Merge branch 'fixes-2.6.38' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq
[pandora-kernel.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 #include <linux/khugepaged.h>
70
71 #include <asm/pgtable.h>
72 #include <asm/pgalloc.h>
73 #include <asm/uaccess.h>
74 #include <asm/mmu_context.h>
75 #include <asm/cacheflush.h>
76 #include <asm/tlbflush.h>
77
78 #include <trace/events/sched.h>
79
80 /*
81  * Protected counters by write_lock_irq(&tasklist_lock)
82  */
83 unsigned long total_forks;      /* Handle normal Linux uptimes. */
84 int nr_threads;                 /* The idle threads do not count.. */
85
86 int max_threads;                /* tunable limit on nr_threads */
87
88 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
89
90 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
91
92 #ifdef CONFIG_PROVE_RCU
93 int lockdep_tasklist_lock_is_held(void)
94 {
95         return lockdep_is_held(&tasklist_lock);
96 }
97 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
98 #endif /* #ifdef CONFIG_PROVE_RCU */
99
100 int nr_processes(void)
101 {
102         int cpu;
103         int total = 0;
104
105         for_each_possible_cpu(cpu)
106                 total += per_cpu(process_counts, cpu);
107
108         return total;
109 }
110
111 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
112 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
113 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
114 static struct kmem_cache *task_struct_cachep;
115 #endif
116
117 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
118 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
119 {
120 #ifdef CONFIG_DEBUG_STACK_USAGE
121         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
122 #else
123         gfp_t mask = GFP_KERNEL;
124 #endif
125         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
126 }
127
128 static inline void free_thread_info(struct thread_info *ti)
129 {
130         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
131 }
132 #endif
133
134 /* SLAB cache for signal_struct structures (tsk->signal) */
135 static struct kmem_cache *signal_cachep;
136
137 /* SLAB cache for sighand_struct structures (tsk->sighand) */
138 struct kmem_cache *sighand_cachep;
139
140 /* SLAB cache for files_struct structures (tsk->files) */
141 struct kmem_cache *files_cachep;
142
143 /* SLAB cache for fs_struct structures (tsk->fs) */
144 struct kmem_cache *fs_cachep;
145
146 /* SLAB cache for vm_area_struct structures */
147 struct kmem_cache *vm_area_cachep;
148
149 /* SLAB cache for mm_struct structures (tsk->mm) */
150 static struct kmem_cache *mm_cachep;
151
152 static void account_kernel_stack(struct thread_info *ti, int account)
153 {
154         struct zone *zone = page_zone(virt_to_page(ti));
155
156         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
157 }
158
159 void free_task(struct task_struct *tsk)
160 {
161         prop_local_destroy_single(&tsk->dirties);
162         account_kernel_stack(tsk->stack, -1);
163         free_thread_info(tsk->stack);
164         rt_mutex_debug_task_free(tsk);
165         ftrace_graph_exit_task(tsk);
166         free_task_struct(tsk);
167 }
168 EXPORT_SYMBOL(free_task);
169
170 static inline void free_signal_struct(struct signal_struct *sig)
171 {
172         taskstats_tgid_free(sig);
173         sched_autogroup_exit(sig);
174         kmem_cache_free(signal_cachep, sig);
175 }
176
177 static inline void put_signal_struct(struct signal_struct *sig)
178 {
179         if (atomic_dec_and_test(&sig->sigcnt))
180                 free_signal_struct(sig);
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         retval = khugepaged_fork(mm, oldmm);
335         if (retval)
336                 goto out;
337
338         prev = NULL;
339         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
340                 struct file *file;
341
342                 if (mpnt->vm_flags & VM_DONTCOPY) {
343                         long pages = vma_pages(mpnt);
344                         mm->total_vm -= pages;
345                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
346                                                                 -pages);
347                         continue;
348                 }
349                 charge = 0;
350                 if (mpnt->vm_flags & VM_ACCOUNT) {
351                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
352                         if (security_vm_enough_memory(len))
353                                 goto fail_nomem;
354                         charge = len;
355                 }
356                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
357                 if (!tmp)
358                         goto fail_nomem;
359                 *tmp = *mpnt;
360                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
361                 pol = mpol_dup(vma_policy(mpnt));
362                 retval = PTR_ERR(pol);
363                 if (IS_ERR(pol))
364                         goto fail_nomem_policy;
365                 vma_set_policy(tmp, pol);
366                 tmp->vm_mm = mm;
367                 if (anon_vma_fork(tmp, mpnt))
368                         goto fail_nomem_anon_vma_fork;
369                 tmp->vm_flags &= ~VM_LOCKED;
370                 tmp->vm_next = tmp->vm_prev = NULL;
371                 file = tmp->vm_file;
372                 if (file) {
373                         struct inode *inode = file->f_path.dentry->d_inode;
374                         struct address_space *mapping = file->f_mapping;
375
376                         get_file(file);
377                         if (tmp->vm_flags & VM_DENYWRITE)
378                                 atomic_dec(&inode->i_writecount);
379                         spin_lock(&mapping->i_mmap_lock);
380                         if (tmp->vm_flags & VM_SHARED)
381                                 mapping->i_mmap_writable++;
382                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
383                         flush_dcache_mmap_lock(mapping);
384                         /* insert tmp into the share list, just after mpnt */
385                         vma_prio_tree_add(tmp, mpnt);
386                         flush_dcache_mmap_unlock(mapping);
387                         spin_unlock(&mapping->i_mmap_lock);
388                 }
389
390                 /*
391                  * Clear hugetlb-related page reserves for children. This only
392                  * affects MAP_PRIVATE mappings. Faults generated by the child
393                  * are not guaranteed to succeed, even if read-only
394                  */
395                 if (is_vm_hugetlb_page(tmp))
396                         reset_vma_resv_huge_pages(tmp);
397
398                 /*
399                  * Link in the new vma and copy the page table entries.
400                  */
401                 *pprev = tmp;
402                 pprev = &tmp->vm_next;
403                 tmp->vm_prev = prev;
404                 prev = tmp;
405
406                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
407                 rb_link = &tmp->vm_rb.rb_right;
408                 rb_parent = &tmp->vm_rb;
409
410                 mm->map_count++;
411                 retval = copy_page_range(mm, oldmm, mpnt);
412
413                 if (tmp->vm_ops && tmp->vm_ops->open)
414                         tmp->vm_ops->open(tmp);
415
416                 if (retval)
417                         goto out;
418         }
419         /* a new mm has just been created */
420         arch_dup_mmap(oldmm, mm);
421         retval = 0;
422 out:
423         up_write(&mm->mmap_sem);
424         flush_tlb_mm(oldmm);
425         up_write(&oldmm->mmap_sem);
426         return retval;
427 fail_nomem_anon_vma_fork:
428         mpol_put(pol);
429 fail_nomem_policy:
430         kmem_cache_free(vm_area_cachep, tmp);
431 fail_nomem:
432         retval = -ENOMEM;
433         vm_unacct_memory(charge);
434         goto out;
435 }
436
437 static inline int mm_alloc_pgd(struct mm_struct * mm)
438 {
439         mm->pgd = pgd_alloc(mm);
440         if (unlikely(!mm->pgd))
441                 return -ENOMEM;
442         return 0;
443 }
444
445 static inline void mm_free_pgd(struct mm_struct * mm)
446 {
447         pgd_free(mm, mm->pgd);
448 }
449 #else
450 #define dup_mmap(mm, oldmm)     (0)
451 #define mm_alloc_pgd(mm)        (0)
452 #define mm_free_pgd(mm)
453 #endif /* CONFIG_MMU */
454
455 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
456
457 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
458 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
459
460 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
461
462 static int __init coredump_filter_setup(char *s)
463 {
464         default_dump_filter =
465                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
466                 MMF_DUMP_FILTER_MASK;
467         return 1;
468 }
469
470 __setup("coredump_filter=", coredump_filter_setup);
471
472 #include <linux/init_task.h>
473
474 static void mm_init_aio(struct mm_struct *mm)
475 {
476 #ifdef CONFIG_AIO
477         spin_lock_init(&mm->ioctx_lock);
478         INIT_HLIST_HEAD(&mm->ioctx_list);
479 #endif
480 }
481
482 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
483 {
484         atomic_set(&mm->mm_users, 1);
485         atomic_set(&mm->mm_count, 1);
486         init_rwsem(&mm->mmap_sem);
487         INIT_LIST_HEAD(&mm->mmlist);
488         mm->flags = (current->mm) ?
489                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
490         mm->core_state = NULL;
491         mm->nr_ptes = 0;
492         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
493         spin_lock_init(&mm->page_table_lock);
494         mm->free_area_cache = TASK_UNMAPPED_BASE;
495         mm->cached_hole_size = ~0UL;
496         mm_init_aio(mm);
497         mm_init_owner(mm, p);
498         atomic_set(&mm->oom_disable_count, 0);
499
500         if (likely(!mm_alloc_pgd(mm))) {
501                 mm->def_flags = 0;
502                 mmu_notifier_mm_init(mm);
503                 return mm;
504         }
505
506         free_mm(mm);
507         return NULL;
508 }
509
510 /*
511  * Allocate and initialize an mm_struct.
512  */
513 struct mm_struct * mm_alloc(void)
514 {
515         struct mm_struct * mm;
516
517         mm = allocate_mm();
518         if (mm) {
519                 memset(mm, 0, sizeof(*mm));
520                 mm = mm_init(mm, current);
521         }
522         return mm;
523 }
524
525 /*
526  * Called when the last reference to the mm
527  * is dropped: either by a lazy thread or by
528  * mmput. Free the page directory and the mm.
529  */
530 void __mmdrop(struct mm_struct *mm)
531 {
532         BUG_ON(mm == &init_mm);
533         mm_free_pgd(mm);
534         destroy_context(mm);
535         mmu_notifier_mm_destroy(mm);
536 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
537         VM_BUG_ON(mm->pmd_huge_pte);
538 #endif
539         free_mm(mm);
540 }
541 EXPORT_SYMBOL_GPL(__mmdrop);
542
543 /*
544  * Decrement the use count and release all resources for an mm.
545  */
546 void mmput(struct mm_struct *mm)
547 {
548         might_sleep();
549
550         if (atomic_dec_and_test(&mm->mm_users)) {
551                 exit_aio(mm);
552                 ksm_exit(mm);
553                 khugepaged_exit(mm); /* must run before exit_mmap */
554                 exit_mmap(mm);
555                 set_mm_exe_file(mm, NULL);
556                 if (!list_empty(&mm->mmlist)) {
557                         spin_lock(&mmlist_lock);
558                         list_del(&mm->mmlist);
559                         spin_unlock(&mmlist_lock);
560                 }
561                 put_swap_token(mm);
562                 if (mm->binfmt)
563                         module_put(mm->binfmt->module);
564                 mmdrop(mm);
565         }
566 }
567 EXPORT_SYMBOL_GPL(mmput);
568
569 /**
570  * get_task_mm - acquire a reference to the task's mm
571  *
572  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
573  * this kernel workthread has transiently adopted a user mm with use_mm,
574  * to do its AIO) is not set and if so returns a reference to it, after
575  * bumping up the use count.  User must release the mm via mmput()
576  * after use.  Typically used by /proc and ptrace.
577  */
578 struct mm_struct *get_task_mm(struct task_struct *task)
579 {
580         struct mm_struct *mm;
581
582         task_lock(task);
583         mm = task->mm;
584         if (mm) {
585                 if (task->flags & PF_KTHREAD)
586                         mm = NULL;
587                 else
588                         atomic_inc(&mm->mm_users);
589         }
590         task_unlock(task);
591         return mm;
592 }
593 EXPORT_SYMBOL_GPL(get_task_mm);
594
595 /* Please note the differences between mmput and mm_release.
596  * mmput is called whenever we stop holding onto a mm_struct,
597  * error success whatever.
598  *
599  * mm_release is called after a mm_struct has been removed
600  * from the current process.
601  *
602  * This difference is important for error handling, when we
603  * only half set up a mm_struct for a new process and need to restore
604  * the old one.  Because we mmput the new mm_struct before
605  * restoring the old one. . .
606  * Eric Biederman 10 January 1998
607  */
608 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
609 {
610         struct completion *vfork_done = tsk->vfork_done;
611
612         /* Get rid of any futexes when releasing the mm */
613 #ifdef CONFIG_FUTEX
614         if (unlikely(tsk->robust_list)) {
615                 exit_robust_list(tsk);
616                 tsk->robust_list = NULL;
617         }
618 #ifdef CONFIG_COMPAT
619         if (unlikely(tsk->compat_robust_list)) {
620                 compat_exit_robust_list(tsk);
621                 tsk->compat_robust_list = NULL;
622         }
623 #endif
624         if (unlikely(!list_empty(&tsk->pi_state_list)))
625                 exit_pi_state_list(tsk);
626 #endif
627
628         /* Get rid of any cached register state */
629         deactivate_mm(tsk, mm);
630
631         /* notify parent sleeping on vfork() */
632         if (vfork_done) {
633                 tsk->vfork_done = NULL;
634                 complete(vfork_done);
635         }
636
637         /*
638          * If we're exiting normally, clear a user-space tid field if
639          * requested.  We leave this alone when dying by signal, to leave
640          * the value intact in a core dump, and to save the unnecessary
641          * trouble otherwise.  Userland only wants this done for a sys_exit.
642          */
643         if (tsk->clear_child_tid) {
644                 if (!(tsk->flags & PF_SIGNALED) &&
645                     atomic_read(&mm->mm_users) > 1) {
646                         /*
647                          * We don't check the error code - if userspace has
648                          * not set up a proper pointer then tough luck.
649                          */
650                         put_user(0, tsk->clear_child_tid);
651                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
652                                         1, NULL, NULL, 0);
653                 }
654                 tsk->clear_child_tid = NULL;
655         }
656 }
657
658 /*
659  * Allocate a new mm structure and copy contents from the
660  * mm structure of the passed in task structure.
661  */
662 struct mm_struct *dup_mm(struct task_struct *tsk)
663 {
664         struct mm_struct *mm, *oldmm = current->mm;
665         int err;
666
667         if (!oldmm)
668                 return NULL;
669
670         mm = allocate_mm();
671         if (!mm)
672                 goto fail_nomem;
673
674         memcpy(mm, oldmm, sizeof(*mm));
675
676         /* Initializing for Swap token stuff */
677         mm->token_priority = 0;
678         mm->last_interval = 0;
679
680 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
681         mm->pmd_huge_pte = NULL;
682 #endif
683
684         if (!mm_init(mm, tsk))
685                 goto fail_nomem;
686
687         if (init_new_context(tsk, mm))
688                 goto fail_nocontext;
689
690         dup_mm_exe_file(oldmm, mm);
691
692         err = dup_mmap(mm, oldmm);
693         if (err)
694                 goto free_pt;
695
696         mm->hiwater_rss = get_mm_rss(mm);
697         mm->hiwater_vm = mm->total_vm;
698
699         if (mm->binfmt && !try_module_get(mm->binfmt->module))
700                 goto free_pt;
701
702         return mm;
703
704 free_pt:
705         /* don't put binfmt in mmput, we haven't got module yet */
706         mm->binfmt = NULL;
707         mmput(mm);
708
709 fail_nomem:
710         return NULL;
711
712 fail_nocontext:
713         /*
714          * If init_new_context() failed, we cannot use mmput() to free the mm
715          * because it calls destroy_context()
716          */
717         mm_free_pgd(mm);
718         free_mm(mm);
719         return NULL;
720 }
721
722 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
723 {
724         struct mm_struct * mm, *oldmm;
725         int retval;
726
727         tsk->min_flt = tsk->maj_flt = 0;
728         tsk->nvcsw = tsk->nivcsw = 0;
729 #ifdef CONFIG_DETECT_HUNG_TASK
730         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
731 #endif
732
733         tsk->mm = NULL;
734         tsk->active_mm = NULL;
735
736         /*
737          * Are we cloning a kernel thread?
738          *
739          * We need to steal a active VM for that..
740          */
741         oldmm = current->mm;
742         if (!oldmm)
743                 return 0;
744
745         if (clone_flags & CLONE_VM) {
746                 atomic_inc(&oldmm->mm_users);
747                 mm = oldmm;
748                 goto good_mm;
749         }
750
751         retval = -ENOMEM;
752         mm = dup_mm(tsk);
753         if (!mm)
754                 goto fail_nomem;
755
756 good_mm:
757         /* Initializing for Swap token stuff */
758         mm->token_priority = 0;
759         mm->last_interval = 0;
760         if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
761                 atomic_inc(&mm->oom_disable_count);
762
763         tsk->mm = mm;
764         tsk->active_mm = mm;
765         return 0;
766
767 fail_nomem:
768         return retval;
769 }
770
771 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
772 {
773         struct fs_struct *fs = current->fs;
774         if (clone_flags & CLONE_FS) {
775                 /* tsk->fs is already what we want */
776                 spin_lock(&fs->lock);
777                 if (fs->in_exec) {
778                         spin_unlock(&fs->lock);
779                         return -EAGAIN;
780                 }
781                 fs->users++;
782                 spin_unlock(&fs->lock);
783                 return 0;
784         }
785         tsk->fs = copy_fs_struct(fs);
786         if (!tsk->fs)
787                 return -ENOMEM;
788         return 0;
789 }
790
791 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
792 {
793         struct files_struct *oldf, *newf;
794         int error = 0;
795
796         /*
797          * A background process may not have any files ...
798          */
799         oldf = current->files;
800         if (!oldf)
801                 goto out;
802
803         if (clone_flags & CLONE_FILES) {
804                 atomic_inc(&oldf->count);
805                 goto out;
806         }
807
808         newf = dup_fd(oldf, &error);
809         if (!newf)
810                 goto out;
811
812         tsk->files = newf;
813         error = 0;
814 out:
815         return error;
816 }
817
818 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
819 {
820 #ifdef CONFIG_BLOCK
821         struct io_context *ioc = current->io_context;
822
823         if (!ioc)
824                 return 0;
825         /*
826          * Share io context with parent, if CLONE_IO is set
827          */
828         if (clone_flags & CLONE_IO) {
829                 tsk->io_context = ioc_task_link(ioc);
830                 if (unlikely(!tsk->io_context))
831                         return -ENOMEM;
832         } else if (ioprio_valid(ioc->ioprio)) {
833                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
834                 if (unlikely(!tsk->io_context))
835                         return -ENOMEM;
836
837                 tsk->io_context->ioprio = ioc->ioprio;
838         }
839 #endif
840         return 0;
841 }
842
843 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
844 {
845         struct sighand_struct *sig;
846
847         if (clone_flags & CLONE_SIGHAND) {
848                 atomic_inc(&current->sighand->count);
849                 return 0;
850         }
851         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
852         rcu_assign_pointer(tsk->sighand, sig);
853         if (!sig)
854                 return -ENOMEM;
855         atomic_set(&sig->count, 1);
856         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
857         return 0;
858 }
859
860 void __cleanup_sighand(struct sighand_struct *sighand)
861 {
862         if (atomic_dec_and_test(&sighand->count))
863                 kmem_cache_free(sighand_cachep, sighand);
864 }
865
866
867 /*
868  * Initialize POSIX timer handling for a thread group.
869  */
870 static void posix_cpu_timers_init_group(struct signal_struct *sig)
871 {
872         unsigned long cpu_limit;
873
874         /* Thread group counters. */
875         thread_group_cputime_init(sig);
876
877         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
878         if (cpu_limit != RLIM_INFINITY) {
879                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
880                 sig->cputimer.running = 1;
881         }
882
883         /* The timer lists. */
884         INIT_LIST_HEAD(&sig->cpu_timers[0]);
885         INIT_LIST_HEAD(&sig->cpu_timers[1]);
886         INIT_LIST_HEAD(&sig->cpu_timers[2]);
887 }
888
889 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
890 {
891         struct signal_struct *sig;
892
893         if (clone_flags & CLONE_THREAD)
894                 return 0;
895
896         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
897         tsk->signal = sig;
898         if (!sig)
899                 return -ENOMEM;
900
901         sig->nr_threads = 1;
902         atomic_set(&sig->live, 1);
903         atomic_set(&sig->sigcnt, 1);
904         init_waitqueue_head(&sig->wait_chldexit);
905         if (clone_flags & CLONE_NEWPID)
906                 sig->flags |= SIGNAL_UNKILLABLE;
907         sig->curr_target = tsk;
908         init_sigpending(&sig->shared_pending);
909         INIT_LIST_HEAD(&sig->posix_timers);
910
911         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
912         sig->real_timer.function = it_real_fn;
913
914         task_lock(current->group_leader);
915         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
916         task_unlock(current->group_leader);
917
918         posix_cpu_timers_init_group(sig);
919
920         tty_audit_fork(sig);
921         sched_autogroup_fork(sig);
922
923         sig->oom_adj = current->signal->oom_adj;
924         sig->oom_score_adj = current->signal->oom_score_adj;
925         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
926
927         mutex_init(&sig->cred_guard_mutex);
928
929         return 0;
930 }
931
932 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
933 {
934         unsigned long new_flags = p->flags;
935
936         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
937         new_flags |= PF_FORKNOEXEC;
938         new_flags |= PF_STARTING;
939         p->flags = new_flags;
940         clear_freeze_flag(p);
941 }
942
943 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
944 {
945         current->clear_child_tid = tidptr;
946
947         return task_pid_vnr(current);
948 }
949
950 static void rt_mutex_init_task(struct task_struct *p)
951 {
952         raw_spin_lock_init(&p->pi_lock);
953 #ifdef CONFIG_RT_MUTEXES
954         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
955         p->pi_blocked_on = NULL;
956 #endif
957 }
958
959 #ifdef CONFIG_MM_OWNER
960 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
961 {
962         mm->owner = p;
963 }
964 #endif /* CONFIG_MM_OWNER */
965
966 /*
967  * Initialize POSIX timer handling for a single task.
968  */
969 static void posix_cpu_timers_init(struct task_struct *tsk)
970 {
971         tsk->cputime_expires.prof_exp = cputime_zero;
972         tsk->cputime_expires.virt_exp = cputime_zero;
973         tsk->cputime_expires.sched_exp = 0;
974         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
975         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
976         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
977 }
978
979 /*
980  * This creates a new process as a copy of the old one,
981  * but does not actually start it yet.
982  *
983  * It copies the registers, and all the appropriate
984  * parts of the process environment (as per the clone
985  * flags). The actual kick-off is left to the caller.
986  */
987 static struct task_struct *copy_process(unsigned long clone_flags,
988                                         unsigned long stack_start,
989                                         struct pt_regs *regs,
990                                         unsigned long stack_size,
991                                         int __user *child_tidptr,
992                                         struct pid *pid,
993                                         int trace)
994 {
995         int retval;
996         struct task_struct *p;
997         int cgroup_callbacks_done = 0;
998
999         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1000                 return ERR_PTR(-EINVAL);
1001
1002         /*
1003          * Thread groups must share signals as well, and detached threads
1004          * can only be started up within the thread group.
1005          */
1006         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1007                 return ERR_PTR(-EINVAL);
1008
1009         /*
1010          * Shared signal handlers imply shared VM. By way of the above,
1011          * thread groups also imply shared VM. Blocking this case allows
1012          * for various simplifications in other code.
1013          */
1014         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1015                 return ERR_PTR(-EINVAL);
1016
1017         /*
1018          * Siblings of global init remain as zombies on exit since they are
1019          * not reaped by their parent (swapper). To solve this and to avoid
1020          * multi-rooted process trees, prevent global and container-inits
1021          * from creating siblings.
1022          */
1023         if ((clone_flags & CLONE_PARENT) &&
1024                                 current->signal->flags & SIGNAL_UNKILLABLE)
1025                 return ERR_PTR(-EINVAL);
1026
1027         retval = security_task_create(clone_flags);
1028         if (retval)
1029                 goto fork_out;
1030
1031         retval = -ENOMEM;
1032         p = dup_task_struct(current);
1033         if (!p)
1034                 goto fork_out;
1035
1036         ftrace_graph_init_task(p);
1037
1038         rt_mutex_init_task(p);
1039
1040 #ifdef CONFIG_PROVE_LOCKING
1041         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1042         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1043 #endif
1044         retval = -EAGAIN;
1045         if (atomic_read(&p->real_cred->user->processes) >=
1046                         task_rlimit(p, RLIMIT_NPROC)) {
1047                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1048                     p->real_cred->user != INIT_USER)
1049                         goto bad_fork_free;
1050         }
1051
1052         retval = copy_creds(p, clone_flags);
1053         if (retval < 0)
1054                 goto bad_fork_free;
1055
1056         /*
1057          * If multiple threads are within copy_process(), then this check
1058          * triggers too late. This doesn't hurt, the check is only there
1059          * to stop root fork bombs.
1060          */
1061         retval = -EAGAIN;
1062         if (nr_threads >= max_threads)
1063                 goto bad_fork_cleanup_count;
1064
1065         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1066                 goto bad_fork_cleanup_count;
1067
1068         p->did_exec = 0;
1069         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1070         copy_flags(clone_flags, p);
1071         INIT_LIST_HEAD(&p->children);
1072         INIT_LIST_HEAD(&p->sibling);
1073         rcu_copy_process(p);
1074         p->vfork_done = NULL;
1075         spin_lock_init(&p->alloc_lock);
1076
1077         init_sigpending(&p->pending);
1078
1079         p->utime = cputime_zero;
1080         p->stime = cputime_zero;
1081         p->gtime = cputime_zero;
1082         p->utimescaled = cputime_zero;
1083         p->stimescaled = cputime_zero;
1084 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1085         p->prev_utime = cputime_zero;
1086         p->prev_stime = cputime_zero;
1087 #endif
1088 #if defined(SPLIT_RSS_COUNTING)
1089         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1090 #endif
1091
1092         p->default_timer_slack_ns = current->timer_slack_ns;
1093
1094         task_io_accounting_init(&p->ioac);
1095         acct_clear_integrals(p);
1096
1097         posix_cpu_timers_init(p);
1098
1099         p->lock_depth = -1;             /* -1 = no lock */
1100         do_posix_clock_monotonic_gettime(&p->start_time);
1101         p->real_start_time = p->start_time;
1102         monotonic_to_bootbased(&p->real_start_time);
1103         p->io_context = NULL;
1104         p->audit_context = NULL;
1105         cgroup_fork(p);
1106 #ifdef CONFIG_NUMA
1107         p->mempolicy = mpol_dup(p->mempolicy);
1108         if (IS_ERR(p->mempolicy)) {
1109                 retval = PTR_ERR(p->mempolicy);
1110                 p->mempolicy = NULL;
1111                 goto bad_fork_cleanup_cgroup;
1112         }
1113         mpol_fix_fork_child_flag(p);
1114 #endif
1115 #ifdef CONFIG_TRACE_IRQFLAGS
1116         p->irq_events = 0;
1117 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1118         p->hardirqs_enabled = 1;
1119 #else
1120         p->hardirqs_enabled = 0;
1121 #endif
1122         p->hardirq_enable_ip = 0;
1123         p->hardirq_enable_event = 0;
1124         p->hardirq_disable_ip = _THIS_IP_;
1125         p->hardirq_disable_event = 0;
1126         p->softirqs_enabled = 1;
1127         p->softirq_enable_ip = _THIS_IP_;
1128         p->softirq_enable_event = 0;
1129         p->softirq_disable_ip = 0;
1130         p->softirq_disable_event = 0;
1131         p->hardirq_context = 0;
1132         p->softirq_context = 0;
1133 #endif
1134 #ifdef CONFIG_LOCKDEP
1135         p->lockdep_depth = 0; /* no locks held yet */
1136         p->curr_chain_key = 0;
1137         p->lockdep_recursion = 0;
1138 #endif
1139
1140 #ifdef CONFIG_DEBUG_MUTEXES
1141         p->blocked_on = NULL; /* not blocked yet */
1142 #endif
1143 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1144         p->memcg_batch.do_batch = 0;
1145         p->memcg_batch.memcg = NULL;
1146 #endif
1147
1148         /* Perform scheduler related setup. Assign this task to a CPU. */
1149         sched_fork(p, clone_flags);
1150
1151         retval = perf_event_init_task(p);
1152         if (retval)
1153                 goto bad_fork_cleanup_policy;
1154
1155         if ((retval = audit_alloc(p)))
1156                 goto bad_fork_cleanup_policy;
1157         /* copy all the process information */
1158         if ((retval = copy_semundo(clone_flags, p)))
1159                 goto bad_fork_cleanup_audit;
1160         if ((retval = copy_files(clone_flags, p)))
1161                 goto bad_fork_cleanup_semundo;
1162         if ((retval = copy_fs(clone_flags, p)))
1163                 goto bad_fork_cleanup_files;
1164         if ((retval = copy_sighand(clone_flags, p)))
1165                 goto bad_fork_cleanup_fs;
1166         if ((retval = copy_signal(clone_flags, p)))
1167                 goto bad_fork_cleanup_sighand;
1168         if ((retval = copy_mm(clone_flags, p)))
1169                 goto bad_fork_cleanup_signal;
1170         if ((retval = copy_namespaces(clone_flags, p)))
1171                 goto bad_fork_cleanup_mm;
1172         if ((retval = copy_io(clone_flags, p)))
1173                 goto bad_fork_cleanup_namespaces;
1174         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1175         if (retval)
1176                 goto bad_fork_cleanup_io;
1177
1178         if (pid != &init_struct_pid) {
1179                 retval = -ENOMEM;
1180                 pid = alloc_pid(p->nsproxy->pid_ns);
1181                 if (!pid)
1182                         goto bad_fork_cleanup_io;
1183
1184                 if (clone_flags & CLONE_NEWPID) {
1185                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1186                         if (retval < 0)
1187                                 goto bad_fork_free_pid;
1188                 }
1189         }
1190
1191         p->pid = pid_nr(pid);
1192         p->tgid = p->pid;
1193         if (clone_flags & CLONE_THREAD)
1194                 p->tgid = current->tgid;
1195
1196         if (current->nsproxy != p->nsproxy) {
1197                 retval = ns_cgroup_clone(p, pid);
1198                 if (retval)
1199                         goto bad_fork_free_pid;
1200         }
1201
1202         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1203         /*
1204          * Clear TID on mm_release()?
1205          */
1206         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1207 #ifdef CONFIG_FUTEX
1208         p->robust_list = NULL;
1209 #ifdef CONFIG_COMPAT
1210         p->compat_robust_list = NULL;
1211 #endif
1212         INIT_LIST_HEAD(&p->pi_state_list);
1213         p->pi_state_cache = NULL;
1214 #endif
1215         /*
1216          * sigaltstack should be cleared when sharing the same VM
1217          */
1218         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1219                 p->sas_ss_sp = p->sas_ss_size = 0;
1220
1221         /*
1222          * Syscall tracing and stepping should be turned off in the
1223          * child regardless of CLONE_PTRACE.
1224          */
1225         user_disable_single_step(p);
1226         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1227 #ifdef TIF_SYSCALL_EMU
1228         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1229 #endif
1230         clear_all_latency_tracing(p);
1231
1232         /* ok, now we should be set up.. */
1233         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1234         p->pdeath_signal = 0;
1235         p->exit_state = 0;
1236
1237         /*
1238          * Ok, make it visible to the rest of the system.
1239          * We dont wake it up yet.
1240          */
1241         p->group_leader = p;
1242         INIT_LIST_HEAD(&p->thread_group);
1243
1244         /* Now that the task is set up, run cgroup callbacks if
1245          * necessary. We need to run them before the task is visible
1246          * on the tasklist. */
1247         cgroup_fork_callbacks(p);
1248         cgroup_callbacks_done = 1;
1249
1250         /* Need tasklist lock for parent etc handling! */
1251         write_lock_irq(&tasklist_lock);
1252
1253         /* CLONE_PARENT re-uses the old parent */
1254         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1255                 p->real_parent = current->real_parent;
1256                 p->parent_exec_id = current->parent_exec_id;
1257         } else {
1258                 p->real_parent = current;
1259                 p->parent_exec_id = current->self_exec_id;
1260         }
1261
1262         spin_lock(&current->sighand->siglock);
1263
1264         /*
1265          * Process group and session signals need to be delivered to just the
1266          * parent before the fork or both the parent and the child after the
1267          * fork. Restart if a signal comes in before we add the new process to
1268          * it's process group.
1269          * A fatal signal pending means that current will exit, so the new
1270          * thread can't slip out of an OOM kill (or normal SIGKILL).
1271          */
1272         recalc_sigpending();
1273         if (signal_pending(current)) {
1274                 spin_unlock(&current->sighand->siglock);
1275                 write_unlock_irq(&tasklist_lock);
1276                 retval = -ERESTARTNOINTR;
1277                 goto bad_fork_free_pid;
1278         }
1279
1280         if (clone_flags & CLONE_THREAD) {
1281                 current->signal->nr_threads++;
1282                 atomic_inc(&current->signal->live);
1283                 atomic_inc(&current->signal->sigcnt);
1284                 p->group_leader = current->group_leader;
1285                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1286         }
1287
1288         if (likely(p->pid)) {
1289                 tracehook_finish_clone(p, clone_flags, trace);
1290
1291                 if (thread_group_leader(p)) {
1292                         if (clone_flags & CLONE_NEWPID)
1293                                 p->nsproxy->pid_ns->child_reaper = p;
1294
1295                         p->signal->leader_pid = pid;
1296                         p->signal->tty = tty_kref_get(current->signal->tty);
1297                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1298                         attach_pid(p, PIDTYPE_SID, task_session(current));
1299                         list_add_tail(&p->sibling, &p->real_parent->children);
1300                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1301                         __this_cpu_inc(process_counts);
1302                 }
1303                 attach_pid(p, PIDTYPE_PID, pid);
1304                 nr_threads++;
1305         }
1306
1307         total_forks++;
1308         spin_unlock(&current->sighand->siglock);
1309         write_unlock_irq(&tasklist_lock);
1310         proc_fork_connector(p);
1311         cgroup_post_fork(p);
1312         perf_event_fork(p);
1313         return p;
1314
1315 bad_fork_free_pid:
1316         if (pid != &init_struct_pid)
1317                 free_pid(pid);
1318 bad_fork_cleanup_io:
1319         if (p->io_context)
1320                 exit_io_context(p);
1321 bad_fork_cleanup_namespaces:
1322         exit_task_namespaces(p);
1323 bad_fork_cleanup_mm:
1324         if (p->mm) {
1325                 task_lock(p);
1326                 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1327                         atomic_dec(&p->mm->oom_disable_count);
1328                 task_unlock(p);
1329                 mmput(p->mm);
1330         }
1331 bad_fork_cleanup_signal:
1332         if (!(clone_flags & CLONE_THREAD))
1333                 free_signal_struct(p->signal);
1334 bad_fork_cleanup_sighand:
1335         __cleanup_sighand(p->sighand);
1336 bad_fork_cleanup_fs:
1337         exit_fs(p); /* blocking */
1338 bad_fork_cleanup_files:
1339         exit_files(p); /* blocking */
1340 bad_fork_cleanup_semundo:
1341         exit_sem(p);
1342 bad_fork_cleanup_audit:
1343         audit_free(p);
1344 bad_fork_cleanup_policy:
1345         perf_event_free_task(p);
1346 #ifdef CONFIG_NUMA
1347         mpol_put(p->mempolicy);
1348 bad_fork_cleanup_cgroup:
1349 #endif
1350         cgroup_exit(p, cgroup_callbacks_done);
1351         delayacct_tsk_free(p);
1352         module_put(task_thread_info(p)->exec_domain->module);
1353 bad_fork_cleanup_count:
1354         atomic_dec(&p->cred->user->processes);
1355         exit_creds(p);
1356 bad_fork_free:
1357         free_task(p);
1358 fork_out:
1359         return ERR_PTR(retval);
1360 }
1361
1362 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1363 {
1364         memset(regs, 0, sizeof(struct pt_regs));
1365         return regs;
1366 }
1367
1368 static inline void init_idle_pids(struct pid_link *links)
1369 {
1370         enum pid_type type;
1371
1372         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1373                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1374                 links[type].pid = &init_struct_pid;
1375         }
1376 }
1377
1378 struct task_struct * __cpuinit fork_idle(int cpu)
1379 {
1380         struct task_struct *task;
1381         struct pt_regs regs;
1382
1383         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1384                             &init_struct_pid, 0);
1385         if (!IS_ERR(task)) {
1386                 init_idle_pids(task->pids);
1387                 init_idle(task, cpu);
1388         }
1389
1390         return task;
1391 }
1392
1393 /*
1394  *  Ok, this is the main fork-routine.
1395  *
1396  * It copies the process, and if successful kick-starts
1397  * it and waits for it to finish using the VM if required.
1398  */
1399 long do_fork(unsigned long clone_flags,
1400               unsigned long stack_start,
1401               struct pt_regs *regs,
1402               unsigned long stack_size,
1403               int __user *parent_tidptr,
1404               int __user *child_tidptr)
1405 {
1406         struct task_struct *p;
1407         int trace = 0;
1408         long nr;
1409
1410         /*
1411          * Do some preliminary argument and permissions checking before we
1412          * actually start allocating stuff
1413          */
1414         if (clone_flags & CLONE_NEWUSER) {
1415                 if (clone_flags & CLONE_THREAD)
1416                         return -EINVAL;
1417                 /* hopefully this check will go away when userns support is
1418                  * complete
1419                  */
1420                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1421                                 !capable(CAP_SETGID))
1422                         return -EPERM;
1423         }
1424
1425         /*
1426          * When called from kernel_thread, don't do user tracing stuff.
1427          */
1428         if (likely(user_mode(regs)))
1429                 trace = tracehook_prepare_clone(clone_flags);
1430
1431         p = copy_process(clone_flags, stack_start, regs, stack_size,
1432                          child_tidptr, NULL, trace);
1433         /*
1434          * Do this prior waking up the new thread - the thread pointer
1435          * might get invalid after that point, if the thread exits quickly.
1436          */
1437         if (!IS_ERR(p)) {
1438                 struct completion vfork;
1439
1440                 trace_sched_process_fork(current, p);
1441
1442                 nr = task_pid_vnr(p);
1443
1444                 if (clone_flags & CLONE_PARENT_SETTID)
1445                         put_user(nr, parent_tidptr);
1446
1447                 if (clone_flags & CLONE_VFORK) {
1448                         p->vfork_done = &vfork;
1449                         init_completion(&vfork);
1450                 }
1451
1452                 audit_finish_fork(p);
1453                 tracehook_report_clone(regs, clone_flags, nr, p);
1454
1455                 /*
1456                  * We set PF_STARTING at creation in case tracing wants to
1457                  * use this to distinguish a fully live task from one that
1458                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1459                  * clear it and set the child going.
1460                  */
1461                 p->flags &= ~PF_STARTING;
1462
1463                 wake_up_new_task(p, clone_flags);
1464
1465                 tracehook_report_clone_complete(trace, regs,
1466                                                 clone_flags, nr, p);
1467
1468                 if (clone_flags & CLONE_VFORK) {
1469                         freezer_do_not_count();
1470                         wait_for_completion(&vfork);
1471                         freezer_count();
1472                         tracehook_report_vfork_done(p, nr);
1473                 }
1474         } else {
1475                 nr = PTR_ERR(p);
1476         }
1477         return nr;
1478 }
1479
1480 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1481 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1482 #endif
1483
1484 static void sighand_ctor(void *data)
1485 {
1486         struct sighand_struct *sighand = data;
1487
1488         spin_lock_init(&sighand->siglock);
1489         init_waitqueue_head(&sighand->signalfd_wqh);
1490 }
1491
1492 void __init proc_caches_init(void)
1493 {
1494         sighand_cachep = kmem_cache_create("sighand_cache",
1495                         sizeof(struct sighand_struct), 0,
1496                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1497                         SLAB_NOTRACK, sighand_ctor);
1498         signal_cachep = kmem_cache_create("signal_cache",
1499                         sizeof(struct signal_struct), 0,
1500                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1501         files_cachep = kmem_cache_create("files_cache",
1502                         sizeof(struct files_struct), 0,
1503                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1504         fs_cachep = kmem_cache_create("fs_cache",
1505                         sizeof(struct fs_struct), 0,
1506                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1507         mm_cachep = kmem_cache_create("mm_struct",
1508                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1509                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1510         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1511         mmap_init();
1512 }
1513
1514 /*
1515  * Check constraints on flags passed to the unshare system call and
1516  * force unsharing of additional process context as appropriate.
1517  */
1518 static void check_unshare_flags(unsigned long *flags_ptr)
1519 {
1520         /*
1521          * If unsharing a thread from a thread group, must also
1522          * unshare vm.
1523          */
1524         if (*flags_ptr & CLONE_THREAD)
1525                 *flags_ptr |= CLONE_VM;
1526
1527         /*
1528          * If unsharing vm, must also unshare signal handlers.
1529          */
1530         if (*flags_ptr & CLONE_VM)
1531                 *flags_ptr |= CLONE_SIGHAND;
1532
1533         /*
1534          * If unsharing namespace, must also unshare filesystem information.
1535          */
1536         if (*flags_ptr & CLONE_NEWNS)
1537                 *flags_ptr |= CLONE_FS;
1538 }
1539
1540 /*
1541  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1542  */
1543 static int unshare_thread(unsigned long unshare_flags)
1544 {
1545         if (unshare_flags & CLONE_THREAD)
1546                 return -EINVAL;
1547
1548         return 0;
1549 }
1550
1551 /*
1552  * Unshare the filesystem structure if it is being shared
1553  */
1554 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1555 {
1556         struct fs_struct *fs = current->fs;
1557
1558         if (!(unshare_flags & CLONE_FS) || !fs)
1559                 return 0;
1560
1561         /* don't need lock here; in the worst case we'll do useless copy */
1562         if (fs->users == 1)
1563                 return 0;
1564
1565         *new_fsp = copy_fs_struct(fs);
1566         if (!*new_fsp)
1567                 return -ENOMEM;
1568
1569         return 0;
1570 }
1571
1572 /*
1573  * Unsharing of sighand is not supported yet
1574  */
1575 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1576 {
1577         struct sighand_struct *sigh = current->sighand;
1578
1579         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1580                 return -EINVAL;
1581         else
1582                 return 0;
1583 }
1584
1585 /*
1586  * Unshare vm if it is being shared
1587  */
1588 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1589 {
1590         struct mm_struct *mm = current->mm;
1591
1592         if ((unshare_flags & CLONE_VM) &&
1593             (mm && atomic_read(&mm->mm_users) > 1)) {
1594                 return -EINVAL;
1595         }
1596
1597         return 0;
1598 }
1599
1600 /*
1601  * Unshare file descriptor table if it is being shared
1602  */
1603 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1604 {
1605         struct files_struct *fd = current->files;
1606         int error = 0;
1607
1608         if ((unshare_flags & CLONE_FILES) &&
1609             (fd && atomic_read(&fd->count) > 1)) {
1610                 *new_fdp = dup_fd(fd, &error);
1611                 if (!*new_fdp)
1612                         return error;
1613         }
1614
1615         return 0;
1616 }
1617
1618 /*
1619  * unshare allows a process to 'unshare' part of the process
1620  * context which was originally shared using clone.  copy_*
1621  * functions used by do_fork() cannot be used here directly
1622  * because they modify an inactive task_struct that is being
1623  * constructed. Here we are modifying the current, active,
1624  * task_struct.
1625  */
1626 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1627 {
1628         int err = 0;
1629         struct fs_struct *fs, *new_fs = NULL;
1630         struct sighand_struct *new_sigh = NULL;
1631         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1632         struct files_struct *fd, *new_fd = NULL;
1633         struct nsproxy *new_nsproxy = NULL;
1634         int do_sysvsem = 0;
1635
1636         check_unshare_flags(&unshare_flags);
1637
1638         /* Return -EINVAL for all unsupported flags */
1639         err = -EINVAL;
1640         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1641                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1642                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1643                 goto bad_unshare_out;
1644
1645         /*
1646          * CLONE_NEWIPC must also detach from the undolist: after switching
1647          * to a new ipc namespace, the semaphore arrays from the old
1648          * namespace are unreachable.
1649          */
1650         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1651                 do_sysvsem = 1;
1652         if ((err = unshare_thread(unshare_flags)))
1653                 goto bad_unshare_out;
1654         if ((err = unshare_fs(unshare_flags, &new_fs)))
1655                 goto bad_unshare_cleanup_thread;
1656         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1657                 goto bad_unshare_cleanup_fs;
1658         if ((err = unshare_vm(unshare_flags, &new_mm)))
1659                 goto bad_unshare_cleanup_sigh;
1660         if ((err = unshare_fd(unshare_flags, &new_fd)))
1661                 goto bad_unshare_cleanup_vm;
1662         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1663                         new_fs)))
1664                 goto bad_unshare_cleanup_fd;
1665
1666         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1667                 if (do_sysvsem) {
1668                         /*
1669                          * CLONE_SYSVSEM is equivalent to sys_exit().
1670                          */
1671                         exit_sem(current);
1672                 }
1673
1674                 if (new_nsproxy) {
1675                         switch_task_namespaces(current, new_nsproxy);
1676                         new_nsproxy = NULL;
1677                 }
1678
1679                 task_lock(current);
1680
1681                 if (new_fs) {
1682                         fs = current->fs;
1683                         spin_lock(&fs->lock);
1684                         current->fs = new_fs;
1685                         if (--fs->users)
1686                                 new_fs = NULL;
1687                         else
1688                                 new_fs = fs;
1689                         spin_unlock(&fs->lock);
1690                 }
1691
1692                 if (new_mm) {
1693                         mm = current->mm;
1694                         active_mm = current->active_mm;
1695                         current->mm = new_mm;
1696                         current->active_mm = new_mm;
1697                         if (current->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
1698                                 atomic_dec(&mm->oom_disable_count);
1699                                 atomic_inc(&new_mm->oom_disable_count);
1700                         }
1701                         activate_mm(active_mm, new_mm);
1702                         new_mm = mm;
1703                 }
1704
1705                 if (new_fd) {
1706                         fd = current->files;
1707                         current->files = new_fd;
1708                         new_fd = fd;
1709                 }
1710
1711                 task_unlock(current);
1712         }
1713
1714         if (new_nsproxy)
1715                 put_nsproxy(new_nsproxy);
1716
1717 bad_unshare_cleanup_fd:
1718         if (new_fd)
1719                 put_files_struct(new_fd);
1720
1721 bad_unshare_cleanup_vm:
1722         if (new_mm)
1723                 mmput(new_mm);
1724
1725 bad_unshare_cleanup_sigh:
1726         if (new_sigh)
1727                 if (atomic_dec_and_test(&new_sigh->count))
1728                         kmem_cache_free(sighand_cachep, new_sigh);
1729
1730 bad_unshare_cleanup_fs:
1731         if (new_fs)
1732                 free_fs_struct(new_fs);
1733
1734 bad_unshare_cleanup_thread:
1735 bad_unshare_out:
1736         return err;
1737 }
1738
1739 /*
1740  *      Helper to unshare the files of the current task.
1741  *      We don't want to expose copy_files internals to
1742  *      the exec layer of the kernel.
1743  */
1744
1745 int unshare_files(struct files_struct **displaced)
1746 {
1747         struct task_struct *task = current;
1748         struct files_struct *copy = NULL;
1749         int error;
1750
1751         error = unshare_fd(CLONE_FILES, &copy);
1752         if (error || !copy) {
1753                 *displaced = NULL;
1754                 return error;
1755         }
1756         *displaced = task->files;
1757         task_lock(task);
1758         task->files = copy;
1759         task_unlock(task);
1760         return 0;
1761 }