xref: /linux/kernel/fork.c (revision 9da8320bb97768e35f2e64fa7642015271d672eb)
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13 
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/mm.h>
32 #include <linux/vmacache.h>
33 #include <linux/nsproxy.h>
34 #include <linux/capability.h>
35 #include <linux/cpu.h>
36 #include <linux/cgroup.h>
37 #include <linux/security.h>
38 #include <linux/hugetlb.h>
39 #include <linux/seccomp.h>
40 #include <linux/swap.h>
41 #include <linux/syscalls.h>
42 #include <linux/jiffies.h>
43 #include <linux/futex.h>
44 #include <linux/compat.h>
45 #include <linux/kthread.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/rcupdate.h>
48 #include <linux/ptrace.h>
49 #include <linux/mount.h>
50 #include <linux/audit.h>
51 #include <linux/memcontrol.h>
52 #include <linux/ftrace.h>
53 #include <linux/proc_fs.h>
54 #include <linux/profile.h>
55 #include <linux/rmap.h>
56 #include <linux/ksm.h>
57 #include <linux/acct.h>
58 #include <linux/tsacct_kern.h>
59 #include <linux/cn_proc.h>
60 #include <linux/freezer.h>
61 #include <linux/delayacct.h>
62 #include <linux/taskstats_kern.h>
63 #include <linux/random.h>
64 #include <linux/tty.h>
65 #include <linux/blkdev.h>
66 #include <linux/fs_struct.h>
67 #include <linux/magic.h>
68 #include <linux/perf_event.h>
69 #include <linux/posix-timers.h>
70 #include <linux/user-return-notifier.h>
71 #include <linux/oom.h>
72 #include <linux/khugepaged.h>
73 #include <linux/signalfd.h>
74 #include <linux/uprobes.h>
75 #include <linux/aio.h>
76 #include <linux/compiler.h>
77 #include <linux/sysctl.h>
78 
79 #include <asm/pgtable.h>
80 #include <asm/pgalloc.h>
81 #include <asm/uaccess.h>
82 #include <asm/mmu_context.h>
83 #include <asm/cacheflush.h>
84 #include <asm/tlbflush.h>
85 
86 #include <trace/events/sched.h>
87 
88 #define CREATE_TRACE_POINTS
89 #include <trace/events/task.h>
90 
91 /*
92  * Minimum number of threads to boot the kernel
93  */
94 #define MIN_THREADS 20
95 
96 /*
97  * Maximum number of threads
98  */
99 #define MAX_THREADS FUTEX_TID_MASK
100 
101 /*
102  * Protected counters by write_lock_irq(&tasklist_lock)
103  */
104 unsigned long total_forks;	/* Handle normal Linux uptimes. */
105 int nr_threads;			/* The idle threads do not count.. */
106 
107 int max_threads;		/* tunable limit on nr_threads */
108 
109 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
110 
111 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
112 
113 #ifdef CONFIG_PROVE_RCU
114 int lockdep_tasklist_lock_is_held(void)
115 {
116 	return lockdep_is_held(&tasklist_lock);
117 }
118 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
119 #endif /* #ifdef CONFIG_PROVE_RCU */
120 
121 int nr_processes(void)
122 {
123 	int cpu;
124 	int total = 0;
125 
126 	for_each_possible_cpu(cpu)
127 		total += per_cpu(process_counts, cpu);
128 
129 	return total;
130 }
131 
132 void __weak arch_release_task_struct(struct task_struct *tsk)
133 {
134 }
135 
136 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
137 static struct kmem_cache *task_struct_cachep;
138 
139 static inline struct task_struct *alloc_task_struct_node(int node)
140 {
141 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
142 }
143 
144 static inline void free_task_struct(struct task_struct *tsk)
145 {
146 	kmem_cache_free(task_struct_cachep, tsk);
147 }
148 #endif
149 
150 void __weak arch_release_thread_info(struct thread_info *ti)
151 {
152 }
153 
154 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
155 
156 /*
157  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
158  * kmemcache based allocator.
159  */
160 # if THREAD_SIZE >= PAGE_SIZE
161 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
162 						  int node)
163 {
164 	struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
165 						  THREAD_SIZE_ORDER);
166 
167 	return page ? page_address(page) : NULL;
168 }
169 
170 static inline void free_thread_info(struct thread_info *ti)
171 {
172 	free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
173 }
174 # else
175 static struct kmem_cache *thread_info_cache;
176 
177 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
178 						  int node)
179 {
180 	return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
181 }
182 
183 static void free_thread_info(struct thread_info *ti)
184 {
185 	kmem_cache_free(thread_info_cache, ti);
186 }
187 
188 void thread_info_cache_init(void)
189 {
190 	thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
191 					      THREAD_SIZE, 0, NULL);
192 	BUG_ON(thread_info_cache == NULL);
193 }
194 # endif
195 #endif
196 
197 /* SLAB cache for signal_struct structures (tsk->signal) */
198 static struct kmem_cache *signal_cachep;
199 
200 /* SLAB cache for sighand_struct structures (tsk->sighand) */
201 struct kmem_cache *sighand_cachep;
202 
203 /* SLAB cache for files_struct structures (tsk->files) */
204 struct kmem_cache *files_cachep;
205 
206 /* SLAB cache for fs_struct structures (tsk->fs) */
207 struct kmem_cache *fs_cachep;
208 
209 /* SLAB cache for vm_area_struct structures */
210 struct kmem_cache *vm_area_cachep;
211 
212 /* SLAB cache for mm_struct structures (tsk->mm) */
213 static struct kmem_cache *mm_cachep;
214 
215 static void account_kernel_stack(struct thread_info *ti, int account)
216 {
217 	struct zone *zone = page_zone(virt_to_page(ti));
218 
219 	mod_zone_page_state(zone, NR_KERNEL_STACK, account);
220 }
221 
222 void free_task(struct task_struct *tsk)
223 {
224 	account_kernel_stack(tsk->stack, -1);
225 	arch_release_thread_info(tsk->stack);
226 	free_thread_info(tsk->stack);
227 	rt_mutex_debug_task_free(tsk);
228 	ftrace_graph_exit_task(tsk);
229 	put_seccomp_filter(tsk);
230 	arch_release_task_struct(tsk);
231 	free_task_struct(tsk);
232 }
233 EXPORT_SYMBOL(free_task);
234 
235 static inline void free_signal_struct(struct signal_struct *sig)
236 {
237 	taskstats_tgid_free(sig);
238 	sched_autogroup_exit(sig);
239 	kmem_cache_free(signal_cachep, sig);
240 }
241 
242 static inline void put_signal_struct(struct signal_struct *sig)
243 {
244 	if (atomic_dec_and_test(&sig->sigcnt))
245 		free_signal_struct(sig);
246 }
247 
248 void __put_task_struct(struct task_struct *tsk)
249 {
250 	WARN_ON(!tsk->exit_state);
251 	WARN_ON(atomic_read(&tsk->usage));
252 	WARN_ON(tsk == current);
253 
254 	cgroup_free(tsk);
255 	task_numa_free(tsk);
256 	security_task_free(tsk);
257 	exit_creds(tsk);
258 	delayacct_tsk_free(tsk);
259 	put_signal_struct(tsk->signal);
260 
261 	if (!profile_handoff_task(tsk))
262 		free_task(tsk);
263 }
264 EXPORT_SYMBOL_GPL(__put_task_struct);
265 
266 void __init __weak arch_task_cache_init(void) { }
267 
268 /*
269  * set_max_threads
270  */
271 static void set_max_threads(unsigned int max_threads_suggested)
272 {
273 	u64 threads;
274 
275 	/*
276 	 * The number of threads shall be limited such that the thread
277 	 * structures may only consume a small part of the available memory.
278 	 */
279 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
280 		threads = MAX_THREADS;
281 	else
282 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
283 				    (u64) THREAD_SIZE * 8UL);
284 
285 	if (threads > max_threads_suggested)
286 		threads = max_threads_suggested;
287 
288 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
289 }
290 
291 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
292 /* Initialized by the architecture: */
293 int arch_task_struct_size __read_mostly;
294 #endif
295 
296 void __init fork_init(void)
297 {
298 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
299 #ifndef ARCH_MIN_TASKALIGN
300 #define ARCH_MIN_TASKALIGN	L1_CACHE_BYTES
301 #endif
302 	/* create a slab on which task_structs can be allocated */
303 	task_struct_cachep =
304 		kmem_cache_create("task_struct", arch_task_struct_size,
305 			ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
306 #endif
307 
308 	/* do the arch specific task caches init */
309 	arch_task_cache_init();
310 
311 	set_max_threads(MAX_THREADS);
312 
313 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
314 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
315 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
316 		init_task.signal->rlim[RLIMIT_NPROC];
317 }
318 
319 int __weak arch_dup_task_struct(struct task_struct *dst,
320 					       struct task_struct *src)
321 {
322 	*dst = *src;
323 	return 0;
324 }
325 
326 void set_task_stack_end_magic(struct task_struct *tsk)
327 {
328 	unsigned long *stackend;
329 
330 	stackend = end_of_stack(tsk);
331 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
332 }
333 
334 static struct task_struct *dup_task_struct(struct task_struct *orig)
335 {
336 	struct task_struct *tsk;
337 	struct thread_info *ti;
338 	int node = tsk_fork_get_node(orig);
339 	int err;
340 
341 	tsk = alloc_task_struct_node(node);
342 	if (!tsk)
343 		return NULL;
344 
345 	ti = alloc_thread_info_node(tsk, node);
346 	if (!ti)
347 		goto free_tsk;
348 
349 	err = arch_dup_task_struct(tsk, orig);
350 	if (err)
351 		goto free_ti;
352 
353 	tsk->stack = ti;
354 #ifdef CONFIG_SECCOMP
355 	/*
356 	 * We must handle setting up seccomp filters once we're under
357 	 * the sighand lock in case orig has changed between now and
358 	 * then. Until then, filter must be NULL to avoid messing up
359 	 * the usage counts on the error path calling free_task.
360 	 */
361 	tsk->seccomp.filter = NULL;
362 #endif
363 
364 	setup_thread_stack(tsk, orig);
365 	clear_user_return_notifier(tsk);
366 	clear_tsk_need_resched(tsk);
367 	set_task_stack_end_magic(tsk);
368 
369 #ifdef CONFIG_CC_STACKPROTECTOR
370 	tsk->stack_canary = get_random_int();
371 #endif
372 
373 	/*
374 	 * One for us, one for whoever does the "release_task()" (usually
375 	 * parent)
376 	 */
377 	atomic_set(&tsk->usage, 2);
378 #ifdef CONFIG_BLK_DEV_IO_TRACE
379 	tsk->btrace_seq = 0;
380 #endif
381 	tsk->splice_pipe = NULL;
382 	tsk->task_frag.page = NULL;
383 	tsk->wake_q.next = NULL;
384 
385 	account_kernel_stack(ti, 1);
386 
387 	return tsk;
388 
389 free_ti:
390 	free_thread_info(ti);
391 free_tsk:
392 	free_task_struct(tsk);
393 	return NULL;
394 }
395 
396 #ifdef CONFIG_MMU
397 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
398 {
399 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
400 	struct rb_node **rb_link, *rb_parent;
401 	int retval;
402 	unsigned long charge;
403 
404 	uprobe_start_dup_mmap();
405 	down_write(&oldmm->mmap_sem);
406 	flush_cache_dup_mm(oldmm);
407 	uprobe_dup_mmap(oldmm, mm);
408 	/*
409 	 * Not linked in yet - no deadlock potential:
410 	 */
411 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
412 
413 	/* No ordering required: file already has been exposed. */
414 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
415 
416 	mm->total_vm = oldmm->total_vm;
417 	mm->shared_vm = oldmm->shared_vm;
418 	mm->exec_vm = oldmm->exec_vm;
419 	mm->stack_vm = oldmm->stack_vm;
420 
421 	rb_link = &mm->mm_rb.rb_node;
422 	rb_parent = NULL;
423 	pprev = &mm->mmap;
424 	retval = ksm_fork(mm, oldmm);
425 	if (retval)
426 		goto out;
427 	retval = khugepaged_fork(mm, oldmm);
428 	if (retval)
429 		goto out;
430 
431 	prev = NULL;
432 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
433 		struct file *file;
434 
435 		if (mpnt->vm_flags & VM_DONTCOPY) {
436 			vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
437 							-vma_pages(mpnt));
438 			continue;
439 		}
440 		charge = 0;
441 		if (mpnt->vm_flags & VM_ACCOUNT) {
442 			unsigned long len = vma_pages(mpnt);
443 
444 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
445 				goto fail_nomem;
446 			charge = len;
447 		}
448 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
449 		if (!tmp)
450 			goto fail_nomem;
451 		*tmp = *mpnt;
452 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
453 		retval = vma_dup_policy(mpnt, tmp);
454 		if (retval)
455 			goto fail_nomem_policy;
456 		tmp->vm_mm = mm;
457 		if (anon_vma_fork(tmp, mpnt))
458 			goto fail_nomem_anon_vma_fork;
459 		tmp->vm_flags &=
460 			~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
461 		tmp->vm_next = tmp->vm_prev = NULL;
462 		tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
463 		file = tmp->vm_file;
464 		if (file) {
465 			struct inode *inode = file_inode(file);
466 			struct address_space *mapping = file->f_mapping;
467 
468 			get_file(file);
469 			if (tmp->vm_flags & VM_DENYWRITE)
470 				atomic_dec(&inode->i_writecount);
471 			i_mmap_lock_write(mapping);
472 			if (tmp->vm_flags & VM_SHARED)
473 				atomic_inc(&mapping->i_mmap_writable);
474 			flush_dcache_mmap_lock(mapping);
475 			/* insert tmp into the share list, just after mpnt */
476 			vma_interval_tree_insert_after(tmp, mpnt,
477 					&mapping->i_mmap);
478 			flush_dcache_mmap_unlock(mapping);
479 			i_mmap_unlock_write(mapping);
480 		}
481 
482 		/*
483 		 * Clear hugetlb-related page reserves for children. This only
484 		 * affects MAP_PRIVATE mappings. Faults generated by the child
485 		 * are not guaranteed to succeed, even if read-only
486 		 */
487 		if (is_vm_hugetlb_page(tmp))
488 			reset_vma_resv_huge_pages(tmp);
489 
490 		/*
491 		 * Link in the new vma and copy the page table entries.
492 		 */
493 		*pprev = tmp;
494 		pprev = &tmp->vm_next;
495 		tmp->vm_prev = prev;
496 		prev = tmp;
497 
498 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
499 		rb_link = &tmp->vm_rb.rb_right;
500 		rb_parent = &tmp->vm_rb;
501 
502 		mm->map_count++;
503 		retval = copy_page_range(mm, oldmm, mpnt);
504 
505 		if (tmp->vm_ops && tmp->vm_ops->open)
506 			tmp->vm_ops->open(tmp);
507 
508 		if (retval)
509 			goto out;
510 	}
511 	/* a new mm has just been created */
512 	arch_dup_mmap(oldmm, mm);
513 	retval = 0;
514 out:
515 	up_write(&mm->mmap_sem);
516 	flush_tlb_mm(oldmm);
517 	up_write(&oldmm->mmap_sem);
518 	uprobe_end_dup_mmap();
519 	return retval;
520 fail_nomem_anon_vma_fork:
521 	mpol_put(vma_policy(tmp));
522 fail_nomem_policy:
523 	kmem_cache_free(vm_area_cachep, tmp);
524 fail_nomem:
525 	retval = -ENOMEM;
526 	vm_unacct_memory(charge);
527 	goto out;
528 }
529 
530 static inline int mm_alloc_pgd(struct mm_struct *mm)
531 {
532 	mm->pgd = pgd_alloc(mm);
533 	if (unlikely(!mm->pgd))
534 		return -ENOMEM;
535 	return 0;
536 }
537 
538 static inline void mm_free_pgd(struct mm_struct *mm)
539 {
540 	pgd_free(mm, mm->pgd);
541 }
542 #else
543 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
544 {
545 	down_write(&oldmm->mmap_sem);
546 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
547 	up_write(&oldmm->mmap_sem);
548 	return 0;
549 }
550 #define mm_alloc_pgd(mm)	(0)
551 #define mm_free_pgd(mm)
552 #endif /* CONFIG_MMU */
553 
554 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
555 
556 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
557 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
558 
559 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
560 
561 static int __init coredump_filter_setup(char *s)
562 {
563 	default_dump_filter =
564 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
565 		MMF_DUMP_FILTER_MASK;
566 	return 1;
567 }
568 
569 __setup("coredump_filter=", coredump_filter_setup);
570 
571 #include <linux/init_task.h>
572 
573 static void mm_init_aio(struct mm_struct *mm)
574 {
575 #ifdef CONFIG_AIO
576 	spin_lock_init(&mm->ioctx_lock);
577 	mm->ioctx_table = NULL;
578 #endif
579 }
580 
581 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
582 {
583 #ifdef CONFIG_MEMCG
584 	mm->owner = p;
585 #endif
586 }
587 
588 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
589 {
590 	mm->mmap = NULL;
591 	mm->mm_rb = RB_ROOT;
592 	mm->vmacache_seqnum = 0;
593 	atomic_set(&mm->mm_users, 1);
594 	atomic_set(&mm->mm_count, 1);
595 	init_rwsem(&mm->mmap_sem);
596 	INIT_LIST_HEAD(&mm->mmlist);
597 	mm->core_state = NULL;
598 	atomic_long_set(&mm->nr_ptes, 0);
599 	mm_nr_pmds_init(mm);
600 	mm->map_count = 0;
601 	mm->locked_vm = 0;
602 	mm->pinned_vm = 0;
603 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
604 	spin_lock_init(&mm->page_table_lock);
605 	mm_init_cpumask(mm);
606 	mm_init_aio(mm);
607 	mm_init_owner(mm, p);
608 	mmu_notifier_mm_init(mm);
609 	clear_tlb_flush_pending(mm);
610 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
611 	mm->pmd_huge_pte = NULL;
612 #endif
613 
614 	if (current->mm) {
615 		mm->flags = current->mm->flags & MMF_INIT_MASK;
616 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
617 	} else {
618 		mm->flags = default_dump_filter;
619 		mm->def_flags = 0;
620 	}
621 
622 	if (mm_alloc_pgd(mm))
623 		goto fail_nopgd;
624 
625 	if (init_new_context(p, mm))
626 		goto fail_nocontext;
627 
628 	return mm;
629 
630 fail_nocontext:
631 	mm_free_pgd(mm);
632 fail_nopgd:
633 	free_mm(mm);
634 	return NULL;
635 }
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 (atomic_long_read(&mm->nr_ptes))
650 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
651 				atomic_long_read(&mm->nr_ptes));
652 	if (mm_nr_pmds(mm))
653 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
654 				mm_nr_pmds(mm));
655 
656 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
657 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
658 #endif
659 }
660 
661 /*
662  * Allocate and initialize an mm_struct.
663  */
664 struct mm_struct *mm_alloc(void)
665 {
666 	struct mm_struct *mm;
667 
668 	mm = allocate_mm();
669 	if (!mm)
670 		return NULL;
671 
672 	memset(mm, 0, sizeof(*mm));
673 	return mm_init(mm, current);
674 }
675 
676 /*
677  * Called when the last reference to the mm
678  * is dropped: either by a lazy thread or by
679  * mmput. Free the page directory and the mm.
680  */
681 void __mmdrop(struct mm_struct *mm)
682 {
683 	BUG_ON(mm == &init_mm);
684 	mm_free_pgd(mm);
685 	destroy_context(mm);
686 	mmu_notifier_mm_destroy(mm);
687 	check_mm(mm);
688 	free_mm(mm);
689 }
690 EXPORT_SYMBOL_GPL(__mmdrop);
691 
692 /*
693  * Decrement the use count and release all resources for an mm.
694  */
695 void mmput(struct mm_struct *mm)
696 {
697 	might_sleep();
698 
699 	if (atomic_dec_and_test(&mm->mm_users)) {
700 		uprobe_clear_state(mm);
701 		exit_aio(mm);
702 		ksm_exit(mm);
703 		khugepaged_exit(mm); /* must run before exit_mmap */
704 		exit_mmap(mm);
705 		set_mm_exe_file(mm, NULL);
706 		if (!list_empty(&mm->mmlist)) {
707 			spin_lock(&mmlist_lock);
708 			list_del(&mm->mmlist);
709 			spin_unlock(&mmlist_lock);
710 		}
711 		if (mm->binfmt)
712 			module_put(mm->binfmt->module);
713 		mmdrop(mm);
714 	}
715 }
716 EXPORT_SYMBOL_GPL(mmput);
717 
718 /**
719  * set_mm_exe_file - change a reference to the mm's executable file
720  *
721  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
722  *
723  * Main users are mmput() and sys_execve(). Callers prevent concurrent
724  * invocations: in mmput() nobody alive left, in execve task is single
725  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
726  * mm->exe_file, but does so without using set_mm_exe_file() in order
727  * to do avoid the need for any locks.
728  */
729 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
730 {
731 	struct file *old_exe_file;
732 
733 	/*
734 	 * It is safe to dereference the exe_file without RCU as
735 	 * this function is only called if nobody else can access
736 	 * this mm -- see comment above for justification.
737 	 */
738 	old_exe_file = rcu_dereference_raw(mm->exe_file);
739 
740 	if (new_exe_file)
741 		get_file(new_exe_file);
742 	rcu_assign_pointer(mm->exe_file, new_exe_file);
743 	if (old_exe_file)
744 		fput(old_exe_file);
745 }
746 
747 /**
748  * get_mm_exe_file - acquire a reference to the mm's executable file
749  *
750  * Returns %NULL if mm has no associated executable file.
751  * User must release file via fput().
752  */
753 struct file *get_mm_exe_file(struct mm_struct *mm)
754 {
755 	struct file *exe_file;
756 
757 	rcu_read_lock();
758 	exe_file = rcu_dereference(mm->exe_file);
759 	if (exe_file && !get_file_rcu(exe_file))
760 		exe_file = NULL;
761 	rcu_read_unlock();
762 	return exe_file;
763 }
764 EXPORT_SYMBOL(get_mm_exe_file);
765 
766 /**
767  * get_task_mm - acquire a reference to the task's mm
768  *
769  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
770  * this kernel workthread has transiently adopted a user mm with use_mm,
771  * to do its AIO) is not set and if so returns a reference to it, after
772  * bumping up the use count.  User must release the mm via mmput()
773  * after use.  Typically used by /proc and ptrace.
774  */
775 struct mm_struct *get_task_mm(struct task_struct *task)
776 {
777 	struct mm_struct *mm;
778 
779 	task_lock(task);
780 	mm = task->mm;
781 	if (mm) {
782 		if (task->flags & PF_KTHREAD)
783 			mm = NULL;
784 		else
785 			atomic_inc(&mm->mm_users);
786 	}
787 	task_unlock(task);
788 	return mm;
789 }
790 EXPORT_SYMBOL_GPL(get_task_mm);
791 
792 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
793 {
794 	struct mm_struct *mm;
795 	int err;
796 
797 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
798 	if (err)
799 		return ERR_PTR(err);
800 
801 	mm = get_task_mm(task);
802 	if (mm && mm != current->mm &&
803 			!ptrace_may_access(task, mode)) {
804 		mmput(mm);
805 		mm = ERR_PTR(-EACCES);
806 	}
807 	mutex_unlock(&task->signal->cred_guard_mutex);
808 
809 	return mm;
810 }
811 
812 static void complete_vfork_done(struct task_struct *tsk)
813 {
814 	struct completion *vfork;
815 
816 	task_lock(tsk);
817 	vfork = tsk->vfork_done;
818 	if (likely(vfork)) {
819 		tsk->vfork_done = NULL;
820 		complete(vfork);
821 	}
822 	task_unlock(tsk);
823 }
824 
825 static int wait_for_vfork_done(struct task_struct *child,
826 				struct completion *vfork)
827 {
828 	int killed;
829 
830 	freezer_do_not_count();
831 	killed = wait_for_completion_killable(vfork);
832 	freezer_count();
833 
834 	if (killed) {
835 		task_lock(child);
836 		child->vfork_done = NULL;
837 		task_unlock(child);
838 	}
839 
840 	put_task_struct(child);
841 	return killed;
842 }
843 
844 /* Please note the differences between mmput and mm_release.
845  * mmput is called whenever we stop holding onto a mm_struct,
846  * error success whatever.
847  *
848  * mm_release is called after a mm_struct has been removed
849  * from the current process.
850  *
851  * This difference is important for error handling, when we
852  * only half set up a mm_struct for a new process and need to restore
853  * the old one.  Because we mmput the new mm_struct before
854  * restoring the old one. . .
855  * Eric Biederman 10 January 1998
856  */
857 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
858 {
859 	/* Get rid of any futexes when releasing the mm */
860 #ifdef CONFIG_FUTEX
861 	if (unlikely(tsk->robust_list)) {
862 		exit_robust_list(tsk);
863 		tsk->robust_list = NULL;
864 	}
865 #ifdef CONFIG_COMPAT
866 	if (unlikely(tsk->compat_robust_list)) {
867 		compat_exit_robust_list(tsk);
868 		tsk->compat_robust_list = NULL;
869 	}
870 #endif
871 	if (unlikely(!list_empty(&tsk->pi_state_list)))
872 		exit_pi_state_list(tsk);
873 #endif
874 
875 	uprobe_free_utask(tsk);
876 
877 	/* Get rid of any cached register state */
878 	deactivate_mm(tsk, mm);
879 
880 	/*
881 	 * If we're exiting normally, clear a user-space tid field if
882 	 * requested.  We leave this alone when dying by signal, to leave
883 	 * the value intact in a core dump, and to save the unnecessary
884 	 * trouble, say, a killed vfork parent shouldn't touch this mm.
885 	 * Userland only wants this done for a sys_exit.
886 	 */
887 	if (tsk->clear_child_tid) {
888 		if (!(tsk->flags & PF_SIGNALED) &&
889 		    atomic_read(&mm->mm_users) > 1) {
890 			/*
891 			 * We don't check the error code - if userspace has
892 			 * not set up a proper pointer then tough luck.
893 			 */
894 			put_user(0, tsk->clear_child_tid);
895 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
896 					1, NULL, NULL, 0);
897 		}
898 		tsk->clear_child_tid = NULL;
899 	}
900 
901 	/*
902 	 * All done, finally we can wake up parent and return this mm to him.
903 	 * Also kthread_stop() uses this completion for synchronization.
904 	 */
905 	if (tsk->vfork_done)
906 		complete_vfork_done(tsk);
907 }
908 
909 /*
910  * Allocate a new mm structure and copy contents from the
911  * mm structure of the passed in task structure.
912  */
913 static struct mm_struct *dup_mm(struct task_struct *tsk)
914 {
915 	struct mm_struct *mm, *oldmm = current->mm;
916 	int err;
917 
918 	mm = allocate_mm();
919 	if (!mm)
920 		goto fail_nomem;
921 
922 	memcpy(mm, oldmm, sizeof(*mm));
923 
924 	if (!mm_init(mm, tsk))
925 		goto fail_nomem;
926 
927 	err = dup_mmap(mm, oldmm);
928 	if (err)
929 		goto free_pt;
930 
931 	mm->hiwater_rss = get_mm_rss(mm);
932 	mm->hiwater_vm = mm->total_vm;
933 
934 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
935 		goto free_pt;
936 
937 	return mm;
938 
939 free_pt:
940 	/* don't put binfmt in mmput, we haven't got module yet */
941 	mm->binfmt = NULL;
942 	mmput(mm);
943 
944 fail_nomem:
945 	return NULL;
946 }
947 
948 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
949 {
950 	struct mm_struct *mm, *oldmm;
951 	int retval;
952 
953 	tsk->min_flt = tsk->maj_flt = 0;
954 	tsk->nvcsw = tsk->nivcsw = 0;
955 #ifdef CONFIG_DETECT_HUNG_TASK
956 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
957 #endif
958 
959 	tsk->mm = NULL;
960 	tsk->active_mm = NULL;
961 
962 	/*
963 	 * Are we cloning a kernel thread?
964 	 *
965 	 * We need to steal a active VM for that..
966 	 */
967 	oldmm = current->mm;
968 	if (!oldmm)
969 		return 0;
970 
971 	/* initialize the new vmacache entries */
972 	vmacache_flush(tsk);
973 
974 	if (clone_flags & CLONE_VM) {
975 		atomic_inc(&oldmm->mm_users);
976 		mm = oldmm;
977 		goto good_mm;
978 	}
979 
980 	retval = -ENOMEM;
981 	mm = dup_mm(tsk);
982 	if (!mm)
983 		goto fail_nomem;
984 
985 good_mm:
986 	tsk->mm = mm;
987 	tsk->active_mm = mm;
988 	return 0;
989 
990 fail_nomem:
991 	return retval;
992 }
993 
994 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
995 {
996 	struct fs_struct *fs = current->fs;
997 	if (clone_flags & CLONE_FS) {
998 		/* tsk->fs is already what we want */
999 		spin_lock(&fs->lock);
1000 		if (fs->in_exec) {
1001 			spin_unlock(&fs->lock);
1002 			return -EAGAIN;
1003 		}
1004 		fs->users++;
1005 		spin_unlock(&fs->lock);
1006 		return 0;
1007 	}
1008 	tsk->fs = copy_fs_struct(fs);
1009 	if (!tsk->fs)
1010 		return -ENOMEM;
1011 	return 0;
1012 }
1013 
1014 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1015 {
1016 	struct files_struct *oldf, *newf;
1017 	int error = 0;
1018 
1019 	/*
1020 	 * A background process may not have any files ...
1021 	 */
1022 	oldf = current->files;
1023 	if (!oldf)
1024 		goto out;
1025 
1026 	if (clone_flags & CLONE_FILES) {
1027 		atomic_inc(&oldf->count);
1028 		goto out;
1029 	}
1030 
1031 	newf = dup_fd(oldf, &error);
1032 	if (!newf)
1033 		goto out;
1034 
1035 	tsk->files = newf;
1036 	error = 0;
1037 out:
1038 	return error;
1039 }
1040 
1041 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1042 {
1043 #ifdef CONFIG_BLOCK
1044 	struct io_context *ioc = current->io_context;
1045 	struct io_context *new_ioc;
1046 
1047 	if (!ioc)
1048 		return 0;
1049 	/*
1050 	 * Share io context with parent, if CLONE_IO is set
1051 	 */
1052 	if (clone_flags & CLONE_IO) {
1053 		ioc_task_link(ioc);
1054 		tsk->io_context = ioc;
1055 	} else if (ioprio_valid(ioc->ioprio)) {
1056 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1057 		if (unlikely(!new_ioc))
1058 			return -ENOMEM;
1059 
1060 		new_ioc->ioprio = ioc->ioprio;
1061 		put_io_context(new_ioc);
1062 	}
1063 #endif
1064 	return 0;
1065 }
1066 
1067 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1068 {
1069 	struct sighand_struct *sig;
1070 
1071 	if (clone_flags & CLONE_SIGHAND) {
1072 		atomic_inc(&current->sighand->count);
1073 		return 0;
1074 	}
1075 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1076 	rcu_assign_pointer(tsk->sighand, sig);
1077 	if (!sig)
1078 		return -ENOMEM;
1079 
1080 	atomic_set(&sig->count, 1);
1081 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1082 	return 0;
1083 }
1084 
1085 void __cleanup_sighand(struct sighand_struct *sighand)
1086 {
1087 	if (atomic_dec_and_test(&sighand->count)) {
1088 		signalfd_cleanup(sighand);
1089 		/*
1090 		 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1091 		 * without an RCU grace period, see __lock_task_sighand().
1092 		 */
1093 		kmem_cache_free(sighand_cachep, sighand);
1094 	}
1095 }
1096 
1097 /*
1098  * Initialize POSIX timer handling for a thread group.
1099  */
1100 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1101 {
1102 	unsigned long cpu_limit;
1103 
1104 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1105 	if (cpu_limit != RLIM_INFINITY) {
1106 		sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1107 		sig->cputimer.running = true;
1108 	}
1109 
1110 	/* The timer lists. */
1111 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1112 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1113 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1114 }
1115 
1116 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1117 {
1118 	struct signal_struct *sig;
1119 
1120 	if (clone_flags & CLONE_THREAD)
1121 		return 0;
1122 
1123 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1124 	tsk->signal = sig;
1125 	if (!sig)
1126 		return -ENOMEM;
1127 
1128 	sig->nr_threads = 1;
1129 	atomic_set(&sig->live, 1);
1130 	atomic_set(&sig->sigcnt, 1);
1131 
1132 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1133 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1134 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1135 
1136 	init_waitqueue_head(&sig->wait_chldexit);
1137 	sig->curr_target = tsk;
1138 	init_sigpending(&sig->shared_pending);
1139 	INIT_LIST_HEAD(&sig->posix_timers);
1140 	seqlock_init(&sig->stats_lock);
1141 	prev_cputime_init(&sig->prev_cputime);
1142 
1143 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1144 	sig->real_timer.function = it_real_fn;
1145 
1146 	task_lock(current->group_leader);
1147 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1148 	task_unlock(current->group_leader);
1149 
1150 	posix_cpu_timers_init_group(sig);
1151 
1152 	tty_audit_fork(sig);
1153 	sched_autogroup_fork(sig);
1154 
1155 	sig->oom_score_adj = current->signal->oom_score_adj;
1156 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1157 
1158 	sig->has_child_subreaper = current->signal->has_child_subreaper ||
1159 				   current->signal->is_child_subreaper;
1160 
1161 	mutex_init(&sig->cred_guard_mutex);
1162 
1163 	return 0;
1164 }
1165 
1166 static void copy_seccomp(struct task_struct *p)
1167 {
1168 #ifdef CONFIG_SECCOMP
1169 	/*
1170 	 * Must be called with sighand->lock held, which is common to
1171 	 * all threads in the group. Holding cred_guard_mutex is not
1172 	 * needed because this new task is not yet running and cannot
1173 	 * be racing exec.
1174 	 */
1175 	assert_spin_locked(&current->sighand->siglock);
1176 
1177 	/* Ref-count the new filter user, and assign it. */
1178 	get_seccomp_filter(current);
1179 	p->seccomp = current->seccomp;
1180 
1181 	/*
1182 	 * Explicitly enable no_new_privs here in case it got set
1183 	 * between the task_struct being duplicated and holding the
1184 	 * sighand lock. The seccomp state and nnp must be in sync.
1185 	 */
1186 	if (task_no_new_privs(current))
1187 		task_set_no_new_privs(p);
1188 
1189 	/*
1190 	 * If the parent gained a seccomp mode after copying thread
1191 	 * flags and between before we held the sighand lock, we have
1192 	 * to manually enable the seccomp thread flag here.
1193 	 */
1194 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1195 		set_tsk_thread_flag(p, TIF_SECCOMP);
1196 #endif
1197 }
1198 
1199 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1200 {
1201 	current->clear_child_tid = tidptr;
1202 
1203 	return task_pid_vnr(current);
1204 }
1205 
1206 static void rt_mutex_init_task(struct task_struct *p)
1207 {
1208 	raw_spin_lock_init(&p->pi_lock);
1209 #ifdef CONFIG_RT_MUTEXES
1210 	p->pi_waiters = RB_ROOT;
1211 	p->pi_waiters_leftmost = NULL;
1212 	p->pi_blocked_on = NULL;
1213 #endif
1214 }
1215 
1216 /*
1217  * Initialize POSIX timer handling for a single task.
1218  */
1219 static void posix_cpu_timers_init(struct task_struct *tsk)
1220 {
1221 	tsk->cputime_expires.prof_exp = 0;
1222 	tsk->cputime_expires.virt_exp = 0;
1223 	tsk->cputime_expires.sched_exp = 0;
1224 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1225 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1226 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1227 }
1228 
1229 static inline void
1230 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1231 {
1232 	 task->pids[type].pid = pid;
1233 }
1234 
1235 /*
1236  * This creates a new process as a copy of the old one,
1237  * but does not actually start it yet.
1238  *
1239  * It copies the registers, and all the appropriate
1240  * parts of the process environment (as per the clone
1241  * flags). The actual kick-off is left to the caller.
1242  */
1243 static struct task_struct *copy_process(unsigned long clone_flags,
1244 					unsigned long stack_start,
1245 					unsigned long stack_size,
1246 					int __user *child_tidptr,
1247 					struct pid *pid,
1248 					int trace,
1249 					unsigned long tls)
1250 {
1251 	int retval;
1252 	struct task_struct *p;
1253 	void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
1254 
1255 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1256 		return ERR_PTR(-EINVAL);
1257 
1258 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1259 		return ERR_PTR(-EINVAL);
1260 
1261 	/*
1262 	 * Thread groups must share signals as well, and detached threads
1263 	 * can only be started up within the thread group.
1264 	 */
1265 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1266 		return ERR_PTR(-EINVAL);
1267 
1268 	/*
1269 	 * Shared signal handlers imply shared VM. By way of the above,
1270 	 * thread groups also imply shared VM. Blocking this case allows
1271 	 * for various simplifications in other code.
1272 	 */
1273 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1274 		return ERR_PTR(-EINVAL);
1275 
1276 	/*
1277 	 * Siblings of global init remain as zombies on exit since they are
1278 	 * not reaped by their parent (swapper). To solve this and to avoid
1279 	 * multi-rooted process trees, prevent global and container-inits
1280 	 * from creating siblings.
1281 	 */
1282 	if ((clone_flags & CLONE_PARENT) &&
1283 				current->signal->flags & SIGNAL_UNKILLABLE)
1284 		return ERR_PTR(-EINVAL);
1285 
1286 	/*
1287 	 * If the new process will be in a different pid or user namespace
1288 	 * do not allow it to share a thread group with the forking task.
1289 	 */
1290 	if (clone_flags & CLONE_THREAD) {
1291 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1292 		    (task_active_pid_ns(current) !=
1293 				current->nsproxy->pid_ns_for_children))
1294 			return ERR_PTR(-EINVAL);
1295 	}
1296 
1297 	retval = security_task_create(clone_flags);
1298 	if (retval)
1299 		goto fork_out;
1300 
1301 	retval = -ENOMEM;
1302 	p = dup_task_struct(current);
1303 	if (!p)
1304 		goto fork_out;
1305 
1306 	ftrace_graph_init_task(p);
1307 
1308 	rt_mutex_init_task(p);
1309 
1310 #ifdef CONFIG_PROVE_LOCKING
1311 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1312 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1313 #endif
1314 	retval = -EAGAIN;
1315 	if (atomic_read(&p->real_cred->user->processes) >=
1316 			task_rlimit(p, RLIMIT_NPROC)) {
1317 		if (p->real_cred->user != INIT_USER &&
1318 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1319 			goto bad_fork_free;
1320 	}
1321 	current->flags &= ~PF_NPROC_EXCEEDED;
1322 
1323 	retval = copy_creds(p, clone_flags);
1324 	if (retval < 0)
1325 		goto bad_fork_free;
1326 
1327 	/*
1328 	 * If multiple threads are within copy_process(), then this check
1329 	 * triggers too late. This doesn't hurt, the check is only there
1330 	 * to stop root fork bombs.
1331 	 */
1332 	retval = -EAGAIN;
1333 	if (nr_threads >= max_threads)
1334 		goto bad_fork_cleanup_count;
1335 
1336 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1337 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1338 	p->flags |= PF_FORKNOEXEC;
1339 	INIT_LIST_HEAD(&p->children);
1340 	INIT_LIST_HEAD(&p->sibling);
1341 	rcu_copy_process(p);
1342 	p->vfork_done = NULL;
1343 	spin_lock_init(&p->alloc_lock);
1344 
1345 	init_sigpending(&p->pending);
1346 
1347 	p->utime = p->stime = p->gtime = 0;
1348 	p->utimescaled = p->stimescaled = 0;
1349 	prev_cputime_init(&p->prev_cputime);
1350 
1351 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1352 	seqlock_init(&p->vtime_seqlock);
1353 	p->vtime_snap = 0;
1354 	p->vtime_snap_whence = VTIME_SLEEPING;
1355 #endif
1356 
1357 #if defined(SPLIT_RSS_COUNTING)
1358 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1359 #endif
1360 
1361 	p->default_timer_slack_ns = current->timer_slack_ns;
1362 
1363 	task_io_accounting_init(&p->ioac);
1364 	acct_clear_integrals(p);
1365 
1366 	posix_cpu_timers_init(p);
1367 
1368 	p->start_time = ktime_get_ns();
1369 	p->real_start_time = ktime_get_boot_ns();
1370 	p->io_context = NULL;
1371 	p->audit_context = NULL;
1372 	threadgroup_change_begin(current);
1373 	cgroup_fork(p);
1374 #ifdef CONFIG_NUMA
1375 	p->mempolicy = mpol_dup(p->mempolicy);
1376 	if (IS_ERR(p->mempolicy)) {
1377 		retval = PTR_ERR(p->mempolicy);
1378 		p->mempolicy = NULL;
1379 		goto bad_fork_cleanup_threadgroup_lock;
1380 	}
1381 #endif
1382 #ifdef CONFIG_CPUSETS
1383 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1384 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1385 	seqcount_init(&p->mems_allowed_seq);
1386 #endif
1387 #ifdef CONFIG_TRACE_IRQFLAGS
1388 	p->irq_events = 0;
1389 	p->hardirqs_enabled = 0;
1390 	p->hardirq_enable_ip = 0;
1391 	p->hardirq_enable_event = 0;
1392 	p->hardirq_disable_ip = _THIS_IP_;
1393 	p->hardirq_disable_event = 0;
1394 	p->softirqs_enabled = 1;
1395 	p->softirq_enable_ip = _THIS_IP_;
1396 	p->softirq_enable_event = 0;
1397 	p->softirq_disable_ip = 0;
1398 	p->softirq_disable_event = 0;
1399 	p->hardirq_context = 0;
1400 	p->softirq_context = 0;
1401 #endif
1402 
1403 	p->pagefault_disabled = 0;
1404 
1405 #ifdef CONFIG_LOCKDEP
1406 	p->lockdep_depth = 0; /* no locks held yet */
1407 	p->curr_chain_key = 0;
1408 	p->lockdep_recursion = 0;
1409 #endif
1410 
1411 #ifdef CONFIG_DEBUG_MUTEXES
1412 	p->blocked_on = NULL; /* not blocked yet */
1413 #endif
1414 #ifdef CONFIG_BCACHE
1415 	p->sequential_io	= 0;
1416 	p->sequential_io_avg	= 0;
1417 #endif
1418 
1419 	/* Perform scheduler related setup. Assign this task to a CPU. */
1420 	retval = sched_fork(clone_flags, p);
1421 	if (retval)
1422 		goto bad_fork_cleanup_policy;
1423 
1424 	retval = perf_event_init_task(p);
1425 	if (retval)
1426 		goto bad_fork_cleanup_policy;
1427 	retval = audit_alloc(p);
1428 	if (retval)
1429 		goto bad_fork_cleanup_perf;
1430 	/* copy all the process information */
1431 	shm_init_task(p);
1432 	retval = copy_semundo(clone_flags, p);
1433 	if (retval)
1434 		goto bad_fork_cleanup_audit;
1435 	retval = copy_files(clone_flags, p);
1436 	if (retval)
1437 		goto bad_fork_cleanup_semundo;
1438 	retval = copy_fs(clone_flags, p);
1439 	if (retval)
1440 		goto bad_fork_cleanup_files;
1441 	retval = copy_sighand(clone_flags, p);
1442 	if (retval)
1443 		goto bad_fork_cleanup_fs;
1444 	retval = copy_signal(clone_flags, p);
1445 	if (retval)
1446 		goto bad_fork_cleanup_sighand;
1447 	retval = copy_mm(clone_flags, p);
1448 	if (retval)
1449 		goto bad_fork_cleanup_signal;
1450 	retval = copy_namespaces(clone_flags, p);
1451 	if (retval)
1452 		goto bad_fork_cleanup_mm;
1453 	retval = copy_io(clone_flags, p);
1454 	if (retval)
1455 		goto bad_fork_cleanup_namespaces;
1456 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1457 	if (retval)
1458 		goto bad_fork_cleanup_io;
1459 
1460 	if (pid != &init_struct_pid) {
1461 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1462 		if (IS_ERR(pid)) {
1463 			retval = PTR_ERR(pid);
1464 			goto bad_fork_cleanup_io;
1465 		}
1466 	}
1467 
1468 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1469 	/*
1470 	 * Clear TID on mm_release()?
1471 	 */
1472 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1473 #ifdef CONFIG_BLOCK
1474 	p->plug = NULL;
1475 #endif
1476 #ifdef CONFIG_FUTEX
1477 	p->robust_list = NULL;
1478 #ifdef CONFIG_COMPAT
1479 	p->compat_robust_list = NULL;
1480 #endif
1481 	INIT_LIST_HEAD(&p->pi_state_list);
1482 	p->pi_state_cache = NULL;
1483 #endif
1484 	/*
1485 	 * sigaltstack should be cleared when sharing the same VM
1486 	 */
1487 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1488 		p->sas_ss_sp = p->sas_ss_size = 0;
1489 
1490 	/*
1491 	 * Syscall tracing and stepping should be turned off in the
1492 	 * child regardless of CLONE_PTRACE.
1493 	 */
1494 	user_disable_single_step(p);
1495 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1496 #ifdef TIF_SYSCALL_EMU
1497 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1498 #endif
1499 	clear_all_latency_tracing(p);
1500 
1501 	/* ok, now we should be set up.. */
1502 	p->pid = pid_nr(pid);
1503 	if (clone_flags & CLONE_THREAD) {
1504 		p->exit_signal = -1;
1505 		p->group_leader = current->group_leader;
1506 		p->tgid = current->tgid;
1507 	} else {
1508 		if (clone_flags & CLONE_PARENT)
1509 			p->exit_signal = current->group_leader->exit_signal;
1510 		else
1511 			p->exit_signal = (clone_flags & CSIGNAL);
1512 		p->group_leader = p;
1513 		p->tgid = p->pid;
1514 	}
1515 
1516 	p->nr_dirtied = 0;
1517 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1518 	p->dirty_paused_when = 0;
1519 
1520 	p->pdeath_signal = 0;
1521 	INIT_LIST_HEAD(&p->thread_group);
1522 	p->task_works = NULL;
1523 
1524 	/*
1525 	 * Ensure that the cgroup subsystem policies allow the new process to be
1526 	 * forked. It should be noted the the new process's css_set can be changed
1527 	 * between here and cgroup_post_fork() if an organisation operation is in
1528 	 * progress.
1529 	 */
1530 	retval = cgroup_can_fork(p, cgrp_ss_priv);
1531 	if (retval)
1532 		goto bad_fork_free_pid;
1533 
1534 	/*
1535 	 * Make it visible to the rest of the system, but dont wake it up yet.
1536 	 * Need tasklist lock for parent etc handling!
1537 	 */
1538 	write_lock_irq(&tasklist_lock);
1539 
1540 	/* CLONE_PARENT re-uses the old parent */
1541 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1542 		p->real_parent = current->real_parent;
1543 		p->parent_exec_id = current->parent_exec_id;
1544 	} else {
1545 		p->real_parent = current;
1546 		p->parent_exec_id = current->self_exec_id;
1547 	}
1548 
1549 	spin_lock(&current->sighand->siglock);
1550 
1551 	/*
1552 	 * Copy seccomp details explicitly here, in case they were changed
1553 	 * before holding sighand lock.
1554 	 */
1555 	copy_seccomp(p);
1556 
1557 	/*
1558 	 * Process group and session signals need to be delivered to just the
1559 	 * parent before the fork or both the parent and the child after the
1560 	 * fork. Restart if a signal comes in before we add the new process to
1561 	 * it's process group.
1562 	 * A fatal signal pending means that current will exit, so the new
1563 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1564 	*/
1565 	recalc_sigpending();
1566 	if (signal_pending(current)) {
1567 		spin_unlock(&current->sighand->siglock);
1568 		write_unlock_irq(&tasklist_lock);
1569 		retval = -ERESTARTNOINTR;
1570 		goto bad_fork_cancel_cgroup;
1571 	}
1572 
1573 	if (likely(p->pid)) {
1574 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1575 
1576 		init_task_pid(p, PIDTYPE_PID, pid);
1577 		if (thread_group_leader(p)) {
1578 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1579 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1580 
1581 			if (is_child_reaper(pid)) {
1582 				ns_of_pid(pid)->child_reaper = p;
1583 				p->signal->flags |= SIGNAL_UNKILLABLE;
1584 			}
1585 
1586 			p->signal->leader_pid = pid;
1587 			p->signal->tty = tty_kref_get(current->signal->tty);
1588 			list_add_tail(&p->sibling, &p->real_parent->children);
1589 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1590 			attach_pid(p, PIDTYPE_PGID);
1591 			attach_pid(p, PIDTYPE_SID);
1592 			__this_cpu_inc(process_counts);
1593 		} else {
1594 			current->signal->nr_threads++;
1595 			atomic_inc(&current->signal->live);
1596 			atomic_inc(&current->signal->sigcnt);
1597 			list_add_tail_rcu(&p->thread_group,
1598 					  &p->group_leader->thread_group);
1599 			list_add_tail_rcu(&p->thread_node,
1600 					  &p->signal->thread_head);
1601 		}
1602 		attach_pid(p, PIDTYPE_PID);
1603 		nr_threads++;
1604 	}
1605 
1606 	total_forks++;
1607 	spin_unlock(&current->sighand->siglock);
1608 	syscall_tracepoint_update(p);
1609 	write_unlock_irq(&tasklist_lock);
1610 
1611 	proc_fork_connector(p);
1612 	cgroup_post_fork(p, cgrp_ss_priv);
1613 	threadgroup_change_end(current);
1614 	perf_event_fork(p);
1615 
1616 	trace_task_newtask(p, clone_flags);
1617 	uprobe_copy_process(p, clone_flags);
1618 
1619 	return p;
1620 
1621 bad_fork_cancel_cgroup:
1622 	cgroup_cancel_fork(p, cgrp_ss_priv);
1623 bad_fork_free_pid:
1624 	if (pid != &init_struct_pid)
1625 		free_pid(pid);
1626 bad_fork_cleanup_io:
1627 	if (p->io_context)
1628 		exit_io_context(p);
1629 bad_fork_cleanup_namespaces:
1630 	exit_task_namespaces(p);
1631 bad_fork_cleanup_mm:
1632 	if (p->mm)
1633 		mmput(p->mm);
1634 bad_fork_cleanup_signal:
1635 	if (!(clone_flags & CLONE_THREAD))
1636 		free_signal_struct(p->signal);
1637 bad_fork_cleanup_sighand:
1638 	__cleanup_sighand(p->sighand);
1639 bad_fork_cleanup_fs:
1640 	exit_fs(p); /* blocking */
1641 bad_fork_cleanup_files:
1642 	exit_files(p); /* blocking */
1643 bad_fork_cleanup_semundo:
1644 	exit_sem(p);
1645 bad_fork_cleanup_audit:
1646 	audit_free(p);
1647 bad_fork_cleanup_perf:
1648 	perf_event_free_task(p);
1649 bad_fork_cleanup_policy:
1650 #ifdef CONFIG_NUMA
1651 	mpol_put(p->mempolicy);
1652 bad_fork_cleanup_threadgroup_lock:
1653 #endif
1654 	threadgroup_change_end(current);
1655 	delayacct_tsk_free(p);
1656 bad_fork_cleanup_count:
1657 	atomic_dec(&p->cred->user->processes);
1658 	exit_creds(p);
1659 bad_fork_free:
1660 	free_task(p);
1661 fork_out:
1662 	return ERR_PTR(retval);
1663 }
1664 
1665 static inline void init_idle_pids(struct pid_link *links)
1666 {
1667 	enum pid_type type;
1668 
1669 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1670 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1671 		links[type].pid = &init_struct_pid;
1672 	}
1673 }
1674 
1675 struct task_struct *fork_idle(int cpu)
1676 {
1677 	struct task_struct *task;
1678 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1679 	if (!IS_ERR(task)) {
1680 		init_idle_pids(task->pids);
1681 		init_idle(task, cpu);
1682 	}
1683 
1684 	return task;
1685 }
1686 
1687 /*
1688  *  Ok, this is the main fork-routine.
1689  *
1690  * It copies the process, and if successful kick-starts
1691  * it and waits for it to finish using the VM if required.
1692  */
1693 long _do_fork(unsigned long clone_flags,
1694 	      unsigned long stack_start,
1695 	      unsigned long stack_size,
1696 	      int __user *parent_tidptr,
1697 	      int __user *child_tidptr,
1698 	      unsigned long tls)
1699 {
1700 	struct task_struct *p;
1701 	int trace = 0;
1702 	long nr;
1703 
1704 	/*
1705 	 * Determine whether and which event to report to ptracer.  When
1706 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
1707 	 * requested, no event is reported; otherwise, report if the event
1708 	 * for the type of forking is enabled.
1709 	 */
1710 	if (!(clone_flags & CLONE_UNTRACED)) {
1711 		if (clone_flags & CLONE_VFORK)
1712 			trace = PTRACE_EVENT_VFORK;
1713 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
1714 			trace = PTRACE_EVENT_CLONE;
1715 		else
1716 			trace = PTRACE_EVENT_FORK;
1717 
1718 		if (likely(!ptrace_event_enabled(current, trace)))
1719 			trace = 0;
1720 	}
1721 
1722 	p = copy_process(clone_flags, stack_start, stack_size,
1723 			 child_tidptr, NULL, trace, tls);
1724 	/*
1725 	 * Do this prior waking up the new thread - the thread pointer
1726 	 * might get invalid after that point, if the thread exits quickly.
1727 	 */
1728 	if (!IS_ERR(p)) {
1729 		struct completion vfork;
1730 		struct pid *pid;
1731 
1732 		trace_sched_process_fork(current, p);
1733 
1734 		pid = get_task_pid(p, PIDTYPE_PID);
1735 		nr = pid_vnr(pid);
1736 
1737 		if (clone_flags & CLONE_PARENT_SETTID)
1738 			put_user(nr, parent_tidptr);
1739 
1740 		if (clone_flags & CLONE_VFORK) {
1741 			p->vfork_done = &vfork;
1742 			init_completion(&vfork);
1743 			get_task_struct(p);
1744 		}
1745 
1746 		wake_up_new_task(p);
1747 
1748 		/* forking complete and child started to run, tell ptracer */
1749 		if (unlikely(trace))
1750 			ptrace_event_pid(trace, pid);
1751 
1752 		if (clone_flags & CLONE_VFORK) {
1753 			if (!wait_for_vfork_done(p, &vfork))
1754 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1755 		}
1756 
1757 		put_pid(pid);
1758 	} else {
1759 		nr = PTR_ERR(p);
1760 	}
1761 	return nr;
1762 }
1763 
1764 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1765 /* For compatibility with architectures that call do_fork directly rather than
1766  * using the syscall entry points below. */
1767 long do_fork(unsigned long clone_flags,
1768 	      unsigned long stack_start,
1769 	      unsigned long stack_size,
1770 	      int __user *parent_tidptr,
1771 	      int __user *child_tidptr)
1772 {
1773 	return _do_fork(clone_flags, stack_start, stack_size,
1774 			parent_tidptr, child_tidptr, 0);
1775 }
1776 #endif
1777 
1778 /*
1779  * Create a kernel thread.
1780  */
1781 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1782 {
1783 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1784 		(unsigned long)arg, NULL, NULL, 0);
1785 }
1786 
1787 #ifdef __ARCH_WANT_SYS_FORK
1788 SYSCALL_DEFINE0(fork)
1789 {
1790 #ifdef CONFIG_MMU
1791 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1792 #else
1793 	/* can not support in nommu mode */
1794 	return -EINVAL;
1795 #endif
1796 }
1797 #endif
1798 
1799 #ifdef __ARCH_WANT_SYS_VFORK
1800 SYSCALL_DEFINE0(vfork)
1801 {
1802 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1803 			0, NULL, NULL, 0);
1804 }
1805 #endif
1806 
1807 #ifdef __ARCH_WANT_SYS_CLONE
1808 #ifdef CONFIG_CLONE_BACKWARDS
1809 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1810 		 int __user *, parent_tidptr,
1811 		 unsigned long, tls,
1812 		 int __user *, child_tidptr)
1813 #elif defined(CONFIG_CLONE_BACKWARDS2)
1814 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1815 		 int __user *, parent_tidptr,
1816 		 int __user *, child_tidptr,
1817 		 unsigned long, tls)
1818 #elif defined(CONFIG_CLONE_BACKWARDS3)
1819 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1820 		int, stack_size,
1821 		int __user *, parent_tidptr,
1822 		int __user *, child_tidptr,
1823 		unsigned long, tls)
1824 #else
1825 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1826 		 int __user *, parent_tidptr,
1827 		 int __user *, child_tidptr,
1828 		 unsigned long, tls)
1829 #endif
1830 {
1831 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1832 }
1833 #endif
1834 
1835 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1836 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1837 #endif
1838 
1839 static void sighand_ctor(void *data)
1840 {
1841 	struct sighand_struct *sighand = data;
1842 
1843 	spin_lock_init(&sighand->siglock);
1844 	init_waitqueue_head(&sighand->signalfd_wqh);
1845 }
1846 
1847 void __init proc_caches_init(void)
1848 {
1849 	sighand_cachep = kmem_cache_create("sighand_cache",
1850 			sizeof(struct sighand_struct), 0,
1851 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1852 			SLAB_NOTRACK, sighand_ctor);
1853 	signal_cachep = kmem_cache_create("signal_cache",
1854 			sizeof(struct signal_struct), 0,
1855 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1856 	files_cachep = kmem_cache_create("files_cache",
1857 			sizeof(struct files_struct), 0,
1858 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1859 	fs_cachep = kmem_cache_create("fs_cache",
1860 			sizeof(struct fs_struct), 0,
1861 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1862 	/*
1863 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1864 	 * whole struct cpumask for the OFFSTACK case. We could change
1865 	 * this to *only* allocate as much of it as required by the
1866 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
1867 	 * is at the end of the structure, exactly for that reason.
1868 	 */
1869 	mm_cachep = kmem_cache_create("mm_struct",
1870 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1871 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1872 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1873 	mmap_init();
1874 	nsproxy_cache_init();
1875 }
1876 
1877 /*
1878  * Check constraints on flags passed to the unshare system call.
1879  */
1880 static int check_unshare_flags(unsigned long unshare_flags)
1881 {
1882 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1883 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1884 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1885 				CLONE_NEWUSER|CLONE_NEWPID))
1886 		return -EINVAL;
1887 	/*
1888 	 * Not implemented, but pretend it works if there is nothing
1889 	 * to unshare.  Note that unsharing the address space or the
1890 	 * signal handlers also need to unshare the signal queues (aka
1891 	 * CLONE_THREAD).
1892 	 */
1893 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1894 		if (!thread_group_empty(current))
1895 			return -EINVAL;
1896 	}
1897 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1898 		if (atomic_read(&current->sighand->count) > 1)
1899 			return -EINVAL;
1900 	}
1901 	if (unshare_flags & CLONE_VM) {
1902 		if (!current_is_single_threaded())
1903 			return -EINVAL;
1904 	}
1905 
1906 	return 0;
1907 }
1908 
1909 /*
1910  * Unshare the filesystem structure if it is being shared
1911  */
1912 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1913 {
1914 	struct fs_struct *fs = current->fs;
1915 
1916 	if (!(unshare_flags & CLONE_FS) || !fs)
1917 		return 0;
1918 
1919 	/* don't need lock here; in the worst case we'll do useless copy */
1920 	if (fs->users == 1)
1921 		return 0;
1922 
1923 	*new_fsp = copy_fs_struct(fs);
1924 	if (!*new_fsp)
1925 		return -ENOMEM;
1926 
1927 	return 0;
1928 }
1929 
1930 /*
1931  * Unshare file descriptor table if it is being shared
1932  */
1933 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1934 {
1935 	struct files_struct *fd = current->files;
1936 	int error = 0;
1937 
1938 	if ((unshare_flags & CLONE_FILES) &&
1939 	    (fd && atomic_read(&fd->count) > 1)) {
1940 		*new_fdp = dup_fd(fd, &error);
1941 		if (!*new_fdp)
1942 			return error;
1943 	}
1944 
1945 	return 0;
1946 }
1947 
1948 /*
1949  * unshare allows a process to 'unshare' part of the process
1950  * context which was originally shared using clone.  copy_*
1951  * functions used by do_fork() cannot be used here directly
1952  * because they modify an inactive task_struct that is being
1953  * constructed. Here we are modifying the current, active,
1954  * task_struct.
1955  */
1956 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1957 {
1958 	struct fs_struct *fs, *new_fs = NULL;
1959 	struct files_struct *fd, *new_fd = NULL;
1960 	struct cred *new_cred = NULL;
1961 	struct nsproxy *new_nsproxy = NULL;
1962 	int do_sysvsem = 0;
1963 	int err;
1964 
1965 	/*
1966 	 * If unsharing a user namespace must also unshare the thread group
1967 	 * and unshare the filesystem root and working directories.
1968 	 */
1969 	if (unshare_flags & CLONE_NEWUSER)
1970 		unshare_flags |= CLONE_THREAD | CLONE_FS;
1971 	/*
1972 	 * If unsharing vm, must also unshare signal handlers.
1973 	 */
1974 	if (unshare_flags & CLONE_VM)
1975 		unshare_flags |= CLONE_SIGHAND;
1976 	/*
1977 	 * If unsharing a signal handlers, must also unshare the signal queues.
1978 	 */
1979 	if (unshare_flags & CLONE_SIGHAND)
1980 		unshare_flags |= CLONE_THREAD;
1981 	/*
1982 	 * If unsharing namespace, must also unshare filesystem information.
1983 	 */
1984 	if (unshare_flags & CLONE_NEWNS)
1985 		unshare_flags |= CLONE_FS;
1986 
1987 	err = check_unshare_flags(unshare_flags);
1988 	if (err)
1989 		goto bad_unshare_out;
1990 	/*
1991 	 * CLONE_NEWIPC must also detach from the undolist: after switching
1992 	 * to a new ipc namespace, the semaphore arrays from the old
1993 	 * namespace are unreachable.
1994 	 */
1995 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1996 		do_sysvsem = 1;
1997 	err = unshare_fs(unshare_flags, &new_fs);
1998 	if (err)
1999 		goto bad_unshare_out;
2000 	err = unshare_fd(unshare_flags, &new_fd);
2001 	if (err)
2002 		goto bad_unshare_cleanup_fs;
2003 	err = unshare_userns(unshare_flags, &new_cred);
2004 	if (err)
2005 		goto bad_unshare_cleanup_fd;
2006 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2007 					 new_cred, new_fs);
2008 	if (err)
2009 		goto bad_unshare_cleanup_cred;
2010 
2011 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2012 		if (do_sysvsem) {
2013 			/*
2014 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2015 			 */
2016 			exit_sem(current);
2017 		}
2018 		if (unshare_flags & CLONE_NEWIPC) {
2019 			/* Orphan segments in old ns (see sem above). */
2020 			exit_shm(current);
2021 			shm_init_task(current);
2022 		}
2023 
2024 		if (new_nsproxy)
2025 			switch_task_namespaces(current, new_nsproxy);
2026 
2027 		task_lock(current);
2028 
2029 		if (new_fs) {
2030 			fs = current->fs;
2031 			spin_lock(&fs->lock);
2032 			current->fs = new_fs;
2033 			if (--fs->users)
2034 				new_fs = NULL;
2035 			else
2036 				new_fs = fs;
2037 			spin_unlock(&fs->lock);
2038 		}
2039 
2040 		if (new_fd) {
2041 			fd = current->files;
2042 			current->files = new_fd;
2043 			new_fd = fd;
2044 		}
2045 
2046 		task_unlock(current);
2047 
2048 		if (new_cred) {
2049 			/* Install the new user namespace */
2050 			commit_creds(new_cred);
2051 			new_cred = NULL;
2052 		}
2053 	}
2054 
2055 bad_unshare_cleanup_cred:
2056 	if (new_cred)
2057 		put_cred(new_cred);
2058 bad_unshare_cleanup_fd:
2059 	if (new_fd)
2060 		put_files_struct(new_fd);
2061 
2062 bad_unshare_cleanup_fs:
2063 	if (new_fs)
2064 		free_fs_struct(new_fs);
2065 
2066 bad_unshare_out:
2067 	return err;
2068 }
2069 
2070 /*
2071  *	Helper to unshare the files of the current task.
2072  *	We don't want to expose copy_files internals to
2073  *	the exec layer of the kernel.
2074  */
2075 
2076 int unshare_files(struct files_struct **displaced)
2077 {
2078 	struct task_struct *task = current;
2079 	struct files_struct *copy = NULL;
2080 	int error;
2081 
2082 	error = unshare_fd(CLONE_FILES, &copy);
2083 	if (error || !copy) {
2084 		*displaced = NULL;
2085 		return error;
2086 	}
2087 	*displaced = task->files;
2088 	task_lock(task);
2089 	task->files = copy;
2090 	task_unlock(task);
2091 	return 0;
2092 }
2093 
2094 int sysctl_max_threads(struct ctl_table *table, int write,
2095 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2096 {
2097 	struct ctl_table t;
2098 	int ret;
2099 	int threads = max_threads;
2100 	int min = MIN_THREADS;
2101 	int max = MAX_THREADS;
2102 
2103 	t = *table;
2104 	t.data = &threads;
2105 	t.extra1 = &min;
2106 	t.extra2 = &max;
2107 
2108 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2109 	if (ret || !write)
2110 		return ret;
2111 
2112 	set_max_threads(threads);
2113 
2114 	return 0;
2115 }
2116