Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux...
[linux-3.10.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/futex.h>
41 #include <linux/compat.h>
42 #include <linux/kthread.h>
43 #include <linux/task_io_accounting_ops.h>
44 #include <linux/rcupdate.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/audit.h>
48 #include <linux/memcontrol.h>
49 #include <linux/ftrace.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/ksm.h>
53 #include <linux/acct.h>
54 #include <linux/tsacct_kern.h>
55 #include <linux/cn_proc.h>
56 #include <linux/freezer.h>
57 #include <linux/delayacct.h>
58 #include <linux/taskstats_kern.h>
59 #include <linux/random.h>
60 #include <linux/tty.h>
61 #include <linux/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_event.h>
65 #include <linux/posix-timers.h>
66 #include <linux/user-return-notifier.h>
67 #include <linux/oom.h>
68 #include <linux/khugepaged.h>
69 #include <linux/signalfd.h>
70
71 #include <asm/pgtable.h>
72 #include <asm/pgalloc.h>
73 #include <asm/uaccess.h>
74 #include <asm/mmu_context.h>
75 #include <asm/cacheflush.h>
76 #include <asm/tlbflush.h>
77
78 #include <trace/events/sched.h>
79
80 #define CREATE_TRACE_POINTS
81 #include <trace/events/task.h>
82
83 /*
84  * Protected counters by write_lock_irq(&tasklist_lock)
85  */
86 unsigned long total_forks;      /* Handle normal Linux uptimes. */
87 int nr_threads;                 /* The idle threads do not count.. */
88
89 int max_threads;                /* tunable limit on nr_threads */
90
91 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
92
93 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
94
95 #ifdef CONFIG_PROVE_RCU
96 int lockdep_tasklist_lock_is_held(void)
97 {
98         return lockdep_is_held(&tasklist_lock);
99 }
100 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
101 #endif /* #ifdef CONFIG_PROVE_RCU */
102
103 int nr_processes(void)
104 {
105         int cpu;
106         int total = 0;
107
108         for_each_possible_cpu(cpu)
109                 total += per_cpu(process_counts, cpu);
110
111         return total;
112 }
113
114 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
115 # define alloc_task_struct_node(node)           \
116                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
117 # define free_task_struct(tsk)                  \
118                 kmem_cache_free(task_struct_cachep, (tsk))
119 static struct kmem_cache *task_struct_cachep;
120 #endif
121
122 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
123 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
124                                                   int node)
125 {
126 #ifdef CONFIG_DEBUG_STACK_USAGE
127         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
128 #else
129         gfp_t mask = GFP_KERNEL;
130 #endif
131         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
132
133         return page ? page_address(page) : NULL;
134 }
135
136 static inline void free_thread_info(struct thread_info *ti)
137 {
138         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
139 }
140 #endif
141
142 /* SLAB cache for signal_struct structures (tsk->signal) */
143 static struct kmem_cache *signal_cachep;
144
145 /* SLAB cache for sighand_struct structures (tsk->sighand) */
146 struct kmem_cache *sighand_cachep;
147
148 /* SLAB cache for files_struct structures (tsk->files) */
149 struct kmem_cache *files_cachep;
150
151 /* SLAB cache for fs_struct structures (tsk->fs) */
152 struct kmem_cache *fs_cachep;
153
154 /* SLAB cache for vm_area_struct structures */
155 struct kmem_cache *vm_area_cachep;
156
157 /* SLAB cache for mm_struct structures (tsk->mm) */
158 static struct kmem_cache *mm_cachep;
159
160 static void account_kernel_stack(struct thread_info *ti, int account)
161 {
162         struct zone *zone = page_zone(virt_to_page(ti));
163
164         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
165 }
166
167 void free_task(struct task_struct *tsk)
168 {
169         account_kernel_stack(tsk->stack, -1);
170         free_thread_info(tsk->stack);
171         rt_mutex_debug_task_free(tsk);
172         ftrace_graph_exit_task(tsk);
173         free_task_struct(tsk);
174 }
175 EXPORT_SYMBOL(free_task);
176
177 static inline void free_signal_struct(struct signal_struct *sig)
178 {
179         taskstats_tgid_free(sig);
180         sched_autogroup_exit(sig);
181         kmem_cache_free(signal_cachep, sig);
182 }
183
184 static inline void put_signal_struct(struct signal_struct *sig)
185 {
186         if (atomic_dec_and_test(&sig->sigcnt))
187                 free_signal_struct(sig);
188 }
189
190 void __put_task_struct(struct task_struct *tsk)
191 {
192         WARN_ON(!tsk->exit_state);
193         WARN_ON(atomic_read(&tsk->usage));
194         WARN_ON(tsk == current);
195
196         security_task_free(tsk);
197         exit_creds(tsk);
198         delayacct_tsk_free(tsk);
199         put_signal_struct(tsk->signal);
200
201         if (!profile_handoff_task(tsk))
202                 free_task(tsk);
203 }
204 EXPORT_SYMBOL_GPL(__put_task_struct);
205
206 /*
207  * macro override instead of weak attribute alias, to workaround
208  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
209  */
210 #ifndef arch_task_cache_init
211 #define arch_task_cache_init()
212 #endif
213
214 void __init fork_init(unsigned long mempages)
215 {
216 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
217 #ifndef ARCH_MIN_TASKALIGN
218 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
219 #endif
220         /* create a slab on which task_structs can be allocated */
221         task_struct_cachep =
222                 kmem_cache_create("task_struct", sizeof(struct task_struct),
223                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
224 #endif
225
226         /* do the arch specific task caches init */
227         arch_task_cache_init();
228
229         /*
230          * The default maximum number of threads is set to a safe
231          * value: the thread structures can take up at most half
232          * of memory.
233          */
234         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
235
236         /*
237          * we need to allow at least 20 threads to boot a system
238          */
239         if (max_threads < 20)
240                 max_threads = 20;
241
242         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
243         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
244         init_task.signal->rlim[RLIMIT_SIGPENDING] =
245                 init_task.signal->rlim[RLIMIT_NPROC];
246 }
247
248 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
249                                                struct task_struct *src)
250 {
251         *dst = *src;
252         return 0;
253 }
254
255 static struct task_struct *dup_task_struct(struct task_struct *orig)
256 {
257         struct task_struct *tsk;
258         struct thread_info *ti;
259         unsigned long *stackend;
260         int node = tsk_fork_get_node(orig);
261         int err;
262
263         prepare_to_copy(orig);
264
265         tsk = alloc_task_struct_node(node);
266         if (!tsk)
267                 return NULL;
268
269         ti = alloc_thread_info_node(tsk, node);
270         if (!ti) {
271                 free_task_struct(tsk);
272                 return NULL;
273         }
274
275         err = arch_dup_task_struct(tsk, orig);
276         if (err)
277                 goto out;
278
279         tsk->stack = ti;
280
281         setup_thread_stack(tsk, orig);
282         clear_user_return_notifier(tsk);
283         clear_tsk_need_resched(tsk);
284         stackend = end_of_stack(tsk);
285         *stackend = STACK_END_MAGIC;    /* for overflow detection */
286
287 #ifdef CONFIG_CC_STACKPROTECTOR
288         tsk->stack_canary = get_random_int();
289 #endif
290
291         /*
292          * One for us, one for whoever does the "release_task()" (usually
293          * parent)
294          */
295         atomic_set(&tsk->usage, 2);
296 #ifdef CONFIG_BLK_DEV_IO_TRACE
297         tsk->btrace_seq = 0;
298 #endif
299         tsk->splice_pipe = NULL;
300
301         account_kernel_stack(ti, 1);
302
303         return tsk;
304
305 out:
306         free_thread_info(ti);
307         free_task_struct(tsk);
308         return NULL;
309 }
310
311 #ifdef CONFIG_MMU
312 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
313 {
314         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
315         struct rb_node **rb_link, *rb_parent;
316         int retval;
317         unsigned long charge;
318         struct mempolicy *pol;
319
320         down_write(&oldmm->mmap_sem);
321         flush_cache_dup_mm(oldmm);
322         /*
323          * Not linked in yet - no deadlock potential:
324          */
325         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
326
327         mm->locked_vm = 0;
328         mm->mmap = NULL;
329         mm->mmap_cache = NULL;
330         mm->free_area_cache = oldmm->mmap_base;
331         mm->cached_hole_size = ~0UL;
332         mm->map_count = 0;
333         cpumask_clear(mm_cpumask(mm));
334         mm->mm_rb = RB_ROOT;
335         rb_link = &mm->mm_rb.rb_node;
336         rb_parent = NULL;
337         pprev = &mm->mmap;
338         retval = ksm_fork(mm, oldmm);
339         if (retval)
340                 goto out;
341         retval = khugepaged_fork(mm, oldmm);
342         if (retval)
343                 goto out;
344
345         prev = NULL;
346         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
347                 struct file *file;
348
349                 if (mpnt->vm_flags & VM_DONTCOPY) {
350                         long pages = vma_pages(mpnt);
351                         mm->total_vm -= pages;
352                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
353                                                                 -pages);
354                         continue;
355                 }
356                 charge = 0;
357                 if (mpnt->vm_flags & VM_ACCOUNT) {
358                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
359                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
360                                 goto fail_nomem;
361                         charge = len;
362                 }
363                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
364                 if (!tmp)
365                         goto fail_nomem;
366                 *tmp = *mpnt;
367                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
368                 pol = mpol_dup(vma_policy(mpnt));
369                 retval = PTR_ERR(pol);
370                 if (IS_ERR(pol))
371                         goto fail_nomem_policy;
372                 vma_set_policy(tmp, pol);
373                 tmp->vm_mm = mm;
374                 if (anon_vma_fork(tmp, mpnt))
375                         goto fail_nomem_anon_vma_fork;
376                 tmp->vm_flags &= ~VM_LOCKED;
377                 tmp->vm_next = tmp->vm_prev = NULL;
378                 file = tmp->vm_file;
379                 if (file) {
380                         struct inode *inode = file->f_path.dentry->d_inode;
381                         struct address_space *mapping = file->f_mapping;
382
383                         get_file(file);
384                         if (tmp->vm_flags & VM_DENYWRITE)
385                                 atomic_dec(&inode->i_writecount);
386                         mutex_lock(&mapping->i_mmap_mutex);
387                         if (tmp->vm_flags & VM_SHARED)
388                                 mapping->i_mmap_writable++;
389                         flush_dcache_mmap_lock(mapping);
390                         /* insert tmp into the share list, just after mpnt */
391                         vma_prio_tree_add(tmp, mpnt);
392                         flush_dcache_mmap_unlock(mapping);
393                         mutex_unlock(&mapping->i_mmap_mutex);
394                 }
395
396                 /*
397                  * Clear hugetlb-related page reserves for children. This only
398                  * affects MAP_PRIVATE mappings. Faults generated by the child
399                  * are not guaranteed to succeed, even if read-only
400                  */
401                 if (is_vm_hugetlb_page(tmp))
402                         reset_vma_resv_huge_pages(tmp);
403
404                 /*
405                  * Link in the new vma and copy the page table entries.
406                  */
407                 *pprev = tmp;
408                 pprev = &tmp->vm_next;
409                 tmp->vm_prev = prev;
410                 prev = tmp;
411
412                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
413                 rb_link = &tmp->vm_rb.rb_right;
414                 rb_parent = &tmp->vm_rb;
415
416                 mm->map_count++;
417                 retval = copy_page_range(mm, oldmm, mpnt);
418
419                 if (tmp->vm_ops && tmp->vm_ops->open)
420                         tmp->vm_ops->open(tmp);
421
422                 if (retval)
423                         goto out;
424         }
425         /* a new mm has just been created */
426         arch_dup_mmap(oldmm, mm);
427         retval = 0;
428 out:
429         up_write(&mm->mmap_sem);
430         flush_tlb_mm(oldmm);
431         up_write(&oldmm->mmap_sem);
432         return retval;
433 fail_nomem_anon_vma_fork:
434         mpol_put(pol);
435 fail_nomem_policy:
436         kmem_cache_free(vm_area_cachep, tmp);
437 fail_nomem:
438         retval = -ENOMEM;
439         vm_unacct_memory(charge);
440         goto out;
441 }
442
443 static inline int mm_alloc_pgd(struct mm_struct *mm)
444 {
445         mm->pgd = pgd_alloc(mm);
446         if (unlikely(!mm->pgd))
447                 return -ENOMEM;
448         return 0;
449 }
450
451 static inline void mm_free_pgd(struct mm_struct *mm)
452 {
453         pgd_free(mm, mm->pgd);
454 }
455 #else
456 #define dup_mmap(mm, oldmm)     (0)
457 #define mm_alloc_pgd(mm)        (0)
458 #define mm_free_pgd(mm)
459 #endif /* CONFIG_MMU */
460
461 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
462
463 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
464 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
465
466 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
467
468 static int __init coredump_filter_setup(char *s)
469 {
470         default_dump_filter =
471                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
472                 MMF_DUMP_FILTER_MASK;
473         return 1;
474 }
475
476 __setup("coredump_filter=", coredump_filter_setup);
477
478 #include <linux/init_task.h>
479
480 static void mm_init_aio(struct mm_struct *mm)
481 {
482 #ifdef CONFIG_AIO
483         spin_lock_init(&mm->ioctx_lock);
484         INIT_HLIST_HEAD(&mm->ioctx_list);
485 #endif
486 }
487
488 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
489 {
490         atomic_set(&mm->mm_users, 1);
491         atomic_set(&mm->mm_count, 1);
492         init_rwsem(&mm->mmap_sem);
493         INIT_LIST_HEAD(&mm->mmlist);
494         mm->flags = (current->mm) ?
495                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
496         mm->core_state = NULL;
497         mm->nr_ptes = 0;
498         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
499         spin_lock_init(&mm->page_table_lock);
500         mm->free_area_cache = TASK_UNMAPPED_BASE;
501         mm->cached_hole_size = ~0UL;
502         mm_init_aio(mm);
503         mm_init_owner(mm, p);
504
505         if (likely(!mm_alloc_pgd(mm))) {
506                 mm->def_flags = 0;
507                 mmu_notifier_mm_init(mm);
508                 return mm;
509         }
510
511         free_mm(mm);
512         return NULL;
513 }
514
515 /*
516  * Allocate and initialize an mm_struct.
517  */
518 struct mm_struct *mm_alloc(void)
519 {
520         struct mm_struct *mm;
521
522         mm = allocate_mm();
523         if (!mm)
524                 return NULL;
525
526         memset(mm, 0, sizeof(*mm));
527         mm_init_cpumask(mm);
528         return mm_init(mm, current);
529 }
530
531 /*
532  * Called when the last reference to the mm
533  * is dropped: either by a lazy thread or by
534  * mmput. Free the page directory and the mm.
535  */
536 void __mmdrop(struct mm_struct *mm)
537 {
538         BUG_ON(mm == &init_mm);
539         mm_free_pgd(mm);
540         destroy_context(mm);
541         mmu_notifier_mm_destroy(mm);
542 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
543         VM_BUG_ON(mm->pmd_huge_pte);
544 #endif
545         free_mm(mm);
546 }
547 EXPORT_SYMBOL_GPL(__mmdrop);
548
549 /*
550  * Decrement the use count and release all resources for an mm.
551  */
552 void mmput(struct mm_struct *mm)
553 {
554         might_sleep();
555
556         if (atomic_dec_and_test(&mm->mm_users)) {
557                 exit_aio(mm);
558                 ksm_exit(mm);
559                 khugepaged_exit(mm); /* must run before exit_mmap */
560                 exit_mmap(mm);
561                 set_mm_exe_file(mm, NULL);
562                 if (!list_empty(&mm->mmlist)) {
563                         spin_lock(&mmlist_lock);
564                         list_del(&mm->mmlist);
565                         spin_unlock(&mmlist_lock);
566                 }
567                 put_swap_token(mm);
568                 if (mm->binfmt)
569                         module_put(mm->binfmt->module);
570                 mmdrop(mm);
571         }
572 }
573 EXPORT_SYMBOL_GPL(mmput);
574
575 /*
576  * We added or removed a vma mapping the executable. The vmas are only mapped
577  * during exec and are not mapped with the mmap system call.
578  * Callers must hold down_write() on the mm's mmap_sem for these
579  */
580 void added_exe_file_vma(struct mm_struct *mm)
581 {
582         mm->num_exe_file_vmas++;
583 }
584
585 void removed_exe_file_vma(struct mm_struct *mm)
586 {
587         mm->num_exe_file_vmas--;
588         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
589                 fput(mm->exe_file);
590                 mm->exe_file = NULL;
591         }
592
593 }
594
595 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
596 {
597         if (new_exe_file)
598                 get_file(new_exe_file);
599         if (mm->exe_file)
600                 fput(mm->exe_file);
601         mm->exe_file = new_exe_file;
602         mm->num_exe_file_vmas = 0;
603 }
604
605 struct file *get_mm_exe_file(struct mm_struct *mm)
606 {
607         struct file *exe_file;
608
609         /* We need mmap_sem to protect against races with removal of
610          * VM_EXECUTABLE vmas */
611         down_read(&mm->mmap_sem);
612         exe_file = mm->exe_file;
613         if (exe_file)
614                 get_file(exe_file);
615         up_read(&mm->mmap_sem);
616         return exe_file;
617 }
618
619 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
620 {
621         /* It's safe to write the exe_file pointer without exe_file_lock because
622          * this is called during fork when the task is not yet in /proc */
623         newmm->exe_file = get_mm_exe_file(oldmm);
624 }
625
626 /**
627  * get_task_mm - acquire a reference to the task's mm
628  *
629  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
630  * this kernel workthread has transiently adopted a user mm with use_mm,
631  * to do its AIO) is not set and if so returns a reference to it, after
632  * bumping up the use count.  User must release the mm via mmput()
633  * after use.  Typically used by /proc and ptrace.
634  */
635 struct mm_struct *get_task_mm(struct task_struct *task)
636 {
637         struct mm_struct *mm;
638
639         task_lock(task);
640         mm = task->mm;
641         if (mm) {
642                 if (task->flags & PF_KTHREAD)
643                         mm = NULL;
644                 else
645                         atomic_inc(&mm->mm_users);
646         }
647         task_unlock(task);
648         return mm;
649 }
650 EXPORT_SYMBOL_GPL(get_task_mm);
651
652 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
653 {
654         struct mm_struct *mm;
655         int err;
656
657         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
658         if (err)
659                 return ERR_PTR(err);
660
661         mm = get_task_mm(task);
662         if (mm && mm != current->mm &&
663                         !ptrace_may_access(task, mode)) {
664                 mmput(mm);
665                 mm = ERR_PTR(-EACCES);
666         }
667         mutex_unlock(&task->signal->cred_guard_mutex);
668
669         return mm;
670 }
671
672 static void complete_vfork_done(struct task_struct *tsk)
673 {
674         struct completion *vfork;
675
676         task_lock(tsk);
677         vfork = tsk->vfork_done;
678         if (likely(vfork)) {
679                 tsk->vfork_done = NULL;
680                 complete(vfork);
681         }
682         task_unlock(tsk);
683 }
684
685 static int wait_for_vfork_done(struct task_struct *child,
686                                 struct completion *vfork)
687 {
688         int killed;
689
690         freezer_do_not_count();
691         killed = wait_for_completion_killable(vfork);
692         freezer_count();
693
694         if (killed) {
695                 task_lock(child);
696                 child->vfork_done = NULL;
697                 task_unlock(child);
698         }
699
700         put_task_struct(child);
701         return killed;
702 }
703
704 /* Please note the differences between mmput and mm_release.
705  * mmput is called whenever we stop holding onto a mm_struct,
706  * error success whatever.
707  *
708  * mm_release is called after a mm_struct has been removed
709  * from the current process.
710  *
711  * This difference is important for error handling, when we
712  * only half set up a mm_struct for a new process and need to restore
713  * the old one.  Because we mmput the new mm_struct before
714  * restoring the old one. . .
715  * Eric Biederman 10 January 1998
716  */
717 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
718 {
719         /* Get rid of any futexes when releasing the mm */
720 #ifdef CONFIG_FUTEX
721         if (unlikely(tsk->robust_list)) {
722                 exit_robust_list(tsk);
723                 tsk->robust_list = NULL;
724         }
725 #ifdef CONFIG_COMPAT
726         if (unlikely(tsk->compat_robust_list)) {
727                 compat_exit_robust_list(tsk);
728                 tsk->compat_robust_list = NULL;
729         }
730 #endif
731         if (unlikely(!list_empty(&tsk->pi_state_list)))
732                 exit_pi_state_list(tsk);
733 #endif
734
735         /* Get rid of any cached register state */
736         deactivate_mm(tsk, mm);
737
738         if (tsk->vfork_done)
739                 complete_vfork_done(tsk);
740
741         /*
742          * If we're exiting normally, clear a user-space tid field if
743          * requested.  We leave this alone when dying by signal, to leave
744          * the value intact in a core dump, and to save the unnecessary
745          * trouble, say, a killed vfork parent shouldn't touch this mm.
746          * Userland only wants this done for a sys_exit.
747          */
748         if (tsk->clear_child_tid) {
749                 if (!(tsk->flags & PF_SIGNALED) &&
750                     atomic_read(&mm->mm_users) > 1) {
751                         /*
752                          * We don't check the error code - if userspace has
753                          * not set up a proper pointer then tough luck.
754                          */
755                         put_user(0, tsk->clear_child_tid);
756                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
757                                         1, NULL, NULL, 0);
758                 }
759                 tsk->clear_child_tid = NULL;
760         }
761 }
762
763 /*
764  * Allocate a new mm structure and copy contents from the
765  * mm structure of the passed in task structure.
766  */
767 struct mm_struct *dup_mm(struct task_struct *tsk)
768 {
769         struct mm_struct *mm, *oldmm = current->mm;
770         int err;
771
772         if (!oldmm)
773                 return NULL;
774
775         mm = allocate_mm();
776         if (!mm)
777                 goto fail_nomem;
778
779         memcpy(mm, oldmm, sizeof(*mm));
780         mm_init_cpumask(mm);
781
782         /* Initializing for Swap token stuff */
783         mm->token_priority = 0;
784         mm->last_interval = 0;
785
786 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
787         mm->pmd_huge_pte = NULL;
788 #endif
789
790         if (!mm_init(mm, tsk))
791                 goto fail_nomem;
792
793         if (init_new_context(tsk, mm))
794                 goto fail_nocontext;
795
796         dup_mm_exe_file(oldmm, mm);
797
798         err = dup_mmap(mm, oldmm);
799         if (err)
800                 goto free_pt;
801
802         mm->hiwater_rss = get_mm_rss(mm);
803         mm->hiwater_vm = mm->total_vm;
804
805         if (mm->binfmt && !try_module_get(mm->binfmt->module))
806                 goto free_pt;
807
808         return mm;
809
810 free_pt:
811         /* don't put binfmt in mmput, we haven't got module yet */
812         mm->binfmt = NULL;
813         mmput(mm);
814
815 fail_nomem:
816         return NULL;
817
818 fail_nocontext:
819         /*
820          * If init_new_context() failed, we cannot use mmput() to free the mm
821          * because it calls destroy_context()
822          */
823         mm_free_pgd(mm);
824         free_mm(mm);
825         return NULL;
826 }
827
828 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
829 {
830         struct mm_struct *mm, *oldmm;
831         int retval;
832
833         tsk->min_flt = tsk->maj_flt = 0;
834         tsk->nvcsw = tsk->nivcsw = 0;
835 #ifdef CONFIG_DETECT_HUNG_TASK
836         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
837 #endif
838
839         tsk->mm = NULL;
840         tsk->active_mm = NULL;
841
842         /*
843          * Are we cloning a kernel thread?
844          *
845          * We need to steal a active VM for that..
846          */
847         oldmm = current->mm;
848         if (!oldmm)
849                 return 0;
850
851         if (clone_flags & CLONE_VM) {
852                 atomic_inc(&oldmm->mm_users);
853                 mm = oldmm;
854                 goto good_mm;
855         }
856
857         retval = -ENOMEM;
858         mm = dup_mm(tsk);
859         if (!mm)
860                 goto fail_nomem;
861
862 good_mm:
863         /* Initializing for Swap token stuff */
864         mm->token_priority = 0;
865         mm->last_interval = 0;
866
867         tsk->mm = mm;
868         tsk->active_mm = mm;
869         return 0;
870
871 fail_nomem:
872         return retval;
873 }
874
875 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
876 {
877         struct fs_struct *fs = current->fs;
878         if (clone_flags & CLONE_FS) {
879                 /* tsk->fs is already what we want */
880                 spin_lock(&fs->lock);
881                 if (fs->in_exec) {
882                         spin_unlock(&fs->lock);
883                         return -EAGAIN;
884                 }
885                 fs->users++;
886                 spin_unlock(&fs->lock);
887                 return 0;
888         }
889         tsk->fs = copy_fs_struct(fs);
890         if (!tsk->fs)
891                 return -ENOMEM;
892         return 0;
893 }
894
895 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
896 {
897         struct files_struct *oldf, *newf;
898         int error = 0;
899
900         /*
901          * A background process may not have any files ...
902          */
903         oldf = current->files;
904         if (!oldf)
905                 goto out;
906
907         if (clone_flags & CLONE_FILES) {
908                 atomic_inc(&oldf->count);
909                 goto out;
910         }
911
912         newf = dup_fd(oldf, &error);
913         if (!newf)
914                 goto out;
915
916         tsk->files = newf;
917         error = 0;
918 out:
919         return error;
920 }
921
922 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
923 {
924 #ifdef CONFIG_BLOCK
925         struct io_context *ioc = current->io_context;
926         struct io_context *new_ioc;
927
928         if (!ioc)
929                 return 0;
930         /*
931          * Share io context with parent, if CLONE_IO is set
932          */
933         if (clone_flags & CLONE_IO) {
934                 tsk->io_context = ioc_task_link(ioc);
935                 if (unlikely(!tsk->io_context))
936                         return -ENOMEM;
937         } else if (ioprio_valid(ioc->ioprio)) {
938                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
939                 if (unlikely(!new_ioc))
940                         return -ENOMEM;
941
942                 new_ioc->ioprio = ioc->ioprio;
943                 put_io_context(new_ioc);
944         }
945 #endif
946         return 0;
947 }
948
949 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
950 {
951         struct sighand_struct *sig;
952
953         if (clone_flags & CLONE_SIGHAND) {
954                 atomic_inc(&current->sighand->count);
955                 return 0;
956         }
957         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
958         rcu_assign_pointer(tsk->sighand, sig);
959         if (!sig)
960                 return -ENOMEM;
961         atomic_set(&sig->count, 1);
962         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
963         return 0;
964 }
965
966 void __cleanup_sighand(struct sighand_struct *sighand)
967 {
968         if (atomic_dec_and_test(&sighand->count)) {
969                 signalfd_cleanup(sighand);
970                 kmem_cache_free(sighand_cachep, sighand);
971         }
972 }
973
974
975 /*
976  * Initialize POSIX timer handling for a thread group.
977  */
978 static void posix_cpu_timers_init_group(struct signal_struct *sig)
979 {
980         unsigned long cpu_limit;
981
982         /* Thread group counters. */
983         thread_group_cputime_init(sig);
984
985         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
986         if (cpu_limit != RLIM_INFINITY) {
987                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
988                 sig->cputimer.running = 1;
989         }
990
991         /* The timer lists. */
992         INIT_LIST_HEAD(&sig->cpu_timers[0]);
993         INIT_LIST_HEAD(&sig->cpu_timers[1]);
994         INIT_LIST_HEAD(&sig->cpu_timers[2]);
995 }
996
997 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
998 {
999         struct signal_struct *sig;
1000
1001         if (clone_flags & CLONE_THREAD)
1002                 return 0;
1003
1004         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1005         tsk->signal = sig;
1006         if (!sig)
1007                 return -ENOMEM;
1008
1009         sig->nr_threads = 1;
1010         atomic_set(&sig->live, 1);
1011         atomic_set(&sig->sigcnt, 1);
1012         init_waitqueue_head(&sig->wait_chldexit);
1013         if (clone_flags & CLONE_NEWPID)
1014                 sig->flags |= SIGNAL_UNKILLABLE;
1015         sig->curr_target = tsk;
1016         init_sigpending(&sig->shared_pending);
1017         INIT_LIST_HEAD(&sig->posix_timers);
1018
1019         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1020         sig->real_timer.function = it_real_fn;
1021
1022         task_lock(current->group_leader);
1023         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1024         task_unlock(current->group_leader);
1025
1026         posix_cpu_timers_init_group(sig);
1027
1028         tty_audit_fork(sig);
1029         sched_autogroup_fork(sig);
1030
1031 #ifdef CONFIG_CGROUPS
1032         init_rwsem(&sig->group_rwsem);
1033 #endif
1034
1035         sig->oom_adj = current->signal->oom_adj;
1036         sig->oom_score_adj = current->signal->oom_score_adj;
1037         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1038
1039         mutex_init(&sig->cred_guard_mutex);
1040
1041         return 0;
1042 }
1043
1044 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1045 {
1046         unsigned long new_flags = p->flags;
1047
1048         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1049         new_flags |= PF_FORKNOEXEC;
1050         p->flags = new_flags;
1051 }
1052
1053 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1054 {
1055         current->clear_child_tid = tidptr;
1056
1057         return task_pid_vnr(current);
1058 }
1059
1060 static void rt_mutex_init_task(struct task_struct *p)
1061 {
1062         raw_spin_lock_init(&p->pi_lock);
1063 #ifdef CONFIG_RT_MUTEXES
1064         plist_head_init(&p->pi_waiters);
1065         p->pi_blocked_on = NULL;
1066 #endif
1067 }
1068
1069 #ifdef CONFIG_MM_OWNER
1070 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1071 {
1072         mm->owner = p;
1073 }
1074 #endif /* CONFIG_MM_OWNER */
1075
1076 /*
1077  * Initialize POSIX timer handling for a single task.
1078  */
1079 static void posix_cpu_timers_init(struct task_struct *tsk)
1080 {
1081         tsk->cputime_expires.prof_exp = 0;
1082         tsk->cputime_expires.virt_exp = 0;
1083         tsk->cputime_expires.sched_exp = 0;
1084         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1085         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1086         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1087 }
1088
1089 /*
1090  * This creates a new process as a copy of the old one,
1091  * but does not actually start it yet.
1092  *
1093  * It copies the registers, and all the appropriate
1094  * parts of the process environment (as per the clone
1095  * flags). The actual kick-off is left to the caller.
1096  */
1097 static struct task_struct *copy_process(unsigned long clone_flags,
1098                                         unsigned long stack_start,
1099                                         struct pt_regs *regs,
1100                                         unsigned long stack_size,
1101                                         int __user *child_tidptr,
1102                                         struct pid *pid,
1103                                         int trace)
1104 {
1105         int retval;
1106         struct task_struct *p;
1107         int cgroup_callbacks_done = 0;
1108
1109         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1110                 return ERR_PTR(-EINVAL);
1111
1112         /*
1113          * Thread groups must share signals as well, and detached threads
1114          * can only be started up within the thread group.
1115          */
1116         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1117                 return ERR_PTR(-EINVAL);
1118
1119         /*
1120          * Shared signal handlers imply shared VM. By way of the above,
1121          * thread groups also imply shared VM. Blocking this case allows
1122          * for various simplifications in other code.
1123          */
1124         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1125                 return ERR_PTR(-EINVAL);
1126
1127         /*
1128          * Siblings of global init remain as zombies on exit since they are
1129          * not reaped by their parent (swapper). To solve this and to avoid
1130          * multi-rooted process trees, prevent global and container-inits
1131          * from creating siblings.
1132          */
1133         if ((clone_flags & CLONE_PARENT) &&
1134                                 current->signal->flags & SIGNAL_UNKILLABLE)
1135                 return ERR_PTR(-EINVAL);
1136
1137         retval = security_task_create(clone_flags);
1138         if (retval)
1139                 goto fork_out;
1140
1141         retval = -ENOMEM;
1142         p = dup_task_struct(current);
1143         if (!p)
1144                 goto fork_out;
1145
1146         ftrace_graph_init_task(p);
1147
1148         rt_mutex_init_task(p);
1149
1150 #ifdef CONFIG_PROVE_LOCKING
1151         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1152         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1153 #endif
1154         retval = -EAGAIN;
1155         if (atomic_read(&p->real_cred->user->processes) >=
1156                         task_rlimit(p, RLIMIT_NPROC)) {
1157                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1158                     p->real_cred->user != INIT_USER)
1159                         goto bad_fork_free;
1160         }
1161         current->flags &= ~PF_NPROC_EXCEEDED;
1162
1163         retval = copy_creds(p, clone_flags);
1164         if (retval < 0)
1165                 goto bad_fork_free;
1166
1167         /*
1168          * If multiple threads are within copy_process(), then this check
1169          * triggers too late. This doesn't hurt, the check is only there
1170          * to stop root fork bombs.
1171          */
1172         retval = -EAGAIN;
1173         if (nr_threads >= max_threads)
1174                 goto bad_fork_cleanup_count;
1175
1176         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1177                 goto bad_fork_cleanup_count;
1178
1179         p->did_exec = 0;
1180         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1181         copy_flags(clone_flags, p);
1182         INIT_LIST_HEAD(&p->children);
1183         INIT_LIST_HEAD(&p->sibling);
1184         rcu_copy_process(p);
1185         p->vfork_done = NULL;
1186         spin_lock_init(&p->alloc_lock);
1187
1188         init_sigpending(&p->pending);
1189
1190         p->utime = p->stime = p->gtime = 0;
1191         p->utimescaled = p->stimescaled = 0;
1192 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1193         p->prev_utime = p->prev_stime = 0;
1194 #endif
1195 #if defined(SPLIT_RSS_COUNTING)
1196         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1197 #endif
1198
1199         p->default_timer_slack_ns = current->timer_slack_ns;
1200
1201         task_io_accounting_init(&p->ioac);
1202         acct_clear_integrals(p);
1203
1204         posix_cpu_timers_init(p);
1205
1206         do_posix_clock_monotonic_gettime(&p->start_time);
1207         p->real_start_time = p->start_time;
1208         monotonic_to_bootbased(&p->real_start_time);
1209         p->io_context = NULL;
1210         p->audit_context = NULL;
1211         if (clone_flags & CLONE_THREAD)
1212                 threadgroup_change_begin(current);
1213         cgroup_fork(p);
1214 #ifdef CONFIG_NUMA
1215         p->mempolicy = mpol_dup(p->mempolicy);
1216         if (IS_ERR(p->mempolicy)) {
1217                 retval = PTR_ERR(p->mempolicy);
1218                 p->mempolicy = NULL;
1219                 goto bad_fork_cleanup_cgroup;
1220         }
1221         mpol_fix_fork_child_flag(p);
1222 #endif
1223 #ifdef CONFIG_CPUSETS
1224         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1225         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1226 #endif
1227 #ifdef CONFIG_TRACE_IRQFLAGS
1228         p->irq_events = 0;
1229 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1230         p->hardirqs_enabled = 1;
1231 #else
1232         p->hardirqs_enabled = 0;
1233 #endif
1234         p->hardirq_enable_ip = 0;
1235         p->hardirq_enable_event = 0;
1236         p->hardirq_disable_ip = _THIS_IP_;
1237         p->hardirq_disable_event = 0;
1238         p->softirqs_enabled = 1;
1239         p->softirq_enable_ip = _THIS_IP_;
1240         p->softirq_enable_event = 0;
1241         p->softirq_disable_ip = 0;
1242         p->softirq_disable_event = 0;
1243         p->hardirq_context = 0;
1244         p->softirq_context = 0;
1245 #endif
1246 #ifdef CONFIG_LOCKDEP
1247         p->lockdep_depth = 0; /* no locks held yet */
1248         p->curr_chain_key = 0;
1249         p->lockdep_recursion = 0;
1250 #endif
1251
1252 #ifdef CONFIG_DEBUG_MUTEXES
1253         p->blocked_on = NULL; /* not blocked yet */
1254 #endif
1255 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1256         p->memcg_batch.do_batch = 0;
1257         p->memcg_batch.memcg = NULL;
1258 #endif
1259
1260         /* Perform scheduler related setup. Assign this task to a CPU. */
1261         sched_fork(p);
1262
1263         retval = perf_event_init_task(p);
1264         if (retval)
1265                 goto bad_fork_cleanup_policy;
1266         retval = audit_alloc(p);
1267         if (retval)
1268                 goto bad_fork_cleanup_policy;
1269         /* copy all the process information */
1270         retval = copy_semundo(clone_flags, p);
1271         if (retval)
1272                 goto bad_fork_cleanup_audit;
1273         retval = copy_files(clone_flags, p);
1274         if (retval)
1275                 goto bad_fork_cleanup_semundo;
1276         retval = copy_fs(clone_flags, p);
1277         if (retval)
1278                 goto bad_fork_cleanup_files;
1279         retval = copy_sighand(clone_flags, p);
1280         if (retval)
1281                 goto bad_fork_cleanup_fs;
1282         retval = copy_signal(clone_flags, p);
1283         if (retval)
1284                 goto bad_fork_cleanup_sighand;
1285         retval = copy_mm(clone_flags, p);
1286         if (retval)
1287                 goto bad_fork_cleanup_signal;
1288         retval = copy_namespaces(clone_flags, p);
1289         if (retval)
1290                 goto bad_fork_cleanup_mm;
1291         retval = copy_io(clone_flags, p);
1292         if (retval)
1293                 goto bad_fork_cleanup_namespaces;
1294         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1295         if (retval)
1296                 goto bad_fork_cleanup_io;
1297
1298         if (pid != &init_struct_pid) {
1299                 retval = -ENOMEM;
1300                 pid = alloc_pid(p->nsproxy->pid_ns);
1301                 if (!pid)
1302                         goto bad_fork_cleanup_io;
1303         }
1304
1305         p->pid = pid_nr(pid);
1306         p->tgid = p->pid;
1307         if (clone_flags & CLONE_THREAD)
1308                 p->tgid = current->tgid;
1309
1310         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1311         /*
1312          * Clear TID on mm_release()?
1313          */
1314         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1315 #ifdef CONFIG_BLOCK
1316         p->plug = NULL;
1317 #endif
1318 #ifdef CONFIG_FUTEX
1319         p->robust_list = NULL;
1320 #ifdef CONFIG_COMPAT
1321         p->compat_robust_list = NULL;
1322 #endif
1323         INIT_LIST_HEAD(&p->pi_state_list);
1324         p->pi_state_cache = NULL;
1325 #endif
1326         /*
1327          * sigaltstack should be cleared when sharing the same VM
1328          */
1329         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1330                 p->sas_ss_sp = p->sas_ss_size = 0;
1331
1332         /*
1333          * Syscall tracing and stepping should be turned off in the
1334          * child regardless of CLONE_PTRACE.
1335          */
1336         user_disable_single_step(p);
1337         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1338 #ifdef TIF_SYSCALL_EMU
1339         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1340 #endif
1341         clear_all_latency_tracing(p);
1342
1343         /* ok, now we should be set up.. */
1344         if (clone_flags & CLONE_THREAD)
1345                 p->exit_signal = -1;
1346         else if (clone_flags & CLONE_PARENT)
1347                 p->exit_signal = current->group_leader->exit_signal;
1348         else
1349                 p->exit_signal = (clone_flags & CSIGNAL);
1350
1351         p->pdeath_signal = 0;
1352         p->exit_state = 0;
1353
1354         p->nr_dirtied = 0;
1355         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1356         p->dirty_paused_when = 0;
1357
1358         /*
1359          * Ok, make it visible to the rest of the system.
1360          * We dont wake it up yet.
1361          */
1362         p->group_leader = p;
1363         INIT_LIST_HEAD(&p->thread_group);
1364
1365         /* Now that the task is set up, run cgroup callbacks if
1366          * necessary. We need to run them before the task is visible
1367          * on the tasklist. */
1368         cgroup_fork_callbacks(p);
1369         cgroup_callbacks_done = 1;
1370
1371         /* Need tasklist lock for parent etc handling! */
1372         write_lock_irq(&tasklist_lock);
1373
1374         /* CLONE_PARENT re-uses the old parent */
1375         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1376                 p->real_parent = current->real_parent;
1377                 p->parent_exec_id = current->parent_exec_id;
1378         } else {
1379                 p->real_parent = current;
1380                 p->parent_exec_id = current->self_exec_id;
1381         }
1382
1383         spin_lock(&current->sighand->siglock);
1384
1385         /*
1386          * Process group and session signals need to be delivered to just the
1387          * parent before the fork or both the parent and the child after the
1388          * fork. Restart if a signal comes in before we add the new process to
1389          * it's process group.
1390          * A fatal signal pending means that current will exit, so the new
1391          * thread can't slip out of an OOM kill (or normal SIGKILL).
1392         */
1393         recalc_sigpending();
1394         if (signal_pending(current)) {
1395                 spin_unlock(&current->sighand->siglock);
1396                 write_unlock_irq(&tasklist_lock);
1397                 retval = -ERESTARTNOINTR;
1398                 goto bad_fork_free_pid;
1399         }
1400
1401         if (clone_flags & CLONE_THREAD) {
1402                 current->signal->nr_threads++;
1403                 atomic_inc(&current->signal->live);
1404                 atomic_inc(&current->signal->sigcnt);
1405                 p->group_leader = current->group_leader;
1406                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1407         }
1408
1409         if (likely(p->pid)) {
1410                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1411
1412                 if (thread_group_leader(p)) {
1413                         if (is_child_reaper(pid))
1414                                 p->nsproxy->pid_ns->child_reaper = p;
1415
1416                         p->signal->leader_pid = pid;
1417                         p->signal->tty = tty_kref_get(current->signal->tty);
1418                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1419                         attach_pid(p, PIDTYPE_SID, task_session(current));
1420                         list_add_tail(&p->sibling, &p->real_parent->children);
1421                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1422                         __this_cpu_inc(process_counts);
1423                 }
1424                 attach_pid(p, PIDTYPE_PID, pid);
1425                 nr_threads++;
1426         }
1427
1428         total_forks++;
1429         spin_unlock(&current->sighand->siglock);
1430         write_unlock_irq(&tasklist_lock);
1431         proc_fork_connector(p);
1432         cgroup_post_fork(p);
1433         if (clone_flags & CLONE_THREAD)
1434                 threadgroup_change_end(current);
1435         perf_event_fork(p);
1436
1437         trace_task_newtask(p, clone_flags);
1438
1439         return p;
1440
1441 bad_fork_free_pid:
1442         if (pid != &init_struct_pid)
1443                 free_pid(pid);
1444 bad_fork_cleanup_io:
1445         if (p->io_context)
1446                 exit_io_context(p);
1447 bad_fork_cleanup_namespaces:
1448         exit_task_namespaces(p);
1449 bad_fork_cleanup_mm:
1450         if (p->mm)
1451                 mmput(p->mm);
1452 bad_fork_cleanup_signal:
1453         if (!(clone_flags & CLONE_THREAD))
1454                 free_signal_struct(p->signal);
1455 bad_fork_cleanup_sighand:
1456         __cleanup_sighand(p->sighand);
1457 bad_fork_cleanup_fs:
1458         exit_fs(p); /* blocking */
1459 bad_fork_cleanup_files:
1460         exit_files(p); /* blocking */
1461 bad_fork_cleanup_semundo:
1462         exit_sem(p);
1463 bad_fork_cleanup_audit:
1464         audit_free(p);
1465 bad_fork_cleanup_policy:
1466         perf_event_free_task(p);
1467 #ifdef CONFIG_NUMA
1468         mpol_put(p->mempolicy);
1469 bad_fork_cleanup_cgroup:
1470 #endif
1471         if (clone_flags & CLONE_THREAD)
1472                 threadgroup_change_end(current);
1473         cgroup_exit(p, cgroup_callbacks_done);
1474         delayacct_tsk_free(p);
1475         module_put(task_thread_info(p)->exec_domain->module);
1476 bad_fork_cleanup_count:
1477         atomic_dec(&p->cred->user->processes);
1478         exit_creds(p);
1479 bad_fork_free:
1480         free_task(p);
1481 fork_out:
1482         return ERR_PTR(retval);
1483 }
1484
1485 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1486 {
1487         memset(regs, 0, sizeof(struct pt_regs));
1488         return regs;
1489 }
1490
1491 static inline void init_idle_pids(struct pid_link *links)
1492 {
1493         enum pid_type type;
1494
1495         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1496                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1497                 links[type].pid = &init_struct_pid;
1498         }
1499 }
1500
1501 struct task_struct * __cpuinit fork_idle(int cpu)
1502 {
1503         struct task_struct *task;
1504         struct pt_regs regs;
1505
1506         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1507                             &init_struct_pid, 0);
1508         if (!IS_ERR(task)) {
1509                 init_idle_pids(task->pids);
1510                 init_idle(task, cpu);
1511         }
1512
1513         return task;
1514 }
1515
1516 /*
1517  *  Ok, this is the main fork-routine.
1518  *
1519  * It copies the process, and if successful kick-starts
1520  * it and waits for it to finish using the VM if required.
1521  */
1522 long do_fork(unsigned long clone_flags,
1523               unsigned long stack_start,
1524               struct pt_regs *regs,
1525               unsigned long stack_size,
1526               int __user *parent_tidptr,
1527               int __user *child_tidptr)
1528 {
1529         struct task_struct *p;
1530         int trace = 0;
1531         long nr;
1532
1533         /*
1534          * Do some preliminary argument and permissions checking before we
1535          * actually start allocating stuff
1536          */
1537         if (clone_flags & CLONE_NEWUSER) {
1538                 if (clone_flags & CLONE_THREAD)
1539                         return -EINVAL;
1540                 /* hopefully this check will go away when userns support is
1541                  * complete
1542                  */
1543                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1544                                 !capable(CAP_SETGID))
1545                         return -EPERM;
1546         }
1547
1548         /*
1549          * Determine whether and which event to report to ptracer.  When
1550          * called from kernel_thread or CLONE_UNTRACED is explicitly
1551          * requested, no event is reported; otherwise, report if the event
1552          * for the type of forking is enabled.
1553          */
1554         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1555                 if (clone_flags & CLONE_VFORK)
1556                         trace = PTRACE_EVENT_VFORK;
1557                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1558                         trace = PTRACE_EVENT_CLONE;
1559                 else
1560                         trace = PTRACE_EVENT_FORK;
1561
1562                 if (likely(!ptrace_event_enabled(current, trace)))
1563                         trace = 0;
1564         }
1565
1566         p = copy_process(clone_flags, stack_start, regs, stack_size,
1567                          child_tidptr, NULL, trace);
1568         /*
1569          * Do this prior waking up the new thread - the thread pointer
1570          * might get invalid after that point, if the thread exits quickly.
1571          */
1572         if (!IS_ERR(p)) {
1573                 struct completion vfork;
1574
1575                 trace_sched_process_fork(current, p);
1576
1577                 nr = task_pid_vnr(p);
1578
1579                 if (clone_flags & CLONE_PARENT_SETTID)
1580                         put_user(nr, parent_tidptr);
1581
1582                 if (clone_flags & CLONE_VFORK) {
1583                         p->vfork_done = &vfork;
1584                         init_completion(&vfork);
1585                         get_task_struct(p);
1586                 }
1587
1588                 wake_up_new_task(p);
1589
1590                 /* forking complete and child started to run, tell ptracer */
1591                 if (unlikely(trace))
1592                         ptrace_event(trace, nr);
1593
1594                 if (clone_flags & CLONE_VFORK) {
1595                         if (!wait_for_vfork_done(p, &vfork))
1596                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1597                 }
1598         } else {
1599                 nr = PTR_ERR(p);
1600         }
1601         return nr;
1602 }
1603
1604 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1605 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1606 #endif
1607
1608 static void sighand_ctor(void *data)
1609 {
1610         struct sighand_struct *sighand = data;
1611
1612         spin_lock_init(&sighand->siglock);
1613         init_waitqueue_head(&sighand->signalfd_wqh);
1614 }
1615
1616 void __init proc_caches_init(void)
1617 {
1618         sighand_cachep = kmem_cache_create("sighand_cache",
1619                         sizeof(struct sighand_struct), 0,
1620                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1621                         SLAB_NOTRACK, sighand_ctor);
1622         signal_cachep = kmem_cache_create("signal_cache",
1623                         sizeof(struct signal_struct), 0,
1624                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1625         files_cachep = kmem_cache_create("files_cache",
1626                         sizeof(struct files_struct), 0,
1627                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1628         fs_cachep = kmem_cache_create("fs_cache",
1629                         sizeof(struct fs_struct), 0,
1630                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1631         /*
1632          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1633          * whole struct cpumask for the OFFSTACK case. We could change
1634          * this to *only* allocate as much of it as required by the
1635          * maximum number of CPU's we can ever have.  The cpumask_allocation
1636          * is at the end of the structure, exactly for that reason.
1637          */
1638         mm_cachep = kmem_cache_create("mm_struct",
1639                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1640                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1641         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1642         mmap_init();
1643         nsproxy_cache_init();
1644 }
1645
1646 /*
1647  * Check constraints on flags passed to the unshare system call.
1648  */
1649 static int check_unshare_flags(unsigned long unshare_flags)
1650 {
1651         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1652                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1653                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1654                 return -EINVAL;
1655         /*
1656          * Not implemented, but pretend it works if there is nothing to
1657          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1658          * needs to unshare vm.
1659          */
1660         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1661                 /* FIXME: get_task_mm() increments ->mm_users */
1662                 if (atomic_read(&current->mm->mm_users) > 1)
1663                         return -EINVAL;
1664         }
1665
1666         return 0;
1667 }
1668
1669 /*
1670  * Unshare the filesystem structure if it is being shared
1671  */
1672 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1673 {
1674         struct fs_struct *fs = current->fs;
1675
1676         if (!(unshare_flags & CLONE_FS) || !fs)
1677                 return 0;
1678
1679         /* don't need lock here; in the worst case we'll do useless copy */
1680         if (fs->users == 1)
1681                 return 0;
1682
1683         *new_fsp = copy_fs_struct(fs);
1684         if (!*new_fsp)
1685                 return -ENOMEM;
1686
1687         return 0;
1688 }
1689
1690 /*
1691  * Unshare file descriptor table if it is being shared
1692  */
1693 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1694 {
1695         struct files_struct *fd = current->files;
1696         int error = 0;
1697
1698         if ((unshare_flags & CLONE_FILES) &&
1699             (fd && atomic_read(&fd->count) > 1)) {
1700                 *new_fdp = dup_fd(fd, &error);
1701                 if (!*new_fdp)
1702                         return error;
1703         }
1704
1705         return 0;
1706 }
1707
1708 /*
1709  * unshare allows a process to 'unshare' part of the process
1710  * context which was originally shared using clone.  copy_*
1711  * functions used by do_fork() cannot be used here directly
1712  * because they modify an inactive task_struct that is being
1713  * constructed. Here we are modifying the current, active,
1714  * task_struct.
1715  */
1716 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1717 {
1718         struct fs_struct *fs, *new_fs = NULL;
1719         struct files_struct *fd, *new_fd = NULL;
1720         struct nsproxy *new_nsproxy = NULL;
1721         int do_sysvsem = 0;
1722         int err;
1723
1724         err = check_unshare_flags(unshare_flags);
1725         if (err)
1726                 goto bad_unshare_out;
1727
1728         /*
1729          * If unsharing namespace, must also unshare filesystem information.
1730          */
1731         if (unshare_flags & CLONE_NEWNS)
1732                 unshare_flags |= CLONE_FS;
1733         /*
1734          * CLONE_NEWIPC must also detach from the undolist: after switching
1735          * to a new ipc namespace, the semaphore arrays from the old
1736          * namespace are unreachable.
1737          */
1738         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1739                 do_sysvsem = 1;
1740         err = unshare_fs(unshare_flags, &new_fs);
1741         if (err)
1742                 goto bad_unshare_out;
1743         err = unshare_fd(unshare_flags, &new_fd);
1744         if (err)
1745                 goto bad_unshare_cleanup_fs;
1746         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1747         if (err)
1748                 goto bad_unshare_cleanup_fd;
1749
1750         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1751                 if (do_sysvsem) {
1752                         /*
1753                          * CLONE_SYSVSEM is equivalent to sys_exit().
1754                          */
1755                         exit_sem(current);
1756                 }
1757
1758                 if (new_nsproxy) {
1759                         switch_task_namespaces(current, new_nsproxy);
1760                         new_nsproxy = NULL;
1761                 }
1762
1763                 task_lock(current);
1764
1765                 if (new_fs) {
1766                         fs = current->fs;
1767                         spin_lock(&fs->lock);
1768                         current->fs = new_fs;
1769                         if (--fs->users)
1770                                 new_fs = NULL;
1771                         else
1772                                 new_fs = fs;
1773                         spin_unlock(&fs->lock);
1774                 }
1775
1776                 if (new_fd) {
1777                         fd = current->files;
1778                         current->files = new_fd;
1779                         new_fd = fd;
1780                 }
1781
1782                 task_unlock(current);
1783         }
1784
1785         if (new_nsproxy)
1786                 put_nsproxy(new_nsproxy);
1787
1788 bad_unshare_cleanup_fd:
1789         if (new_fd)
1790                 put_files_struct(new_fd);
1791
1792 bad_unshare_cleanup_fs:
1793         if (new_fs)
1794                 free_fs_struct(new_fs);
1795
1796 bad_unshare_out:
1797         return err;
1798 }
1799
1800 /*
1801  *      Helper to unshare the files of the current task.
1802  *      We don't want to expose copy_files internals to
1803  *      the exec layer of the kernel.
1804  */
1805
1806 int unshare_files(struct files_struct **displaced)
1807 {
1808         struct task_struct *task = current;
1809         struct files_struct *copy = NULL;
1810         int error;
1811
1812         error = unshare_fd(CLONE_FILES, &copy);
1813         if (error || !copy) {
1814                 *displaced = NULL;
1815                 return error;
1816         }
1817         *displaced = task->files;
1818         task_lock(task);
1819         task->files = copy;
1820         task_unlock(task);
1821         return 0;
1822 }