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