Merge tag 'stackleak-v4.20-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git...
[muen/linux.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/sched/autogroup.h>
16 #include <linux/sched/mm.h>
17 #include <linux/sched/coredump.h>
18 #include <linux/sched/user.h>
19 #include <linux/sched/numa_balancing.h>
20 #include <linux/sched/stat.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/task_stack.h>
23 #include <linux/sched/cputime.h>
24 #include <linux/rtmutex.h>
25 #include <linux/init.h>
26 #include <linux/unistd.h>
27 #include <linux/module.h>
28 #include <linux/vmalloc.h>
29 #include <linux/completion.h>
30 #include <linux/personality.h>
31 #include <linux/mempolicy.h>
32 #include <linux/sem.h>
33 #include <linux/file.h>
34 #include <linux/fdtable.h>
35 #include <linux/iocontext.h>
36 #include <linux/key.h>
37 #include <linux/binfmts.h>
38 #include <linux/mman.h>
39 #include <linux/mmu_notifier.h>
40 #include <linux/hmm.h>
41 #include <linux/fs.h>
42 #include <linux/mm.h>
43 #include <linux/vmacache.h>
44 #include <linux/nsproxy.h>
45 #include <linux/capability.h>
46 #include <linux/cpu.h>
47 #include <linux/cgroup.h>
48 #include <linux/security.h>
49 #include <linux/hugetlb.h>
50 #include <linux/seccomp.h>
51 #include <linux/swap.h>
52 #include <linux/syscalls.h>
53 #include <linux/jiffies.h>
54 #include <linux/futex.h>
55 #include <linux/compat.h>
56 #include <linux/kthread.h>
57 #include <linux/task_io_accounting_ops.h>
58 #include <linux/rcupdate.h>
59 #include <linux/ptrace.h>
60 #include <linux/mount.h>
61 #include <linux/audit.h>
62 #include <linux/memcontrol.h>
63 #include <linux/ftrace.h>
64 #include <linux/proc_fs.h>
65 #include <linux/profile.h>
66 #include <linux/rmap.h>
67 #include <linux/ksm.h>
68 #include <linux/acct.h>
69 #include <linux/userfaultfd_k.h>
70 #include <linux/tsacct_kern.h>
71 #include <linux/cn_proc.h>
72 #include <linux/freezer.h>
73 #include <linux/delayacct.h>
74 #include <linux/taskstats_kern.h>
75 #include <linux/random.h>
76 #include <linux/tty.h>
77 #include <linux/blkdev.h>
78 #include <linux/fs_struct.h>
79 #include <linux/magic.h>
80 #include <linux/sched/mm.h>
81 #include <linux/perf_event.h>
82 #include <linux/posix-timers.h>
83 #include <linux/user-return-notifier.h>
84 #include <linux/oom.h>
85 #include <linux/khugepaged.h>
86 #include <linux/signalfd.h>
87 #include <linux/uprobes.h>
88 #include <linux/aio.h>
89 #include <linux/compiler.h>
90 #include <linux/sysctl.h>
91 #include <linux/kcov.h>
92 #include <linux/livepatch.h>
93 #include <linux/thread_info.h>
94 #include <linux/stackleak.h>
95
96 #include <asm/pgtable.h>
97 #include <asm/pgalloc.h>
98 #include <linux/uaccess.h>
99 #include <asm/mmu_context.h>
100 #include <asm/cacheflush.h>
101 #include <asm/tlbflush.h>
102
103 #include <trace/events/sched.h>
104
105 #define CREATE_TRACE_POINTS
106 #include <trace/events/task.h>
107
108 /*
109  * Minimum number of threads to boot the kernel
110  */
111 #define MIN_THREADS 20
112
113 /*
114  * Maximum number of threads
115  */
116 #define MAX_THREADS FUTEX_TID_MASK
117
118 /*
119  * Protected counters by write_lock_irq(&tasklist_lock)
120  */
121 unsigned long total_forks;      /* Handle normal Linux uptimes. */
122 int nr_threads;                 /* The idle threads do not count.. */
123
124 int max_threads;                /* tunable limit on nr_threads */
125
126 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
127
128 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
129
130 #ifdef CONFIG_PROVE_RCU
131 int lockdep_tasklist_lock_is_held(void)
132 {
133         return lockdep_is_held(&tasklist_lock);
134 }
135 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
136 #endif /* #ifdef CONFIG_PROVE_RCU */
137
138 int nr_processes(void)
139 {
140         int cpu;
141         int total = 0;
142
143         for_each_possible_cpu(cpu)
144                 total += per_cpu(process_counts, cpu);
145
146         return total;
147 }
148
149 void __weak arch_release_task_struct(struct task_struct *tsk)
150 {
151 }
152
153 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
154 static struct kmem_cache *task_struct_cachep;
155
156 static inline struct task_struct *alloc_task_struct_node(int node)
157 {
158         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
159 }
160
161 static inline void free_task_struct(struct task_struct *tsk)
162 {
163         kmem_cache_free(task_struct_cachep, tsk);
164 }
165 #endif
166
167 void __weak arch_release_thread_stack(unsigned long *stack)
168 {
169 }
170
171 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
172
173 /*
174  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
175  * kmemcache based allocator.
176  */
177 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
178
179 #ifdef CONFIG_VMAP_STACK
180 /*
181  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
182  * flush.  Try to minimize the number of calls by caching stacks.
183  */
184 #define NR_CACHED_STACKS 2
185 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
186
187 static int free_vm_stack_cache(unsigned int cpu)
188 {
189         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
190         int i;
191
192         for (i = 0; i < NR_CACHED_STACKS; i++) {
193                 struct vm_struct *vm_stack = cached_vm_stacks[i];
194
195                 if (!vm_stack)
196                         continue;
197
198                 vfree(vm_stack->addr);
199                 cached_vm_stacks[i] = NULL;
200         }
201
202         return 0;
203 }
204 #endif
205
206 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
207 {
208 #ifdef CONFIG_VMAP_STACK
209         void *stack;
210         int i;
211
212         for (i = 0; i < NR_CACHED_STACKS; i++) {
213                 struct vm_struct *s;
214
215                 s = this_cpu_xchg(cached_stacks[i], NULL);
216
217                 if (!s)
218                         continue;
219
220                 /* Clear stale pointers from reused stack. */
221                 memset(s->addr, 0, THREAD_SIZE);
222
223                 tsk->stack_vm_area = s;
224                 return s->addr;
225         }
226
227         /*
228          * Allocated stacks are cached and later reused by new threads,
229          * so memcg accounting is performed manually on assigning/releasing
230          * stacks to tasks. Drop __GFP_ACCOUNT.
231          */
232         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
233                                      VMALLOC_START, VMALLOC_END,
234                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
235                                      PAGE_KERNEL,
236                                      0, node, __builtin_return_address(0));
237
238         /*
239          * We can't call find_vm_area() in interrupt context, and
240          * free_thread_stack() can be called in interrupt context,
241          * so cache the vm_struct.
242          */
243         if (stack)
244                 tsk->stack_vm_area = find_vm_area(stack);
245         return stack;
246 #else
247         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
248                                              THREAD_SIZE_ORDER);
249
250         return page ? page_address(page) : NULL;
251 #endif
252 }
253
254 static inline void free_thread_stack(struct task_struct *tsk)
255 {
256 #ifdef CONFIG_VMAP_STACK
257         struct vm_struct *vm = task_stack_vm_area(tsk);
258
259         if (vm) {
260                 int i;
261
262                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
263                         mod_memcg_page_state(vm->pages[i],
264                                              MEMCG_KERNEL_STACK_KB,
265                                              -(int)(PAGE_SIZE / 1024));
266
267                         memcg_kmem_uncharge(vm->pages[i], 0);
268                 }
269
270                 for (i = 0; i < NR_CACHED_STACKS; i++) {
271                         if (this_cpu_cmpxchg(cached_stacks[i],
272                                         NULL, tsk->stack_vm_area) != NULL)
273                                 continue;
274
275                         return;
276                 }
277
278                 vfree_atomic(tsk->stack);
279                 return;
280         }
281 #endif
282
283         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
284 }
285 # else
286 static struct kmem_cache *thread_stack_cache;
287
288 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
289                                                   int node)
290 {
291         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
292 }
293
294 static void free_thread_stack(struct task_struct *tsk)
295 {
296         kmem_cache_free(thread_stack_cache, tsk->stack);
297 }
298
299 void thread_stack_cache_init(void)
300 {
301         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
302                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
303                                         THREAD_SIZE, NULL);
304         BUG_ON(thread_stack_cache == NULL);
305 }
306 # endif
307 #endif
308
309 /* SLAB cache for signal_struct structures (tsk->signal) */
310 static struct kmem_cache *signal_cachep;
311
312 /* SLAB cache for sighand_struct structures (tsk->sighand) */
313 struct kmem_cache *sighand_cachep;
314
315 /* SLAB cache for files_struct structures (tsk->files) */
316 struct kmem_cache *files_cachep;
317
318 /* SLAB cache for fs_struct structures (tsk->fs) */
319 struct kmem_cache *fs_cachep;
320
321 /* SLAB cache for vm_area_struct structures */
322 static struct kmem_cache *vm_area_cachep;
323
324 /* SLAB cache for mm_struct structures (tsk->mm) */
325 static struct kmem_cache *mm_cachep;
326
327 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
328 {
329         struct vm_area_struct *vma;
330
331         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
332         if (vma)
333                 vma_init(vma, mm);
334         return vma;
335 }
336
337 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
338 {
339         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
340
341         if (new) {
342                 *new = *orig;
343                 INIT_LIST_HEAD(&new->anon_vma_chain);
344         }
345         return new;
346 }
347
348 void vm_area_free(struct vm_area_struct *vma)
349 {
350         kmem_cache_free(vm_area_cachep, vma);
351 }
352
353 static void account_kernel_stack(struct task_struct *tsk, int account)
354 {
355         void *stack = task_stack_page(tsk);
356         struct vm_struct *vm = task_stack_vm_area(tsk);
357
358         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
359
360         if (vm) {
361                 int i;
362
363                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
364
365                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
366                         mod_zone_page_state(page_zone(vm->pages[i]),
367                                             NR_KERNEL_STACK_KB,
368                                             PAGE_SIZE / 1024 * account);
369                 }
370         } else {
371                 /*
372                  * All stack pages are in the same zone and belong to the
373                  * same memcg.
374                  */
375                 struct page *first_page = virt_to_page(stack);
376
377                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
378                                     THREAD_SIZE / 1024 * account);
379
380                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
381                                      account * (THREAD_SIZE / 1024));
382         }
383 }
384
385 static int memcg_charge_kernel_stack(struct task_struct *tsk)
386 {
387 #ifdef CONFIG_VMAP_STACK
388         struct vm_struct *vm = task_stack_vm_area(tsk);
389         int ret;
390
391         if (vm) {
392                 int i;
393
394                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
395                         /*
396                          * If memcg_kmem_charge() fails, page->mem_cgroup
397                          * pointer is NULL, and both memcg_kmem_uncharge()
398                          * and mod_memcg_page_state() in free_thread_stack()
399                          * will ignore this page. So it's safe.
400                          */
401                         ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
402                         if (ret)
403                                 return ret;
404
405                         mod_memcg_page_state(vm->pages[i],
406                                              MEMCG_KERNEL_STACK_KB,
407                                              PAGE_SIZE / 1024);
408                 }
409         }
410 #endif
411         return 0;
412 }
413
414 static void release_task_stack(struct task_struct *tsk)
415 {
416         if (WARN_ON(tsk->state != TASK_DEAD))
417                 return;  /* Better to leak the stack than to free prematurely */
418
419         account_kernel_stack(tsk, -1);
420         arch_release_thread_stack(tsk->stack);
421         free_thread_stack(tsk);
422         tsk->stack = NULL;
423 #ifdef CONFIG_VMAP_STACK
424         tsk->stack_vm_area = NULL;
425 #endif
426 }
427
428 #ifdef CONFIG_THREAD_INFO_IN_TASK
429 void put_task_stack(struct task_struct *tsk)
430 {
431         if (atomic_dec_and_test(&tsk->stack_refcount))
432                 release_task_stack(tsk);
433 }
434 #endif
435
436 void free_task(struct task_struct *tsk)
437 {
438 #ifndef CONFIG_THREAD_INFO_IN_TASK
439         /*
440          * The task is finally done with both the stack and thread_info,
441          * so free both.
442          */
443         release_task_stack(tsk);
444 #else
445         /*
446          * If the task had a separate stack allocation, it should be gone
447          * by now.
448          */
449         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
450 #endif
451         rt_mutex_debug_task_free(tsk);
452         ftrace_graph_exit_task(tsk);
453         put_seccomp_filter(tsk);
454         arch_release_task_struct(tsk);
455         if (tsk->flags & PF_KTHREAD)
456                 free_kthread_struct(tsk);
457         free_task_struct(tsk);
458 }
459 EXPORT_SYMBOL(free_task);
460
461 #ifdef CONFIG_MMU
462 static __latent_entropy int dup_mmap(struct mm_struct *mm,
463                                         struct mm_struct *oldmm)
464 {
465         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
466         struct rb_node **rb_link, *rb_parent;
467         int retval;
468         unsigned long charge;
469         LIST_HEAD(uf);
470
471         uprobe_start_dup_mmap();
472         if (down_write_killable(&oldmm->mmap_sem)) {
473                 retval = -EINTR;
474                 goto fail_uprobe_end;
475         }
476         flush_cache_dup_mm(oldmm);
477         uprobe_dup_mmap(oldmm, mm);
478         /*
479          * Not linked in yet - no deadlock potential:
480          */
481         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
482
483         /* No ordering required: file already has been exposed. */
484         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
485
486         mm->total_vm = oldmm->total_vm;
487         mm->data_vm = oldmm->data_vm;
488         mm->exec_vm = oldmm->exec_vm;
489         mm->stack_vm = oldmm->stack_vm;
490
491         rb_link = &mm->mm_rb.rb_node;
492         rb_parent = NULL;
493         pprev = &mm->mmap;
494         retval = ksm_fork(mm, oldmm);
495         if (retval)
496                 goto out;
497         retval = khugepaged_fork(mm, oldmm);
498         if (retval)
499                 goto out;
500
501         prev = NULL;
502         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
503                 struct file *file;
504
505                 if (mpnt->vm_flags & VM_DONTCOPY) {
506                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
507                         continue;
508                 }
509                 charge = 0;
510                 /*
511                  * Don't duplicate many vmas if we've been oom-killed (for
512                  * example)
513                  */
514                 if (fatal_signal_pending(current)) {
515                         retval = -EINTR;
516                         goto out;
517                 }
518                 if (mpnt->vm_flags & VM_ACCOUNT) {
519                         unsigned long len = vma_pages(mpnt);
520
521                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
522                                 goto fail_nomem;
523                         charge = len;
524                 }
525                 tmp = vm_area_dup(mpnt);
526                 if (!tmp)
527                         goto fail_nomem;
528                 retval = vma_dup_policy(mpnt, tmp);
529                 if (retval)
530                         goto fail_nomem_policy;
531                 tmp->vm_mm = mm;
532                 retval = dup_userfaultfd(tmp, &uf);
533                 if (retval)
534                         goto fail_nomem_anon_vma_fork;
535                 if (tmp->vm_flags & VM_WIPEONFORK) {
536                         /* VM_WIPEONFORK gets a clean slate in the child. */
537                         tmp->anon_vma = NULL;
538                         if (anon_vma_prepare(tmp))
539                                 goto fail_nomem_anon_vma_fork;
540                 } else if (anon_vma_fork(tmp, mpnt))
541                         goto fail_nomem_anon_vma_fork;
542                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
543                 tmp->vm_next = tmp->vm_prev = NULL;
544                 file = tmp->vm_file;
545                 if (file) {
546                         struct inode *inode = file_inode(file);
547                         struct address_space *mapping = file->f_mapping;
548
549                         get_file(file);
550                         if (tmp->vm_flags & VM_DENYWRITE)
551                                 atomic_dec(&inode->i_writecount);
552                         i_mmap_lock_write(mapping);
553                         if (tmp->vm_flags & VM_SHARED)
554                                 atomic_inc(&mapping->i_mmap_writable);
555                         flush_dcache_mmap_lock(mapping);
556                         /* insert tmp into the share list, just after mpnt */
557                         vma_interval_tree_insert_after(tmp, mpnt,
558                                         &mapping->i_mmap);
559                         flush_dcache_mmap_unlock(mapping);
560                         i_mmap_unlock_write(mapping);
561                 }
562
563                 /*
564                  * Clear hugetlb-related page reserves for children. This only
565                  * affects MAP_PRIVATE mappings. Faults generated by the child
566                  * are not guaranteed to succeed, even if read-only
567                  */
568                 if (is_vm_hugetlb_page(tmp))
569                         reset_vma_resv_huge_pages(tmp);
570
571                 /*
572                  * Link in the new vma and copy the page table entries.
573                  */
574                 *pprev = tmp;
575                 pprev = &tmp->vm_next;
576                 tmp->vm_prev = prev;
577                 prev = tmp;
578
579                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
580                 rb_link = &tmp->vm_rb.rb_right;
581                 rb_parent = &tmp->vm_rb;
582
583                 mm->map_count++;
584                 if (!(tmp->vm_flags & VM_WIPEONFORK))
585                         retval = copy_page_range(mm, oldmm, mpnt);
586
587                 if (tmp->vm_ops && tmp->vm_ops->open)
588                         tmp->vm_ops->open(tmp);
589
590                 if (retval)
591                         goto out;
592         }
593         /* a new mm has just been created */
594         retval = arch_dup_mmap(oldmm, mm);
595 out:
596         up_write(&mm->mmap_sem);
597         flush_tlb_mm(oldmm);
598         up_write(&oldmm->mmap_sem);
599         dup_userfaultfd_complete(&uf);
600 fail_uprobe_end:
601         uprobe_end_dup_mmap();
602         return retval;
603 fail_nomem_anon_vma_fork:
604         mpol_put(vma_policy(tmp));
605 fail_nomem_policy:
606         vm_area_free(tmp);
607 fail_nomem:
608         retval = -ENOMEM;
609         vm_unacct_memory(charge);
610         goto out;
611 }
612
613 static inline int mm_alloc_pgd(struct mm_struct *mm)
614 {
615         mm->pgd = pgd_alloc(mm);
616         if (unlikely(!mm->pgd))
617                 return -ENOMEM;
618         return 0;
619 }
620
621 static inline void mm_free_pgd(struct mm_struct *mm)
622 {
623         pgd_free(mm, mm->pgd);
624 }
625 #else
626 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
627 {
628         down_write(&oldmm->mmap_sem);
629         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
630         up_write(&oldmm->mmap_sem);
631         return 0;
632 }
633 #define mm_alloc_pgd(mm)        (0)
634 #define mm_free_pgd(mm)
635 #endif /* CONFIG_MMU */
636
637 static void check_mm(struct mm_struct *mm)
638 {
639         int i;
640
641         for (i = 0; i < NR_MM_COUNTERS; i++) {
642                 long x = atomic_long_read(&mm->rss_stat.count[i]);
643
644                 if (unlikely(x))
645                         printk(KERN_ALERT "BUG: Bad rss-counter state "
646                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
647         }
648
649         if (mm_pgtables_bytes(mm))
650                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
651                                 mm_pgtables_bytes(mm));
652
653 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
654         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
655 #endif
656 }
657
658 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
659 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
660
661 /*
662  * Called when the last reference to the mm
663  * is dropped: either by a lazy thread or by
664  * mmput. Free the page directory and the mm.
665  */
666 void __mmdrop(struct mm_struct *mm)
667 {
668         BUG_ON(mm == &init_mm);
669         WARN_ON_ONCE(mm == current->mm);
670         WARN_ON_ONCE(mm == current->active_mm);
671         mm_free_pgd(mm);
672         destroy_context(mm);
673         hmm_mm_destroy(mm);
674         mmu_notifier_mm_destroy(mm);
675         check_mm(mm);
676         put_user_ns(mm->user_ns);
677         free_mm(mm);
678 }
679 EXPORT_SYMBOL_GPL(__mmdrop);
680
681 static void mmdrop_async_fn(struct work_struct *work)
682 {
683         struct mm_struct *mm;
684
685         mm = container_of(work, struct mm_struct, async_put_work);
686         __mmdrop(mm);
687 }
688
689 static void mmdrop_async(struct mm_struct *mm)
690 {
691         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
692                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
693                 schedule_work(&mm->async_put_work);
694         }
695 }
696
697 static inline void free_signal_struct(struct signal_struct *sig)
698 {
699         taskstats_tgid_free(sig);
700         sched_autogroup_exit(sig);
701         /*
702          * __mmdrop is not safe to call from softirq context on x86 due to
703          * pgd_dtor so postpone it to the async context
704          */
705         if (sig->oom_mm)
706                 mmdrop_async(sig->oom_mm);
707         kmem_cache_free(signal_cachep, sig);
708 }
709
710 static inline void put_signal_struct(struct signal_struct *sig)
711 {
712         if (atomic_dec_and_test(&sig->sigcnt))
713                 free_signal_struct(sig);
714 }
715
716 void __put_task_struct(struct task_struct *tsk)
717 {
718         WARN_ON(!tsk->exit_state);
719         WARN_ON(atomic_read(&tsk->usage));
720         WARN_ON(tsk == current);
721
722         cgroup_free(tsk);
723         task_numa_free(tsk);
724         security_task_free(tsk);
725         exit_creds(tsk);
726         delayacct_tsk_free(tsk);
727         put_signal_struct(tsk->signal);
728
729         if (!profile_handoff_task(tsk))
730                 free_task(tsk);
731 }
732 EXPORT_SYMBOL_GPL(__put_task_struct);
733
734 void __init __weak arch_task_cache_init(void) { }
735
736 /*
737  * set_max_threads
738  */
739 static void set_max_threads(unsigned int max_threads_suggested)
740 {
741         u64 threads;
742
743         /*
744          * The number of threads shall be limited such that the thread
745          * structures may only consume a small part of the available memory.
746          */
747         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
748                 threads = MAX_THREADS;
749         else
750                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
751                                     (u64) THREAD_SIZE * 8UL);
752
753         if (threads > max_threads_suggested)
754                 threads = max_threads_suggested;
755
756         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
757 }
758
759 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
760 /* Initialized by the architecture: */
761 int arch_task_struct_size __read_mostly;
762 #endif
763
764 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
765 {
766         /* Fetch thread_struct whitelist for the architecture. */
767         arch_thread_struct_whitelist(offset, size);
768
769         /*
770          * Handle zero-sized whitelist or empty thread_struct, otherwise
771          * adjust offset to position of thread_struct in task_struct.
772          */
773         if (unlikely(*size == 0))
774                 *offset = 0;
775         else
776                 *offset += offsetof(struct task_struct, thread);
777 }
778
779 void __init fork_init(void)
780 {
781         int i;
782 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
783 #ifndef ARCH_MIN_TASKALIGN
784 #define ARCH_MIN_TASKALIGN      0
785 #endif
786         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
787         unsigned long useroffset, usersize;
788
789         /* create a slab on which task_structs can be allocated */
790         task_struct_whitelist(&useroffset, &usersize);
791         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
792                         arch_task_struct_size, align,
793                         SLAB_PANIC|SLAB_ACCOUNT,
794                         useroffset, usersize, NULL);
795 #endif
796
797         /* do the arch specific task caches init */
798         arch_task_cache_init();
799
800         set_max_threads(MAX_THREADS);
801
802         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
803         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
804         init_task.signal->rlim[RLIMIT_SIGPENDING] =
805                 init_task.signal->rlim[RLIMIT_NPROC];
806
807         for (i = 0; i < UCOUNT_COUNTS; i++) {
808                 init_user_ns.ucount_max[i] = max_threads/2;
809         }
810
811 #ifdef CONFIG_VMAP_STACK
812         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
813                           NULL, free_vm_stack_cache);
814 #endif
815
816         lockdep_init_task(&init_task);
817 }
818
819 int __weak arch_dup_task_struct(struct task_struct *dst,
820                                                struct task_struct *src)
821 {
822         *dst = *src;
823         return 0;
824 }
825
826 void set_task_stack_end_magic(struct task_struct *tsk)
827 {
828         unsigned long *stackend;
829
830         stackend = end_of_stack(tsk);
831         *stackend = STACK_END_MAGIC;    /* for overflow detection */
832 }
833
834 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
835 {
836         struct task_struct *tsk;
837         unsigned long *stack;
838         struct vm_struct *stack_vm_area;
839         int err;
840
841         if (node == NUMA_NO_NODE)
842                 node = tsk_fork_get_node(orig);
843         tsk = alloc_task_struct_node(node);
844         if (!tsk)
845                 return NULL;
846
847         stack = alloc_thread_stack_node(tsk, node);
848         if (!stack)
849                 goto free_tsk;
850
851         if (memcg_charge_kernel_stack(tsk))
852                 goto free_stack;
853
854         stack_vm_area = task_stack_vm_area(tsk);
855
856         err = arch_dup_task_struct(tsk, orig);
857
858         /*
859          * arch_dup_task_struct() clobbers the stack-related fields.  Make
860          * sure they're properly initialized before using any stack-related
861          * functions again.
862          */
863         tsk->stack = stack;
864 #ifdef CONFIG_VMAP_STACK
865         tsk->stack_vm_area = stack_vm_area;
866 #endif
867 #ifdef CONFIG_THREAD_INFO_IN_TASK
868         atomic_set(&tsk->stack_refcount, 1);
869 #endif
870
871         if (err)
872                 goto free_stack;
873
874 #ifdef CONFIG_SECCOMP
875         /*
876          * We must handle setting up seccomp filters once we're under
877          * the sighand lock in case orig has changed between now and
878          * then. Until then, filter must be NULL to avoid messing up
879          * the usage counts on the error path calling free_task.
880          */
881         tsk->seccomp.filter = NULL;
882 #endif
883
884         setup_thread_stack(tsk, orig);
885         clear_user_return_notifier(tsk);
886         clear_tsk_need_resched(tsk);
887         set_task_stack_end_magic(tsk);
888
889 #ifdef CONFIG_STACKPROTECTOR
890         tsk->stack_canary = get_random_canary();
891 #endif
892
893         /*
894          * One for us, one for whoever does the "release_task()" (usually
895          * parent)
896          */
897         atomic_set(&tsk->usage, 2);
898 #ifdef CONFIG_BLK_DEV_IO_TRACE
899         tsk->btrace_seq = 0;
900 #endif
901         tsk->splice_pipe = NULL;
902         tsk->task_frag.page = NULL;
903         tsk->wake_q.next = NULL;
904
905         account_kernel_stack(tsk, 1);
906
907         kcov_task_init(tsk);
908
909 #ifdef CONFIG_FAULT_INJECTION
910         tsk->fail_nth = 0;
911 #endif
912
913 #ifdef CONFIG_BLK_CGROUP
914         tsk->throttle_queue = NULL;
915         tsk->use_memdelay = 0;
916 #endif
917
918 #ifdef CONFIG_MEMCG
919         tsk->active_memcg = NULL;
920 #endif
921         return tsk;
922
923 free_stack:
924         free_thread_stack(tsk);
925 free_tsk:
926         free_task_struct(tsk);
927         return NULL;
928 }
929
930 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
931
932 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
933
934 static int __init coredump_filter_setup(char *s)
935 {
936         default_dump_filter =
937                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
938                 MMF_DUMP_FILTER_MASK;
939         return 1;
940 }
941
942 __setup("coredump_filter=", coredump_filter_setup);
943
944 #include <linux/init_task.h>
945
946 static void mm_init_aio(struct mm_struct *mm)
947 {
948 #ifdef CONFIG_AIO
949         spin_lock_init(&mm->ioctx_lock);
950         mm->ioctx_table = NULL;
951 #endif
952 }
953
954 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
955 {
956 #ifdef CONFIG_MEMCG
957         mm->owner = p;
958 #endif
959 }
960
961 static void mm_init_uprobes_state(struct mm_struct *mm)
962 {
963 #ifdef CONFIG_UPROBES
964         mm->uprobes_state.xol_area = NULL;
965 #endif
966 }
967
968 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
969         struct user_namespace *user_ns)
970 {
971         mm->mmap = NULL;
972         mm->mm_rb = RB_ROOT;
973         mm->vmacache_seqnum = 0;
974         atomic_set(&mm->mm_users, 1);
975         atomic_set(&mm->mm_count, 1);
976         init_rwsem(&mm->mmap_sem);
977         INIT_LIST_HEAD(&mm->mmlist);
978         mm->core_state = NULL;
979         mm_pgtables_bytes_init(mm);
980         mm->map_count = 0;
981         mm->locked_vm = 0;
982         mm->pinned_vm = 0;
983         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
984         spin_lock_init(&mm->page_table_lock);
985         spin_lock_init(&mm->arg_lock);
986         mm_init_cpumask(mm);
987         mm_init_aio(mm);
988         mm_init_owner(mm, p);
989         RCU_INIT_POINTER(mm->exe_file, NULL);
990         mmu_notifier_mm_init(mm);
991         hmm_mm_init(mm);
992         init_tlb_flush_pending(mm);
993 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
994         mm->pmd_huge_pte = NULL;
995 #endif
996         mm_init_uprobes_state(mm);
997
998         if (current->mm) {
999                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1000                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1001         } else {
1002                 mm->flags = default_dump_filter;
1003                 mm->def_flags = 0;
1004         }
1005
1006         if (mm_alloc_pgd(mm))
1007                 goto fail_nopgd;
1008
1009         if (init_new_context(p, mm))
1010                 goto fail_nocontext;
1011
1012         mm->user_ns = get_user_ns(user_ns);
1013         return mm;
1014
1015 fail_nocontext:
1016         mm_free_pgd(mm);
1017 fail_nopgd:
1018         free_mm(mm);
1019         return NULL;
1020 }
1021
1022 /*
1023  * Allocate and initialize an mm_struct.
1024  */
1025 struct mm_struct *mm_alloc(void)
1026 {
1027         struct mm_struct *mm;
1028
1029         mm = allocate_mm();
1030         if (!mm)
1031                 return NULL;
1032
1033         memset(mm, 0, sizeof(*mm));
1034         return mm_init(mm, current, current_user_ns());
1035 }
1036
1037 static inline void __mmput(struct mm_struct *mm)
1038 {
1039         VM_BUG_ON(atomic_read(&mm->mm_users));
1040
1041         uprobe_clear_state(mm);
1042         exit_aio(mm);
1043         ksm_exit(mm);
1044         khugepaged_exit(mm); /* must run before exit_mmap */
1045         exit_mmap(mm);
1046         mm_put_huge_zero_page(mm);
1047         set_mm_exe_file(mm, NULL);
1048         if (!list_empty(&mm->mmlist)) {
1049                 spin_lock(&mmlist_lock);
1050                 list_del(&mm->mmlist);
1051                 spin_unlock(&mmlist_lock);
1052         }
1053         if (mm->binfmt)
1054                 module_put(mm->binfmt->module);
1055         mmdrop(mm);
1056 }
1057
1058 /*
1059  * Decrement the use count and release all resources for an mm.
1060  */
1061 void mmput(struct mm_struct *mm)
1062 {
1063         might_sleep();
1064
1065         if (atomic_dec_and_test(&mm->mm_users))
1066                 __mmput(mm);
1067 }
1068 EXPORT_SYMBOL_GPL(mmput);
1069
1070 #ifdef CONFIG_MMU
1071 static void mmput_async_fn(struct work_struct *work)
1072 {
1073         struct mm_struct *mm = container_of(work, struct mm_struct,
1074                                             async_put_work);
1075
1076         __mmput(mm);
1077 }
1078
1079 void mmput_async(struct mm_struct *mm)
1080 {
1081         if (atomic_dec_and_test(&mm->mm_users)) {
1082                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1083                 schedule_work(&mm->async_put_work);
1084         }
1085 }
1086 #endif
1087
1088 /**
1089  * set_mm_exe_file - change a reference to the mm's executable file
1090  *
1091  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1092  *
1093  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1094  * invocations: in mmput() nobody alive left, in execve task is single
1095  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1096  * mm->exe_file, but does so without using set_mm_exe_file() in order
1097  * to do avoid the need for any locks.
1098  */
1099 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1100 {
1101         struct file *old_exe_file;
1102
1103         /*
1104          * It is safe to dereference the exe_file without RCU as
1105          * this function is only called if nobody else can access
1106          * this mm -- see comment above for justification.
1107          */
1108         old_exe_file = rcu_dereference_raw(mm->exe_file);
1109
1110         if (new_exe_file)
1111                 get_file(new_exe_file);
1112         rcu_assign_pointer(mm->exe_file, new_exe_file);
1113         if (old_exe_file)
1114                 fput(old_exe_file);
1115 }
1116
1117 /**
1118  * get_mm_exe_file - acquire a reference to the mm's executable file
1119  *
1120  * Returns %NULL if mm has no associated executable file.
1121  * User must release file via fput().
1122  */
1123 struct file *get_mm_exe_file(struct mm_struct *mm)
1124 {
1125         struct file *exe_file;
1126
1127         rcu_read_lock();
1128         exe_file = rcu_dereference(mm->exe_file);
1129         if (exe_file && !get_file_rcu(exe_file))
1130                 exe_file = NULL;
1131         rcu_read_unlock();
1132         return exe_file;
1133 }
1134 EXPORT_SYMBOL(get_mm_exe_file);
1135
1136 /**
1137  * get_task_exe_file - acquire a reference to the task's executable file
1138  *
1139  * Returns %NULL if task's mm (if any) has no associated executable file or
1140  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1141  * User must release file via fput().
1142  */
1143 struct file *get_task_exe_file(struct task_struct *task)
1144 {
1145         struct file *exe_file = NULL;
1146         struct mm_struct *mm;
1147
1148         task_lock(task);
1149         mm = task->mm;
1150         if (mm) {
1151                 if (!(task->flags & PF_KTHREAD))
1152                         exe_file = get_mm_exe_file(mm);
1153         }
1154         task_unlock(task);
1155         return exe_file;
1156 }
1157 EXPORT_SYMBOL(get_task_exe_file);
1158
1159 /**
1160  * get_task_mm - acquire a reference to the task's mm
1161  *
1162  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1163  * this kernel workthread has transiently adopted a user mm with use_mm,
1164  * to do its AIO) is not set and if so returns a reference to it, after
1165  * bumping up the use count.  User must release the mm via mmput()
1166  * after use.  Typically used by /proc and ptrace.
1167  */
1168 struct mm_struct *get_task_mm(struct task_struct *task)
1169 {
1170         struct mm_struct *mm;
1171
1172         task_lock(task);
1173         mm = task->mm;
1174         if (mm) {
1175                 if (task->flags & PF_KTHREAD)
1176                         mm = NULL;
1177                 else
1178                         mmget(mm);
1179         }
1180         task_unlock(task);
1181         return mm;
1182 }
1183 EXPORT_SYMBOL_GPL(get_task_mm);
1184
1185 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1186 {
1187         struct mm_struct *mm;
1188         int err;
1189
1190         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1191         if (err)
1192                 return ERR_PTR(err);
1193
1194         mm = get_task_mm(task);
1195         if (mm && mm != current->mm &&
1196                         !ptrace_may_access(task, mode)) {
1197                 mmput(mm);
1198                 mm = ERR_PTR(-EACCES);
1199         }
1200         mutex_unlock(&task->signal->cred_guard_mutex);
1201
1202         return mm;
1203 }
1204
1205 static void complete_vfork_done(struct task_struct *tsk)
1206 {
1207         struct completion *vfork;
1208
1209         task_lock(tsk);
1210         vfork = tsk->vfork_done;
1211         if (likely(vfork)) {
1212                 tsk->vfork_done = NULL;
1213                 complete(vfork);
1214         }
1215         task_unlock(tsk);
1216 }
1217
1218 static int wait_for_vfork_done(struct task_struct *child,
1219                                 struct completion *vfork)
1220 {
1221         int killed;
1222
1223         freezer_do_not_count();
1224         killed = wait_for_completion_killable(vfork);
1225         freezer_count();
1226
1227         if (killed) {
1228                 task_lock(child);
1229                 child->vfork_done = NULL;
1230                 task_unlock(child);
1231         }
1232
1233         put_task_struct(child);
1234         return killed;
1235 }
1236
1237 /* Please note the differences between mmput and mm_release.
1238  * mmput is called whenever we stop holding onto a mm_struct,
1239  * error success whatever.
1240  *
1241  * mm_release is called after a mm_struct has been removed
1242  * from the current process.
1243  *
1244  * This difference is important for error handling, when we
1245  * only half set up a mm_struct for a new process and need to restore
1246  * the old one.  Because we mmput the new mm_struct before
1247  * restoring the old one. . .
1248  * Eric Biederman 10 January 1998
1249  */
1250 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1251 {
1252         /* Get rid of any futexes when releasing the mm */
1253 #ifdef CONFIG_FUTEX
1254         if (unlikely(tsk->robust_list)) {
1255                 exit_robust_list(tsk);
1256                 tsk->robust_list = NULL;
1257         }
1258 #ifdef CONFIG_COMPAT
1259         if (unlikely(tsk->compat_robust_list)) {
1260                 compat_exit_robust_list(tsk);
1261                 tsk->compat_robust_list = NULL;
1262         }
1263 #endif
1264         if (unlikely(!list_empty(&tsk->pi_state_list)))
1265                 exit_pi_state_list(tsk);
1266 #endif
1267
1268         uprobe_free_utask(tsk);
1269
1270         /* Get rid of any cached register state */
1271         deactivate_mm(tsk, mm);
1272
1273         /*
1274          * Signal userspace if we're not exiting with a core dump
1275          * because we want to leave the value intact for debugging
1276          * purposes.
1277          */
1278         if (tsk->clear_child_tid) {
1279                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1280                     atomic_read(&mm->mm_users) > 1) {
1281                         /*
1282                          * We don't check the error code - if userspace has
1283                          * not set up a proper pointer then tough luck.
1284                          */
1285                         put_user(0, tsk->clear_child_tid);
1286                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1287                                         1, NULL, NULL, 0, 0);
1288                 }
1289                 tsk->clear_child_tid = NULL;
1290         }
1291
1292         /*
1293          * All done, finally we can wake up parent and return this mm to him.
1294          * Also kthread_stop() uses this completion for synchronization.
1295          */
1296         if (tsk->vfork_done)
1297                 complete_vfork_done(tsk);
1298 }
1299
1300 /*
1301  * Allocate a new mm structure and copy contents from the
1302  * mm structure of the passed in task structure.
1303  */
1304 static struct mm_struct *dup_mm(struct task_struct *tsk)
1305 {
1306         struct mm_struct *mm, *oldmm = current->mm;
1307         int err;
1308
1309         mm = allocate_mm();
1310         if (!mm)
1311                 goto fail_nomem;
1312
1313         memcpy(mm, oldmm, sizeof(*mm));
1314
1315         if (!mm_init(mm, tsk, mm->user_ns))
1316                 goto fail_nomem;
1317
1318         err = dup_mmap(mm, oldmm);
1319         if (err)
1320                 goto free_pt;
1321
1322         mm->hiwater_rss = get_mm_rss(mm);
1323         mm->hiwater_vm = mm->total_vm;
1324
1325         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1326                 goto free_pt;
1327
1328         return mm;
1329
1330 free_pt:
1331         /* don't put binfmt in mmput, we haven't got module yet */
1332         mm->binfmt = NULL;
1333         mmput(mm);
1334
1335 fail_nomem:
1336         return NULL;
1337 }
1338
1339 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1340 {
1341         struct mm_struct *mm, *oldmm;
1342         int retval;
1343
1344         tsk->min_flt = tsk->maj_flt = 0;
1345         tsk->nvcsw = tsk->nivcsw = 0;
1346 #ifdef CONFIG_DETECT_HUNG_TASK
1347         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1348         tsk->last_switch_time = 0;
1349 #endif
1350
1351         tsk->mm = NULL;
1352         tsk->active_mm = NULL;
1353
1354         /*
1355          * Are we cloning a kernel thread?
1356          *
1357          * We need to steal a active VM for that..
1358          */
1359         oldmm = current->mm;
1360         if (!oldmm)
1361                 return 0;
1362
1363         /* initialize the new vmacache entries */
1364         vmacache_flush(tsk);
1365
1366         if (clone_flags & CLONE_VM) {
1367                 mmget(oldmm);
1368                 mm = oldmm;
1369                 goto good_mm;
1370         }
1371
1372         retval = -ENOMEM;
1373         mm = dup_mm(tsk);
1374         if (!mm)
1375                 goto fail_nomem;
1376
1377 good_mm:
1378         tsk->mm = mm;
1379         tsk->active_mm = mm;
1380         return 0;
1381
1382 fail_nomem:
1383         return retval;
1384 }
1385
1386 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1387 {
1388         struct fs_struct *fs = current->fs;
1389         if (clone_flags & CLONE_FS) {
1390                 /* tsk->fs is already what we want */
1391                 spin_lock(&fs->lock);
1392                 if (fs->in_exec) {
1393                         spin_unlock(&fs->lock);
1394                         return -EAGAIN;
1395                 }
1396                 fs->users++;
1397                 spin_unlock(&fs->lock);
1398                 return 0;
1399         }
1400         tsk->fs = copy_fs_struct(fs);
1401         if (!tsk->fs)
1402                 return -ENOMEM;
1403         return 0;
1404 }
1405
1406 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1407 {
1408         struct files_struct *oldf, *newf;
1409         int error = 0;
1410
1411         /*
1412          * A background process may not have any files ...
1413          */
1414         oldf = current->files;
1415         if (!oldf)
1416                 goto out;
1417
1418         if (clone_flags & CLONE_FILES) {
1419                 atomic_inc(&oldf->count);
1420                 goto out;
1421         }
1422
1423         newf = dup_fd(oldf, &error);
1424         if (!newf)
1425                 goto out;
1426
1427         tsk->files = newf;
1428         error = 0;
1429 out:
1430         return error;
1431 }
1432
1433 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1434 {
1435 #ifdef CONFIG_BLOCK
1436         struct io_context *ioc = current->io_context;
1437         struct io_context *new_ioc;
1438
1439         if (!ioc)
1440                 return 0;
1441         /*
1442          * Share io context with parent, if CLONE_IO is set
1443          */
1444         if (clone_flags & CLONE_IO) {
1445                 ioc_task_link(ioc);
1446                 tsk->io_context = ioc;
1447         } else if (ioprio_valid(ioc->ioprio)) {
1448                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1449                 if (unlikely(!new_ioc))
1450                         return -ENOMEM;
1451
1452                 new_ioc->ioprio = ioc->ioprio;
1453                 put_io_context(new_ioc);
1454         }
1455 #endif
1456         return 0;
1457 }
1458
1459 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1460 {
1461         struct sighand_struct *sig;
1462
1463         if (clone_flags & CLONE_SIGHAND) {
1464                 atomic_inc(&current->sighand->count);
1465                 return 0;
1466         }
1467         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1468         rcu_assign_pointer(tsk->sighand, sig);
1469         if (!sig)
1470                 return -ENOMEM;
1471
1472         atomic_set(&sig->count, 1);
1473         spin_lock_irq(&current->sighand->siglock);
1474         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1475         spin_unlock_irq(&current->sighand->siglock);
1476         return 0;
1477 }
1478
1479 void __cleanup_sighand(struct sighand_struct *sighand)
1480 {
1481         if (atomic_dec_and_test(&sighand->count)) {
1482                 signalfd_cleanup(sighand);
1483                 /*
1484                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1485                  * without an RCU grace period, see __lock_task_sighand().
1486                  */
1487                 kmem_cache_free(sighand_cachep, sighand);
1488         }
1489 }
1490
1491 #ifdef CONFIG_POSIX_TIMERS
1492 /*
1493  * Initialize POSIX timer handling for a thread group.
1494  */
1495 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1496 {
1497         unsigned long cpu_limit;
1498
1499         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1500         if (cpu_limit != RLIM_INFINITY) {
1501                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1502                 sig->cputimer.running = true;
1503         }
1504
1505         /* The timer lists. */
1506         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1507         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1508         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1509 }
1510 #else
1511 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1512 #endif
1513
1514 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1515 {
1516         struct signal_struct *sig;
1517
1518         if (clone_flags & CLONE_THREAD)
1519                 return 0;
1520
1521         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1522         tsk->signal = sig;
1523         if (!sig)
1524                 return -ENOMEM;
1525
1526         sig->nr_threads = 1;
1527         atomic_set(&sig->live, 1);
1528         atomic_set(&sig->sigcnt, 1);
1529
1530         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1531         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1532         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1533
1534         init_waitqueue_head(&sig->wait_chldexit);
1535         sig->curr_target = tsk;
1536         init_sigpending(&sig->shared_pending);
1537         INIT_HLIST_HEAD(&sig->multiprocess);
1538         seqlock_init(&sig->stats_lock);
1539         prev_cputime_init(&sig->prev_cputime);
1540
1541 #ifdef CONFIG_POSIX_TIMERS
1542         INIT_LIST_HEAD(&sig->posix_timers);
1543         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1544         sig->real_timer.function = it_real_fn;
1545 #endif
1546
1547         task_lock(current->group_leader);
1548         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1549         task_unlock(current->group_leader);
1550
1551         posix_cpu_timers_init_group(sig);
1552
1553         tty_audit_fork(sig);
1554         sched_autogroup_fork(sig);
1555
1556         sig->oom_score_adj = current->signal->oom_score_adj;
1557         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1558
1559         mutex_init(&sig->cred_guard_mutex);
1560
1561         return 0;
1562 }
1563
1564 static void copy_seccomp(struct task_struct *p)
1565 {
1566 #ifdef CONFIG_SECCOMP
1567         /*
1568          * Must be called with sighand->lock held, which is common to
1569          * all threads in the group. Holding cred_guard_mutex is not
1570          * needed because this new task is not yet running and cannot
1571          * be racing exec.
1572          */
1573         assert_spin_locked(&current->sighand->siglock);
1574
1575         /* Ref-count the new filter user, and assign it. */
1576         get_seccomp_filter(current);
1577         p->seccomp = current->seccomp;
1578
1579         /*
1580          * Explicitly enable no_new_privs here in case it got set
1581          * between the task_struct being duplicated and holding the
1582          * sighand lock. The seccomp state and nnp must be in sync.
1583          */
1584         if (task_no_new_privs(current))
1585                 task_set_no_new_privs(p);
1586
1587         /*
1588          * If the parent gained a seccomp mode after copying thread
1589          * flags and between before we held the sighand lock, we have
1590          * to manually enable the seccomp thread flag here.
1591          */
1592         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1593                 set_tsk_thread_flag(p, TIF_SECCOMP);
1594 #endif
1595 }
1596
1597 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1598 {
1599         current->clear_child_tid = tidptr;
1600
1601         return task_pid_vnr(current);
1602 }
1603
1604 static void rt_mutex_init_task(struct task_struct *p)
1605 {
1606         raw_spin_lock_init(&p->pi_lock);
1607 #ifdef CONFIG_RT_MUTEXES
1608         p->pi_waiters = RB_ROOT_CACHED;
1609         p->pi_top_task = NULL;
1610         p->pi_blocked_on = NULL;
1611 #endif
1612 }
1613
1614 #ifdef CONFIG_POSIX_TIMERS
1615 /*
1616  * Initialize POSIX timer handling for a single task.
1617  */
1618 static void posix_cpu_timers_init(struct task_struct *tsk)
1619 {
1620         tsk->cputime_expires.prof_exp = 0;
1621         tsk->cputime_expires.virt_exp = 0;
1622         tsk->cputime_expires.sched_exp = 0;
1623         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1624         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1625         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1626 }
1627 #else
1628 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1629 #endif
1630
1631 static inline void init_task_pid_links(struct task_struct *task)
1632 {
1633         enum pid_type type;
1634
1635         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1636                 INIT_HLIST_NODE(&task->pid_links[type]);
1637         }
1638 }
1639
1640 static inline void
1641 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1642 {
1643         if (type == PIDTYPE_PID)
1644                 task->thread_pid = pid;
1645         else
1646                 task->signal->pids[type] = pid;
1647 }
1648
1649 static inline void rcu_copy_process(struct task_struct *p)
1650 {
1651 #ifdef CONFIG_PREEMPT_RCU
1652         p->rcu_read_lock_nesting = 0;
1653         p->rcu_read_unlock_special.s = 0;
1654         p->rcu_blocked_node = NULL;
1655         INIT_LIST_HEAD(&p->rcu_node_entry);
1656 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1657 #ifdef CONFIG_TASKS_RCU
1658         p->rcu_tasks_holdout = false;
1659         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1660         p->rcu_tasks_idle_cpu = -1;
1661 #endif /* #ifdef CONFIG_TASKS_RCU */
1662 }
1663
1664 /*
1665  * This creates a new process as a copy of the old one,
1666  * but does not actually start it yet.
1667  *
1668  * It copies the registers, and all the appropriate
1669  * parts of the process environment (as per the clone
1670  * flags). The actual kick-off is left to the caller.
1671  */
1672 static __latent_entropy struct task_struct *copy_process(
1673                                         unsigned long clone_flags,
1674                                         unsigned long stack_start,
1675                                         unsigned long stack_size,
1676                                         int __user *child_tidptr,
1677                                         struct pid *pid,
1678                                         int trace,
1679                                         unsigned long tls,
1680                                         int node)
1681 {
1682         int retval;
1683         struct task_struct *p;
1684         struct multiprocess_signals delayed;
1685
1686         /*
1687          * Don't allow sharing the root directory with processes in a different
1688          * namespace
1689          */
1690         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1691                 return ERR_PTR(-EINVAL);
1692
1693         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1694                 return ERR_PTR(-EINVAL);
1695
1696         /*
1697          * Thread groups must share signals as well, and detached threads
1698          * can only be started up within the thread group.
1699          */
1700         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1701                 return ERR_PTR(-EINVAL);
1702
1703         /*
1704          * Shared signal handlers imply shared VM. By way of the above,
1705          * thread groups also imply shared VM. Blocking this case allows
1706          * for various simplifications in other code.
1707          */
1708         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1709                 return ERR_PTR(-EINVAL);
1710
1711         /*
1712          * Siblings of global init remain as zombies on exit since they are
1713          * not reaped by their parent (swapper). To solve this and to avoid
1714          * multi-rooted process trees, prevent global and container-inits
1715          * from creating siblings.
1716          */
1717         if ((clone_flags & CLONE_PARENT) &&
1718                                 current->signal->flags & SIGNAL_UNKILLABLE)
1719                 return ERR_PTR(-EINVAL);
1720
1721         /*
1722          * If the new process will be in a different pid or user namespace
1723          * do not allow it to share a thread group with the forking task.
1724          */
1725         if (clone_flags & CLONE_THREAD) {
1726                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1727                     (task_active_pid_ns(current) !=
1728                                 current->nsproxy->pid_ns_for_children))
1729                         return ERR_PTR(-EINVAL);
1730         }
1731
1732         /*
1733          * Force any signals received before this point to be delivered
1734          * before the fork happens.  Collect up signals sent to multiple
1735          * processes that happen during the fork and delay them so that
1736          * they appear to happen after the fork.
1737          */
1738         sigemptyset(&delayed.signal);
1739         INIT_HLIST_NODE(&delayed.node);
1740
1741         spin_lock_irq(&current->sighand->siglock);
1742         if (!(clone_flags & CLONE_THREAD))
1743                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1744         recalc_sigpending();
1745         spin_unlock_irq(&current->sighand->siglock);
1746         retval = -ERESTARTNOINTR;
1747         if (signal_pending(current))
1748                 goto fork_out;
1749
1750         retval = -ENOMEM;
1751         p = dup_task_struct(current, node);
1752         if (!p)
1753                 goto fork_out;
1754
1755         /*
1756          * This _must_ happen before we call free_task(), i.e. before we jump
1757          * to any of the bad_fork_* labels. This is to avoid freeing
1758          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1759          * kernel threads (PF_KTHREAD).
1760          */
1761         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1762         /*
1763          * Clear TID on mm_release()?
1764          */
1765         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1766
1767         ftrace_graph_init_task(p);
1768
1769         rt_mutex_init_task(p);
1770
1771 #ifdef CONFIG_PROVE_LOCKING
1772         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1773         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1774 #endif
1775         retval = -EAGAIN;
1776         if (atomic_read(&p->real_cred->user->processes) >=
1777                         task_rlimit(p, RLIMIT_NPROC)) {
1778                 if (p->real_cred->user != INIT_USER &&
1779                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1780                         goto bad_fork_free;
1781         }
1782         current->flags &= ~PF_NPROC_EXCEEDED;
1783
1784         retval = copy_creds(p, clone_flags);
1785         if (retval < 0)
1786                 goto bad_fork_free;
1787
1788         /*
1789          * If multiple threads are within copy_process(), then this check
1790          * triggers too late. This doesn't hurt, the check is only there
1791          * to stop root fork bombs.
1792          */
1793         retval = -EAGAIN;
1794         if (nr_threads >= max_threads)
1795                 goto bad_fork_cleanup_count;
1796
1797         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1798         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1799         p->flags |= PF_FORKNOEXEC;
1800         INIT_LIST_HEAD(&p->children);
1801         INIT_LIST_HEAD(&p->sibling);
1802         rcu_copy_process(p);
1803         p->vfork_done = NULL;
1804         spin_lock_init(&p->alloc_lock);
1805
1806         init_sigpending(&p->pending);
1807
1808         p->utime = p->stime = p->gtime = 0;
1809 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1810         p->utimescaled = p->stimescaled = 0;
1811 #endif
1812         prev_cputime_init(&p->prev_cputime);
1813
1814 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1815         seqcount_init(&p->vtime.seqcount);
1816         p->vtime.starttime = 0;
1817         p->vtime.state = VTIME_INACTIVE;
1818 #endif
1819
1820 #if defined(SPLIT_RSS_COUNTING)
1821         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1822 #endif
1823
1824         p->default_timer_slack_ns = current->timer_slack_ns;
1825
1826 #ifdef CONFIG_PSI
1827         p->psi_flags = 0;
1828 #endif
1829
1830         task_io_accounting_init(&p->ioac);
1831         acct_clear_integrals(p);
1832
1833         posix_cpu_timers_init(p);
1834
1835         p->start_time = ktime_get_ns();
1836         p->real_start_time = ktime_get_boot_ns();
1837         p->io_context = NULL;
1838         audit_set_context(p, NULL);
1839         cgroup_fork(p);
1840 #ifdef CONFIG_NUMA
1841         p->mempolicy = mpol_dup(p->mempolicy);
1842         if (IS_ERR(p->mempolicy)) {
1843                 retval = PTR_ERR(p->mempolicy);
1844                 p->mempolicy = NULL;
1845                 goto bad_fork_cleanup_threadgroup_lock;
1846         }
1847 #endif
1848 #ifdef CONFIG_CPUSETS
1849         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1850         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1851         seqcount_init(&p->mems_allowed_seq);
1852 #endif
1853 #ifdef CONFIG_TRACE_IRQFLAGS
1854         p->irq_events = 0;
1855         p->hardirqs_enabled = 0;
1856         p->hardirq_enable_ip = 0;
1857         p->hardirq_enable_event = 0;
1858         p->hardirq_disable_ip = _THIS_IP_;
1859         p->hardirq_disable_event = 0;
1860         p->softirqs_enabled = 1;
1861         p->softirq_enable_ip = _THIS_IP_;
1862         p->softirq_enable_event = 0;
1863         p->softirq_disable_ip = 0;
1864         p->softirq_disable_event = 0;
1865         p->hardirq_context = 0;
1866         p->softirq_context = 0;
1867 #endif
1868
1869         p->pagefault_disabled = 0;
1870
1871 #ifdef CONFIG_LOCKDEP
1872         p->lockdep_depth = 0; /* no locks held yet */
1873         p->curr_chain_key = 0;
1874         p->lockdep_recursion = 0;
1875         lockdep_init_task(p);
1876 #endif
1877
1878 #ifdef CONFIG_DEBUG_MUTEXES
1879         p->blocked_on = NULL; /* not blocked yet */
1880 #endif
1881 #ifdef CONFIG_BCACHE
1882         p->sequential_io        = 0;
1883         p->sequential_io_avg    = 0;
1884 #endif
1885
1886         /* Perform scheduler related setup. Assign this task to a CPU. */
1887         retval = sched_fork(clone_flags, p);
1888         if (retval)
1889                 goto bad_fork_cleanup_policy;
1890
1891         retval = perf_event_init_task(p);
1892         if (retval)
1893                 goto bad_fork_cleanup_policy;
1894         retval = audit_alloc(p);
1895         if (retval)
1896                 goto bad_fork_cleanup_perf;
1897         /* copy all the process information */
1898         shm_init_task(p);
1899         retval = security_task_alloc(p, clone_flags);
1900         if (retval)
1901                 goto bad_fork_cleanup_audit;
1902         retval = copy_semundo(clone_flags, p);
1903         if (retval)
1904                 goto bad_fork_cleanup_security;
1905         retval = copy_files(clone_flags, p);
1906         if (retval)
1907                 goto bad_fork_cleanup_semundo;
1908         retval = copy_fs(clone_flags, p);
1909         if (retval)
1910                 goto bad_fork_cleanup_files;
1911         retval = copy_sighand(clone_flags, p);
1912         if (retval)
1913                 goto bad_fork_cleanup_fs;
1914         retval = copy_signal(clone_flags, p);
1915         if (retval)
1916                 goto bad_fork_cleanup_sighand;
1917         retval = copy_mm(clone_flags, p);
1918         if (retval)
1919                 goto bad_fork_cleanup_signal;
1920         retval = copy_namespaces(clone_flags, p);
1921         if (retval)
1922                 goto bad_fork_cleanup_mm;
1923         retval = copy_io(clone_flags, p);
1924         if (retval)
1925                 goto bad_fork_cleanup_namespaces;
1926         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1927         if (retval)
1928                 goto bad_fork_cleanup_io;
1929
1930         stackleak_task_init(p);
1931
1932         if (pid != &init_struct_pid) {
1933                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1934                 if (IS_ERR(pid)) {
1935                         retval = PTR_ERR(pid);
1936                         goto bad_fork_cleanup_thread;
1937                 }
1938         }
1939
1940 #ifdef CONFIG_BLOCK
1941         p->plug = NULL;
1942 #endif
1943 #ifdef CONFIG_FUTEX
1944         p->robust_list = NULL;
1945 #ifdef CONFIG_COMPAT
1946         p->compat_robust_list = NULL;
1947 #endif
1948         INIT_LIST_HEAD(&p->pi_state_list);
1949         p->pi_state_cache = NULL;
1950 #endif
1951         /*
1952          * sigaltstack should be cleared when sharing the same VM
1953          */
1954         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1955                 sas_ss_reset(p);
1956
1957         /*
1958          * Syscall tracing and stepping should be turned off in the
1959          * child regardless of CLONE_PTRACE.
1960          */
1961         user_disable_single_step(p);
1962         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1963 #ifdef TIF_SYSCALL_EMU
1964         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1965 #endif
1966         clear_all_latency_tracing(p);
1967
1968         /* ok, now we should be set up.. */
1969         p->pid = pid_nr(pid);
1970         if (clone_flags & CLONE_THREAD) {
1971                 p->exit_signal = -1;
1972                 p->group_leader = current->group_leader;
1973                 p->tgid = current->tgid;
1974         } else {
1975                 if (clone_flags & CLONE_PARENT)
1976                         p->exit_signal = current->group_leader->exit_signal;
1977                 else
1978                         p->exit_signal = (clone_flags & CSIGNAL);
1979                 p->group_leader = p;
1980                 p->tgid = p->pid;
1981         }
1982
1983         p->nr_dirtied = 0;
1984         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1985         p->dirty_paused_when = 0;
1986
1987         p->pdeath_signal = 0;
1988         INIT_LIST_HEAD(&p->thread_group);
1989         p->task_works = NULL;
1990
1991         cgroup_threadgroup_change_begin(current);
1992         /*
1993          * Ensure that the cgroup subsystem policies allow the new process to be
1994          * forked. It should be noted the the new process's css_set can be changed
1995          * between here and cgroup_post_fork() if an organisation operation is in
1996          * progress.
1997          */
1998         retval = cgroup_can_fork(p);
1999         if (retval)
2000                 goto bad_fork_free_pid;
2001
2002         /*
2003          * Make it visible to the rest of the system, but dont wake it up yet.
2004          * Need tasklist lock for parent etc handling!
2005          */
2006         write_lock_irq(&tasklist_lock);
2007
2008         /* CLONE_PARENT re-uses the old parent */
2009         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2010                 p->real_parent = current->real_parent;
2011                 p->parent_exec_id = current->parent_exec_id;
2012         } else {
2013                 p->real_parent = current;
2014                 p->parent_exec_id = current->self_exec_id;
2015         }
2016
2017         klp_copy_process(p);
2018
2019         spin_lock(&current->sighand->siglock);
2020
2021         /*
2022          * Copy seccomp details explicitly here, in case they were changed
2023          * before holding sighand lock.
2024          */
2025         copy_seccomp(p);
2026
2027         rseq_fork(p, clone_flags);
2028
2029         /* Don't start children in a dying pid namespace */
2030         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2031                 retval = -ENOMEM;
2032                 goto bad_fork_cancel_cgroup;
2033         }
2034
2035         /* Let kill terminate clone/fork in the middle */
2036         if (fatal_signal_pending(current)) {
2037                 retval = -EINTR;
2038                 goto bad_fork_cancel_cgroup;
2039         }
2040
2041
2042         init_task_pid_links(p);
2043         if (likely(p->pid)) {
2044                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2045
2046                 init_task_pid(p, PIDTYPE_PID, pid);
2047                 if (thread_group_leader(p)) {
2048                         init_task_pid(p, PIDTYPE_TGID, pid);
2049                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2050                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2051
2052                         if (is_child_reaper(pid)) {
2053                                 ns_of_pid(pid)->child_reaper = p;
2054                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2055                         }
2056                         p->signal->shared_pending.signal = delayed.signal;
2057                         p->signal->tty = tty_kref_get(current->signal->tty);
2058                         /*
2059                          * Inherit has_child_subreaper flag under the same
2060                          * tasklist_lock with adding child to the process tree
2061                          * for propagate_has_child_subreaper optimization.
2062                          */
2063                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2064                                                          p->real_parent->signal->is_child_subreaper;
2065                         list_add_tail(&p->sibling, &p->real_parent->children);
2066                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2067                         attach_pid(p, PIDTYPE_TGID);
2068                         attach_pid(p, PIDTYPE_PGID);
2069                         attach_pid(p, PIDTYPE_SID);
2070                         __this_cpu_inc(process_counts);
2071                 } else {
2072                         current->signal->nr_threads++;
2073                         atomic_inc(&current->signal->live);
2074                         atomic_inc(&current->signal->sigcnt);
2075                         task_join_group_stop(p);
2076                         list_add_tail_rcu(&p->thread_group,
2077                                           &p->group_leader->thread_group);
2078                         list_add_tail_rcu(&p->thread_node,
2079                                           &p->signal->thread_head);
2080                 }
2081                 attach_pid(p, PIDTYPE_PID);
2082                 nr_threads++;
2083         }
2084         total_forks++;
2085         hlist_del_init(&delayed.node);
2086         spin_unlock(&current->sighand->siglock);
2087         syscall_tracepoint_update(p);
2088         write_unlock_irq(&tasklist_lock);
2089
2090         proc_fork_connector(p);
2091         cgroup_post_fork(p);
2092         cgroup_threadgroup_change_end(current);
2093         perf_event_fork(p);
2094
2095         trace_task_newtask(p, clone_flags);
2096         uprobe_copy_process(p, clone_flags);
2097
2098         return p;
2099
2100 bad_fork_cancel_cgroup:
2101         spin_unlock(&current->sighand->siglock);
2102         write_unlock_irq(&tasklist_lock);
2103         cgroup_cancel_fork(p);
2104 bad_fork_free_pid:
2105         cgroup_threadgroup_change_end(current);
2106         if (pid != &init_struct_pid)
2107                 free_pid(pid);
2108 bad_fork_cleanup_thread:
2109         exit_thread(p);
2110 bad_fork_cleanup_io:
2111         if (p->io_context)
2112                 exit_io_context(p);
2113 bad_fork_cleanup_namespaces:
2114         exit_task_namespaces(p);
2115 bad_fork_cleanup_mm:
2116         if (p->mm)
2117                 mmput(p->mm);
2118 bad_fork_cleanup_signal:
2119         if (!(clone_flags & CLONE_THREAD))
2120                 free_signal_struct(p->signal);
2121 bad_fork_cleanup_sighand:
2122         __cleanup_sighand(p->sighand);
2123 bad_fork_cleanup_fs:
2124         exit_fs(p); /* blocking */
2125 bad_fork_cleanup_files:
2126         exit_files(p); /* blocking */
2127 bad_fork_cleanup_semundo:
2128         exit_sem(p);
2129 bad_fork_cleanup_security:
2130         security_task_free(p);
2131 bad_fork_cleanup_audit:
2132         audit_free(p);
2133 bad_fork_cleanup_perf:
2134         perf_event_free_task(p);
2135 bad_fork_cleanup_policy:
2136         lockdep_free_task(p);
2137 #ifdef CONFIG_NUMA
2138         mpol_put(p->mempolicy);
2139 bad_fork_cleanup_threadgroup_lock:
2140 #endif
2141         delayacct_tsk_free(p);
2142 bad_fork_cleanup_count:
2143         atomic_dec(&p->cred->user->processes);
2144         exit_creds(p);
2145 bad_fork_free:
2146         p->state = TASK_DEAD;
2147         put_task_stack(p);
2148         free_task(p);
2149 fork_out:
2150         spin_lock_irq(&current->sighand->siglock);
2151         hlist_del_init(&delayed.node);
2152         spin_unlock_irq(&current->sighand->siglock);
2153         return ERR_PTR(retval);
2154 }
2155
2156 static inline void init_idle_pids(struct task_struct *idle)
2157 {
2158         enum pid_type type;
2159
2160         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2161                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2162                 init_task_pid(idle, type, &init_struct_pid);
2163         }
2164 }
2165
2166 struct task_struct *fork_idle(int cpu)
2167 {
2168         struct task_struct *task;
2169         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2170                             cpu_to_node(cpu));
2171         if (!IS_ERR(task)) {
2172                 init_idle_pids(task);
2173                 init_idle(task, cpu);
2174         }
2175
2176         return task;
2177 }
2178
2179 /*
2180  *  Ok, this is the main fork-routine.
2181  *
2182  * It copies the process, and if successful kick-starts
2183  * it and waits for it to finish using the VM if required.
2184  */
2185 long _do_fork(unsigned long clone_flags,
2186               unsigned long stack_start,
2187               unsigned long stack_size,
2188               int __user *parent_tidptr,
2189               int __user *child_tidptr,
2190               unsigned long tls)
2191 {
2192         struct completion vfork;
2193         struct pid *pid;
2194         struct task_struct *p;
2195         int trace = 0;
2196         long nr;
2197
2198         /*
2199          * Determine whether and which event to report to ptracer.  When
2200          * called from kernel_thread or CLONE_UNTRACED is explicitly
2201          * requested, no event is reported; otherwise, report if the event
2202          * for the type of forking is enabled.
2203          */
2204         if (!(clone_flags & CLONE_UNTRACED)) {
2205                 if (clone_flags & CLONE_VFORK)
2206                         trace = PTRACE_EVENT_VFORK;
2207                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2208                         trace = PTRACE_EVENT_CLONE;
2209                 else
2210                         trace = PTRACE_EVENT_FORK;
2211
2212                 if (likely(!ptrace_event_enabled(current, trace)))
2213                         trace = 0;
2214         }
2215
2216         p = copy_process(clone_flags, stack_start, stack_size,
2217                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2218         add_latent_entropy();
2219
2220         if (IS_ERR(p))
2221                 return PTR_ERR(p);
2222
2223         /*
2224          * Do this prior waking up the new thread - the thread pointer
2225          * might get invalid after that point, if the thread exits quickly.
2226          */
2227         trace_sched_process_fork(current, p);
2228
2229         pid = get_task_pid(p, PIDTYPE_PID);
2230         nr = pid_vnr(pid);
2231
2232         if (clone_flags & CLONE_PARENT_SETTID)
2233                 put_user(nr, parent_tidptr);
2234
2235         if (clone_flags & CLONE_VFORK) {
2236                 p->vfork_done = &vfork;
2237                 init_completion(&vfork);
2238                 get_task_struct(p);
2239         }
2240
2241         wake_up_new_task(p);
2242
2243         /* forking complete and child started to run, tell ptracer */
2244         if (unlikely(trace))
2245                 ptrace_event_pid(trace, pid);
2246
2247         if (clone_flags & CLONE_VFORK) {
2248                 if (!wait_for_vfork_done(p, &vfork))
2249                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2250         }
2251
2252         put_pid(pid);
2253         return nr;
2254 }
2255
2256 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2257 /* For compatibility with architectures that call do_fork directly rather than
2258  * using the syscall entry points below. */
2259 long do_fork(unsigned long clone_flags,
2260               unsigned long stack_start,
2261               unsigned long stack_size,
2262               int __user *parent_tidptr,
2263               int __user *child_tidptr)
2264 {
2265         return _do_fork(clone_flags, stack_start, stack_size,
2266                         parent_tidptr, child_tidptr, 0);
2267 }
2268 #endif
2269
2270 /*
2271  * Create a kernel thread.
2272  */
2273 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2274 {
2275         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2276                 (unsigned long)arg, NULL, NULL, 0);
2277 }
2278
2279 #ifdef __ARCH_WANT_SYS_FORK
2280 SYSCALL_DEFINE0(fork)
2281 {
2282 #ifdef CONFIG_MMU
2283         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2284 #else
2285         /* can not support in nommu mode */
2286         return -EINVAL;
2287 #endif
2288 }
2289 #endif
2290
2291 #ifdef __ARCH_WANT_SYS_VFORK
2292 SYSCALL_DEFINE0(vfork)
2293 {
2294         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2295                         0, NULL, NULL, 0);
2296 }
2297 #endif
2298
2299 #ifdef __ARCH_WANT_SYS_CLONE
2300 #ifdef CONFIG_CLONE_BACKWARDS
2301 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2302                  int __user *, parent_tidptr,
2303                  unsigned long, tls,
2304                  int __user *, child_tidptr)
2305 #elif defined(CONFIG_CLONE_BACKWARDS2)
2306 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2307                  int __user *, parent_tidptr,
2308                  int __user *, child_tidptr,
2309                  unsigned long, tls)
2310 #elif defined(CONFIG_CLONE_BACKWARDS3)
2311 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2312                 int, stack_size,
2313                 int __user *, parent_tidptr,
2314                 int __user *, child_tidptr,
2315                 unsigned long, tls)
2316 #else
2317 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2318                  int __user *, parent_tidptr,
2319                  int __user *, child_tidptr,
2320                  unsigned long, tls)
2321 #endif
2322 {
2323         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2324 }
2325 #endif
2326
2327 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2328 {
2329         struct task_struct *leader, *parent, *child;
2330         int res;
2331
2332         read_lock(&tasklist_lock);
2333         leader = top = top->group_leader;
2334 down:
2335         for_each_thread(leader, parent) {
2336                 list_for_each_entry(child, &parent->children, sibling) {
2337                         res = visitor(child, data);
2338                         if (res) {
2339                                 if (res < 0)
2340                                         goto out;
2341                                 leader = child;
2342                                 goto down;
2343                         }
2344 up:
2345                         ;
2346                 }
2347         }
2348
2349         if (leader != top) {
2350                 child = leader;
2351                 parent = child->real_parent;
2352                 leader = parent->group_leader;
2353                 goto up;
2354         }
2355 out:
2356         read_unlock(&tasklist_lock);
2357 }
2358
2359 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2360 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2361 #endif
2362
2363 static void sighand_ctor(void *data)
2364 {
2365         struct sighand_struct *sighand = data;
2366
2367         spin_lock_init(&sighand->siglock);
2368         init_waitqueue_head(&sighand->signalfd_wqh);
2369 }
2370
2371 void __init proc_caches_init(void)
2372 {
2373         unsigned int mm_size;
2374
2375         sighand_cachep = kmem_cache_create("sighand_cache",
2376                         sizeof(struct sighand_struct), 0,
2377                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2378                         SLAB_ACCOUNT, sighand_ctor);
2379         signal_cachep = kmem_cache_create("signal_cache",
2380                         sizeof(struct signal_struct), 0,
2381                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2382                         NULL);
2383         files_cachep = kmem_cache_create("files_cache",
2384                         sizeof(struct files_struct), 0,
2385                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2386                         NULL);
2387         fs_cachep = kmem_cache_create("fs_cache",
2388                         sizeof(struct fs_struct), 0,
2389                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2390                         NULL);
2391
2392         /*
2393          * The mm_cpumask is located at the end of mm_struct, and is
2394          * dynamically sized based on the maximum CPU number this system
2395          * can have, taking hotplug into account (nr_cpu_ids).
2396          */
2397         mm_size = sizeof(struct mm_struct) + cpumask_size();
2398
2399         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2400                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2401                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2402                         offsetof(struct mm_struct, saved_auxv),
2403                         sizeof_field(struct mm_struct, saved_auxv),
2404                         NULL);
2405         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2406         mmap_init();
2407         nsproxy_cache_init();
2408 }
2409
2410 /*
2411  * Check constraints on flags passed to the unshare system call.
2412  */
2413 static int check_unshare_flags(unsigned long unshare_flags)
2414 {
2415         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2416                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2417                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2418                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2419                 return -EINVAL;
2420         /*
2421          * Not implemented, but pretend it works if there is nothing
2422          * to unshare.  Note that unsharing the address space or the
2423          * signal handlers also need to unshare the signal queues (aka
2424          * CLONE_THREAD).
2425          */
2426         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2427                 if (!thread_group_empty(current))
2428                         return -EINVAL;
2429         }
2430         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2431                 if (atomic_read(&current->sighand->count) > 1)
2432                         return -EINVAL;
2433         }
2434         if (unshare_flags & CLONE_VM) {
2435                 if (!current_is_single_threaded())
2436                         return -EINVAL;
2437         }
2438
2439         return 0;
2440 }
2441
2442 /*
2443  * Unshare the filesystem structure if it is being shared
2444  */
2445 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2446 {
2447         struct fs_struct *fs = current->fs;
2448
2449         if (!(unshare_flags & CLONE_FS) || !fs)
2450                 return 0;
2451
2452         /* don't need lock here; in the worst case we'll do useless copy */
2453         if (fs->users == 1)
2454                 return 0;
2455
2456         *new_fsp = copy_fs_struct(fs);
2457         if (!*new_fsp)
2458                 return -ENOMEM;
2459
2460         return 0;
2461 }
2462
2463 /*
2464  * Unshare file descriptor table if it is being shared
2465  */
2466 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2467 {
2468         struct files_struct *fd = current->files;
2469         int error = 0;
2470
2471         if ((unshare_flags & CLONE_FILES) &&
2472             (fd && atomic_read(&fd->count) > 1)) {
2473                 *new_fdp = dup_fd(fd, &error);
2474                 if (!*new_fdp)
2475                         return error;
2476         }
2477
2478         return 0;
2479 }
2480
2481 /*
2482  * unshare allows a process to 'unshare' part of the process
2483  * context which was originally shared using clone.  copy_*
2484  * functions used by do_fork() cannot be used here directly
2485  * because they modify an inactive task_struct that is being
2486  * constructed. Here we are modifying the current, active,
2487  * task_struct.
2488  */
2489 int ksys_unshare(unsigned long unshare_flags)
2490 {
2491         struct fs_struct *fs, *new_fs = NULL;
2492         struct files_struct *fd, *new_fd = NULL;
2493         struct cred *new_cred = NULL;
2494         struct nsproxy *new_nsproxy = NULL;
2495         int do_sysvsem = 0;
2496         int err;
2497
2498         /*
2499          * If unsharing a user namespace must also unshare the thread group
2500          * and unshare the filesystem root and working directories.
2501          */
2502         if (unshare_flags & CLONE_NEWUSER)
2503                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2504         /*
2505          * If unsharing vm, must also unshare signal handlers.
2506          */
2507         if (unshare_flags & CLONE_VM)
2508                 unshare_flags |= CLONE_SIGHAND;
2509         /*
2510          * If unsharing a signal handlers, must also unshare the signal queues.
2511          */
2512         if (unshare_flags & CLONE_SIGHAND)
2513                 unshare_flags |= CLONE_THREAD;
2514         /*
2515          * If unsharing namespace, must also unshare filesystem information.
2516          */
2517         if (unshare_flags & CLONE_NEWNS)
2518                 unshare_flags |= CLONE_FS;
2519
2520         err = check_unshare_flags(unshare_flags);
2521         if (err)
2522                 goto bad_unshare_out;
2523         /*
2524          * CLONE_NEWIPC must also detach from the undolist: after switching
2525          * to a new ipc namespace, the semaphore arrays from the old
2526          * namespace are unreachable.
2527          */
2528         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2529                 do_sysvsem = 1;
2530         err = unshare_fs(unshare_flags, &new_fs);
2531         if (err)
2532                 goto bad_unshare_out;
2533         err = unshare_fd(unshare_flags, &new_fd);
2534         if (err)
2535                 goto bad_unshare_cleanup_fs;
2536         err = unshare_userns(unshare_flags, &new_cred);
2537         if (err)
2538                 goto bad_unshare_cleanup_fd;
2539         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2540                                          new_cred, new_fs);
2541         if (err)
2542                 goto bad_unshare_cleanup_cred;
2543
2544         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2545                 if (do_sysvsem) {
2546                         /*
2547                          * CLONE_SYSVSEM is equivalent to sys_exit().
2548                          */
2549                         exit_sem(current);
2550                 }
2551                 if (unshare_flags & CLONE_NEWIPC) {
2552                         /* Orphan segments in old ns (see sem above). */
2553                         exit_shm(current);
2554                         shm_init_task(current);
2555                 }
2556
2557                 if (new_nsproxy)
2558                         switch_task_namespaces(current, new_nsproxy);
2559
2560                 task_lock(current);
2561
2562                 if (new_fs) {
2563                         fs = current->fs;
2564                         spin_lock(&fs->lock);
2565                         current->fs = new_fs;
2566                         if (--fs->users)
2567                                 new_fs = NULL;
2568                         else
2569                                 new_fs = fs;
2570                         spin_unlock(&fs->lock);
2571                 }
2572
2573                 if (new_fd) {
2574                         fd = current->files;
2575                         current->files = new_fd;
2576                         new_fd = fd;
2577                 }
2578
2579                 task_unlock(current);
2580
2581                 if (new_cred) {
2582                         /* Install the new user namespace */
2583                         commit_creds(new_cred);
2584                         new_cred = NULL;
2585                 }
2586         }
2587
2588         perf_event_namespaces(current);
2589
2590 bad_unshare_cleanup_cred:
2591         if (new_cred)
2592                 put_cred(new_cred);
2593 bad_unshare_cleanup_fd:
2594         if (new_fd)
2595                 put_files_struct(new_fd);
2596
2597 bad_unshare_cleanup_fs:
2598         if (new_fs)
2599                 free_fs_struct(new_fs);
2600
2601 bad_unshare_out:
2602         return err;
2603 }
2604
2605 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2606 {
2607         return ksys_unshare(unshare_flags);
2608 }
2609
2610 /*
2611  *      Helper to unshare the files of the current task.
2612  *      We don't want to expose copy_files internals to
2613  *      the exec layer of the kernel.
2614  */
2615
2616 int unshare_files(struct files_struct **displaced)
2617 {
2618         struct task_struct *task = current;
2619         struct files_struct *copy = NULL;
2620         int error;
2621
2622         error = unshare_fd(CLONE_FILES, &copy);
2623         if (error || !copy) {
2624                 *displaced = NULL;
2625                 return error;
2626         }
2627         *displaced = task->files;
2628         task_lock(task);
2629         task->files = copy;
2630         task_unlock(task);
2631         return 0;
2632 }
2633
2634 int sysctl_max_threads(struct ctl_table *table, int write,
2635                        void __user *buffer, size_t *lenp, loff_t *ppos)
2636 {
2637         struct ctl_table t;
2638         int ret;
2639         int threads = max_threads;
2640         int min = MIN_THREADS;
2641         int max = MAX_THREADS;
2642
2643         t = *table;
2644         t.data = &threads;
2645         t.extra1 = &min;
2646         t.extra2 = &max;
2647
2648         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2649         if (ret || !write)
2650                 return ret;
2651
2652         set_max_threads(threads);
2653
2654         return 0;
2655 }