sched: Move sched_autogroup_exit() to free_signal_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 #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         sched_autogroup_exit(sig);
173         kmem_cache_free(signal_cachep, sig);
174 }
175
176 static inline void put_signal_struct(struct signal_struct *sig)
177 {
178         if (atomic_dec_and_test(&sig->sigcnt))
179                 free_signal_struct(sig);
180 }
181
182 void __put_task_struct(struct task_struct *tsk)
183 {
184         WARN_ON(!tsk->exit_state);
185         WARN_ON(atomic_read(&tsk->usage));
186         WARN_ON(tsk == current);
187
188         exit_creds(tsk);
189         delayacct_tsk_free(tsk);
190         put_signal_struct(tsk->signal);
191
192         if (!profile_handoff_task(tsk))
193                 free_task(tsk);
194 }
195
196 /*
197  * macro override instead of weak attribute alias, to workaround
198  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
199  */
200 #ifndef arch_task_cache_init
201 #define arch_task_cache_init()
202 #endif
203
204 void __init fork_init(unsigned long mempages)
205 {
206 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
207 #ifndef ARCH_MIN_TASKALIGN
208 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
209 #endif
210         /* create a slab on which task_structs can be allocated */
211         task_struct_cachep =
212                 kmem_cache_create("task_struct", sizeof(struct task_struct),
213                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
214 #endif
215
216         /* do the arch specific task caches init */
217         arch_task_cache_init();
218
219         /*
220          * The default maximum number of threads is set to a safe
221          * value: the thread structures can take up at most half
222          * of memory.
223          */
224         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
225
226         /*
227          * we need to allow at least 20 threads to boot a system
228          */
229         if(max_threads < 20)
230                 max_threads = 20;
231
232         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
233         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
234         init_task.signal->rlim[RLIMIT_SIGPENDING] =
235                 init_task.signal->rlim[RLIMIT_NPROC];
236 }
237
238 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
239                                                struct task_struct *src)
240 {
241         *dst = *src;
242         return 0;
243 }
244
245 static struct task_struct *dup_task_struct(struct task_struct *orig)
246 {
247         struct task_struct *tsk;
248         struct thread_info *ti;
249         unsigned long *stackend;
250
251         int err;
252
253         prepare_to_copy(orig);
254
255         tsk = alloc_task_struct();
256         if (!tsk)
257                 return NULL;
258
259         ti = alloc_thread_info(tsk);
260         if (!ti) {
261                 free_task_struct(tsk);
262                 return NULL;
263         }
264
265         err = arch_dup_task_struct(tsk, orig);
266         if (err)
267                 goto out;
268
269         tsk->stack = ti;
270
271         err = prop_local_init_single(&tsk->dirties);
272         if (err)
273                 goto out;
274
275         setup_thread_stack(tsk, orig);
276         clear_user_return_notifier(tsk);
277         clear_tsk_need_resched(tsk);
278         stackend = end_of_stack(tsk);
279         *stackend = STACK_END_MAGIC;    /* for overflow detection */
280
281 #ifdef CONFIG_CC_STACKPROTECTOR
282         tsk->stack_canary = get_random_int();
283 #endif
284
285         /* One for us, one for whoever does the "release_task()" (usually parent) */
286         atomic_set(&tsk->usage,2);
287         atomic_set(&tsk->fs_excl, 0);
288 #ifdef CONFIG_BLK_DEV_IO_TRACE
289         tsk->btrace_seq = 0;
290 #endif
291         tsk->splice_pipe = NULL;
292
293         account_kernel_stack(ti, 1);
294
295         return tsk;
296
297 out:
298         free_thread_info(ti);
299         free_task_struct(tsk);
300         return NULL;
301 }
302
303 #ifdef CONFIG_MMU
304 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
305 {
306         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
307         struct rb_node **rb_link, *rb_parent;
308         int retval;
309         unsigned long charge;
310         struct mempolicy *pol;
311
312         down_write(&oldmm->mmap_sem);
313         flush_cache_dup_mm(oldmm);
314         /*
315          * Not linked in yet - no deadlock potential:
316          */
317         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
318
319         mm->locked_vm = 0;
320         mm->mmap = NULL;
321         mm->mmap_cache = NULL;
322         mm->free_area_cache = oldmm->mmap_base;
323         mm->cached_hole_size = ~0UL;
324         mm->map_count = 0;
325         cpumask_clear(mm_cpumask(mm));
326         mm->mm_rb = RB_ROOT;
327         rb_link = &mm->mm_rb.rb_node;
328         rb_parent = NULL;
329         pprev = &mm->mmap;
330         retval = ksm_fork(mm, oldmm);
331         if (retval)
332                 goto out;
333
334         prev = NULL;
335         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
336                 struct file *file;
337
338                 if (mpnt->vm_flags & VM_DONTCOPY) {
339                         long pages = vma_pages(mpnt);
340                         mm->total_vm -= pages;
341                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
342                                                                 -pages);
343                         continue;
344                 }
345                 charge = 0;
346                 if (mpnt->vm_flags & VM_ACCOUNT) {
347                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
348                         if (security_vm_enough_memory(len))
349                                 goto fail_nomem;
350                         charge = len;
351                 }
352                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
353                 if (!tmp)
354                         goto fail_nomem;
355                 *tmp = *mpnt;
356                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
357                 pol = mpol_dup(vma_policy(mpnt));
358                 retval = PTR_ERR(pol);
359                 if (IS_ERR(pol))
360                         goto fail_nomem_policy;
361                 vma_set_policy(tmp, pol);
362                 tmp->vm_mm = mm;
363                 if (anon_vma_fork(tmp, mpnt))
364                         goto fail_nomem_anon_vma_fork;
365                 tmp->vm_flags &= ~VM_LOCKED;
366                 tmp->vm_next = tmp->vm_prev = NULL;
367                 file = tmp->vm_file;
368                 if (file) {
369                         struct inode *inode = file->f_path.dentry->d_inode;
370                         struct address_space *mapping = file->f_mapping;
371
372                         get_file(file);
373                         if (tmp->vm_flags & VM_DENYWRITE)
374                                 atomic_dec(&inode->i_writecount);
375                         spin_lock(&mapping->i_mmap_lock);
376                         if (tmp->vm_flags & VM_SHARED)
377                                 mapping->i_mmap_writable++;
378                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
379                         flush_dcache_mmap_lock(mapping);
380                         /* insert tmp into the share list, just after mpnt */
381                         vma_prio_tree_add(tmp, mpnt);
382                         flush_dcache_mmap_unlock(mapping);
383                         spin_unlock(&mapping->i_mmap_lock);
384                 }
385
386                 /*
387                  * Clear hugetlb-related page reserves for children. This only
388                  * affects MAP_PRIVATE mappings. Faults generated by the child
389                  * are not guaranteed to succeed, even if read-only
390                  */
391                 if (is_vm_hugetlb_page(tmp))
392                         reset_vma_resv_huge_pages(tmp);
393
394                 /*
395                  * Link in the new vma and copy the page table entries.
396                  */
397                 *pprev = tmp;
398                 pprev = &tmp->vm_next;
399                 tmp->vm_prev = prev;
400                 prev = tmp;
401
402                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
403                 rb_link = &tmp->vm_rb.rb_right;
404                 rb_parent = &tmp->vm_rb;
405
406                 mm->map_count++;
407                 retval = copy_page_range(mm, oldmm, mpnt);
408
409                 if (tmp->vm_ops && tmp->vm_ops->open)
410                         tmp->vm_ops->open(tmp);
411
412                 if (retval)
413                         goto out;
414         }
415         /* a new mm has just been created */
416         arch_dup_mmap(oldmm, mm);
417         retval = 0;
418 out:
419         up_write(&mm->mmap_sem);
420         flush_tlb_mm(oldmm);
421         up_write(&oldmm->mmap_sem);
422         return retval;
423 fail_nomem_anon_vma_fork:
424         mpol_put(pol);
425 fail_nomem_policy:
426         kmem_cache_free(vm_area_cachep, tmp);
427 fail_nomem:
428         retval = -ENOMEM;
429         vm_unacct_memory(charge);
430         goto out;
431 }
432
433 static inline int mm_alloc_pgd(struct mm_struct * mm)
434 {
435         mm->pgd = pgd_alloc(mm);
436         if (unlikely(!mm->pgd))
437                 return -ENOMEM;
438         return 0;
439 }
440
441 static inline void mm_free_pgd(struct mm_struct * mm)
442 {
443         pgd_free(mm, mm->pgd);
444 }
445 #else
446 #define dup_mmap(mm, oldmm)     (0)
447 #define mm_alloc_pgd(mm)        (0)
448 #define mm_free_pgd(mm)
449 #endif /* CONFIG_MMU */
450
451 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
452
453 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
454 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
455
456 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
457
458 static int __init coredump_filter_setup(char *s)
459 {
460         default_dump_filter =
461                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
462                 MMF_DUMP_FILTER_MASK;
463         return 1;
464 }
465
466 __setup("coredump_filter=", coredump_filter_setup);
467
468 #include <linux/init_task.h>
469
470 static void mm_init_aio(struct mm_struct *mm)
471 {
472 #ifdef CONFIG_AIO
473         spin_lock_init(&mm->ioctx_lock);
474         INIT_HLIST_HEAD(&mm->ioctx_list);
475 #endif
476 }
477
478 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
479 {
480         atomic_set(&mm->mm_users, 1);
481         atomic_set(&mm->mm_count, 1);
482         init_rwsem(&mm->mmap_sem);
483         INIT_LIST_HEAD(&mm->mmlist);
484         mm->flags = (current->mm) ?
485                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
486         mm->core_state = NULL;
487         mm->nr_ptes = 0;
488         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
489         spin_lock_init(&mm->page_table_lock);
490         mm->free_area_cache = TASK_UNMAPPED_BASE;
491         mm->cached_hole_size = ~0UL;
492         mm_init_aio(mm);
493         mm_init_owner(mm, p);
494         atomic_set(&mm->oom_disable_count, 0);
495
496         if (likely(!mm_alloc_pgd(mm))) {
497                 mm->def_flags = 0;
498                 mmu_notifier_mm_init(mm);
499                 return mm;
500         }
501
502         free_mm(mm);
503         return NULL;
504 }
505
506 /*
507  * Allocate and initialize an mm_struct.
508  */
509 struct mm_struct * mm_alloc(void)
510 {
511         struct mm_struct * mm;
512
513         mm = allocate_mm();
514         if (mm) {
515                 memset(mm, 0, sizeof(*mm));
516                 mm = mm_init(mm, current);
517         }
518         return mm;
519 }
520
521 /*
522  * Called when the last reference to the mm
523  * is dropped: either by a lazy thread or by
524  * mmput. Free the page directory and the mm.
525  */
526 void __mmdrop(struct mm_struct *mm)
527 {
528         BUG_ON(mm == &init_mm);
529         mm_free_pgd(mm);
530         destroy_context(mm);
531         mmu_notifier_mm_destroy(mm);
532         free_mm(mm);
533 }
534 EXPORT_SYMBOL_GPL(__mmdrop);
535
536 /*
537  * Decrement the use count and release all resources for an mm.
538  */
539 void mmput(struct mm_struct *mm)
540 {
541         might_sleep();
542
543         if (atomic_dec_and_test(&mm->mm_users)) {
544                 exit_aio(mm);
545                 ksm_exit(mm);
546                 exit_mmap(mm);
547                 set_mm_exe_file(mm, NULL);
548                 if (!list_empty(&mm->mmlist)) {
549                         spin_lock(&mmlist_lock);
550                         list_del(&mm->mmlist);
551                         spin_unlock(&mmlist_lock);
552                 }
553                 put_swap_token(mm);
554                 if (mm->binfmt)
555                         module_put(mm->binfmt->module);
556                 mmdrop(mm);
557         }
558 }
559 EXPORT_SYMBOL_GPL(mmput);
560
561 /**
562  * get_task_mm - acquire a reference to the task's mm
563  *
564  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
565  * this kernel workthread has transiently adopted a user mm with use_mm,
566  * to do its AIO) is not set and if so returns a reference to it, after
567  * bumping up the use count.  User must release the mm via mmput()
568  * after use.  Typically used by /proc and ptrace.
569  */
570 struct mm_struct *get_task_mm(struct task_struct *task)
571 {
572         struct mm_struct *mm;
573
574         task_lock(task);
575         mm = task->mm;
576         if (mm) {
577                 if (task->flags & PF_KTHREAD)
578                         mm = NULL;
579                 else
580                         atomic_inc(&mm->mm_users);
581         }
582         task_unlock(task);
583         return mm;
584 }
585 EXPORT_SYMBOL_GPL(get_task_mm);
586
587 /* Please note the differences between mmput and mm_release.
588  * mmput is called whenever we stop holding onto a mm_struct,
589  * error success whatever.
590  *
591  * mm_release is called after a mm_struct has been removed
592  * from the current process.
593  *
594  * This difference is important for error handling, when we
595  * only half set up a mm_struct for a new process and need to restore
596  * the old one.  Because we mmput the new mm_struct before
597  * restoring the old one. . .
598  * Eric Biederman 10 January 1998
599  */
600 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
601 {
602         struct completion *vfork_done = tsk->vfork_done;
603
604         /* Get rid of any futexes when releasing the mm */
605 #ifdef CONFIG_FUTEX
606         if (unlikely(tsk->robust_list)) {
607                 exit_robust_list(tsk);
608                 tsk->robust_list = NULL;
609         }
610 #ifdef CONFIG_COMPAT
611         if (unlikely(tsk->compat_robust_list)) {
612                 compat_exit_robust_list(tsk);
613                 tsk->compat_robust_list = NULL;
614         }
615 #endif
616         if (unlikely(!list_empty(&tsk->pi_state_list)))
617                 exit_pi_state_list(tsk);
618 #endif
619
620         /* Get rid of any cached register state */
621         deactivate_mm(tsk, mm);
622
623         /* notify parent sleeping on vfork() */
624         if (vfork_done) {
625                 tsk->vfork_done = NULL;
626                 complete(vfork_done);
627         }
628
629         /*
630          * If we're exiting normally, clear a user-space tid field if
631          * requested.  We leave this alone when dying by signal, to leave
632          * the value intact in a core dump, and to save the unnecessary
633          * trouble otherwise.  Userland only wants this done for a sys_exit.
634          */
635         if (tsk->clear_child_tid) {
636                 if (!(tsk->flags & PF_SIGNALED) &&
637                     atomic_read(&mm->mm_users) > 1) {
638                         /*
639                          * We don't check the error code - if userspace has
640                          * not set up a proper pointer then tough luck.
641                          */
642                         put_user(0, tsk->clear_child_tid);
643                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
644                                         1, NULL, NULL, 0);
645                 }
646                 tsk->clear_child_tid = NULL;
647         }
648 }
649
650 /*
651  * Allocate a new mm structure and copy contents from the
652  * mm structure of the passed in task structure.
653  */
654 struct mm_struct *dup_mm(struct task_struct *tsk)
655 {
656         struct mm_struct *mm, *oldmm = current->mm;
657         int err;
658
659         if (!oldmm)
660                 return NULL;
661
662         mm = allocate_mm();
663         if (!mm)
664                 goto fail_nomem;
665
666         memcpy(mm, oldmm, sizeof(*mm));
667
668         /* Initializing for Swap token stuff */
669         mm->token_priority = 0;
670         mm->last_interval = 0;
671
672         if (!mm_init(mm, tsk))
673                 goto fail_nomem;
674
675         if (init_new_context(tsk, mm))
676                 goto fail_nocontext;
677
678         dup_mm_exe_file(oldmm, mm);
679
680         err = dup_mmap(mm, oldmm);
681         if (err)
682                 goto free_pt;
683
684         mm->hiwater_rss = get_mm_rss(mm);
685         mm->hiwater_vm = mm->total_vm;
686
687         if (mm->binfmt && !try_module_get(mm->binfmt->module))
688                 goto free_pt;
689
690         return mm;
691
692 free_pt:
693         /* don't put binfmt in mmput, we haven't got module yet */
694         mm->binfmt = NULL;
695         mmput(mm);
696
697 fail_nomem:
698         return NULL;
699
700 fail_nocontext:
701         /*
702          * If init_new_context() failed, we cannot use mmput() to free the mm
703          * because it calls destroy_context()
704          */
705         mm_free_pgd(mm);
706         free_mm(mm);
707         return NULL;
708 }
709
710 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
711 {
712         struct mm_struct * mm, *oldmm;
713         int retval;
714
715         tsk->min_flt = tsk->maj_flt = 0;
716         tsk->nvcsw = tsk->nivcsw = 0;
717 #ifdef CONFIG_DETECT_HUNG_TASK
718         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
719 #endif
720
721         tsk->mm = NULL;
722         tsk->active_mm = NULL;
723
724         /*
725          * Are we cloning a kernel thread?
726          *
727          * We need to steal a active VM for that..
728          */
729         oldmm = current->mm;
730         if (!oldmm)
731                 return 0;
732
733         if (clone_flags & CLONE_VM) {
734                 atomic_inc(&oldmm->mm_users);
735                 mm = oldmm;
736                 goto good_mm;
737         }
738
739         retval = -ENOMEM;
740         mm = dup_mm(tsk);
741         if (!mm)
742                 goto fail_nomem;
743
744 good_mm:
745         /* Initializing for Swap token stuff */
746         mm->token_priority = 0;
747         mm->last_interval = 0;
748         if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
749                 atomic_inc(&mm->oom_disable_count);
750
751         tsk->mm = mm;
752         tsk->active_mm = mm;
753         return 0;
754
755 fail_nomem:
756         return retval;
757 }
758
759 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
760 {
761         struct fs_struct *fs = current->fs;
762         if (clone_flags & CLONE_FS) {
763                 /* tsk->fs is already what we want */
764                 spin_lock(&fs->lock);
765                 if (fs->in_exec) {
766                         spin_unlock(&fs->lock);
767                         return -EAGAIN;
768                 }
769                 fs->users++;
770                 spin_unlock(&fs->lock);
771                 return 0;
772         }
773         tsk->fs = copy_fs_struct(fs);
774         if (!tsk->fs)
775                 return -ENOMEM;
776         return 0;
777 }
778
779 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
780 {
781         struct files_struct *oldf, *newf;
782         int error = 0;
783
784         /*
785          * A background process may not have any files ...
786          */
787         oldf = current->files;
788         if (!oldf)
789                 goto out;
790
791         if (clone_flags & CLONE_FILES) {
792                 atomic_inc(&oldf->count);
793                 goto out;
794         }
795
796         newf = dup_fd(oldf, &error);
797         if (!newf)
798                 goto out;
799
800         tsk->files = newf;
801         error = 0;
802 out:
803         return error;
804 }
805
806 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
807 {
808 #ifdef CONFIG_BLOCK
809         struct io_context *ioc = current->io_context;
810
811         if (!ioc)
812                 return 0;
813         /*
814          * Share io context with parent, if CLONE_IO is set
815          */
816         if (clone_flags & CLONE_IO) {
817                 tsk->io_context = ioc_task_link(ioc);
818                 if (unlikely(!tsk->io_context))
819                         return -ENOMEM;
820         } else if (ioprio_valid(ioc->ioprio)) {
821                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
822                 if (unlikely(!tsk->io_context))
823                         return -ENOMEM;
824
825                 tsk->io_context->ioprio = ioc->ioprio;
826         }
827 #endif
828         return 0;
829 }
830
831 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
832 {
833         struct sighand_struct *sig;
834
835         if (clone_flags & CLONE_SIGHAND) {
836                 atomic_inc(&current->sighand->count);
837                 return 0;
838         }
839         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
840         rcu_assign_pointer(tsk->sighand, sig);
841         if (!sig)
842                 return -ENOMEM;
843         atomic_set(&sig->count, 1);
844         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
845         return 0;
846 }
847
848 void __cleanup_sighand(struct sighand_struct *sighand)
849 {
850         if (atomic_dec_and_test(&sighand->count))
851                 kmem_cache_free(sighand_cachep, sighand);
852 }
853
854
855 /*
856  * Initialize POSIX timer handling for a thread group.
857  */
858 static void posix_cpu_timers_init_group(struct signal_struct *sig)
859 {
860         unsigned long cpu_limit;
861
862         /* Thread group counters. */
863         thread_group_cputime_init(sig);
864
865         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
866         if (cpu_limit != RLIM_INFINITY) {
867                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
868                 sig->cputimer.running = 1;
869         }
870
871         /* The timer lists. */
872         INIT_LIST_HEAD(&sig->cpu_timers[0]);
873         INIT_LIST_HEAD(&sig->cpu_timers[1]);
874         INIT_LIST_HEAD(&sig->cpu_timers[2]);
875 }
876
877 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
878 {
879         struct signal_struct *sig;
880
881         if (clone_flags & CLONE_THREAD)
882                 return 0;
883
884         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
885         tsk->signal = sig;
886         if (!sig)
887                 return -ENOMEM;
888
889         sig->nr_threads = 1;
890         atomic_set(&sig->live, 1);
891         atomic_set(&sig->sigcnt, 1);
892         init_waitqueue_head(&sig->wait_chldexit);
893         if (clone_flags & CLONE_NEWPID)
894                 sig->flags |= SIGNAL_UNKILLABLE;
895         sig->curr_target = tsk;
896         init_sigpending(&sig->shared_pending);
897         INIT_LIST_HEAD(&sig->posix_timers);
898
899         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
900         sig->real_timer.function = it_real_fn;
901
902         task_lock(current->group_leader);
903         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
904         task_unlock(current->group_leader);
905
906         posix_cpu_timers_init_group(sig);
907
908         tty_audit_fork(sig);
909         sched_autogroup_fork(sig);
910
911         sig->oom_adj = current->signal->oom_adj;
912         sig->oom_score_adj = current->signal->oom_score_adj;
913
914         mutex_init(&sig->cred_guard_mutex);
915
916         return 0;
917 }
918
919 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
920 {
921         unsigned long new_flags = p->flags;
922
923         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
924         new_flags |= PF_FORKNOEXEC;
925         new_flags |= PF_STARTING;
926         p->flags = new_flags;
927         clear_freeze_flag(p);
928 }
929
930 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
931 {
932         current->clear_child_tid = tidptr;
933
934         return task_pid_vnr(current);
935 }
936
937 static void rt_mutex_init_task(struct task_struct *p)
938 {
939         raw_spin_lock_init(&p->pi_lock);
940 #ifdef CONFIG_RT_MUTEXES
941         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
942         p->pi_blocked_on = NULL;
943 #endif
944 }
945
946 #ifdef CONFIG_MM_OWNER
947 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
948 {
949         mm->owner = p;
950 }
951 #endif /* CONFIG_MM_OWNER */
952
953 /*
954  * Initialize POSIX timer handling for a single task.
955  */
956 static void posix_cpu_timers_init(struct task_struct *tsk)
957 {
958         tsk->cputime_expires.prof_exp = cputime_zero;
959         tsk->cputime_expires.virt_exp = cputime_zero;
960         tsk->cputime_expires.sched_exp = 0;
961         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
962         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
963         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
964 }
965
966 /*
967  * This creates a new process as a copy of the old one,
968  * but does not actually start it yet.
969  *
970  * It copies the registers, and all the appropriate
971  * parts of the process environment (as per the clone
972  * flags). The actual kick-off is left to the caller.
973  */
974 static struct task_struct *copy_process(unsigned long clone_flags,
975                                         unsigned long stack_start,
976                                         struct pt_regs *regs,
977                                         unsigned long stack_size,
978                                         int __user *child_tidptr,
979                                         struct pid *pid,
980                                         int trace)
981 {
982         int retval;
983         struct task_struct *p;
984         int cgroup_callbacks_done = 0;
985
986         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
987                 return ERR_PTR(-EINVAL);
988
989         /*
990          * Thread groups must share signals as well, and detached threads
991          * can only be started up within the thread group.
992          */
993         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
994                 return ERR_PTR(-EINVAL);
995
996         /*
997          * Shared signal handlers imply shared VM. By way of the above,
998          * thread groups also imply shared VM. Blocking this case allows
999          * for various simplifications in other code.
1000          */
1001         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1002                 return ERR_PTR(-EINVAL);
1003
1004         /*
1005          * Siblings of global init remain as zombies on exit since they are
1006          * not reaped by their parent (swapper). To solve this and to avoid
1007          * multi-rooted process trees, prevent global and container-inits
1008          * from creating siblings.
1009          */
1010         if ((clone_flags & CLONE_PARENT) &&
1011                                 current->signal->flags & SIGNAL_UNKILLABLE)
1012                 return ERR_PTR(-EINVAL);
1013
1014         retval = security_task_create(clone_flags);
1015         if (retval)
1016                 goto fork_out;
1017
1018         retval = -ENOMEM;
1019         p = dup_task_struct(current);
1020         if (!p)
1021                 goto fork_out;
1022
1023         ftrace_graph_init_task(p);
1024
1025         rt_mutex_init_task(p);
1026
1027 #ifdef CONFIG_PROVE_LOCKING
1028         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1029         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1030 #endif
1031         retval = -EAGAIN;
1032         if (atomic_read(&p->real_cred->user->processes) >=
1033                         task_rlimit(p, RLIMIT_NPROC)) {
1034                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1035                     p->real_cred->user != INIT_USER)
1036                         goto bad_fork_free;
1037         }
1038
1039         retval = copy_creds(p, clone_flags);
1040         if (retval < 0)
1041                 goto bad_fork_free;
1042
1043         /*
1044          * If multiple threads are within copy_process(), then this check
1045          * triggers too late. This doesn't hurt, the check is only there
1046          * to stop root fork bombs.
1047          */
1048         retval = -EAGAIN;
1049         if (nr_threads >= max_threads)
1050                 goto bad_fork_cleanup_count;
1051
1052         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1053                 goto bad_fork_cleanup_count;
1054
1055         p->did_exec = 0;
1056         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1057         copy_flags(clone_flags, p);
1058         INIT_LIST_HEAD(&p->children);
1059         INIT_LIST_HEAD(&p->sibling);
1060         rcu_copy_process(p);
1061         p->vfork_done = NULL;
1062         spin_lock_init(&p->alloc_lock);
1063
1064         init_sigpending(&p->pending);
1065
1066         p->utime = cputime_zero;
1067         p->stime = cputime_zero;
1068         p->gtime = cputime_zero;
1069         p->utimescaled = cputime_zero;
1070         p->stimescaled = cputime_zero;
1071 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1072         p->prev_utime = cputime_zero;
1073         p->prev_stime = cputime_zero;
1074 #endif
1075 #if defined(SPLIT_RSS_COUNTING)
1076         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1077 #endif
1078
1079         p->default_timer_slack_ns = current->timer_slack_ns;
1080
1081         task_io_accounting_init(&p->ioac);
1082         acct_clear_integrals(p);
1083
1084         posix_cpu_timers_init(p);
1085
1086         p->lock_depth = -1;             /* -1 = no lock */
1087         do_posix_clock_monotonic_gettime(&p->start_time);
1088         p->real_start_time = p->start_time;
1089         monotonic_to_bootbased(&p->real_start_time);
1090         p->io_context = NULL;
1091         p->audit_context = NULL;
1092         cgroup_fork(p);
1093 #ifdef CONFIG_NUMA
1094         p->mempolicy = mpol_dup(p->mempolicy);
1095         if (IS_ERR(p->mempolicy)) {
1096                 retval = PTR_ERR(p->mempolicy);
1097                 p->mempolicy = NULL;
1098                 goto bad_fork_cleanup_cgroup;
1099         }
1100         mpol_fix_fork_child_flag(p);
1101 #endif
1102 #ifdef CONFIG_TRACE_IRQFLAGS
1103         p->irq_events = 0;
1104 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1105         p->hardirqs_enabled = 1;
1106 #else
1107         p->hardirqs_enabled = 0;
1108 #endif
1109         p->hardirq_enable_ip = 0;
1110         p->hardirq_enable_event = 0;
1111         p->hardirq_disable_ip = _THIS_IP_;
1112         p->hardirq_disable_event = 0;
1113         p->softirqs_enabled = 1;
1114         p->softirq_enable_ip = _THIS_IP_;
1115         p->softirq_enable_event = 0;
1116         p->softirq_disable_ip = 0;
1117         p->softirq_disable_event = 0;
1118         p->hardirq_context = 0;
1119         p->softirq_context = 0;
1120 #endif
1121 #ifdef CONFIG_LOCKDEP
1122         p->lockdep_depth = 0; /* no locks held yet */
1123         p->curr_chain_key = 0;
1124         p->lockdep_recursion = 0;
1125 #endif
1126
1127 #ifdef CONFIG_DEBUG_MUTEXES
1128         p->blocked_on = NULL; /* not blocked yet */
1129 #endif
1130 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1131         p->memcg_batch.do_batch = 0;
1132         p->memcg_batch.memcg = NULL;
1133 #endif
1134
1135         /* Perform scheduler related setup. Assign this task to a CPU. */
1136         sched_fork(p, clone_flags);
1137
1138         retval = perf_event_init_task(p);
1139         if (retval)
1140                 goto bad_fork_cleanup_policy;
1141
1142         if ((retval = audit_alloc(p)))
1143                 goto bad_fork_cleanup_policy;
1144         /* copy all the process information */
1145         if ((retval = copy_semundo(clone_flags, p)))
1146                 goto bad_fork_cleanup_audit;
1147         if ((retval = copy_files(clone_flags, p)))
1148                 goto bad_fork_cleanup_semundo;
1149         if ((retval = copy_fs(clone_flags, p)))
1150                 goto bad_fork_cleanup_files;
1151         if ((retval = copy_sighand(clone_flags, p)))
1152                 goto bad_fork_cleanup_fs;
1153         if ((retval = copy_signal(clone_flags, p)))
1154                 goto bad_fork_cleanup_sighand;
1155         if ((retval = copy_mm(clone_flags, p)))
1156                 goto bad_fork_cleanup_signal;
1157         if ((retval = copy_namespaces(clone_flags, p)))
1158                 goto bad_fork_cleanup_mm;
1159         if ((retval = copy_io(clone_flags, p)))
1160                 goto bad_fork_cleanup_namespaces;
1161         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1162         if (retval)
1163                 goto bad_fork_cleanup_io;
1164
1165         if (pid != &init_struct_pid) {
1166                 retval = -ENOMEM;
1167                 pid = alloc_pid(p->nsproxy->pid_ns);
1168                 if (!pid)
1169                         goto bad_fork_cleanup_io;
1170
1171                 if (clone_flags & CLONE_NEWPID) {
1172                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1173                         if (retval < 0)
1174                                 goto bad_fork_free_pid;
1175                 }
1176         }
1177
1178         p->pid = pid_nr(pid);
1179         p->tgid = p->pid;
1180         if (clone_flags & CLONE_THREAD)
1181                 p->tgid = current->tgid;
1182
1183         if (current->nsproxy != p->nsproxy) {
1184                 retval = ns_cgroup_clone(p, pid);
1185                 if (retval)
1186                         goto bad_fork_free_pid;
1187         }
1188
1189         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1190         /*
1191          * Clear TID on mm_release()?
1192          */
1193         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1194 #ifdef CONFIG_FUTEX
1195         p->robust_list = NULL;
1196 #ifdef CONFIG_COMPAT
1197         p->compat_robust_list = NULL;
1198 #endif
1199         INIT_LIST_HEAD(&p->pi_state_list);
1200         p->pi_state_cache = NULL;
1201 #endif
1202         /*
1203          * sigaltstack should be cleared when sharing the same VM
1204          */
1205         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1206                 p->sas_ss_sp = p->sas_ss_size = 0;
1207
1208         /*
1209          * Syscall tracing and stepping should be turned off in the
1210          * child regardless of CLONE_PTRACE.
1211          */
1212         user_disable_single_step(p);
1213         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1214 #ifdef TIF_SYSCALL_EMU
1215         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1216 #endif
1217         clear_all_latency_tracing(p);
1218
1219         /* ok, now we should be set up.. */
1220         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1221         p->pdeath_signal = 0;
1222         p->exit_state = 0;
1223
1224         /*
1225          * Ok, make it visible to the rest of the system.
1226          * We dont wake it up yet.
1227          */
1228         p->group_leader = p;
1229         INIT_LIST_HEAD(&p->thread_group);
1230
1231         /* Now that the task is set up, run cgroup callbacks if
1232          * necessary. We need to run them before the task is visible
1233          * on the tasklist. */
1234         cgroup_fork_callbacks(p);
1235         cgroup_callbacks_done = 1;
1236
1237         /* Need tasklist lock for parent etc handling! */
1238         write_lock_irq(&tasklist_lock);
1239
1240         /* CLONE_PARENT re-uses the old parent */
1241         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1242                 p->real_parent = current->real_parent;
1243                 p->parent_exec_id = current->parent_exec_id;
1244         } else {
1245                 p->real_parent = current;
1246                 p->parent_exec_id = current->self_exec_id;
1247         }
1248
1249         spin_lock(&current->sighand->siglock);
1250
1251         /*
1252          * Process group and session signals need to be delivered to just the
1253          * parent before the fork or both the parent and the child after the
1254          * fork. Restart if a signal comes in before we add the new process to
1255          * it's process group.
1256          * A fatal signal pending means that current will exit, so the new
1257          * thread can't slip out of an OOM kill (or normal SIGKILL).
1258          */
1259         recalc_sigpending();
1260         if (signal_pending(current)) {
1261                 spin_unlock(&current->sighand->siglock);
1262                 write_unlock_irq(&tasklist_lock);
1263                 retval = -ERESTARTNOINTR;
1264                 goto bad_fork_free_pid;
1265         }
1266
1267         if (clone_flags & CLONE_THREAD) {
1268                 current->signal->nr_threads++;
1269                 atomic_inc(&current->signal->live);
1270                 atomic_inc(&current->signal->sigcnt);
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                 tracehook_finish_clone(p, clone_flags, trace);
1277
1278                 if (thread_group_leader(p)) {
1279                         if (clone_flags & CLONE_NEWPID)
1280                                 p->nsproxy->pid_ns->child_reaper = p;
1281
1282                         p->signal->leader_pid = pid;
1283                         p->signal->tty = tty_kref_get(current->signal->tty);
1284                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1285                         attach_pid(p, PIDTYPE_SID, task_session(current));
1286                         list_add_tail(&p->sibling, &p->real_parent->children);
1287                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1288                         __get_cpu_var(process_counts)++;
1289                 }
1290                 attach_pid(p, PIDTYPE_PID, pid);
1291                 nr_threads++;
1292         }
1293
1294         total_forks++;
1295         spin_unlock(&current->sighand->siglock);
1296         write_unlock_irq(&tasklist_lock);
1297         proc_fork_connector(p);
1298         cgroup_post_fork(p);
1299         perf_event_fork(p);
1300         return p;
1301
1302 bad_fork_free_pid:
1303         if (pid != &init_struct_pid)
1304                 free_pid(pid);
1305 bad_fork_cleanup_io:
1306         if (p->io_context)
1307                 exit_io_context(p);
1308 bad_fork_cleanup_namespaces:
1309         exit_task_namespaces(p);
1310 bad_fork_cleanup_mm:
1311         if (p->mm) {
1312                 task_lock(p);
1313                 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1314                         atomic_dec(&p->mm->oom_disable_count);
1315                 task_unlock(p);
1316                 mmput(p->mm);
1317         }
1318 bad_fork_cleanup_signal:
1319         if (!(clone_flags & CLONE_THREAD))
1320                 free_signal_struct(p->signal);
1321 bad_fork_cleanup_sighand:
1322         __cleanup_sighand(p->sighand);
1323 bad_fork_cleanup_fs:
1324         exit_fs(p); /* blocking */
1325 bad_fork_cleanup_files:
1326         exit_files(p); /* blocking */
1327 bad_fork_cleanup_semundo:
1328         exit_sem(p);
1329 bad_fork_cleanup_audit:
1330         audit_free(p);
1331 bad_fork_cleanup_policy:
1332         perf_event_free_task(p);
1333 #ifdef CONFIG_NUMA
1334         mpol_put(p->mempolicy);
1335 bad_fork_cleanup_cgroup:
1336 #endif
1337         cgroup_exit(p, cgroup_callbacks_done);
1338         delayacct_tsk_free(p);
1339         module_put(task_thread_info(p)->exec_domain->module);
1340 bad_fork_cleanup_count:
1341         atomic_dec(&p->cred->user->processes);
1342         exit_creds(p);
1343 bad_fork_free:
1344         free_task(p);
1345 fork_out:
1346         return ERR_PTR(retval);
1347 }
1348
1349 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1350 {
1351         memset(regs, 0, sizeof(struct pt_regs));
1352         return regs;
1353 }
1354
1355 static inline void init_idle_pids(struct pid_link *links)
1356 {
1357         enum pid_type type;
1358
1359         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1360                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1361                 links[type].pid = &init_struct_pid;
1362         }
1363 }
1364
1365 struct task_struct * __cpuinit fork_idle(int cpu)
1366 {
1367         struct task_struct *task;
1368         struct pt_regs regs;
1369
1370         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1371                             &init_struct_pid, 0);
1372         if (!IS_ERR(task)) {
1373                 init_idle_pids(task->pids);
1374                 init_idle(task, cpu);
1375         }
1376
1377         return task;
1378 }
1379
1380 /*
1381  *  Ok, this is the main fork-routine.
1382  *
1383  * It copies the process, and if successful kick-starts
1384  * it and waits for it to finish using the VM if required.
1385  */
1386 long do_fork(unsigned long clone_flags,
1387               unsigned long stack_start,
1388               struct pt_regs *regs,
1389               unsigned long stack_size,
1390               int __user *parent_tidptr,
1391               int __user *child_tidptr)
1392 {
1393         struct task_struct *p;
1394         int trace = 0;
1395         long nr;
1396
1397         /*
1398          * Do some preliminary argument and permissions checking before we
1399          * actually start allocating stuff
1400          */
1401         if (clone_flags & CLONE_NEWUSER) {
1402                 if (clone_flags & CLONE_THREAD)
1403                         return -EINVAL;
1404                 /* hopefully this check will go away when userns support is
1405                  * complete
1406                  */
1407                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1408                                 !capable(CAP_SETGID))
1409                         return -EPERM;
1410         }
1411
1412         /*
1413          * We hope to recycle these flags after 2.6.26
1414          */
1415         if (unlikely(clone_flags & CLONE_STOPPED)) {
1416                 static int __read_mostly count = 100;
1417
1418                 if (count > 0 && printk_ratelimit()) {
1419                         char comm[TASK_COMM_LEN];
1420
1421                         count--;
1422                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1423                                         "clone flags 0x%lx\n",
1424                                 get_task_comm(comm, current),
1425                                 clone_flags & CLONE_STOPPED);
1426                 }
1427         }
1428
1429         /*
1430          * When called from kernel_thread, don't do user tracing stuff.
1431          */
1432         if (likely(user_mode(regs)))
1433                 trace = tracehook_prepare_clone(clone_flags);
1434
1435         p = copy_process(clone_flags, stack_start, regs, stack_size,
1436                          child_tidptr, NULL, trace);
1437         /*
1438          * Do this prior waking up the new thread - the thread pointer
1439          * might get invalid after that point, if the thread exits quickly.
1440          */
1441         if (!IS_ERR(p)) {
1442                 struct completion vfork;
1443
1444                 trace_sched_process_fork(current, p);
1445
1446                 nr = task_pid_vnr(p);
1447
1448                 if (clone_flags & CLONE_PARENT_SETTID)
1449                         put_user(nr, parent_tidptr);
1450
1451                 if (clone_flags & CLONE_VFORK) {
1452                         p->vfork_done = &vfork;
1453                         init_completion(&vfork);
1454                 }
1455
1456                 audit_finish_fork(p);
1457                 tracehook_report_clone(regs, clone_flags, nr, p);
1458
1459                 /*
1460                  * We set PF_STARTING at creation in case tracing wants to
1461                  * use this to distinguish a fully live task from one that
1462                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1463                  * clear it and set the child going.
1464                  */
1465                 p->flags &= ~PF_STARTING;
1466
1467                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1468                         /*
1469                          * We'll start up with an immediate SIGSTOP.
1470                          */
1471                         sigaddset(&p->pending.signal, SIGSTOP);
1472                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1473                         __set_task_state(p, TASK_STOPPED);
1474                 } else {
1475                         wake_up_new_task(p, clone_flags);
1476                 }
1477
1478                 tracehook_report_clone_complete(trace, regs,
1479                                                 clone_flags, nr, p);
1480
1481                 if (clone_flags & CLONE_VFORK) {
1482                         freezer_do_not_count();
1483                         wait_for_completion(&vfork);
1484                         freezer_count();
1485                         tracehook_report_vfork_done(p, nr);
1486                 }
1487         } else {
1488                 nr = PTR_ERR(p);
1489         }
1490         return nr;
1491 }
1492
1493 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1494 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1495 #endif
1496
1497 static void sighand_ctor(void *data)
1498 {
1499         struct sighand_struct *sighand = data;
1500
1501         spin_lock_init(&sighand->siglock);
1502         init_waitqueue_head(&sighand->signalfd_wqh);
1503 }
1504
1505 void __init proc_caches_init(void)
1506 {
1507         sighand_cachep = kmem_cache_create("sighand_cache",
1508                         sizeof(struct sighand_struct), 0,
1509                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1510                         SLAB_NOTRACK, sighand_ctor);
1511         signal_cachep = kmem_cache_create("signal_cache",
1512                         sizeof(struct signal_struct), 0,
1513                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1514         files_cachep = kmem_cache_create("files_cache",
1515                         sizeof(struct files_struct), 0,
1516                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1517         fs_cachep = kmem_cache_create("fs_cache",
1518                         sizeof(struct fs_struct), 0,
1519                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1520         mm_cachep = kmem_cache_create("mm_struct",
1521                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1522                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1523         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1524         mmap_init();
1525 }
1526
1527 /*
1528  * Check constraints on flags passed to the unshare system call and
1529  * force unsharing of additional process context as appropriate.
1530  */
1531 static void check_unshare_flags(unsigned long *flags_ptr)
1532 {
1533         /*
1534          * If unsharing a thread from a thread group, must also
1535          * unshare vm.
1536          */
1537         if (*flags_ptr & CLONE_THREAD)
1538                 *flags_ptr |= CLONE_VM;
1539
1540         /*
1541          * If unsharing vm, must also unshare signal handlers.
1542          */
1543         if (*flags_ptr & CLONE_VM)
1544                 *flags_ptr |= CLONE_SIGHAND;
1545
1546         /*
1547          * If unsharing namespace, must also unshare filesystem information.
1548          */
1549         if (*flags_ptr & CLONE_NEWNS)
1550                 *flags_ptr |= CLONE_FS;
1551 }
1552
1553 /*
1554  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1555  */
1556 static int unshare_thread(unsigned long unshare_flags)
1557 {
1558         if (unshare_flags & CLONE_THREAD)
1559                 return -EINVAL;
1560
1561         return 0;
1562 }
1563
1564 /*
1565  * Unshare the filesystem structure if it is being shared
1566  */
1567 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1568 {
1569         struct fs_struct *fs = current->fs;
1570
1571         if (!(unshare_flags & CLONE_FS) || !fs)
1572                 return 0;
1573
1574         /* don't need lock here; in the worst case we'll do useless copy */
1575         if (fs->users == 1)
1576                 return 0;
1577
1578         *new_fsp = copy_fs_struct(fs);
1579         if (!*new_fsp)
1580                 return -ENOMEM;
1581
1582         return 0;
1583 }
1584
1585 /*
1586  * Unsharing of sighand is not supported yet
1587  */
1588 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1589 {
1590         struct sighand_struct *sigh = current->sighand;
1591
1592         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1593                 return -EINVAL;
1594         else
1595                 return 0;
1596 }
1597
1598 /*
1599  * Unshare vm if it is being shared
1600  */
1601 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1602 {
1603         struct mm_struct *mm = current->mm;
1604
1605         if ((unshare_flags & CLONE_VM) &&
1606             (mm && atomic_read(&mm->mm_users) > 1)) {
1607                 return -EINVAL;
1608         }
1609
1610         return 0;
1611 }
1612
1613 /*
1614  * Unshare file descriptor table if it is being shared
1615  */
1616 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1617 {
1618         struct files_struct *fd = current->files;
1619         int error = 0;
1620
1621         if ((unshare_flags & CLONE_FILES) &&
1622             (fd && atomic_read(&fd->count) > 1)) {
1623                 *new_fdp = dup_fd(fd, &error);
1624                 if (!*new_fdp)
1625                         return error;
1626         }
1627
1628         return 0;
1629 }
1630
1631 /*
1632  * unshare allows a process to 'unshare' part of the process
1633  * context which was originally shared using clone.  copy_*
1634  * functions used by do_fork() cannot be used here directly
1635  * because they modify an inactive task_struct that is being
1636  * constructed. Here we are modifying the current, active,
1637  * task_struct.
1638  */
1639 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1640 {
1641         int err = 0;
1642         struct fs_struct *fs, *new_fs = NULL;
1643         struct sighand_struct *new_sigh = NULL;
1644         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1645         struct files_struct *fd, *new_fd = NULL;
1646         struct nsproxy *new_nsproxy = NULL;
1647         int do_sysvsem = 0;
1648
1649         check_unshare_flags(&unshare_flags);
1650
1651         /* Return -EINVAL for all unsupported flags */
1652         err = -EINVAL;
1653         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1654                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1655                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1656                 goto bad_unshare_out;
1657
1658         /*
1659          * CLONE_NEWIPC must also detach from the undolist: after switching
1660          * to a new ipc namespace, the semaphore arrays from the old
1661          * namespace are unreachable.
1662          */
1663         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1664                 do_sysvsem = 1;
1665         if ((err = unshare_thread(unshare_flags)))
1666                 goto bad_unshare_out;
1667         if ((err = unshare_fs(unshare_flags, &new_fs)))
1668                 goto bad_unshare_cleanup_thread;
1669         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1670                 goto bad_unshare_cleanup_fs;
1671         if ((err = unshare_vm(unshare_flags, &new_mm)))
1672                 goto bad_unshare_cleanup_sigh;
1673         if ((err = unshare_fd(unshare_flags, &new_fd)))
1674                 goto bad_unshare_cleanup_vm;
1675         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1676                         new_fs)))
1677                 goto bad_unshare_cleanup_fd;
1678
1679         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1680                 if (do_sysvsem) {
1681                         /*
1682                          * CLONE_SYSVSEM is equivalent to sys_exit().
1683                          */
1684                         exit_sem(current);
1685                 }
1686
1687                 if (new_nsproxy) {
1688                         switch_task_namespaces(current, new_nsproxy);
1689                         new_nsproxy = NULL;
1690                 }
1691
1692                 task_lock(current);
1693
1694                 if (new_fs) {
1695                         fs = current->fs;
1696                         spin_lock(&fs->lock);
1697                         current->fs = new_fs;
1698                         if (--fs->users)
1699                                 new_fs = NULL;
1700                         else
1701                                 new_fs = fs;
1702                         spin_unlock(&fs->lock);
1703                 }
1704
1705                 if (new_mm) {
1706                         mm = current->mm;
1707                         active_mm = current->active_mm;
1708                         current->mm = new_mm;
1709                         current->active_mm = new_mm;
1710                         if (current->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
1711                                 atomic_dec(&mm->oom_disable_count);
1712                                 atomic_inc(&new_mm->oom_disable_count);
1713                         }
1714                         activate_mm(active_mm, new_mm);
1715                         new_mm = mm;
1716                 }
1717
1718                 if (new_fd) {
1719                         fd = current->files;
1720                         current->files = new_fd;
1721                         new_fd = fd;
1722                 }
1723
1724                 task_unlock(current);
1725         }
1726
1727         if (new_nsproxy)
1728                 put_nsproxy(new_nsproxy);
1729
1730 bad_unshare_cleanup_fd:
1731         if (new_fd)
1732                 put_files_struct(new_fd);
1733
1734 bad_unshare_cleanup_vm:
1735         if (new_mm)
1736                 mmput(new_mm);
1737
1738 bad_unshare_cleanup_sigh:
1739         if (new_sigh)
1740                 if (atomic_dec_and_test(&new_sigh->count))
1741                         kmem_cache_free(sighand_cachep, new_sigh);
1742
1743 bad_unshare_cleanup_fs:
1744         if (new_fs)
1745                 free_fs_struct(new_fs);
1746
1747 bad_unshare_cleanup_thread:
1748 bad_unshare_out:
1749         return err;
1750 }
1751
1752 /*
1753  *      Helper to unshare the files of the current task.
1754  *      We don't want to expose copy_files internals to
1755  *      the exec layer of the kernel.
1756  */
1757
1758 int unshare_files(struct files_struct **displaced)
1759 {
1760         struct task_struct *task = current;
1761         struct files_struct *copy = NULL;
1762         int error;
1763
1764         error = unshare_fd(CLONE_FILES, &copy);
1765         if (error || !copy) {
1766                 *displaced = NULL;
1767                 return error;
1768         }
1769         *displaced = task->files;
1770         task_lock(task);
1771         task->files = copy;
1772         task_unlock(task);
1773         return 0;
1774 }