f6083561dfe0a9f8d2a13138f7332bc358a51653
[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/mnt_namespace.h>
21 #include <linux/personality.h>
22 #include <linux/mempolicy.h>
23 #include <linux/sem.h>
24 #include <linux/file.h>
25 #include <linux/fdtable.h>
26 #include <linux/iocontext.h>
27 #include <linux/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.h>
30 #include <linux/mmu_notifier.h>
31 #include <linux/fs.h>
32 #include <linux/nsproxy.h>
33 #include <linux/capability.h>
34 #include <linux/cpu.h>
35 #include <linux/cgroup.h>
36 #include <linux/security.h>
37 #include <linux/hugetlb.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/tracehook.h>
42 #include <linux/futex.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/profile.h>
50 #include <linux/rmap.h>
51 #include <linux/acct.h>
52 #include <linux/tsacct_kern.h>
53 #include <linux/cn_proc.h>
54 #include <linux/freezer.h>
55 #include <linux/delayacct.h>
56 #include <linux/taskstats_kern.h>
57 #include <linux/random.h>
58 #include <linux/tty.h>
59 #include <linux/proc_fs.h>
60 #include <linux/blkdev.h>
61 #include <trace/sched.h>
62
63 #include <asm/pgtable.h>
64 #include <asm/pgalloc.h>
65 #include <asm/uaccess.h>
66 #include <asm/mmu_context.h>
67 #include <asm/cacheflush.h>
68 #include <asm/tlbflush.h>
69
70 /*
71  * Protected counters by write_lock_irq(&tasklist_lock)
72  */
73 unsigned long total_forks;      /* Handle normal Linux uptimes. */
74 int nr_threads;                 /* The idle threads do not count.. */
75
76 int max_threads;                /* tunable limit on nr_threads */
77
78 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
79
80 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
81
82 int nr_processes(void)
83 {
84         int cpu;
85         int total = 0;
86
87         for_each_online_cpu(cpu)
88                 total += per_cpu(process_counts, cpu);
89
90         return total;
91 }
92
93 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
94 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
95 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
96 static struct kmem_cache *task_struct_cachep;
97 #endif
98
99 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
100 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
101 {
102 #ifdef CONFIG_DEBUG_STACK_USAGE
103         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
104 #else
105         gfp_t mask = GFP_KERNEL;
106 #endif
107         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
108 }
109
110 static inline void free_thread_info(struct thread_info *ti)
111 {
112         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
113 }
114 #endif
115
116 /* SLAB cache for signal_struct structures (tsk->signal) */
117 static struct kmem_cache *signal_cachep;
118
119 /* SLAB cache for sighand_struct structures (tsk->sighand) */
120 struct kmem_cache *sighand_cachep;
121
122 /* SLAB cache for files_struct structures (tsk->files) */
123 struct kmem_cache *files_cachep;
124
125 /* SLAB cache for fs_struct structures (tsk->fs) */
126 struct kmem_cache *fs_cachep;
127
128 /* SLAB cache for vm_area_struct structures */
129 struct kmem_cache *vm_area_cachep;
130
131 /* SLAB cache for mm_struct structures (tsk->mm) */
132 static struct kmem_cache *mm_cachep;
133
134 void free_task(struct task_struct *tsk)
135 {
136         prop_local_destroy_single(&tsk->dirties);
137         free_thread_info(tsk->stack);
138         rt_mutex_debug_task_free(tsk);
139         free_task_struct(tsk);
140 }
141 EXPORT_SYMBOL(free_task);
142
143 void __put_task_struct(struct task_struct *tsk)
144 {
145         WARN_ON(!tsk->exit_state);
146         WARN_ON(atomic_read(&tsk->usage));
147         WARN_ON(tsk == current);
148
149         security_task_free(tsk);
150         free_uid(tsk->user);
151         put_group_info(tsk->group_info);
152         delayacct_tsk_free(tsk);
153
154         if (!profile_handoff_task(tsk))
155                 free_task(tsk);
156 }
157
158 /*
159  * macro override instead of weak attribute alias, to workaround
160  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
161  */
162 #ifndef arch_task_cache_init
163 #define arch_task_cache_init()
164 #endif
165
166 void __init fork_init(unsigned long mempages)
167 {
168 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
169 #ifndef ARCH_MIN_TASKALIGN
170 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
171 #endif
172         /* create a slab on which task_structs can be allocated */
173         task_struct_cachep =
174                 kmem_cache_create("task_struct", sizeof(struct task_struct),
175                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
176 #endif
177
178         /* do the arch specific task caches init */
179         arch_task_cache_init();
180
181         /*
182          * The default maximum number of threads is set to a safe
183          * value: the thread structures can take up at most half
184          * of memory.
185          */
186         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
187
188         /*
189          * we need to allow at least 20 threads to boot a system
190          */
191         if(max_threads < 20)
192                 max_threads = 20;
193
194         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
195         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
196         init_task.signal->rlim[RLIMIT_SIGPENDING] =
197                 init_task.signal->rlim[RLIMIT_NPROC];
198 }
199
200 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
201                                                struct task_struct *src)
202 {
203         *dst = *src;
204         return 0;
205 }
206
207 static struct task_struct *dup_task_struct(struct task_struct *orig)
208 {
209         struct task_struct *tsk;
210         struct thread_info *ti;
211         int err;
212
213         prepare_to_copy(orig);
214
215         tsk = alloc_task_struct();
216         if (!tsk)
217                 return NULL;
218
219         ti = alloc_thread_info(tsk);
220         if (!ti) {
221                 free_task_struct(tsk);
222                 return NULL;
223         }
224
225         err = arch_dup_task_struct(tsk, orig);
226         if (err)
227                 goto out;
228
229         tsk->stack = ti;
230
231         err = prop_local_init_single(&tsk->dirties);
232         if (err)
233                 goto out;
234
235         setup_thread_stack(tsk, orig);
236
237 #ifdef CONFIG_CC_STACKPROTECTOR
238         tsk->stack_canary = get_random_int();
239 #endif
240
241         /* One for us, one for whoever does the "release_task()" (usually parent) */
242         atomic_set(&tsk->usage,2);
243         atomic_set(&tsk->fs_excl, 0);
244 #ifdef CONFIG_BLK_DEV_IO_TRACE
245         tsk->btrace_seq = 0;
246 #endif
247         tsk->splice_pipe = NULL;
248         return tsk;
249
250 out:
251         free_thread_info(ti);
252         free_task_struct(tsk);
253         return NULL;
254 }
255
256 #ifdef CONFIG_MMU
257 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
258 {
259         struct vm_area_struct *mpnt, *tmp, **pprev;
260         struct rb_node **rb_link, *rb_parent;
261         int retval;
262         unsigned long charge;
263         struct mempolicy *pol;
264
265         down_write(&oldmm->mmap_sem);
266         flush_cache_dup_mm(oldmm);
267         /*
268          * Not linked in yet - no deadlock potential:
269          */
270         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
271
272         mm->locked_vm = 0;
273         mm->mmap = NULL;
274         mm->mmap_cache = NULL;
275         mm->free_area_cache = oldmm->mmap_base;
276         mm->cached_hole_size = ~0UL;
277         mm->map_count = 0;
278         cpus_clear(mm->cpu_vm_mask);
279         mm->mm_rb = RB_ROOT;
280         rb_link = &mm->mm_rb.rb_node;
281         rb_parent = NULL;
282         pprev = &mm->mmap;
283
284         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
285                 struct file *file;
286
287                 if (mpnt->vm_flags & VM_DONTCOPY) {
288                         long pages = vma_pages(mpnt);
289                         mm->total_vm -= pages;
290                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
291                                                                 -pages);
292                         continue;
293                 }
294                 charge = 0;
295                 if (mpnt->vm_flags & VM_ACCOUNT) {
296                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
297                         if (security_vm_enough_memory(len))
298                                 goto fail_nomem;
299                         charge = len;
300                 }
301                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
302                 if (!tmp)
303                         goto fail_nomem;
304                 *tmp = *mpnt;
305                 pol = mpol_dup(vma_policy(mpnt));
306                 retval = PTR_ERR(pol);
307                 if (IS_ERR(pol))
308                         goto fail_nomem_policy;
309                 vma_set_policy(tmp, pol);
310                 tmp->vm_flags &= ~VM_LOCKED;
311                 tmp->vm_mm = mm;
312                 tmp->vm_next = NULL;
313                 anon_vma_link(tmp);
314                 file = tmp->vm_file;
315                 if (file) {
316                         struct inode *inode = file->f_path.dentry->d_inode;
317                         get_file(file);
318                         if (tmp->vm_flags & VM_DENYWRITE)
319                                 atomic_dec(&inode->i_writecount);
320
321                         /* insert tmp into the share list, just after mpnt */
322                         spin_lock(&file->f_mapping->i_mmap_lock);
323                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
324                         flush_dcache_mmap_lock(file->f_mapping);
325                         vma_prio_tree_add(tmp, mpnt);
326                         flush_dcache_mmap_unlock(file->f_mapping);
327                         spin_unlock(&file->f_mapping->i_mmap_lock);
328                 }
329
330                 /*
331                  * Clear hugetlb-related page reserves for children. This only
332                  * affects MAP_PRIVATE mappings. Faults generated by the child
333                  * are not guaranteed to succeed, even if read-only
334                  */
335                 if (is_vm_hugetlb_page(tmp))
336                         reset_vma_resv_huge_pages(tmp);
337
338                 /*
339                  * Link in the new vma and copy the page table entries.
340                  */
341                 *pprev = tmp;
342                 pprev = &tmp->vm_next;
343
344                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
345                 rb_link = &tmp->vm_rb.rb_right;
346                 rb_parent = &tmp->vm_rb;
347
348                 mm->map_count++;
349                 retval = copy_page_range(mm, oldmm, mpnt);
350
351                 if (tmp->vm_ops && tmp->vm_ops->open)
352                         tmp->vm_ops->open(tmp);
353
354                 if (retval)
355                         goto out;
356         }
357         /* a new mm has just been created */
358         arch_dup_mmap(oldmm, mm);
359         retval = 0;
360 out:
361         up_write(&mm->mmap_sem);
362         flush_tlb_mm(oldmm);
363         up_write(&oldmm->mmap_sem);
364         return retval;
365 fail_nomem_policy:
366         kmem_cache_free(vm_area_cachep, tmp);
367 fail_nomem:
368         retval = -ENOMEM;
369         vm_unacct_memory(charge);
370         goto out;
371 }
372
373 static inline int mm_alloc_pgd(struct mm_struct * mm)
374 {
375         mm->pgd = pgd_alloc(mm);
376         if (unlikely(!mm->pgd))
377                 return -ENOMEM;
378         return 0;
379 }
380
381 static inline void mm_free_pgd(struct mm_struct * mm)
382 {
383         pgd_free(mm, mm->pgd);
384 }
385 #else
386 #define dup_mmap(mm, oldmm)     (0)
387 #define mm_alloc_pgd(mm)        (0)
388 #define mm_free_pgd(mm)
389 #endif /* CONFIG_MMU */
390
391 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
392
393 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
394 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
395
396 #include <linux/init_task.h>
397
398 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
399 {
400         atomic_set(&mm->mm_users, 1);
401         atomic_set(&mm->mm_count, 1);
402         init_rwsem(&mm->mmap_sem);
403         INIT_LIST_HEAD(&mm->mmlist);
404         mm->flags = (current->mm) ? current->mm->flags
405                                   : MMF_DUMP_FILTER_DEFAULT;
406         mm->core_state = NULL;
407         mm->nr_ptes = 0;
408         set_mm_counter(mm, file_rss, 0);
409         set_mm_counter(mm, anon_rss, 0);
410         spin_lock_init(&mm->page_table_lock);
411         rwlock_init(&mm->ioctx_list_lock);
412         mm->ioctx_list = NULL;
413         mm->free_area_cache = TASK_UNMAPPED_BASE;
414         mm->cached_hole_size = ~0UL;
415         mm_init_owner(mm, p);
416
417         if (likely(!mm_alloc_pgd(mm))) {
418                 mm->def_flags = 0;
419                 mmu_notifier_mm_init(mm);
420                 return mm;
421         }
422
423         free_mm(mm);
424         return NULL;
425 }
426
427 /*
428  * Allocate and initialize an mm_struct.
429  */
430 struct mm_struct * mm_alloc(void)
431 {
432         struct mm_struct * mm;
433
434         mm = allocate_mm();
435         if (mm) {
436                 memset(mm, 0, sizeof(*mm));
437                 mm = mm_init(mm, current);
438         }
439         return mm;
440 }
441
442 /*
443  * Called when the last reference to the mm
444  * is dropped: either by a lazy thread or by
445  * mmput. Free the page directory and the mm.
446  */
447 void __mmdrop(struct mm_struct *mm)
448 {
449         BUG_ON(mm == &init_mm);
450         mm_free_pgd(mm);
451         destroy_context(mm);
452         mmu_notifier_mm_destroy(mm);
453         free_mm(mm);
454 }
455 EXPORT_SYMBOL_GPL(__mmdrop);
456
457 /*
458  * Decrement the use count and release all resources for an mm.
459  */
460 void mmput(struct mm_struct *mm)
461 {
462         might_sleep();
463
464         if (atomic_dec_and_test(&mm->mm_users)) {
465                 exit_aio(mm);
466                 exit_mmap(mm);
467                 set_mm_exe_file(mm, NULL);
468                 if (!list_empty(&mm->mmlist)) {
469                         spin_lock(&mmlist_lock);
470                         list_del(&mm->mmlist);
471                         spin_unlock(&mmlist_lock);
472                 }
473                 put_swap_token(mm);
474                 mmdrop(mm);
475         }
476 }
477 EXPORT_SYMBOL_GPL(mmput);
478
479 /**
480  * get_task_mm - acquire a reference to the task's mm
481  *
482  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
483  * this kernel workthread has transiently adopted a user mm with use_mm,
484  * to do its AIO) is not set and if so returns a reference to it, after
485  * bumping up the use count.  User must release the mm via mmput()
486  * after use.  Typically used by /proc and ptrace.
487  */
488 struct mm_struct *get_task_mm(struct task_struct *task)
489 {
490         struct mm_struct *mm;
491
492         task_lock(task);
493         mm = task->mm;
494         if (mm) {
495                 if (task->flags & PF_KTHREAD)
496                         mm = NULL;
497                 else
498                         atomic_inc(&mm->mm_users);
499         }
500         task_unlock(task);
501         return mm;
502 }
503 EXPORT_SYMBOL_GPL(get_task_mm);
504
505 /* Please note the differences between mmput and mm_release.
506  * mmput is called whenever we stop holding onto a mm_struct,
507  * error success whatever.
508  *
509  * mm_release is called after a mm_struct has been removed
510  * from the current process.
511  *
512  * This difference is important for error handling, when we
513  * only half set up a mm_struct for a new process and need to restore
514  * the old one.  Because we mmput the new mm_struct before
515  * restoring the old one. . .
516  * Eric Biederman 10 January 1998
517  */
518 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
519 {
520         struct completion *vfork_done = tsk->vfork_done;
521
522         /* Get rid of any cached register state */
523         deactivate_mm(tsk, mm);
524
525         /* notify parent sleeping on vfork() */
526         if (vfork_done) {
527                 tsk->vfork_done = NULL;
528                 complete(vfork_done);
529         }
530
531         /*
532          * If we're exiting normally, clear a user-space tid field if
533          * requested.  We leave this alone when dying by signal, to leave
534          * the value intact in a core dump, and to save the unnecessary
535          * trouble otherwise.  Userland only wants this done for a sys_exit.
536          */
537         if (tsk->clear_child_tid
538             && !(tsk->flags & PF_SIGNALED)
539             && atomic_read(&mm->mm_users) > 1) {
540                 u32 __user * tidptr = tsk->clear_child_tid;
541                 tsk->clear_child_tid = NULL;
542
543                 /*
544                  * We don't check the error code - if userspace has
545                  * not set up a proper pointer then tough luck.
546                  */
547                 put_user(0, tidptr);
548                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
549         }
550 }
551
552 /*
553  * Allocate a new mm structure and copy contents from the
554  * mm structure of the passed in task structure.
555  */
556 struct mm_struct *dup_mm(struct task_struct *tsk)
557 {
558         struct mm_struct *mm, *oldmm = current->mm;
559         int err;
560
561         if (!oldmm)
562                 return NULL;
563
564         mm = allocate_mm();
565         if (!mm)
566                 goto fail_nomem;
567
568         memcpy(mm, oldmm, sizeof(*mm));
569
570         /* Initializing for Swap token stuff */
571         mm->token_priority = 0;
572         mm->last_interval = 0;
573
574         if (!mm_init(mm, tsk))
575                 goto fail_nomem;
576
577         if (init_new_context(tsk, mm))
578                 goto fail_nocontext;
579
580         dup_mm_exe_file(oldmm, mm);
581
582         err = dup_mmap(mm, oldmm);
583         if (err)
584                 goto free_pt;
585
586         mm->hiwater_rss = get_mm_rss(mm);
587         mm->hiwater_vm = mm->total_vm;
588
589         return mm;
590
591 free_pt:
592         mmput(mm);
593
594 fail_nomem:
595         return NULL;
596
597 fail_nocontext:
598         /*
599          * If init_new_context() failed, we cannot use mmput() to free the mm
600          * because it calls destroy_context()
601          */
602         mm_free_pgd(mm);
603         free_mm(mm);
604         return NULL;
605 }
606
607 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
608 {
609         struct mm_struct * mm, *oldmm;
610         int retval;
611
612         tsk->min_flt = tsk->maj_flt = 0;
613         tsk->nvcsw = tsk->nivcsw = 0;
614
615         tsk->mm = NULL;
616         tsk->active_mm = NULL;
617
618         /*
619          * Are we cloning a kernel thread?
620          *
621          * We need to steal a active VM for that..
622          */
623         oldmm = current->mm;
624         if (!oldmm)
625                 return 0;
626
627         if (clone_flags & CLONE_VM) {
628                 atomic_inc(&oldmm->mm_users);
629                 mm = oldmm;
630                 goto good_mm;
631         }
632
633         retval = -ENOMEM;
634         mm = dup_mm(tsk);
635         if (!mm)
636                 goto fail_nomem;
637
638 good_mm:
639         /* Initializing for Swap token stuff */
640         mm->token_priority = 0;
641         mm->last_interval = 0;
642
643         tsk->mm = mm;
644         tsk->active_mm = mm;
645         return 0;
646
647 fail_nomem:
648         return retval;
649 }
650
651 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
652 {
653         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
654         /* We don't need to lock fs - think why ;-) */
655         if (fs) {
656                 atomic_set(&fs->count, 1);
657                 rwlock_init(&fs->lock);
658                 fs->umask = old->umask;
659                 read_lock(&old->lock);
660                 fs->root = old->root;
661                 path_get(&old->root);
662                 fs->pwd = old->pwd;
663                 path_get(&old->pwd);
664                 read_unlock(&old->lock);
665         }
666         return fs;
667 }
668
669 struct fs_struct *copy_fs_struct(struct fs_struct *old)
670 {
671         return __copy_fs_struct(old);
672 }
673
674 EXPORT_SYMBOL_GPL(copy_fs_struct);
675
676 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
677 {
678         if (clone_flags & CLONE_FS) {
679                 atomic_inc(&current->fs->count);
680                 return 0;
681         }
682         tsk->fs = __copy_fs_struct(current->fs);
683         if (!tsk->fs)
684                 return -ENOMEM;
685         return 0;
686 }
687
688 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
689 {
690         struct files_struct *oldf, *newf;
691         int error = 0;
692
693         /*
694          * A background process may not have any files ...
695          */
696         oldf = current->files;
697         if (!oldf)
698                 goto out;
699
700         if (clone_flags & CLONE_FILES) {
701                 atomic_inc(&oldf->count);
702                 goto out;
703         }
704
705         newf = dup_fd(oldf, &error);
706         if (!newf)
707                 goto out;
708
709         tsk->files = newf;
710         error = 0;
711 out:
712         return error;
713 }
714
715 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
716 {
717 #ifdef CONFIG_BLOCK
718         struct io_context *ioc = current->io_context;
719
720         if (!ioc)
721                 return 0;
722         /*
723          * Share io context with parent, if CLONE_IO is set
724          */
725         if (clone_flags & CLONE_IO) {
726                 tsk->io_context = ioc_task_link(ioc);
727                 if (unlikely(!tsk->io_context))
728                         return -ENOMEM;
729         } else if (ioprio_valid(ioc->ioprio)) {
730                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
731                 if (unlikely(!tsk->io_context))
732                         return -ENOMEM;
733
734                 tsk->io_context->ioprio = ioc->ioprio;
735         }
736 #endif
737         return 0;
738 }
739
740 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
741 {
742         struct sighand_struct *sig;
743
744         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
745                 atomic_inc(&current->sighand->count);
746                 return 0;
747         }
748         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
749         rcu_assign_pointer(tsk->sighand, sig);
750         if (!sig)
751                 return -ENOMEM;
752         atomic_set(&sig->count, 1);
753         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
754         return 0;
755 }
756
757 void __cleanup_sighand(struct sighand_struct *sighand)
758 {
759         if (atomic_dec_and_test(&sighand->count))
760                 kmem_cache_free(sighand_cachep, sighand);
761 }
762
763
764 /*
765  * Initialize POSIX timer handling for a thread group.
766  */
767 static void posix_cpu_timers_init_group(struct signal_struct *sig)
768 {
769         /* Thread group counters. */
770         thread_group_cputime_init(sig);
771
772         /* Expiration times and increments. */
773         sig->it_virt_expires = cputime_zero;
774         sig->it_virt_incr = cputime_zero;
775         sig->it_prof_expires = cputime_zero;
776         sig->it_prof_incr = cputime_zero;
777
778         /* Cached expiration times. */
779         sig->cputime_expires.prof_exp = cputime_zero;
780         sig->cputime_expires.virt_exp = cputime_zero;
781         sig->cputime_expires.sched_exp = 0;
782
783         /* The timer lists. */
784         INIT_LIST_HEAD(&sig->cpu_timers[0]);
785         INIT_LIST_HEAD(&sig->cpu_timers[1]);
786         INIT_LIST_HEAD(&sig->cpu_timers[2]);
787 }
788
789 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
790 {
791         struct signal_struct *sig;
792         int ret;
793
794         if (clone_flags & CLONE_THREAD) {
795                 ret = thread_group_cputime_clone_thread(current);
796                 if (likely(!ret)) {
797                         atomic_inc(&current->signal->count);
798                         atomic_inc(&current->signal->live);
799                 }
800                 return ret;
801         }
802         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
803         tsk->signal = sig;
804         if (!sig)
805                 return -ENOMEM;
806
807         ret = copy_thread_group_keys(tsk);
808         if (ret < 0) {
809                 kmem_cache_free(signal_cachep, sig);
810                 return ret;
811         }
812
813         atomic_set(&sig->count, 1);
814         atomic_set(&sig->live, 1);
815         init_waitqueue_head(&sig->wait_chldexit);
816         sig->flags = 0;
817         sig->group_exit_code = 0;
818         sig->group_exit_task = NULL;
819         sig->group_stop_count = 0;
820         sig->curr_target = tsk;
821         init_sigpending(&sig->shared_pending);
822         INIT_LIST_HEAD(&sig->posix_timers);
823
824         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
825         sig->it_real_incr.tv64 = 0;
826         sig->real_timer.function = it_real_fn;
827
828         sig->leader = 0;        /* session leadership doesn't inherit */
829         sig->tty_old_pgrp = NULL;
830         sig->tty = NULL;
831
832         sig->cutime = sig->cstime = cputime_zero;
833         sig->gtime = cputime_zero;
834         sig->cgtime = cputime_zero;
835         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
836         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
837         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
838         task_io_accounting_init(&sig->ioac);
839         taskstats_tgid_init(sig);
840
841         task_lock(current->group_leader);
842         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
843         task_unlock(current->group_leader);
844
845         posix_cpu_timers_init_group(sig);
846
847         acct_init_pacct(&sig->pacct);
848
849         tty_audit_fork(sig);
850
851         return 0;
852 }
853
854 void __cleanup_signal(struct signal_struct *sig)
855 {
856         thread_group_cputime_free(sig);
857         exit_thread_group_keys(sig);
858         tty_kref_put(sig->tty);
859         kmem_cache_free(signal_cachep, sig);
860 }
861
862 static void cleanup_signal(struct task_struct *tsk)
863 {
864         struct signal_struct *sig = tsk->signal;
865
866         atomic_dec(&sig->live);
867
868         if (atomic_dec_and_test(&sig->count))
869                 __cleanup_signal(sig);
870 }
871
872 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
873 {
874         unsigned long new_flags = p->flags;
875
876         new_flags &= ~PF_SUPERPRIV;
877         new_flags |= PF_FORKNOEXEC;
878         new_flags |= PF_STARTING;
879         p->flags = new_flags;
880         clear_freeze_flag(p);
881 }
882
883 asmlinkage long sys_set_tid_address(int __user *tidptr)
884 {
885         current->clear_child_tid = tidptr;
886
887         return task_pid_vnr(current);
888 }
889
890 static void rt_mutex_init_task(struct task_struct *p)
891 {
892         spin_lock_init(&p->pi_lock);
893 #ifdef CONFIG_RT_MUTEXES
894         plist_head_init(&p->pi_waiters, &p->pi_lock);
895         p->pi_blocked_on = NULL;
896 #endif
897 }
898
899 #ifdef CONFIG_MM_OWNER
900 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
901 {
902         mm->owner = p;
903 }
904 #endif /* CONFIG_MM_OWNER */
905
906 /*
907  * Initialize POSIX timer handling for a single task.
908  */
909 static void posix_cpu_timers_init(struct task_struct *tsk)
910 {
911         tsk->cputime_expires.prof_exp = cputime_zero;
912         tsk->cputime_expires.virt_exp = cputime_zero;
913         tsk->cputime_expires.sched_exp = 0;
914         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
915         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
916         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
917 }
918
919 /*
920  * This creates a new process as a copy of the old one,
921  * but does not actually start it yet.
922  *
923  * It copies the registers, and all the appropriate
924  * parts of the process environment (as per the clone
925  * flags). The actual kick-off is left to the caller.
926  */
927 static struct task_struct *copy_process(unsigned long clone_flags,
928                                         unsigned long stack_start,
929                                         struct pt_regs *regs,
930                                         unsigned long stack_size,
931                                         int __user *child_tidptr,
932                                         struct pid *pid,
933                                         int trace)
934 {
935         int retval;
936         struct task_struct *p;
937         int cgroup_callbacks_done = 0;
938
939         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
940                 return ERR_PTR(-EINVAL);
941
942         /*
943          * Thread groups must share signals as well, and detached threads
944          * can only be started up within the thread group.
945          */
946         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
947                 return ERR_PTR(-EINVAL);
948
949         /*
950          * Shared signal handlers imply shared VM. By way of the above,
951          * thread groups also imply shared VM. Blocking this case allows
952          * for various simplifications in other code.
953          */
954         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
955                 return ERR_PTR(-EINVAL);
956
957         retval = security_task_create(clone_flags);
958         if (retval)
959                 goto fork_out;
960
961         retval = -ENOMEM;
962         p = dup_task_struct(current);
963         if (!p)
964                 goto fork_out;
965
966         rt_mutex_init_task(p);
967
968 #ifdef CONFIG_PROVE_LOCKING
969         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
970         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
971 #endif
972         retval = -EAGAIN;
973         if (atomic_read(&p->user->processes) >=
974                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
975                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
976                     p->user != current->nsproxy->user_ns->root_user)
977                         goto bad_fork_free;
978         }
979
980         atomic_inc(&p->user->__count);
981         atomic_inc(&p->user->processes);
982         get_group_info(p->group_info);
983
984         /*
985          * If multiple threads are within copy_process(), then this check
986          * triggers too late. This doesn't hurt, the check is only there
987          * to stop root fork bombs.
988          */
989         if (nr_threads >= max_threads)
990                 goto bad_fork_cleanup_count;
991
992         if (!try_module_get(task_thread_info(p)->exec_domain->module))
993                 goto bad_fork_cleanup_count;
994
995         if (p->binfmt && !try_module_get(p->binfmt->module))
996                 goto bad_fork_cleanup_put_domain;
997
998         p->did_exec = 0;
999         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1000         copy_flags(clone_flags, p);
1001         INIT_LIST_HEAD(&p->children);
1002         INIT_LIST_HEAD(&p->sibling);
1003 #ifdef CONFIG_PREEMPT_RCU
1004         p->rcu_read_lock_nesting = 0;
1005         p->rcu_flipctr_idx = 0;
1006 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1007         p->vfork_done = NULL;
1008         spin_lock_init(&p->alloc_lock);
1009
1010         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1011         init_sigpending(&p->pending);
1012
1013         p->utime = cputime_zero;
1014         p->stime = cputime_zero;
1015         p->gtime = cputime_zero;
1016         p->utimescaled = cputime_zero;
1017         p->stimescaled = cputime_zero;
1018         p->prev_utime = cputime_zero;
1019         p->prev_stime = cputime_zero;
1020
1021         p->default_timer_slack_ns = current->timer_slack_ns;
1022
1023 #ifdef CONFIG_DETECT_SOFTLOCKUP
1024         p->last_switch_count = 0;
1025         p->last_switch_timestamp = 0;
1026 #endif
1027
1028         task_io_accounting_init(&p->ioac);
1029         acct_clear_integrals(p);
1030
1031         posix_cpu_timers_init(p);
1032
1033         p->lock_depth = -1;             /* -1 = no lock */
1034         do_posix_clock_monotonic_gettime(&p->start_time);
1035         p->real_start_time = p->start_time;
1036         monotonic_to_bootbased(&p->real_start_time);
1037 #ifdef CONFIG_SECURITY
1038         p->security = NULL;
1039 #endif
1040         p->cap_bset = current->cap_bset;
1041         p->io_context = NULL;
1042         p->audit_context = NULL;
1043         cgroup_fork(p);
1044 #ifdef CONFIG_NUMA
1045         p->mempolicy = mpol_dup(p->mempolicy);
1046         if (IS_ERR(p->mempolicy)) {
1047                 retval = PTR_ERR(p->mempolicy);
1048                 p->mempolicy = NULL;
1049                 goto bad_fork_cleanup_cgroup;
1050         }
1051         mpol_fix_fork_child_flag(p);
1052 #endif
1053 #ifdef CONFIG_TRACE_IRQFLAGS
1054         p->irq_events = 0;
1055 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1056         p->hardirqs_enabled = 1;
1057 #else
1058         p->hardirqs_enabled = 0;
1059 #endif
1060         p->hardirq_enable_ip = 0;
1061         p->hardirq_enable_event = 0;
1062         p->hardirq_disable_ip = _THIS_IP_;
1063         p->hardirq_disable_event = 0;
1064         p->softirqs_enabled = 1;
1065         p->softirq_enable_ip = _THIS_IP_;
1066         p->softirq_enable_event = 0;
1067         p->softirq_disable_ip = 0;
1068         p->softirq_disable_event = 0;
1069         p->hardirq_context = 0;
1070         p->softirq_context = 0;
1071 #endif
1072 #ifdef CONFIG_LOCKDEP
1073         p->lockdep_depth = 0; /* no locks held yet */
1074         p->curr_chain_key = 0;
1075         p->lockdep_recursion = 0;
1076 #endif
1077
1078 #ifdef CONFIG_DEBUG_MUTEXES
1079         p->blocked_on = NULL; /* not blocked yet */
1080 #endif
1081
1082         /* Perform scheduler related setup. Assign this task to a CPU. */
1083         sched_fork(p, clone_flags);
1084
1085         if ((retval = security_task_alloc(p)))
1086                 goto bad_fork_cleanup_policy;
1087         if ((retval = audit_alloc(p)))
1088                 goto bad_fork_cleanup_security;
1089         /* copy all the process information */
1090         if ((retval = copy_semundo(clone_flags, p)))
1091                 goto bad_fork_cleanup_audit;
1092         if ((retval = copy_files(clone_flags, p)))
1093                 goto bad_fork_cleanup_semundo;
1094         if ((retval = copy_fs(clone_flags, p)))
1095                 goto bad_fork_cleanup_files;
1096         if ((retval = copy_sighand(clone_flags, p)))
1097                 goto bad_fork_cleanup_fs;
1098         if ((retval = copy_signal(clone_flags, p)))
1099                 goto bad_fork_cleanup_sighand;
1100         if ((retval = copy_mm(clone_flags, p)))
1101                 goto bad_fork_cleanup_signal;
1102         if ((retval = copy_keys(clone_flags, p)))
1103                 goto bad_fork_cleanup_mm;
1104         if ((retval = copy_namespaces(clone_flags, p)))
1105                 goto bad_fork_cleanup_keys;
1106         if ((retval = copy_io(clone_flags, p)))
1107                 goto bad_fork_cleanup_namespaces;
1108         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1109         if (retval)
1110                 goto bad_fork_cleanup_io;
1111
1112         if (pid != &init_struct_pid) {
1113                 retval = -ENOMEM;
1114                 pid = alloc_pid(task_active_pid_ns(p));
1115                 if (!pid)
1116                         goto bad_fork_cleanup_io;
1117
1118                 if (clone_flags & CLONE_NEWPID) {
1119                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1120                         if (retval < 0)
1121                                 goto bad_fork_free_pid;
1122                 }
1123         }
1124
1125         p->pid = pid_nr(pid);
1126         p->tgid = p->pid;
1127         if (clone_flags & CLONE_THREAD)
1128                 p->tgid = current->tgid;
1129
1130         if (current->nsproxy != p->nsproxy) {
1131                 retval = ns_cgroup_clone(p, pid);
1132                 if (retval)
1133                         goto bad_fork_free_pid;
1134         }
1135
1136         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1137         /*
1138          * Clear TID on mm_release()?
1139          */
1140         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1141 #ifdef CONFIG_FUTEX
1142         p->robust_list = NULL;
1143 #ifdef CONFIG_COMPAT
1144         p->compat_robust_list = NULL;
1145 #endif
1146         INIT_LIST_HEAD(&p->pi_state_list);
1147         p->pi_state_cache = NULL;
1148 #endif
1149         /*
1150          * sigaltstack should be cleared when sharing the same VM
1151          */
1152         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1153                 p->sas_ss_sp = p->sas_ss_size = 0;
1154
1155         /*
1156          * Syscall tracing should be turned off in the child regardless
1157          * of CLONE_PTRACE.
1158          */
1159         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1160 #ifdef TIF_SYSCALL_EMU
1161         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1162 #endif
1163         clear_all_latency_tracing(p);
1164
1165         /* Our parent execution domain becomes current domain
1166            These must match for thread signalling to apply */
1167         p->parent_exec_id = p->self_exec_id;
1168
1169         /* ok, now we should be set up.. */
1170         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1171         p->pdeath_signal = 0;
1172         p->exit_state = 0;
1173
1174         /*
1175          * Ok, make it visible to the rest of the system.
1176          * We dont wake it up yet.
1177          */
1178         p->group_leader = p;
1179         INIT_LIST_HEAD(&p->thread_group);
1180
1181         /* Now that the task is set up, run cgroup callbacks if
1182          * necessary. We need to run them before the task is visible
1183          * on the tasklist. */
1184         cgroup_fork_callbacks(p);
1185         cgroup_callbacks_done = 1;
1186
1187         /* Need tasklist lock for parent etc handling! */
1188         write_lock_irq(&tasklist_lock);
1189
1190         /*
1191          * The task hasn't been attached yet, so its cpus_allowed mask will
1192          * not be changed, nor will its assigned CPU.
1193          *
1194          * The cpus_allowed mask of the parent may have changed after it was
1195          * copied first time - so re-copy it here, then check the child's CPU
1196          * to ensure it is on a valid CPU (and if not, just force it back to
1197          * parent's CPU). This avoids alot of nasty races.
1198          */
1199         p->cpus_allowed = current->cpus_allowed;
1200         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1201         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1202                         !cpu_online(task_cpu(p))))
1203                 set_task_cpu(p, smp_processor_id());
1204
1205         /* CLONE_PARENT re-uses the old parent */
1206         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1207                 p->real_parent = current->real_parent;
1208         else
1209                 p->real_parent = current;
1210
1211         spin_lock(&current->sighand->siglock);
1212
1213         /*
1214          * Process group and session signals need to be delivered to just the
1215          * parent before the fork or both the parent and the child after the
1216          * fork. Restart if a signal comes in before we add the new process to
1217          * it's process group.
1218          * A fatal signal pending means that current will exit, so the new
1219          * thread can't slip out of an OOM kill (or normal SIGKILL).
1220          */
1221         recalc_sigpending();
1222         if (signal_pending(current)) {
1223                 spin_unlock(&current->sighand->siglock);
1224                 write_unlock_irq(&tasklist_lock);
1225                 retval = -ERESTARTNOINTR;
1226                 goto bad_fork_free_pid;
1227         }
1228
1229         if (clone_flags & CLONE_THREAD) {
1230                 p->group_leader = current->group_leader;
1231                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1232         }
1233
1234         if (likely(p->pid)) {
1235                 list_add_tail(&p->sibling, &p->real_parent->children);
1236                 tracehook_finish_clone(p, clone_flags, trace);
1237
1238                 if (thread_group_leader(p)) {
1239                         if (clone_flags & CLONE_NEWPID)
1240                                 p->nsproxy->pid_ns->child_reaper = p;
1241
1242                         p->signal->leader_pid = pid;
1243                         tty_kref_put(p->signal->tty);
1244                         p->signal->tty = tty_kref_get(current->signal->tty);
1245                         set_task_pgrp(p, task_pgrp_nr(current));
1246                         set_task_session(p, task_session_nr(current));
1247                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1248                         attach_pid(p, PIDTYPE_SID, task_session(current));
1249                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1250                         __get_cpu_var(process_counts)++;
1251                 }
1252                 attach_pid(p, PIDTYPE_PID, pid);
1253                 nr_threads++;
1254         }
1255
1256         total_forks++;
1257         spin_unlock(&current->sighand->siglock);
1258         write_unlock_irq(&tasklist_lock);
1259         proc_fork_connector(p);
1260         cgroup_post_fork(p);
1261         return p;
1262
1263 bad_fork_free_pid:
1264         if (pid != &init_struct_pid)
1265                 free_pid(pid);
1266 bad_fork_cleanup_io:
1267         put_io_context(p->io_context);
1268 bad_fork_cleanup_namespaces:
1269         exit_task_namespaces(p);
1270 bad_fork_cleanup_keys:
1271         exit_keys(p);
1272 bad_fork_cleanup_mm:
1273         if (p->mm)
1274                 mmput(p->mm);
1275 bad_fork_cleanup_signal:
1276         cleanup_signal(p);
1277 bad_fork_cleanup_sighand:
1278         __cleanup_sighand(p->sighand);
1279 bad_fork_cleanup_fs:
1280         exit_fs(p); /* blocking */
1281 bad_fork_cleanup_files:
1282         exit_files(p); /* blocking */
1283 bad_fork_cleanup_semundo:
1284         exit_sem(p);
1285 bad_fork_cleanup_audit:
1286         audit_free(p);
1287 bad_fork_cleanup_security:
1288         security_task_free(p);
1289 bad_fork_cleanup_policy:
1290 #ifdef CONFIG_NUMA
1291         mpol_put(p->mempolicy);
1292 bad_fork_cleanup_cgroup:
1293 #endif
1294         cgroup_exit(p, cgroup_callbacks_done);
1295         delayacct_tsk_free(p);
1296         if (p->binfmt)
1297                 module_put(p->binfmt->module);
1298 bad_fork_cleanup_put_domain:
1299         module_put(task_thread_info(p)->exec_domain->module);
1300 bad_fork_cleanup_count:
1301         put_group_info(p->group_info);
1302         atomic_dec(&p->user->processes);
1303         free_uid(p->user);
1304 bad_fork_free:
1305         free_task(p);
1306 fork_out:
1307         return ERR_PTR(retval);
1308 }
1309
1310 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1311 {
1312         memset(regs, 0, sizeof(struct pt_regs));
1313         return regs;
1314 }
1315
1316 struct task_struct * __cpuinit fork_idle(int cpu)
1317 {
1318         struct task_struct *task;
1319         struct pt_regs regs;
1320
1321         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1322                             &init_struct_pid, 0);
1323         if (!IS_ERR(task))
1324                 init_idle(task, cpu);
1325
1326         return task;
1327 }
1328
1329 /*
1330  *  Ok, this is the main fork-routine.
1331  *
1332  * It copies the process, and if successful kick-starts
1333  * it and waits for it to finish using the VM if required.
1334  */
1335 long do_fork(unsigned long clone_flags,
1336               unsigned long stack_start,
1337               struct pt_regs *regs,
1338               unsigned long stack_size,
1339               int __user *parent_tidptr,
1340               int __user *child_tidptr)
1341 {
1342         struct task_struct *p;
1343         int trace = 0;
1344         long nr;
1345
1346         /*
1347          * We hope to recycle these flags after 2.6.26
1348          */
1349         if (unlikely(clone_flags & CLONE_STOPPED)) {
1350                 static int __read_mostly count = 100;
1351
1352                 if (count > 0 && printk_ratelimit()) {
1353                         char comm[TASK_COMM_LEN];
1354
1355                         count--;
1356                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1357                                         "clone flags 0x%lx\n",
1358                                 get_task_comm(comm, current),
1359                                 clone_flags & CLONE_STOPPED);
1360                 }
1361         }
1362
1363         /*
1364          * When called from kernel_thread, don't do user tracing stuff.
1365          */
1366         if (likely(user_mode(regs)))
1367                 trace = tracehook_prepare_clone(clone_flags);
1368
1369         p = copy_process(clone_flags, stack_start, regs, stack_size,
1370                          child_tidptr, NULL, trace);
1371         /*
1372          * Do this prior waking up the new thread - the thread pointer
1373          * might get invalid after that point, if the thread exits quickly.
1374          */
1375         if (!IS_ERR(p)) {
1376                 struct completion vfork;
1377
1378                 trace_sched_process_fork(current, p);
1379
1380                 nr = task_pid_vnr(p);
1381
1382                 if (clone_flags & CLONE_PARENT_SETTID)
1383                         put_user(nr, parent_tidptr);
1384
1385                 if (clone_flags & CLONE_VFORK) {
1386                         p->vfork_done = &vfork;
1387                         init_completion(&vfork);
1388                 }
1389
1390                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1391
1392                 /*
1393                  * We set PF_STARTING at creation in case tracing wants to
1394                  * use this to distinguish a fully live task from one that
1395                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1396                  * clear it and set the child going.
1397                  */
1398                 p->flags &= ~PF_STARTING;
1399
1400                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1401                         /*
1402                          * We'll start up with an immediate SIGSTOP.
1403                          */
1404                         sigaddset(&p->pending.signal, SIGSTOP);
1405                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1406                         __set_task_state(p, TASK_STOPPED);
1407                 } else {
1408                         wake_up_new_task(p, clone_flags);
1409                 }
1410
1411                 tracehook_report_clone_complete(trace, regs,
1412                                                 clone_flags, nr, p);
1413
1414                 if (clone_flags & CLONE_VFORK) {
1415                         freezer_do_not_count();
1416                         wait_for_completion(&vfork);
1417                         freezer_count();
1418                         tracehook_report_vfork_done(p, nr);
1419                 }
1420         } else {
1421                 nr = PTR_ERR(p);
1422         }
1423         return nr;
1424 }
1425
1426 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1427 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1428 #endif
1429
1430 static void sighand_ctor(void *data)
1431 {
1432         struct sighand_struct *sighand = data;
1433
1434         spin_lock_init(&sighand->siglock);
1435         init_waitqueue_head(&sighand->signalfd_wqh);
1436 }
1437
1438 void __init proc_caches_init(void)
1439 {
1440         sighand_cachep = kmem_cache_create("sighand_cache",
1441                         sizeof(struct sighand_struct), 0,
1442                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1443                         sighand_ctor);
1444         signal_cachep = kmem_cache_create("signal_cache",
1445                         sizeof(struct signal_struct), 0,
1446                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1447         files_cachep = kmem_cache_create("files_cache",
1448                         sizeof(struct files_struct), 0,
1449                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1450         fs_cachep = kmem_cache_create("fs_cache",
1451                         sizeof(struct fs_struct), 0,
1452                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1453         vm_area_cachep = kmem_cache_create("vm_area_struct",
1454                         sizeof(struct vm_area_struct), 0,
1455                         SLAB_PANIC, NULL);
1456         mm_cachep = kmem_cache_create("mm_struct",
1457                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1458                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1459 }
1460
1461 /*
1462  * Check constraints on flags passed to the unshare system call and
1463  * force unsharing of additional process context as appropriate.
1464  */
1465 static void check_unshare_flags(unsigned long *flags_ptr)
1466 {
1467         /*
1468          * If unsharing a thread from a thread group, must also
1469          * unshare vm.
1470          */
1471         if (*flags_ptr & CLONE_THREAD)
1472                 *flags_ptr |= CLONE_VM;
1473
1474         /*
1475          * If unsharing vm, must also unshare signal handlers.
1476          */
1477         if (*flags_ptr & CLONE_VM)
1478                 *flags_ptr |= CLONE_SIGHAND;
1479
1480         /*
1481          * If unsharing signal handlers and the task was created
1482          * using CLONE_THREAD, then must unshare the thread
1483          */
1484         if ((*flags_ptr & CLONE_SIGHAND) &&
1485             (atomic_read(&current->signal->count) > 1))
1486                 *flags_ptr |= CLONE_THREAD;
1487
1488         /*
1489          * If unsharing namespace, must also unshare filesystem information.
1490          */
1491         if (*flags_ptr & CLONE_NEWNS)
1492                 *flags_ptr |= CLONE_FS;
1493 }
1494
1495 /*
1496  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1497  */
1498 static int unshare_thread(unsigned long unshare_flags)
1499 {
1500         if (unshare_flags & CLONE_THREAD)
1501                 return -EINVAL;
1502
1503         return 0;
1504 }
1505
1506 /*
1507  * Unshare the filesystem structure if it is being shared
1508  */
1509 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1510 {
1511         struct fs_struct *fs = current->fs;
1512
1513         if ((unshare_flags & CLONE_FS) &&
1514             (fs && atomic_read(&fs->count) > 1)) {
1515                 *new_fsp = __copy_fs_struct(current->fs);
1516                 if (!*new_fsp)
1517                         return -ENOMEM;
1518         }
1519
1520         return 0;
1521 }
1522
1523 /*
1524  * Unsharing of sighand is not supported yet
1525  */
1526 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1527 {
1528         struct sighand_struct *sigh = current->sighand;
1529
1530         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1531                 return -EINVAL;
1532         else
1533                 return 0;
1534 }
1535
1536 /*
1537  * Unshare vm if it is being shared
1538  */
1539 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1540 {
1541         struct mm_struct *mm = current->mm;
1542
1543         if ((unshare_flags & CLONE_VM) &&
1544             (mm && atomic_read(&mm->mm_users) > 1)) {
1545                 return -EINVAL;
1546         }
1547
1548         return 0;
1549 }
1550
1551 /*
1552  * Unshare file descriptor table if it is being shared
1553  */
1554 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1555 {
1556         struct files_struct *fd = current->files;
1557         int error = 0;
1558
1559         if ((unshare_flags & CLONE_FILES) &&
1560             (fd && atomic_read(&fd->count) > 1)) {
1561                 *new_fdp = dup_fd(fd, &error);
1562                 if (!*new_fdp)
1563                         return error;
1564         }
1565
1566         return 0;
1567 }
1568
1569 /*
1570  * unshare allows a process to 'unshare' part of the process
1571  * context which was originally shared using clone.  copy_*
1572  * functions used by do_fork() cannot be used here directly
1573  * because they modify an inactive task_struct that is being
1574  * constructed. Here we are modifying the current, active,
1575  * task_struct.
1576  */
1577 asmlinkage long sys_unshare(unsigned long unshare_flags)
1578 {
1579         int err = 0;
1580         struct fs_struct *fs, *new_fs = NULL;
1581         struct sighand_struct *new_sigh = NULL;
1582         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1583         struct files_struct *fd, *new_fd = NULL;
1584         struct nsproxy *new_nsproxy = NULL;
1585         int do_sysvsem = 0;
1586
1587         check_unshare_flags(&unshare_flags);
1588
1589         /* Return -EINVAL for all unsupported flags */
1590         err = -EINVAL;
1591         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1592                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1593                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1594                                 CLONE_NEWNET))
1595                 goto bad_unshare_out;
1596
1597         /*
1598          * CLONE_NEWIPC must also detach from the undolist: after switching
1599          * to a new ipc namespace, the semaphore arrays from the old
1600          * namespace are unreachable.
1601          */
1602         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1603                 do_sysvsem = 1;
1604         if ((err = unshare_thread(unshare_flags)))
1605                 goto bad_unshare_out;
1606         if ((err = unshare_fs(unshare_flags, &new_fs)))
1607                 goto bad_unshare_cleanup_thread;
1608         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1609                 goto bad_unshare_cleanup_fs;
1610         if ((err = unshare_vm(unshare_flags, &new_mm)))
1611                 goto bad_unshare_cleanup_sigh;
1612         if ((err = unshare_fd(unshare_flags, &new_fd)))
1613                 goto bad_unshare_cleanup_vm;
1614         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1615                         new_fs)))
1616                 goto bad_unshare_cleanup_fd;
1617
1618         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1619                 if (do_sysvsem) {
1620                         /*
1621                          * CLONE_SYSVSEM is equivalent to sys_exit().
1622                          */
1623                         exit_sem(current);
1624                 }
1625
1626                 if (new_nsproxy) {
1627                         switch_task_namespaces(current, new_nsproxy);
1628                         new_nsproxy = NULL;
1629                 }
1630
1631                 task_lock(current);
1632
1633                 if (new_fs) {
1634                         fs = current->fs;
1635                         current->fs = new_fs;
1636                         new_fs = fs;
1637                 }
1638
1639                 if (new_mm) {
1640                         mm = current->mm;
1641                         active_mm = current->active_mm;
1642                         current->mm = new_mm;
1643                         current->active_mm = new_mm;
1644                         activate_mm(active_mm, new_mm);
1645                         new_mm = mm;
1646                 }
1647
1648                 if (new_fd) {
1649                         fd = current->files;
1650                         current->files = new_fd;
1651                         new_fd = fd;
1652                 }
1653
1654                 task_unlock(current);
1655         }
1656
1657         if (new_nsproxy)
1658                 put_nsproxy(new_nsproxy);
1659
1660 bad_unshare_cleanup_fd:
1661         if (new_fd)
1662                 put_files_struct(new_fd);
1663
1664 bad_unshare_cleanup_vm:
1665         if (new_mm)
1666                 mmput(new_mm);
1667
1668 bad_unshare_cleanup_sigh:
1669         if (new_sigh)
1670                 if (atomic_dec_and_test(&new_sigh->count))
1671                         kmem_cache_free(sighand_cachep, new_sigh);
1672
1673 bad_unshare_cleanup_fs:
1674         if (new_fs)
1675                 put_fs_struct(new_fs);
1676
1677 bad_unshare_cleanup_thread:
1678 bad_unshare_out:
1679         return err;
1680 }
1681
1682 /*
1683  *      Helper to unshare the files of the current task.
1684  *      We don't want to expose copy_files internals to
1685  *      the exec layer of the kernel.
1686  */
1687
1688 int unshare_files(struct files_struct **displaced)
1689 {
1690         struct task_struct *task = current;
1691         struct files_struct *copy = NULL;
1692         int error;
1693
1694         error = unshare_fd(CLONE_FILES, &copy);
1695         if (error || !copy) {
1696                 *displaced = NULL;
1697                 return error;
1698         }
1699         *displaced = task->files;
1700         task_lock(task);
1701         task->files = copy;
1702         task_unlock(task);
1703         return 0;
1704 }