function-graph: move initialization of new tasks up in 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/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/compat.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/profile.h>
52 #include <linux/rmap.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 <trace/sched.h>
65 #include <linux/magic.h>
66
67 #include <asm/pgtable.h>
68 #include <asm/pgalloc.h>
69 #include <asm/uaccess.h>
70 #include <asm/mmu_context.h>
71 #include <asm/cacheflush.h>
72 #include <asm/tlbflush.h>
73
74 /*
75  * Protected counters by write_lock_irq(&tasklist_lock)
76  */
77 unsigned long total_forks;      /* Handle normal Linux uptimes. */
78 int nr_threads;                 /* The idle threads do not count.. */
79
80 int max_threads;                /* tunable limit on nr_threads */
81
82 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
83
84 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
85
86 DEFINE_TRACE(sched_process_fork);
87
88 int nr_processes(void)
89 {
90         int cpu;
91         int total = 0;
92
93         for_each_online_cpu(cpu)
94                 total += per_cpu(process_counts, cpu);
95
96         return total;
97 }
98
99 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
100 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
101 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
102 static struct kmem_cache *task_struct_cachep;
103 #endif
104
105 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
106 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
107 {
108 #ifdef CONFIG_DEBUG_STACK_USAGE
109         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
110 #else
111         gfp_t mask = GFP_KERNEL;
112 #endif
113         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
114 }
115
116 static inline void free_thread_info(struct thread_info *ti)
117 {
118         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
119 }
120 #endif
121
122 /* SLAB cache for signal_struct structures (tsk->signal) */
123 static struct kmem_cache *signal_cachep;
124
125 /* SLAB cache for sighand_struct structures (tsk->sighand) */
126 struct kmem_cache *sighand_cachep;
127
128 /* SLAB cache for files_struct structures (tsk->files) */
129 struct kmem_cache *files_cachep;
130
131 /* SLAB cache for fs_struct structures (tsk->fs) */
132 struct kmem_cache *fs_cachep;
133
134 /* SLAB cache for vm_area_struct structures */
135 struct kmem_cache *vm_area_cachep;
136
137 /* SLAB cache for mm_struct structures (tsk->mm) */
138 static struct kmem_cache *mm_cachep;
139
140 void free_task(struct task_struct *tsk)
141 {
142         prop_local_destroy_single(&tsk->dirties);
143         free_thread_info(tsk->stack);
144         rt_mutex_debug_task_free(tsk);
145         ftrace_graph_exit_task(tsk);
146         free_task_struct(tsk);
147 }
148 EXPORT_SYMBOL(free_task);
149
150 void __put_task_struct(struct task_struct *tsk)
151 {
152         WARN_ON(!tsk->exit_state);
153         WARN_ON(atomic_read(&tsk->usage));
154         WARN_ON(tsk == current);
155
156         put_cred(tsk->real_cred);
157         put_cred(tsk->cred);
158         delayacct_tsk_free(tsk);
159
160         if (!profile_handoff_task(tsk))
161                 free_task(tsk);
162 }
163
164 /*
165  * macro override instead of weak attribute alias, to workaround
166  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
167  */
168 #ifndef arch_task_cache_init
169 #define arch_task_cache_init()
170 #endif
171
172 void __init fork_init(unsigned long mempages)
173 {
174 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
175 #ifndef ARCH_MIN_TASKALIGN
176 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
177 #endif
178         /* create a slab on which task_structs can be allocated */
179         task_struct_cachep =
180                 kmem_cache_create("task_struct", sizeof(struct task_struct),
181                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
182 #endif
183
184         /* do the arch specific task caches init */
185         arch_task_cache_init();
186
187         /*
188          * The default maximum number of threads is set to a safe
189          * value: the thread structures can take up at most half
190          * of memory.
191          */
192         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
193
194         /*
195          * we need to allow at least 20 threads to boot a system
196          */
197         if(max_threads < 20)
198                 max_threads = 20;
199
200         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
201         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
202         init_task.signal->rlim[RLIMIT_SIGPENDING] =
203                 init_task.signal->rlim[RLIMIT_NPROC];
204 }
205
206 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
207                                                struct task_struct *src)
208 {
209         *dst = *src;
210         return 0;
211 }
212
213 static struct task_struct *dup_task_struct(struct task_struct *orig)
214 {
215         struct task_struct *tsk;
216         struct thread_info *ti;
217         unsigned long *stackend;
218
219         int err;
220
221         prepare_to_copy(orig);
222
223         tsk = alloc_task_struct();
224         if (!tsk)
225                 return NULL;
226
227         ti = alloc_thread_info(tsk);
228         if (!ti) {
229                 free_task_struct(tsk);
230                 return NULL;
231         }
232
233         err = arch_dup_task_struct(tsk, orig);
234         if (err)
235                 goto out;
236
237         tsk->stack = ti;
238
239         err = prop_local_init_single(&tsk->dirties);
240         if (err)
241                 goto out;
242
243         setup_thread_stack(tsk, orig);
244         stackend = end_of_stack(tsk);
245         *stackend = STACK_END_MAGIC;    /* for overflow detection */
246
247 #ifdef CONFIG_CC_STACKPROTECTOR
248         tsk->stack_canary = get_random_int();
249 #endif
250
251         /* One for us, one for whoever does the "release_task()" (usually parent) */
252         atomic_set(&tsk->usage,2);
253         atomic_set(&tsk->fs_excl, 0);
254 #ifdef CONFIG_BLK_DEV_IO_TRACE
255         tsk->btrace_seq = 0;
256 #endif
257         tsk->splice_pipe = NULL;
258         return tsk;
259
260 out:
261         free_thread_info(ti);
262         free_task_struct(tsk);
263         return NULL;
264 }
265
266 #ifdef CONFIG_MMU
267 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
268 {
269         struct vm_area_struct *mpnt, *tmp, **pprev;
270         struct rb_node **rb_link, *rb_parent;
271         int retval;
272         unsigned long charge;
273         struct mempolicy *pol;
274
275         down_write(&oldmm->mmap_sem);
276         flush_cache_dup_mm(oldmm);
277         /*
278          * Not linked in yet - no deadlock potential:
279          */
280         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
281
282         mm->locked_vm = 0;
283         mm->mmap = NULL;
284         mm->mmap_cache = NULL;
285         mm->free_area_cache = oldmm->mmap_base;
286         mm->cached_hole_size = ~0UL;
287         mm->map_count = 0;
288         cpumask_clear(mm_cpumask(mm));
289         mm->mm_rb = RB_ROOT;
290         rb_link = &mm->mm_rb.rb_node;
291         rb_parent = NULL;
292         pprev = &mm->mmap;
293
294         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
295                 struct file *file;
296
297                 if (mpnt->vm_flags & VM_DONTCOPY) {
298                         long pages = vma_pages(mpnt);
299                         mm->total_vm -= pages;
300                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
301                                                                 -pages);
302                         continue;
303                 }
304                 charge = 0;
305                 if (mpnt->vm_flags & VM_ACCOUNT) {
306                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
307                         if (security_vm_enough_memory(len))
308                                 goto fail_nomem;
309                         charge = len;
310                 }
311                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
312                 if (!tmp)
313                         goto fail_nomem;
314                 *tmp = *mpnt;
315                 pol = mpol_dup(vma_policy(mpnt));
316                 retval = PTR_ERR(pol);
317                 if (IS_ERR(pol))
318                         goto fail_nomem_policy;
319                 vma_set_policy(tmp, pol);
320                 tmp->vm_flags &= ~VM_LOCKED;
321                 tmp->vm_mm = mm;
322                 tmp->vm_next = NULL;
323                 anon_vma_link(tmp);
324                 file = tmp->vm_file;
325                 if (file) {
326                         struct inode *inode = file->f_path.dentry->d_inode;
327                         struct address_space *mapping = file->f_mapping;
328
329                         get_file(file);
330                         if (tmp->vm_flags & VM_DENYWRITE)
331                                 atomic_dec(&inode->i_writecount);
332                         spin_lock(&mapping->i_mmap_lock);
333                         if (tmp->vm_flags & VM_SHARED)
334                                 mapping->i_mmap_writable++;
335                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
336                         flush_dcache_mmap_lock(mapping);
337                         /* insert tmp into the share list, just after mpnt */
338                         vma_prio_tree_add(tmp, mpnt);
339                         flush_dcache_mmap_unlock(mapping);
340                         spin_unlock(&mapping->i_mmap_lock);
341                 }
342
343                 /*
344                  * Clear hugetlb-related page reserves for children. This only
345                  * affects MAP_PRIVATE mappings. Faults generated by the child
346                  * are not guaranteed to succeed, even if read-only
347                  */
348                 if (is_vm_hugetlb_page(tmp))
349                         reset_vma_resv_huge_pages(tmp);
350
351                 /*
352                  * Link in the new vma and copy the page table entries.
353                  */
354                 *pprev = tmp;
355                 pprev = &tmp->vm_next;
356
357                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
358                 rb_link = &tmp->vm_rb.rb_right;
359                 rb_parent = &tmp->vm_rb;
360
361                 mm->map_count++;
362                 retval = copy_page_range(mm, oldmm, mpnt);
363
364                 if (tmp->vm_ops && tmp->vm_ops->open)
365                         tmp->vm_ops->open(tmp);
366
367                 if (retval)
368                         goto out;
369         }
370         /* a new mm has just been created */
371         arch_dup_mmap(oldmm, mm);
372         retval = 0;
373 out:
374         up_write(&mm->mmap_sem);
375         flush_tlb_mm(oldmm);
376         up_write(&oldmm->mmap_sem);
377         return retval;
378 fail_nomem_policy:
379         kmem_cache_free(vm_area_cachep, tmp);
380 fail_nomem:
381         retval = -ENOMEM;
382         vm_unacct_memory(charge);
383         goto out;
384 }
385
386 static inline int mm_alloc_pgd(struct mm_struct * mm)
387 {
388         mm->pgd = pgd_alloc(mm);
389         if (unlikely(!mm->pgd))
390                 return -ENOMEM;
391         return 0;
392 }
393
394 static inline void mm_free_pgd(struct mm_struct * mm)
395 {
396         pgd_free(mm, mm->pgd);
397 }
398 #else
399 #define dup_mmap(mm, oldmm)     (0)
400 #define mm_alloc_pgd(mm)        (0)
401 #define mm_free_pgd(mm)
402 #endif /* CONFIG_MMU */
403
404 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
405
406 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
407 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
408
409 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
410
411 static int __init coredump_filter_setup(char *s)
412 {
413         default_dump_filter =
414                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
415                 MMF_DUMP_FILTER_MASK;
416         return 1;
417 }
418
419 __setup("coredump_filter=", coredump_filter_setup);
420
421 #include <linux/init_task.h>
422
423 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
424 {
425         atomic_set(&mm->mm_users, 1);
426         atomic_set(&mm->mm_count, 1);
427         init_rwsem(&mm->mmap_sem);
428         INIT_LIST_HEAD(&mm->mmlist);
429         mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
430         mm->core_state = NULL;
431         mm->nr_ptes = 0;
432         set_mm_counter(mm, file_rss, 0);
433         set_mm_counter(mm, anon_rss, 0);
434         spin_lock_init(&mm->page_table_lock);
435         spin_lock_init(&mm->ioctx_lock);
436         INIT_HLIST_HEAD(&mm->ioctx_list);
437         mm->free_area_cache = TASK_UNMAPPED_BASE;
438         mm->cached_hole_size = ~0UL;
439         mm_init_owner(mm, p);
440
441         if (likely(!mm_alloc_pgd(mm))) {
442                 mm->def_flags = 0;
443                 mmu_notifier_mm_init(mm);
444                 return mm;
445         }
446
447         free_mm(mm);
448         return NULL;
449 }
450
451 /*
452  * Allocate and initialize an mm_struct.
453  */
454 struct mm_struct * mm_alloc(void)
455 {
456         struct mm_struct * mm;
457
458         mm = allocate_mm();
459         if (mm) {
460                 memset(mm, 0, sizeof(*mm));
461                 mm = mm_init(mm, current);
462         }
463         return mm;
464 }
465
466 /*
467  * Called when the last reference to the mm
468  * is dropped: either by a lazy thread or by
469  * mmput. Free the page directory and the mm.
470  */
471 void __mmdrop(struct mm_struct *mm)
472 {
473         BUG_ON(mm == &init_mm);
474         mm_free_pgd(mm);
475         destroy_context(mm);
476         mmu_notifier_mm_destroy(mm);
477         free_mm(mm);
478 }
479 EXPORT_SYMBOL_GPL(__mmdrop);
480
481 /*
482  * Decrement the use count and release all resources for an mm.
483  */
484 void mmput(struct mm_struct *mm)
485 {
486         might_sleep();
487
488         if (atomic_dec_and_test(&mm->mm_users)) {
489                 exit_aio(mm);
490                 exit_mmap(mm);
491                 set_mm_exe_file(mm, NULL);
492                 if (!list_empty(&mm->mmlist)) {
493                         spin_lock(&mmlist_lock);
494                         list_del(&mm->mmlist);
495                         spin_unlock(&mmlist_lock);
496                 }
497                 put_swap_token(mm);
498                 mmdrop(mm);
499         }
500 }
501 EXPORT_SYMBOL_GPL(mmput);
502
503 /**
504  * get_task_mm - acquire a reference to the task's mm
505  *
506  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
507  * this kernel workthread has transiently adopted a user mm with use_mm,
508  * to do its AIO) is not set and if so returns a reference to it, after
509  * bumping up the use count.  User must release the mm via mmput()
510  * after use.  Typically used by /proc and ptrace.
511  */
512 struct mm_struct *get_task_mm(struct task_struct *task)
513 {
514         struct mm_struct *mm;
515
516         task_lock(task);
517         mm = task->mm;
518         if (mm) {
519                 if (task->flags & PF_KTHREAD)
520                         mm = NULL;
521                 else
522                         atomic_inc(&mm->mm_users);
523         }
524         task_unlock(task);
525         return mm;
526 }
527 EXPORT_SYMBOL_GPL(get_task_mm);
528
529 /* Please note the differences between mmput and mm_release.
530  * mmput is called whenever we stop holding onto a mm_struct,
531  * error success whatever.
532  *
533  * mm_release is called after a mm_struct has been removed
534  * from the current process.
535  *
536  * This difference is important for error handling, when we
537  * only half set up a mm_struct for a new process and need to restore
538  * the old one.  Because we mmput the new mm_struct before
539  * restoring the old one. . .
540  * Eric Biederman 10 January 1998
541  */
542 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
543 {
544         struct completion *vfork_done = tsk->vfork_done;
545
546         /* Get rid of any futexes when releasing the mm */
547 #ifdef CONFIG_FUTEX
548         if (unlikely(tsk->robust_list))
549                 exit_robust_list(tsk);
550 #ifdef CONFIG_COMPAT
551         if (unlikely(tsk->compat_robust_list))
552                 compat_exit_robust_list(tsk);
553 #endif
554 #endif
555
556         /* Get rid of any cached register state */
557         deactivate_mm(tsk, mm);
558
559         /* notify parent sleeping on vfork() */
560         if (vfork_done) {
561                 tsk->vfork_done = NULL;
562                 complete(vfork_done);
563         }
564
565         /*
566          * If we're exiting normally, clear a user-space tid field if
567          * requested.  We leave this alone when dying by signal, to leave
568          * the value intact in a core dump, and to save the unnecessary
569          * trouble otherwise.  Userland only wants this done for a sys_exit.
570          */
571         if (tsk->clear_child_tid
572             && !(tsk->flags & PF_SIGNALED)
573             && atomic_read(&mm->mm_users) > 1) {
574                 u32 __user * tidptr = tsk->clear_child_tid;
575                 tsk->clear_child_tid = NULL;
576
577                 /*
578                  * We don't check the error code - if userspace has
579                  * not set up a proper pointer then tough luck.
580                  */
581                 put_user(0, tidptr);
582                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
583         }
584 }
585
586 /*
587  * Allocate a new mm structure and copy contents from the
588  * mm structure of the passed in task structure.
589  */
590 struct mm_struct *dup_mm(struct task_struct *tsk)
591 {
592         struct mm_struct *mm, *oldmm = current->mm;
593         int err;
594
595         if (!oldmm)
596                 return NULL;
597
598         mm = allocate_mm();
599         if (!mm)
600                 goto fail_nomem;
601
602         memcpy(mm, oldmm, sizeof(*mm));
603
604         /* Initializing for Swap token stuff */
605         mm->token_priority = 0;
606         mm->last_interval = 0;
607
608         if (!mm_init(mm, tsk))
609                 goto fail_nomem;
610
611         if (init_new_context(tsk, mm))
612                 goto fail_nocontext;
613
614         dup_mm_exe_file(oldmm, mm);
615
616         err = dup_mmap(mm, oldmm);
617         if (err)
618                 goto free_pt;
619
620         mm->hiwater_rss = get_mm_rss(mm);
621         mm->hiwater_vm = mm->total_vm;
622
623         return mm;
624
625 free_pt:
626         mmput(mm);
627
628 fail_nomem:
629         return NULL;
630
631 fail_nocontext:
632         /*
633          * If init_new_context() failed, we cannot use mmput() to free the mm
634          * because it calls destroy_context()
635          */
636         mm_free_pgd(mm);
637         free_mm(mm);
638         return NULL;
639 }
640
641 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
642 {
643         struct mm_struct * mm, *oldmm;
644         int retval;
645
646         tsk->min_flt = tsk->maj_flt = 0;
647         tsk->nvcsw = tsk->nivcsw = 0;
648 #ifdef CONFIG_DETECT_HUNG_TASK
649         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
650 #endif
651
652         tsk->mm = NULL;
653         tsk->active_mm = NULL;
654
655         /*
656          * Are we cloning a kernel thread?
657          *
658          * We need to steal a active VM for that..
659          */
660         oldmm = current->mm;
661         if (!oldmm)
662                 return 0;
663
664         if (clone_flags & CLONE_VM) {
665                 atomic_inc(&oldmm->mm_users);
666                 mm = oldmm;
667                 goto good_mm;
668         }
669
670         retval = -ENOMEM;
671         mm = dup_mm(tsk);
672         if (!mm)
673                 goto fail_nomem;
674
675 good_mm:
676         /* Initializing for Swap token stuff */
677         mm->token_priority = 0;
678         mm->last_interval = 0;
679
680         tsk->mm = mm;
681         tsk->active_mm = mm;
682         return 0;
683
684 fail_nomem:
685         return retval;
686 }
687
688 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
689 {
690         struct fs_struct *fs = current->fs;
691         if (clone_flags & CLONE_FS) {
692                 /* tsk->fs is already what we want */
693                 write_lock(&fs->lock);
694                 if (fs->in_exec) {
695                         write_unlock(&fs->lock);
696                         return -EAGAIN;
697                 }
698                 fs->users++;
699                 write_unlock(&fs->lock);
700                 return 0;
701         }
702         tsk->fs = copy_fs_struct(fs);
703         if (!tsk->fs)
704                 return -ENOMEM;
705         return 0;
706 }
707
708 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
709 {
710         struct files_struct *oldf, *newf;
711         int error = 0;
712
713         /*
714          * A background process may not have any files ...
715          */
716         oldf = current->files;
717         if (!oldf)
718                 goto out;
719
720         if (clone_flags & CLONE_FILES) {
721                 atomic_inc(&oldf->count);
722                 goto out;
723         }
724
725         newf = dup_fd(oldf, &error);
726         if (!newf)
727                 goto out;
728
729         tsk->files = newf;
730         error = 0;
731 out:
732         return error;
733 }
734
735 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
736 {
737 #ifdef CONFIG_BLOCK
738         struct io_context *ioc = current->io_context;
739
740         if (!ioc)
741                 return 0;
742         /*
743          * Share io context with parent, if CLONE_IO is set
744          */
745         if (clone_flags & CLONE_IO) {
746                 tsk->io_context = ioc_task_link(ioc);
747                 if (unlikely(!tsk->io_context))
748                         return -ENOMEM;
749         } else if (ioprio_valid(ioc->ioprio)) {
750                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
751                 if (unlikely(!tsk->io_context))
752                         return -ENOMEM;
753
754                 tsk->io_context->ioprio = ioc->ioprio;
755         }
756 #endif
757         return 0;
758 }
759
760 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
761 {
762         struct sighand_struct *sig;
763
764         if (clone_flags & CLONE_SIGHAND) {
765                 atomic_inc(&current->sighand->count);
766                 return 0;
767         }
768         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
769         rcu_assign_pointer(tsk->sighand, sig);
770         if (!sig)
771                 return -ENOMEM;
772         atomic_set(&sig->count, 1);
773         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
774         return 0;
775 }
776
777 void __cleanup_sighand(struct sighand_struct *sighand)
778 {
779         if (atomic_dec_and_test(&sighand->count))
780                 kmem_cache_free(sighand_cachep, sighand);
781 }
782
783
784 /*
785  * Initialize POSIX timer handling for a thread group.
786  */
787 static void posix_cpu_timers_init_group(struct signal_struct *sig)
788 {
789         /* Thread group counters. */
790         thread_group_cputime_init(sig);
791
792         /* Expiration times and increments. */
793         sig->it_virt_expires = cputime_zero;
794         sig->it_virt_incr = cputime_zero;
795         sig->it_prof_expires = cputime_zero;
796         sig->it_prof_incr = cputime_zero;
797
798         /* Cached expiration times. */
799         sig->cputime_expires.prof_exp = cputime_zero;
800         sig->cputime_expires.virt_exp = cputime_zero;
801         sig->cputime_expires.sched_exp = 0;
802
803         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
804                 sig->cputime_expires.prof_exp =
805                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
806                 sig->cputimer.running = 1;
807         }
808
809         /* The timer lists. */
810         INIT_LIST_HEAD(&sig->cpu_timers[0]);
811         INIT_LIST_HEAD(&sig->cpu_timers[1]);
812         INIT_LIST_HEAD(&sig->cpu_timers[2]);
813 }
814
815 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
816 {
817         struct signal_struct *sig;
818
819         if (clone_flags & CLONE_THREAD) {
820                 atomic_inc(&current->signal->count);
821                 atomic_inc(&current->signal->live);
822                 return 0;
823         }
824
825         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
826         tsk->signal = sig;
827         if (!sig)
828                 return -ENOMEM;
829
830         atomic_set(&sig->count, 1);
831         atomic_set(&sig->live, 1);
832         init_waitqueue_head(&sig->wait_chldexit);
833         sig->flags = 0;
834         if (clone_flags & CLONE_NEWPID)
835                 sig->flags |= SIGNAL_UNKILLABLE;
836         sig->group_exit_code = 0;
837         sig->group_exit_task = NULL;
838         sig->group_stop_count = 0;
839         sig->curr_target = tsk;
840         init_sigpending(&sig->shared_pending);
841         INIT_LIST_HEAD(&sig->posix_timers);
842
843         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
844         sig->it_real_incr.tv64 = 0;
845         sig->real_timer.function = it_real_fn;
846
847         sig->leader = 0;        /* session leadership doesn't inherit */
848         sig->tty_old_pgrp = NULL;
849         sig->tty = NULL;
850
851         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
852         sig->gtime = cputime_zero;
853         sig->cgtime = cputime_zero;
854         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
855         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
856         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
857         task_io_accounting_init(&sig->ioac);
858         sig->sum_sched_runtime = 0;
859         taskstats_tgid_init(sig);
860
861         task_lock(current->group_leader);
862         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
863         task_unlock(current->group_leader);
864
865         posix_cpu_timers_init_group(sig);
866
867         acct_init_pacct(&sig->pacct);
868
869         tty_audit_fork(sig);
870
871         return 0;
872 }
873
874 void __cleanup_signal(struct signal_struct *sig)
875 {
876         thread_group_cputime_free(sig);
877         tty_kref_put(sig->tty);
878         kmem_cache_free(signal_cachep, sig);
879 }
880
881 static void cleanup_signal(struct task_struct *tsk)
882 {
883         struct signal_struct *sig = tsk->signal;
884
885         atomic_dec(&sig->live);
886
887         if (atomic_dec_and_test(&sig->count))
888                 __cleanup_signal(sig);
889 }
890
891 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
892 {
893         unsigned long new_flags = p->flags;
894
895         new_flags &= ~PF_SUPERPRIV;
896         new_flags |= PF_FORKNOEXEC;
897         new_flags |= PF_STARTING;
898         p->flags = new_flags;
899         clear_freeze_flag(p);
900 }
901
902 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
903 {
904         current->clear_child_tid = tidptr;
905
906         return task_pid_vnr(current);
907 }
908
909 static void rt_mutex_init_task(struct task_struct *p)
910 {
911         spin_lock_init(&p->pi_lock);
912 #ifdef CONFIG_RT_MUTEXES
913         plist_head_init(&p->pi_waiters, &p->pi_lock);
914         p->pi_blocked_on = NULL;
915 #endif
916 }
917
918 #ifdef CONFIG_MM_OWNER
919 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
920 {
921         mm->owner = p;
922 }
923 #endif /* CONFIG_MM_OWNER */
924
925 /*
926  * Initialize POSIX timer handling for a single task.
927  */
928 static void posix_cpu_timers_init(struct task_struct *tsk)
929 {
930         tsk->cputime_expires.prof_exp = cputime_zero;
931         tsk->cputime_expires.virt_exp = cputime_zero;
932         tsk->cputime_expires.sched_exp = 0;
933         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
934         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
935         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
936 }
937
938 /*
939  * This creates a new process as a copy of the old one,
940  * but does not actually start it yet.
941  *
942  * It copies the registers, and all the appropriate
943  * parts of the process environment (as per the clone
944  * flags). The actual kick-off is left to the caller.
945  */
946 static struct task_struct *copy_process(unsigned long clone_flags,
947                                         unsigned long stack_start,
948                                         struct pt_regs *regs,
949                                         unsigned long stack_size,
950                                         int __user *child_tidptr,
951                                         struct pid *pid,
952                                         int trace)
953 {
954         int retval;
955         struct task_struct *p;
956         int cgroup_callbacks_done = 0;
957
958         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
959                 return ERR_PTR(-EINVAL);
960
961         /*
962          * Thread groups must share signals as well, and detached threads
963          * can only be started up within the thread group.
964          */
965         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
966                 return ERR_PTR(-EINVAL);
967
968         /*
969          * Shared signal handlers imply shared VM. By way of the above,
970          * thread groups also imply shared VM. Blocking this case allows
971          * for various simplifications in other code.
972          */
973         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
974                 return ERR_PTR(-EINVAL);
975
976         retval = security_task_create(clone_flags);
977         if (retval)
978                 goto fork_out;
979
980         retval = -ENOMEM;
981         p = dup_task_struct(current);
982         if (!p)
983                 goto fork_out;
984
985         ftrace_graph_init_task(p);
986
987         rt_mutex_init_task(p);
988
989 #ifdef CONFIG_PROVE_LOCKING
990         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
991         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
992 #endif
993         retval = -EAGAIN;
994         if (atomic_read(&p->real_cred->user->processes) >=
995                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
996                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
997                     p->real_cred->user != INIT_USER)
998                         goto bad_fork_free;
999         }
1000
1001         retval = copy_creds(p, clone_flags);
1002         if (retval < 0)
1003                 goto bad_fork_free;
1004
1005         /*
1006          * If multiple threads are within copy_process(), then this check
1007          * triggers too late. This doesn't hurt, the check is only there
1008          * to stop root fork bombs.
1009          */
1010         retval = -EAGAIN;
1011         if (nr_threads >= max_threads)
1012                 goto bad_fork_cleanup_count;
1013
1014         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1015                 goto bad_fork_cleanup_count;
1016
1017         if (p->binfmt && !try_module_get(p->binfmt->module))
1018                 goto bad_fork_cleanup_put_domain;
1019
1020         p->did_exec = 0;
1021         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1022         copy_flags(clone_flags, p);
1023         INIT_LIST_HEAD(&p->children);
1024         INIT_LIST_HEAD(&p->sibling);
1025 #ifdef CONFIG_PREEMPT_RCU
1026         p->rcu_read_lock_nesting = 0;
1027         p->rcu_flipctr_idx = 0;
1028 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1029         p->vfork_done = NULL;
1030         spin_lock_init(&p->alloc_lock);
1031
1032         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1033         init_sigpending(&p->pending);
1034
1035         p->utime = cputime_zero;
1036         p->stime = cputime_zero;
1037         p->gtime = cputime_zero;
1038         p->utimescaled = cputime_zero;
1039         p->stimescaled = cputime_zero;
1040         p->prev_utime = cputime_zero;
1041         p->prev_stime = cputime_zero;
1042
1043         p->default_timer_slack_ns = current->timer_slack_ns;
1044
1045         task_io_accounting_init(&p->ioac);
1046         acct_clear_integrals(p);
1047
1048         posix_cpu_timers_init(p);
1049
1050         p->lock_depth = -1;             /* -1 = no lock */
1051         do_posix_clock_monotonic_gettime(&p->start_time);
1052         p->real_start_time = p->start_time;
1053         monotonic_to_bootbased(&p->real_start_time);
1054         p->io_context = NULL;
1055         p->audit_context = NULL;
1056         cgroup_fork(p);
1057 #ifdef CONFIG_NUMA
1058         p->mempolicy = mpol_dup(p->mempolicy);
1059         if (IS_ERR(p->mempolicy)) {
1060                 retval = PTR_ERR(p->mempolicy);
1061                 p->mempolicy = NULL;
1062                 goto bad_fork_cleanup_cgroup;
1063         }
1064         mpol_fix_fork_child_flag(p);
1065 #endif
1066 #ifdef CONFIG_TRACE_IRQFLAGS
1067         p->irq_events = 0;
1068 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1069         p->hardirqs_enabled = 1;
1070 #else
1071         p->hardirqs_enabled = 0;
1072 #endif
1073         p->hardirq_enable_ip = 0;
1074         p->hardirq_enable_event = 0;
1075         p->hardirq_disable_ip = _THIS_IP_;
1076         p->hardirq_disable_event = 0;
1077         p->softirqs_enabled = 1;
1078         p->softirq_enable_ip = _THIS_IP_;
1079         p->softirq_enable_event = 0;
1080         p->softirq_disable_ip = 0;
1081         p->softirq_disable_event = 0;
1082         p->hardirq_context = 0;
1083         p->softirq_context = 0;
1084 #endif
1085 #ifdef CONFIG_LOCKDEP
1086         p->lockdep_depth = 0; /* no locks held yet */
1087         p->curr_chain_key = 0;
1088         p->lockdep_recursion = 0;
1089 #endif
1090
1091 #ifdef CONFIG_DEBUG_MUTEXES
1092         p->blocked_on = NULL; /* not blocked yet */
1093 #endif
1094         if (unlikely(current->ptrace))
1095                 ptrace_fork(p, clone_flags);
1096
1097         /* Perform scheduler related setup. Assign this task to a CPU. */
1098         sched_fork(p, clone_flags);
1099
1100         if ((retval = audit_alloc(p)))
1101                 goto bad_fork_cleanup_policy;
1102         /* copy all the process information */
1103         if ((retval = copy_semundo(clone_flags, p)))
1104                 goto bad_fork_cleanup_audit;
1105         if ((retval = copy_files(clone_flags, p)))
1106                 goto bad_fork_cleanup_semundo;
1107         if ((retval = copy_fs(clone_flags, p)))
1108                 goto bad_fork_cleanup_files;
1109         if ((retval = copy_sighand(clone_flags, p)))
1110                 goto bad_fork_cleanup_fs;
1111         if ((retval = copy_signal(clone_flags, p)))
1112                 goto bad_fork_cleanup_sighand;
1113         if ((retval = copy_mm(clone_flags, p)))
1114                 goto bad_fork_cleanup_signal;
1115         if ((retval = copy_namespaces(clone_flags, p)))
1116                 goto bad_fork_cleanup_mm;
1117         if ((retval = copy_io(clone_flags, p)))
1118                 goto bad_fork_cleanup_namespaces;
1119         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1120         if (retval)
1121                 goto bad_fork_cleanup_io;
1122
1123         if (pid != &init_struct_pid) {
1124                 retval = -ENOMEM;
1125                 pid = alloc_pid(p->nsproxy->pid_ns);
1126                 if (!pid)
1127                         goto bad_fork_cleanup_io;
1128
1129                 if (clone_flags & CLONE_NEWPID) {
1130                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1131                         if (retval < 0)
1132                                 goto bad_fork_free_pid;
1133                 }
1134         }
1135
1136         p->pid = pid_nr(pid);
1137         p->tgid = p->pid;
1138         if (clone_flags & CLONE_THREAD)
1139                 p->tgid = current->tgid;
1140
1141         if (current->nsproxy != p->nsproxy) {
1142                 retval = ns_cgroup_clone(p, pid);
1143                 if (retval)
1144                         goto bad_fork_free_pid;
1145         }
1146
1147         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1148         /*
1149          * Clear TID on mm_release()?
1150          */
1151         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1152 #ifdef CONFIG_FUTEX
1153         p->robust_list = NULL;
1154 #ifdef CONFIG_COMPAT
1155         p->compat_robust_list = NULL;
1156 #endif
1157         INIT_LIST_HEAD(&p->pi_state_list);
1158         p->pi_state_cache = NULL;
1159 #endif
1160         /*
1161          * sigaltstack should be cleared when sharing the same VM
1162          */
1163         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1164                 p->sas_ss_sp = p->sas_ss_size = 0;
1165
1166         /*
1167          * Syscall tracing should be turned off in the child regardless
1168          * of CLONE_PTRACE.
1169          */
1170         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1171 #ifdef TIF_SYSCALL_EMU
1172         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1173 #endif
1174         clear_all_latency_tracing(p);
1175
1176         /* ok, now we should be set up.. */
1177         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1178         p->pdeath_signal = 0;
1179         p->exit_state = 0;
1180
1181         /*
1182          * Ok, make it visible to the rest of the system.
1183          * We dont wake it up yet.
1184          */
1185         p->group_leader = p;
1186         INIT_LIST_HEAD(&p->thread_group);
1187
1188         /* Now that the task is set up, run cgroup callbacks if
1189          * necessary. We need to run them before the task is visible
1190          * on the tasklist. */
1191         cgroup_fork_callbacks(p);
1192         cgroup_callbacks_done = 1;
1193
1194         /* Need tasklist lock for parent etc handling! */
1195         write_lock_irq(&tasklist_lock);
1196
1197         /*
1198          * The task hasn't been attached yet, so its cpus_allowed mask will
1199          * not be changed, nor will its assigned CPU.
1200          *
1201          * The cpus_allowed mask of the parent may have changed after it was
1202          * copied first time - so re-copy it here, then check the child's CPU
1203          * to ensure it is on a valid CPU (and if not, just force it back to
1204          * parent's CPU). This avoids alot of nasty races.
1205          */
1206         p->cpus_allowed = current->cpus_allowed;
1207         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1208         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1209                         !cpu_online(task_cpu(p))))
1210                 set_task_cpu(p, smp_processor_id());
1211
1212         /* CLONE_PARENT re-uses the old parent */
1213         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1214                 p->real_parent = current->real_parent;
1215                 p->parent_exec_id = current->parent_exec_id;
1216         } else {
1217                 p->real_parent = current;
1218                 p->parent_exec_id = current->self_exec_id;
1219         }
1220
1221         spin_lock(&current->sighand->siglock);
1222
1223         /*
1224          * Process group and session signals need to be delivered to just the
1225          * parent before the fork or both the parent and the child after the
1226          * fork. Restart if a signal comes in before we add the new process to
1227          * it's process group.
1228          * A fatal signal pending means that current will exit, so the new
1229          * thread can't slip out of an OOM kill (or normal SIGKILL).
1230          */
1231         recalc_sigpending();
1232         if (signal_pending(current)) {
1233                 spin_unlock(&current->sighand->siglock);
1234                 write_unlock_irq(&tasklist_lock);
1235                 retval = -ERESTARTNOINTR;
1236                 goto bad_fork_free_pid;
1237         }
1238
1239         if (clone_flags & CLONE_THREAD) {
1240                 p->group_leader = current->group_leader;
1241                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1242         }
1243
1244         if (likely(p->pid)) {
1245                 list_add_tail(&p->sibling, &p->real_parent->children);
1246                 tracehook_finish_clone(p, clone_flags, trace);
1247
1248                 if (thread_group_leader(p)) {
1249                         if (clone_flags & CLONE_NEWPID)
1250                                 p->nsproxy->pid_ns->child_reaper = p;
1251
1252                         p->signal->leader_pid = pid;
1253                         tty_kref_put(p->signal->tty);
1254                         p->signal->tty = tty_kref_get(current->signal->tty);
1255                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1256                         attach_pid(p, PIDTYPE_SID, task_session(current));
1257                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1258                         __get_cpu_var(process_counts)++;
1259                 }
1260                 attach_pid(p, PIDTYPE_PID, pid);
1261                 nr_threads++;
1262         }
1263
1264         total_forks++;
1265         spin_unlock(&current->sighand->siglock);
1266         write_unlock_irq(&tasklist_lock);
1267         proc_fork_connector(p);
1268         cgroup_post_fork(p);
1269         return p;
1270
1271 bad_fork_free_pid:
1272         if (pid != &init_struct_pid)
1273                 free_pid(pid);
1274 bad_fork_cleanup_io:
1275         put_io_context(p->io_context);
1276 bad_fork_cleanup_namespaces:
1277         exit_task_namespaces(p);
1278 bad_fork_cleanup_mm:
1279         if (p->mm)
1280                 mmput(p->mm);
1281 bad_fork_cleanup_signal:
1282         cleanup_signal(p);
1283 bad_fork_cleanup_sighand:
1284         __cleanup_sighand(p->sighand);
1285 bad_fork_cleanup_fs:
1286         exit_fs(p); /* blocking */
1287 bad_fork_cleanup_files:
1288         exit_files(p); /* blocking */
1289 bad_fork_cleanup_semundo:
1290         exit_sem(p);
1291 bad_fork_cleanup_audit:
1292         audit_free(p);
1293 bad_fork_cleanup_policy:
1294 #ifdef CONFIG_NUMA
1295         mpol_put(p->mempolicy);
1296 bad_fork_cleanup_cgroup:
1297 #endif
1298         cgroup_exit(p, cgroup_callbacks_done);
1299         delayacct_tsk_free(p);
1300         if (p->binfmt)
1301                 module_put(p->binfmt->module);
1302 bad_fork_cleanup_put_domain:
1303         module_put(task_thread_info(p)->exec_domain->module);
1304 bad_fork_cleanup_count:
1305         atomic_dec(&p->cred->user->processes);
1306         put_cred(p->real_cred);
1307         put_cred(p->cred);
1308 bad_fork_free:
1309         free_task(p);
1310 fork_out:
1311         return ERR_PTR(retval);
1312 }
1313
1314 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1315 {
1316         memset(regs, 0, sizeof(struct pt_regs));
1317         return regs;
1318 }
1319
1320 struct task_struct * __cpuinit fork_idle(int cpu)
1321 {
1322         struct task_struct *task;
1323         struct pt_regs regs;
1324
1325         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1326                             &init_struct_pid, 0);
1327         if (!IS_ERR(task))
1328                 init_idle(task, cpu);
1329
1330         return task;
1331 }
1332
1333 /*
1334  *  Ok, this is the main fork-routine.
1335  *
1336  * It copies the process, and if successful kick-starts
1337  * it and waits for it to finish using the VM if required.
1338  */
1339 long do_fork(unsigned long clone_flags,
1340               unsigned long stack_start,
1341               struct pt_regs *regs,
1342               unsigned long stack_size,
1343               int __user *parent_tidptr,
1344               int __user *child_tidptr)
1345 {
1346         struct task_struct *p;
1347         int trace = 0;
1348         long nr;
1349
1350         /*
1351          * Do some preliminary argument and permissions checking before we
1352          * actually start allocating stuff
1353          */
1354         if (clone_flags & CLONE_NEWUSER) {
1355                 if (clone_flags & CLONE_THREAD)
1356                         return -EINVAL;
1357                 /* hopefully this check will go away when userns support is
1358                  * complete
1359                  */
1360                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1361                                 !capable(CAP_SETGID))
1362                         return -EPERM;
1363         }
1364
1365         /*
1366          * We hope to recycle these flags after 2.6.26
1367          */
1368         if (unlikely(clone_flags & CLONE_STOPPED)) {
1369                 static int __read_mostly count = 100;
1370
1371                 if (count > 0 && printk_ratelimit()) {
1372                         char comm[TASK_COMM_LEN];
1373
1374                         count--;
1375                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1376                                         "clone flags 0x%lx\n",
1377                                 get_task_comm(comm, current),
1378                                 clone_flags & CLONE_STOPPED);
1379                 }
1380         }
1381
1382         /*
1383          * When called from kernel_thread, don't do user tracing stuff.
1384          */
1385         if (likely(user_mode(regs)))
1386                 trace = tracehook_prepare_clone(clone_flags);
1387
1388         p = copy_process(clone_flags, stack_start, regs, stack_size,
1389                          child_tidptr, NULL, trace);
1390         /*
1391          * Do this prior waking up the new thread - the thread pointer
1392          * might get invalid after that point, if the thread exits quickly.
1393          */
1394         if (!IS_ERR(p)) {
1395                 struct completion vfork;
1396
1397                 trace_sched_process_fork(current, p);
1398
1399                 nr = task_pid_vnr(p);
1400
1401                 if (clone_flags & CLONE_PARENT_SETTID)
1402                         put_user(nr, parent_tidptr);
1403
1404                 if (clone_flags & CLONE_VFORK) {
1405                         p->vfork_done = &vfork;
1406                         init_completion(&vfork);
1407                 }
1408
1409                 audit_finish_fork(p);
1410                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1411
1412                 /*
1413                  * We set PF_STARTING at creation in case tracing wants to
1414                  * use this to distinguish a fully live task from one that
1415                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1416                  * clear it and set the child going.
1417                  */
1418                 p->flags &= ~PF_STARTING;
1419
1420                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1421                         /*
1422                          * We'll start up with an immediate SIGSTOP.
1423                          */
1424                         sigaddset(&p->pending.signal, SIGSTOP);
1425                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1426                         __set_task_state(p, TASK_STOPPED);
1427                 } else {
1428                         wake_up_new_task(p, clone_flags);
1429                 }
1430
1431                 tracehook_report_clone_complete(trace, regs,
1432                                                 clone_flags, nr, p);
1433
1434                 if (clone_flags & CLONE_VFORK) {
1435                         freezer_do_not_count();
1436                         wait_for_completion(&vfork);
1437                         freezer_count();
1438                         tracehook_report_vfork_done(p, nr);
1439                 }
1440         } else {
1441                 nr = PTR_ERR(p);
1442         }
1443         return nr;
1444 }
1445
1446 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1447 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1448 #endif
1449
1450 static void sighand_ctor(void *data)
1451 {
1452         struct sighand_struct *sighand = data;
1453
1454         spin_lock_init(&sighand->siglock);
1455         init_waitqueue_head(&sighand->signalfd_wqh);
1456 }
1457
1458 void __init proc_caches_init(void)
1459 {
1460         sighand_cachep = kmem_cache_create("sighand_cache",
1461                         sizeof(struct sighand_struct), 0,
1462                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1463                         sighand_ctor);
1464         signal_cachep = kmem_cache_create("signal_cache",
1465                         sizeof(struct signal_struct), 0,
1466                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1467         files_cachep = kmem_cache_create("files_cache",
1468                         sizeof(struct files_struct), 0,
1469                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1470         fs_cachep = kmem_cache_create("fs_cache",
1471                         sizeof(struct fs_struct), 0,
1472                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1473         mm_cachep = kmem_cache_create("mm_struct",
1474                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1475                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1476         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1477         mmap_init();
1478 }
1479
1480 /*
1481  * Check constraints on flags passed to the unshare system call and
1482  * force unsharing of additional process context as appropriate.
1483  */
1484 static void check_unshare_flags(unsigned long *flags_ptr)
1485 {
1486         /*
1487          * If unsharing a thread from a thread group, must also
1488          * unshare vm.
1489          */
1490         if (*flags_ptr & CLONE_THREAD)
1491                 *flags_ptr |= CLONE_VM;
1492
1493         /*
1494          * If unsharing vm, must also unshare signal handlers.
1495          */
1496         if (*flags_ptr & CLONE_VM)
1497                 *flags_ptr |= CLONE_SIGHAND;
1498
1499         /*
1500          * If unsharing signal handlers and the task was created
1501          * using CLONE_THREAD, then must unshare the thread
1502          */
1503         if ((*flags_ptr & CLONE_SIGHAND) &&
1504             (atomic_read(&current->signal->count) > 1))
1505                 *flags_ptr |= CLONE_THREAD;
1506
1507         /*
1508          * If unsharing namespace, must also unshare filesystem information.
1509          */
1510         if (*flags_ptr & CLONE_NEWNS)
1511                 *flags_ptr |= CLONE_FS;
1512 }
1513
1514 /*
1515  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1516  */
1517 static int unshare_thread(unsigned long unshare_flags)
1518 {
1519         if (unshare_flags & CLONE_THREAD)
1520                 return -EINVAL;
1521
1522         return 0;
1523 }
1524
1525 /*
1526  * Unshare the filesystem structure if it is being shared
1527  */
1528 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1529 {
1530         struct fs_struct *fs = current->fs;
1531
1532         if (!(unshare_flags & CLONE_FS) || !fs)
1533                 return 0;
1534
1535         /* don't need lock here; in the worst case we'll do useless copy */
1536         if (fs->users == 1)
1537                 return 0;
1538
1539         *new_fsp = copy_fs_struct(fs);
1540         if (!*new_fsp)
1541                 return -ENOMEM;
1542
1543         return 0;
1544 }
1545
1546 /*
1547  * Unsharing of sighand is not supported yet
1548  */
1549 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1550 {
1551         struct sighand_struct *sigh = current->sighand;
1552
1553         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1554                 return -EINVAL;
1555         else
1556                 return 0;
1557 }
1558
1559 /*
1560  * Unshare vm if it is being shared
1561  */
1562 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1563 {
1564         struct mm_struct *mm = current->mm;
1565
1566         if ((unshare_flags & CLONE_VM) &&
1567             (mm && atomic_read(&mm->mm_users) > 1)) {
1568                 return -EINVAL;
1569         }
1570
1571         return 0;
1572 }
1573
1574 /*
1575  * Unshare file descriptor table if it is being shared
1576  */
1577 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1578 {
1579         struct files_struct *fd = current->files;
1580         int error = 0;
1581
1582         if ((unshare_flags & CLONE_FILES) &&
1583             (fd && atomic_read(&fd->count) > 1)) {
1584                 *new_fdp = dup_fd(fd, &error);
1585                 if (!*new_fdp)
1586                         return error;
1587         }
1588
1589         return 0;
1590 }
1591
1592 /*
1593  * unshare allows a process to 'unshare' part of the process
1594  * context which was originally shared using clone.  copy_*
1595  * functions used by do_fork() cannot be used here directly
1596  * because they modify an inactive task_struct that is being
1597  * constructed. Here we are modifying the current, active,
1598  * task_struct.
1599  */
1600 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1601 {
1602         int err = 0;
1603         struct fs_struct *fs, *new_fs = NULL;
1604         struct sighand_struct *new_sigh = NULL;
1605         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1606         struct files_struct *fd, *new_fd = NULL;
1607         struct nsproxy *new_nsproxy = NULL;
1608         int do_sysvsem = 0;
1609
1610         check_unshare_flags(&unshare_flags);
1611
1612         /* Return -EINVAL for all unsupported flags */
1613         err = -EINVAL;
1614         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1615                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1616                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1617                 goto bad_unshare_out;
1618
1619         /*
1620          * CLONE_NEWIPC must also detach from the undolist: after switching
1621          * to a new ipc namespace, the semaphore arrays from the old
1622          * namespace are unreachable.
1623          */
1624         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1625                 do_sysvsem = 1;
1626         if ((err = unshare_thread(unshare_flags)))
1627                 goto bad_unshare_out;
1628         if ((err = unshare_fs(unshare_flags, &new_fs)))
1629                 goto bad_unshare_cleanup_thread;
1630         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1631                 goto bad_unshare_cleanup_fs;
1632         if ((err = unshare_vm(unshare_flags, &new_mm)))
1633                 goto bad_unshare_cleanup_sigh;
1634         if ((err = unshare_fd(unshare_flags, &new_fd)))
1635                 goto bad_unshare_cleanup_vm;
1636         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1637                         new_fs)))
1638                 goto bad_unshare_cleanup_fd;
1639
1640         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1641                 if (do_sysvsem) {
1642                         /*
1643                          * CLONE_SYSVSEM is equivalent to sys_exit().
1644                          */
1645                         exit_sem(current);
1646                 }
1647
1648                 if (new_nsproxy) {
1649                         switch_task_namespaces(current, new_nsproxy);
1650                         new_nsproxy = NULL;
1651                 }
1652
1653                 task_lock(current);
1654
1655                 if (new_fs) {
1656                         fs = current->fs;
1657                         write_lock(&fs->lock);
1658                         current->fs = new_fs;
1659                         if (--fs->users)
1660                                 new_fs = NULL;
1661                         else
1662                                 new_fs = fs;
1663                         write_unlock(&fs->lock);
1664                 }
1665
1666                 if (new_mm) {
1667                         mm = current->mm;
1668                         active_mm = current->active_mm;
1669                         current->mm = new_mm;
1670                         current->active_mm = new_mm;
1671                         activate_mm(active_mm, new_mm);
1672                         new_mm = mm;
1673                 }
1674
1675                 if (new_fd) {
1676                         fd = current->files;
1677                         current->files = new_fd;
1678                         new_fd = fd;
1679                 }
1680
1681                 task_unlock(current);
1682         }
1683
1684         if (new_nsproxy)
1685                 put_nsproxy(new_nsproxy);
1686
1687 bad_unshare_cleanup_fd:
1688         if (new_fd)
1689                 put_files_struct(new_fd);
1690
1691 bad_unshare_cleanup_vm:
1692         if (new_mm)
1693                 mmput(new_mm);
1694
1695 bad_unshare_cleanup_sigh:
1696         if (new_sigh)
1697                 if (atomic_dec_and_test(&new_sigh->count))
1698                         kmem_cache_free(sighand_cachep, new_sigh);
1699
1700 bad_unshare_cleanup_fs:
1701         if (new_fs)
1702                 free_fs_struct(new_fs);
1703
1704 bad_unshare_cleanup_thread:
1705 bad_unshare_out:
1706         return err;
1707 }
1708
1709 /*
1710  *      Helper to unshare the files of the current task.
1711  *      We don't want to expose copy_files internals to
1712  *      the exec layer of the kernel.
1713  */
1714
1715 int unshare_files(struct files_struct **displaced)
1716 {
1717         struct task_struct *task = current;
1718         struct files_struct *copy = NULL;
1719         int error;
1720
1721         error = unshare_fd(CLONE_FILES, &copy);
1722         if (error || !copy) {
1723                 *displaced = NULL;
1724                 return error;
1725         }
1726         *displaced = task->files;
1727         task_lock(task);
1728         task->files = copy;
1729         task_unlock(task);
1730         return 0;
1731 }