%lx\n",
1425                                 get_task_comm(comm, current),
1426                                 clone_flags & CLONE_STOPPED);
1427                 }
1428         }
1429
1430         /*
1431          * When called from kernel_thread, don't do user tracing stuff.
1432          */
1433         if (likely(user_mode(regs)))
1434                 trace = tracehook_prepare_clone(clone_flags);
1435
1436         p = copy_process(clone_flags, stack_start, regs, stack_size,
1437                          child_tidptr, NULL, trace);
1438         /*
1439          * Do this prior waking up the new thread - the thread pointer
1440          * might get invalid after that point, if the thread exits quickly.
1441          */
1442         if (!IS_ERR(p)) {
1443                 struct completion vfork;
1444
1445                 trace_sched_process_fork(current, p);
1446
1447                 nr = task_pid_vnr(p);
1448
1449                 if (clone_flags & CLONE_PARENT_SETTID)
1450                         put_user(nr, parent_tidptr);
1451
1452                 if (clone_flags & CLONE_VFORK) {
1453                         p->vfork_done = &vfork;
1454                         init_completion(&vfork);
1455                 }
1456
1457                 audit_finish_fork(p);
1458                 tracehook_report_clone(regs, clone_flags, nr, p);
1459
1460                 /*
1461                  * We set PF_STARTING at creation in case tracing wants to
1462                  * use this to distinguish a fully live task from one that
1463                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1464                  * clear it and set the child going.
1465                  */
1466                 p->flags &= ~PF_STARTING;
1467
1468                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1469                         /*
1470                          * We'll start up with an immediate SIGSTOP.
1471                          */
1472                         sigaddset(&p->pending.signal, SIGSTOP);
1473                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1474                         __set_task_state(p, TASK_STOPPED);
1475                 } else {
1476                         wake_up_new_task(p, clone_flags);
1477                 }
1478
1479                 tracehook_report_clone_complete(trace, regs,
1480                                                 clone_flags, nr, p);
1481
1482                 if (clone_flags & CLONE_VFORK) {
1483                         freezer_do_not_count();
1484                         wait_for_completion(&vfork);
1485                         freezer_count();
1486                         tracehook_report_vfork_done(p, nr);
1487                 }
1488         } else {
1489                 nr = PTR_ERR(p);
1490         }
1491         return nr;
1492 }
1493
1494 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1495 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1496 #endif
1497
1498 static void sighand_ctor(void *data)
1499 {
1500         struct sighand_struct *sighand = data;
1501
1502         spin_lock_init(&sighand->siglock);
1503         init_waitqueue_head(&sighand->signalfd_wqh);
1504 }
1505
1506 void __init proc_caches_init(void)
1507 {
1508         sighand_cachep = kmem_cache_create("sighand_cache",
1509                         sizeof(struct sighand_struct), 0,
1510                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1511                         SLAB_NOTRACK, sighand_ctor);
1512         signal_cachep = kmem_cache_create("signal_cache",
1513                         sizeof(struct signal_struct), 0,
1514                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1515         files_cachep = kmem_cache_create("files_cache",
1516                         sizeof(struct files_struct), 0,
1517                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1518         fs_cachep = kmem_cache_create("fs_cache",
1519                         sizeof(struct fs_struct), 0,
1520                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1521         mm_cachep = kmem_cache_create("mm_struct",
1522                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1523                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1524         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1525         mmap_init();
1526 }
1527
1528 /*
1529  * Check constraints on flags passed to the unshare system call and
1530  * force unsharing of additional process context as appropriate.
1531  */
1532 static void check_unshare_flags(unsigned long *flags_ptr)
1533 {
1534         /*
1535          * If unsharing a thread from a thread group, must also
1536          * unshare vm.
1537          */
1538         if (*flags_ptr & CLONE_THREAD)
1539                 *flags_ptr |= CLONE_VM;
1540
1541         /*
1542          * If unsharing vm, must also unshare signal handlers.
1543          */
1544         if (*flags_ptr & CLONE_VM)
1545                 *flags_ptr |= CLONE_SIGHAND;
1546
1547         /*
1548          * If unsharing namespace, must also unshare filesystem information.
1549          */
1550         if (*flags_ptr & CLONE_NEWNS)
1551                 *flags_ptr |= CLONE_FS;
1552 }
1553
1554 /*
1555  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1556  */
1557 static int unshare_thread(unsigned long unshare_flags)
1558 {
1559         if (unshare_flags & CLONE_THREAD)
1560                 return -EINVAL;
1561
1562         return 0;
1563 }
1564
1565 /*
1566  * Unshare the filesystem structure if it is being shared
1567  */
1568 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1569 {
1570         struct fs_struct *fs = current->fs;
1571
1572         if (!(unshare_flags & CLONE_FS) || !fs)
1573                 return 0;
1574
1575         /* don't need lock here; in the worst case we'll do useless copy */
1576         if (fs->users == 1)
1577                 return 0;
1578
1579         *new_fsp = copy_fs_struct(fs);
1580         if (!*new_fsp)
1581                 return -ENOMEM;
1582
1583         return 0;
1584 }
1585
1586 /*
1587  * Unsharing of sighand is not supported yet
1588  */
1589 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1590 {
1591         struct sighand_struct *sigh = current->sighand;
1592
1593         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1594                 return -EINVAL;
1595         else
1596                 return 0;
1597 }
1598
1599 /*
1600  * Unshare vm if it is being shared
1601  */
1602 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1603 {
1604         struct mm_struct *mm = current->mm;
1605
1606         if ((unshare_flags & CLONE_VM) &&
1607             (mm && atomic_read(&mm->mm_users) > 1)) {
1608                 return -EINVAL;
1609         }
1610
1611         return 0;
1612 }
1613
1614 /*
1615  * Unshare file descriptor table if it is being shared
1616  */
1617 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1618 {
1619         struct files_struct *fd = current->files;
1620         int error = 0;
1621
1622         if ((unshare_flags & CLONE_FILES) &&
1623             (fd && atomic_read(&fd->count) > 1)) {
1624                 *new_fdp = dup_fd(fd, &error);
1625                 if (!*new_fdp)
1626                         return error;
1627         }
1628
1629         return 0;
1630 }
1631
1632 /*
1633  * unshare allows a process to 'unshare' part of the process
1634  * context which was originally shared using clone.  copy_*
1635  * functions used by do_fork() cannot be used here directly
1636  * because they modify an inactive task_struct that is being
1637  * constructed. Here we are modifying the current, active,
1638  * task_struct.
1639  */
1640 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1641 {
1642         int err = 0;
1643         struct fs_struct *fs, *new_fs = NULL;
1644         struct sighand_struct *new_sigh = NULL;
1645         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1646         struct files_struct *fd, *new_fd = NULL;
1647         struct nsproxy *new_nsproxy = NULL;
1648         int do_sysvsem = 0;
1649
1650         check_unshare_flags(&unshare_flags);
1651
1652         /* Return -EINVAL for all unsupported flags */
1653         err = -EINVAL;
1654         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1655                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1656                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1657                 goto bad_unshare_out;
1658
1659         /*
1660          * CLONE_NEWIPC must also detach from the undolist: after switching
1661          * to a new ipc namespace, the semaphore arrays from the old
1662          * namespace are unreachable.
1663          */
1664         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1665                 do_sysvsem = 1;
1666         if ((err = unshare_thread(unshare_flags)))
1667                 goto bad_unshare_out;
1668         if ((err = unshare_fs(unshare_flags, &new_fs)))
1669                 goto bad_unshare_cleanup_thread;
1670         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1671                 goto bad_unshare_cleanup_fs;
1672         if ((err = unshare_vm(unshare_flags, &new_mm)))
1673                 goto bad_unshare_cleanup_sigh;
1674         if ((err = unshare_fd(unshare_flags, &new_fd)))
1675                 goto bad_unshare_cleanup_vm;
1676         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1677                         new_fs)))
1678                 goto bad_unshare_cleanup_fd;
1679
1680         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1681                 if (do_sysvsem) {
1682                         /*
1683                          * CLONE_SYSVSEM is equivalent to sys_exit().
1684                          */
1685                         exit_sem(current);
1686                 }
1687
1688                 if (new_nsproxy) {
1689                         switch_task_namespaces(current, new_nsproxy);
1690                         new_nsproxy = NULL;
1691                 }
1692
1693                 task_lock(current);
1694
1695                 if (new_fs) {
1696                         fs = current->fs;
1697                         spin_lock(&fs->lock);
1698                         current->fs = new_fs;
1699                         if (--fs->users)
1700                                 new_fs = NULL;
1701                         else
1702                                 new_fs = fs;
1703                         spin_unlock(&fs->lock);
1704                 }
1705
1706                 if (new_mm) {
1707                         mm = current->mm;
1708                         active_mm = current->active_mm;
1709                         current->mm = new_mm;
1710                         current->active_mm = new_mm;
1711                         if (current->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
1712                                 atomic_dec(&mm->oom_disable_count);
1713                                 atomic_inc(&new_mm->oom_disable_count);
1714                         }
1715                         activate_mm(active_mm, new_mm);
1716                         new_mm = mm;
1717                 }
1718
1719                 if (new_fd) {
1720                         fd = current->files;
1721                         current->files = new_fd;
1722                         new_fd = fd;
1723                 }
1724
1725                 task_unlock(current);
1726         }
1727
1728         if (new_nsproxy)
1729                 put_nsproxy(new_nsproxy);
1730
1731 bad_unshare_cleanup_fd:
1732         if (new_fd)
1733                 put_files_struct(new_fd);
1734
1735 bad_unshare_cleanup_vm:
1736         if (new_mm)
1737                 mmput(new_mm);
1738
1739 bad_unshare_cleanup_sigh:
1740         if (new_sigh)
1741                 if (atomic_dec_and_test(&new_sigh->count))
1742                         kmem_cache_free(sighand_cachep, new_sigh);
1743
1744 bad_unshare_cleanup_fs:
1745         if (new_fs)
1746                 free_fs_struct(new_fs);
1747
1748 bad_unshare_cleanup_thread:
1749 bad_unshare_out:
1750         return err;
1751 }
1752
1753 /*
1754  *      Helper to unshare the files of the current task.
1755  *      We don't want to expose copy_files internals to
1756  *      the exec layer of the kernel.
1757  */
1758
1759 int unshare_files(struct files_struct **displaced)
1760 {
1761         struct task_struct *task = current;
1762         struct files_struct *copy = NULL;
1763         int error;
1764
1765         error = unshare_fd(CLONE_FILES, &copy);
1766         if (error || !copy) {
1767                 *displaced = NULL;
1768                 return error;
1769         }
1770         *displaced = task->files;
1771         task_lock(task);
1772         task->files = copy;
1773         task_unlock(task);
1774         return 0;
1775 }