xref: /linux/kernel/fork.c (revision 372e2db7210df7c45ead46429aeb1443ba148060)
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 #include <linux/kcov.h>
79 
80 #include <asm/pgtable.h>
81 #include <asm/pgalloc.h>
82 #include <asm/uaccess.h>
83 #include <asm/mmu_context.h>
84 #include <asm/cacheflush.h>
85 #include <asm/tlbflush.h>
86 
87 #include <trace/events/sched.h>
88 
89 #define CREATE_TRACE_POINTS
90 #include <trace/events/task.h>
91 
92 /*
93  * Minimum number of threads to boot the kernel
94  */
95 #define MIN_THREADS 20
96 
97 /*
98  * Maximum number of threads
99  */
100 #define MAX_THREADS FUTEX_TID_MASK
101 
102 /*
103  * Protected counters by write_lock_irq(&tasklist_lock)
104  */
105 unsigned long total_forks;	/* Handle normal Linux uptimes. */
106 int nr_threads;			/* The idle threads do not count.. */
107 
108 int max_threads;		/* tunable limit on nr_threads */
109 
110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
111 
112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
113 
114 #ifdef CONFIG_PROVE_RCU
115 int lockdep_tasklist_lock_is_held(void)
116 {
117 	return lockdep_is_held(&tasklist_lock);
118 }
119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
120 #endif /* #ifdef CONFIG_PROVE_RCU */
121 
122 int nr_processes(void)
123 {
124 	int cpu;
125 	int total = 0;
126 
127 	for_each_possible_cpu(cpu)
128 		total += per_cpu(process_counts, cpu);
129 
130 	return total;
131 }
132 
133 void __weak arch_release_task_struct(struct task_struct *tsk)
134 {
135 }
136 
137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
138 static struct kmem_cache *task_struct_cachep;
139 
140 static inline struct task_struct *alloc_task_struct_node(int node)
141 {
142 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
143 }
144 
145 static inline void free_task_struct(struct task_struct *tsk)
146 {
147 	kmem_cache_free(task_struct_cachep, tsk);
148 }
149 #endif
150 
151 void __weak arch_release_thread_stack(unsigned long *stack)
152 {
153 }
154 
155 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
156 
157 /*
158  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
159  * kmemcache based allocator.
160  */
161 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
162 
163 #ifdef CONFIG_VMAP_STACK
164 /*
165  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
166  * flush.  Try to minimize the number of calls by caching stacks.
167  */
168 #define NR_CACHED_STACKS 2
169 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
170 #endif
171 
172 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
173 {
174 #ifdef CONFIG_VMAP_STACK
175 	void *stack;
176 	int i;
177 
178 	local_irq_disable();
179 	for (i = 0; i < NR_CACHED_STACKS; i++) {
180 		struct vm_struct *s = this_cpu_read(cached_stacks[i]);
181 
182 		if (!s)
183 			continue;
184 		this_cpu_write(cached_stacks[i], NULL);
185 
186 		tsk->stack_vm_area = s;
187 		local_irq_enable();
188 		return s->addr;
189 	}
190 	local_irq_enable();
191 
192 	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
193 				     VMALLOC_START, VMALLOC_END,
194 				     THREADINFO_GFP | __GFP_HIGHMEM,
195 				     PAGE_KERNEL,
196 				     0, node, __builtin_return_address(0));
197 
198 	/*
199 	 * We can't call find_vm_area() in interrupt context, and
200 	 * free_thread_stack() can be called in interrupt context,
201 	 * so cache the vm_struct.
202 	 */
203 	if (stack)
204 		tsk->stack_vm_area = find_vm_area(stack);
205 	return stack;
206 #else
207 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
208 					     THREAD_SIZE_ORDER);
209 
210 	return page ? page_address(page) : NULL;
211 #endif
212 }
213 
214 static inline void free_thread_stack(struct task_struct *tsk)
215 {
216 #ifdef CONFIG_VMAP_STACK
217 	if (task_stack_vm_area(tsk)) {
218 		unsigned long flags;
219 		int i;
220 
221 		local_irq_save(flags);
222 		for (i = 0; i < NR_CACHED_STACKS; i++) {
223 			if (this_cpu_read(cached_stacks[i]))
224 				continue;
225 
226 			this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
227 			local_irq_restore(flags);
228 			return;
229 		}
230 		local_irq_restore(flags);
231 
232 		vfree_atomic(tsk->stack);
233 		return;
234 	}
235 #endif
236 
237 	__free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
238 }
239 # else
240 static struct kmem_cache *thread_stack_cache;
241 
242 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
243 						  int node)
244 {
245 	return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
246 }
247 
248 static void free_thread_stack(struct task_struct *tsk)
249 {
250 	kmem_cache_free(thread_stack_cache, tsk->stack);
251 }
252 
253 void thread_stack_cache_init(void)
254 {
255 	thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
256 					      THREAD_SIZE, 0, NULL);
257 	BUG_ON(thread_stack_cache == NULL);
258 }
259 # endif
260 #endif
261 
262 /* SLAB cache for signal_struct structures (tsk->signal) */
263 static struct kmem_cache *signal_cachep;
264 
265 /* SLAB cache for sighand_struct structures (tsk->sighand) */
266 struct kmem_cache *sighand_cachep;
267 
268 /* SLAB cache for files_struct structures (tsk->files) */
269 struct kmem_cache *files_cachep;
270 
271 /* SLAB cache for fs_struct structures (tsk->fs) */
272 struct kmem_cache *fs_cachep;
273 
274 /* SLAB cache for vm_area_struct structures */
275 struct kmem_cache *vm_area_cachep;
276 
277 /* SLAB cache for mm_struct structures (tsk->mm) */
278 static struct kmem_cache *mm_cachep;
279 
280 static void account_kernel_stack(struct task_struct *tsk, int account)
281 {
282 	void *stack = task_stack_page(tsk);
283 	struct vm_struct *vm = task_stack_vm_area(tsk);
284 
285 	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
286 
287 	if (vm) {
288 		int i;
289 
290 		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
291 
292 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
293 			mod_zone_page_state(page_zone(vm->pages[i]),
294 					    NR_KERNEL_STACK_KB,
295 					    PAGE_SIZE / 1024 * account);
296 		}
297 
298 		/* All stack pages belong to the same memcg. */
299 		memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
300 					    account * (THREAD_SIZE / 1024));
301 	} else {
302 		/*
303 		 * All stack pages are in the same zone and belong to the
304 		 * same memcg.
305 		 */
306 		struct page *first_page = virt_to_page(stack);
307 
308 		mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
309 				    THREAD_SIZE / 1024 * account);
310 
311 		memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
312 					    account * (THREAD_SIZE / 1024));
313 	}
314 }
315 
316 static void release_task_stack(struct task_struct *tsk)
317 {
318 	if (WARN_ON(tsk->state != TASK_DEAD))
319 		return;  /* Better to leak the stack than to free prematurely */
320 
321 	account_kernel_stack(tsk, -1);
322 	arch_release_thread_stack(tsk->stack);
323 	free_thread_stack(tsk);
324 	tsk->stack = NULL;
325 #ifdef CONFIG_VMAP_STACK
326 	tsk->stack_vm_area = NULL;
327 #endif
328 }
329 
330 #ifdef CONFIG_THREAD_INFO_IN_TASK
331 void put_task_stack(struct task_struct *tsk)
332 {
333 	if (atomic_dec_and_test(&tsk->stack_refcount))
334 		release_task_stack(tsk);
335 }
336 #endif
337 
338 void free_task(struct task_struct *tsk)
339 {
340 #ifndef CONFIG_THREAD_INFO_IN_TASK
341 	/*
342 	 * The task is finally done with both the stack and thread_info,
343 	 * so free both.
344 	 */
345 	release_task_stack(tsk);
346 #else
347 	/*
348 	 * If the task had a separate stack allocation, it should be gone
349 	 * by now.
350 	 */
351 	WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
352 #endif
353 	rt_mutex_debug_task_free(tsk);
354 	ftrace_graph_exit_task(tsk);
355 	put_seccomp_filter(tsk);
356 	arch_release_task_struct(tsk);
357 	if (tsk->flags & PF_KTHREAD)
358 		free_kthread_struct(tsk);
359 	free_task_struct(tsk);
360 }
361 EXPORT_SYMBOL(free_task);
362 
363 static inline void free_signal_struct(struct signal_struct *sig)
364 {
365 	taskstats_tgid_free(sig);
366 	sched_autogroup_exit(sig);
367 	/*
368 	 * __mmdrop is not safe to call from softirq context on x86 due to
369 	 * pgd_dtor so postpone it to the async context
370 	 */
371 	if (sig->oom_mm)
372 		mmdrop_async(sig->oom_mm);
373 	kmem_cache_free(signal_cachep, sig);
374 }
375 
376 static inline void put_signal_struct(struct signal_struct *sig)
377 {
378 	if (atomic_dec_and_test(&sig->sigcnt))
379 		free_signal_struct(sig);
380 }
381 
382 void __put_task_struct(struct task_struct *tsk)
383 {
384 	WARN_ON(!tsk->exit_state);
385 	WARN_ON(atomic_read(&tsk->usage));
386 	WARN_ON(tsk == current);
387 
388 	cgroup_free(tsk);
389 	task_numa_free(tsk);
390 	security_task_free(tsk);
391 	exit_creds(tsk);
392 	delayacct_tsk_free(tsk);
393 	put_signal_struct(tsk->signal);
394 
395 	if (!profile_handoff_task(tsk))
396 		free_task(tsk);
397 }
398 EXPORT_SYMBOL_GPL(__put_task_struct);
399 
400 void __init __weak arch_task_cache_init(void) { }
401 
402 /*
403  * set_max_threads
404  */
405 static void set_max_threads(unsigned int max_threads_suggested)
406 {
407 	u64 threads;
408 
409 	/*
410 	 * The number of threads shall be limited such that the thread
411 	 * structures may only consume a small part of the available memory.
412 	 */
413 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
414 		threads = MAX_THREADS;
415 	else
416 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
417 				    (u64) THREAD_SIZE * 8UL);
418 
419 	if (threads > max_threads_suggested)
420 		threads = max_threads_suggested;
421 
422 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
423 }
424 
425 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
426 /* Initialized by the architecture: */
427 int arch_task_struct_size __read_mostly;
428 #endif
429 
430 void __init fork_init(void)
431 {
432 	int i;
433 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
434 #ifndef ARCH_MIN_TASKALIGN
435 #define ARCH_MIN_TASKALIGN	L1_CACHE_BYTES
436 #endif
437 	/* create a slab on which task_structs can be allocated */
438 	task_struct_cachep = kmem_cache_create("task_struct",
439 			arch_task_struct_size, ARCH_MIN_TASKALIGN,
440 			SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
441 #endif
442 
443 	/* do the arch specific task caches init */
444 	arch_task_cache_init();
445 
446 	set_max_threads(MAX_THREADS);
447 
448 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
449 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
450 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
451 		init_task.signal->rlim[RLIMIT_NPROC];
452 
453 	for (i = 0; i < UCOUNT_COUNTS; i++) {
454 		init_user_ns.ucount_max[i] = max_threads/2;
455 	}
456 }
457 
458 int __weak arch_dup_task_struct(struct task_struct *dst,
459 					       struct task_struct *src)
460 {
461 	*dst = *src;
462 	return 0;
463 }
464 
465 void set_task_stack_end_magic(struct task_struct *tsk)
466 {
467 	unsigned long *stackend;
468 
469 	stackend = end_of_stack(tsk);
470 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
471 }
472 
473 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
474 {
475 	struct task_struct *tsk;
476 	unsigned long *stack;
477 	struct vm_struct *stack_vm_area;
478 	int err;
479 
480 	if (node == NUMA_NO_NODE)
481 		node = tsk_fork_get_node(orig);
482 	tsk = alloc_task_struct_node(node);
483 	if (!tsk)
484 		return NULL;
485 
486 	stack = alloc_thread_stack_node(tsk, node);
487 	if (!stack)
488 		goto free_tsk;
489 
490 	stack_vm_area = task_stack_vm_area(tsk);
491 
492 	err = arch_dup_task_struct(tsk, orig);
493 
494 	/*
495 	 * arch_dup_task_struct() clobbers the stack-related fields.  Make
496 	 * sure they're properly initialized before using any stack-related
497 	 * functions again.
498 	 */
499 	tsk->stack = stack;
500 #ifdef CONFIG_VMAP_STACK
501 	tsk->stack_vm_area = stack_vm_area;
502 #endif
503 #ifdef CONFIG_THREAD_INFO_IN_TASK
504 	atomic_set(&tsk->stack_refcount, 1);
505 #endif
506 
507 	if (err)
508 		goto free_stack;
509 
510 #ifdef CONFIG_SECCOMP
511 	/*
512 	 * We must handle setting up seccomp filters once we're under
513 	 * the sighand lock in case orig has changed between now and
514 	 * then. Until then, filter must be NULL to avoid messing up
515 	 * the usage counts on the error path calling free_task.
516 	 */
517 	tsk->seccomp.filter = NULL;
518 #endif
519 
520 	setup_thread_stack(tsk, orig);
521 	clear_user_return_notifier(tsk);
522 	clear_tsk_need_resched(tsk);
523 	set_task_stack_end_magic(tsk);
524 
525 #ifdef CONFIG_CC_STACKPROTECTOR
526 	tsk->stack_canary = get_random_int();
527 #endif
528 
529 	/*
530 	 * One for us, one for whoever does the "release_task()" (usually
531 	 * parent)
532 	 */
533 	atomic_set(&tsk->usage, 2);
534 #ifdef CONFIG_BLK_DEV_IO_TRACE
535 	tsk->btrace_seq = 0;
536 #endif
537 	tsk->splice_pipe = NULL;
538 	tsk->task_frag.page = NULL;
539 	tsk->wake_q.next = NULL;
540 
541 	account_kernel_stack(tsk, 1);
542 
543 	kcov_task_init(tsk);
544 
545 	return tsk;
546 
547 free_stack:
548 	free_thread_stack(tsk);
549 free_tsk:
550 	free_task_struct(tsk);
551 	return NULL;
552 }
553 
554 #ifdef CONFIG_MMU
555 static __latent_entropy int dup_mmap(struct mm_struct *mm,
556 					struct mm_struct *oldmm)
557 {
558 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
559 	struct rb_node **rb_link, *rb_parent;
560 	int retval;
561 	unsigned long charge;
562 
563 	uprobe_start_dup_mmap();
564 	if (down_write_killable(&oldmm->mmap_sem)) {
565 		retval = -EINTR;
566 		goto fail_uprobe_end;
567 	}
568 	flush_cache_dup_mm(oldmm);
569 	uprobe_dup_mmap(oldmm, mm);
570 	/*
571 	 * Not linked in yet - no deadlock potential:
572 	 */
573 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
574 
575 	/* No ordering required: file already has been exposed. */
576 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
577 
578 	mm->total_vm = oldmm->total_vm;
579 	mm->data_vm = oldmm->data_vm;
580 	mm->exec_vm = oldmm->exec_vm;
581 	mm->stack_vm = oldmm->stack_vm;
582 
583 	rb_link = &mm->mm_rb.rb_node;
584 	rb_parent = NULL;
585 	pprev = &mm->mmap;
586 	retval = ksm_fork(mm, oldmm);
587 	if (retval)
588 		goto out;
589 	retval = khugepaged_fork(mm, oldmm);
590 	if (retval)
591 		goto out;
592 
593 	prev = NULL;
594 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
595 		struct file *file;
596 
597 		if (mpnt->vm_flags & VM_DONTCOPY) {
598 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
599 			continue;
600 		}
601 		charge = 0;
602 		if (mpnt->vm_flags & VM_ACCOUNT) {
603 			unsigned long len = vma_pages(mpnt);
604 
605 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
606 				goto fail_nomem;
607 			charge = len;
608 		}
609 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
610 		if (!tmp)
611 			goto fail_nomem;
612 		*tmp = *mpnt;
613 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
614 		retval = vma_dup_policy(mpnt, tmp);
615 		if (retval)
616 			goto fail_nomem_policy;
617 		tmp->vm_mm = mm;
618 		if (anon_vma_fork(tmp, mpnt))
619 			goto fail_nomem_anon_vma_fork;
620 		tmp->vm_flags &=
621 			~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
622 		tmp->vm_next = tmp->vm_prev = NULL;
623 		tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
624 		file = tmp->vm_file;
625 		if (file) {
626 			struct inode *inode = file_inode(file);
627 			struct address_space *mapping = file->f_mapping;
628 
629 			get_file(file);
630 			if (tmp->vm_flags & VM_DENYWRITE)
631 				atomic_dec(&inode->i_writecount);
632 			i_mmap_lock_write(mapping);
633 			if (tmp->vm_flags & VM_SHARED)
634 				atomic_inc(&mapping->i_mmap_writable);
635 			flush_dcache_mmap_lock(mapping);
636 			/* insert tmp into the share list, just after mpnt */
637 			vma_interval_tree_insert_after(tmp, mpnt,
638 					&mapping->i_mmap);
639 			flush_dcache_mmap_unlock(mapping);
640 			i_mmap_unlock_write(mapping);
641 		}
642 
643 		/*
644 		 * Clear hugetlb-related page reserves for children. This only
645 		 * affects MAP_PRIVATE mappings. Faults generated by the child
646 		 * are not guaranteed to succeed, even if read-only
647 		 */
648 		if (is_vm_hugetlb_page(tmp))
649 			reset_vma_resv_huge_pages(tmp);
650 
651 		/*
652 		 * Link in the new vma and copy the page table entries.
653 		 */
654 		*pprev = tmp;
655 		pprev = &tmp->vm_next;
656 		tmp->vm_prev = prev;
657 		prev = tmp;
658 
659 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
660 		rb_link = &tmp->vm_rb.rb_right;
661 		rb_parent = &tmp->vm_rb;
662 
663 		mm->map_count++;
664 		retval = copy_page_range(mm, oldmm, mpnt);
665 
666 		if (tmp->vm_ops && tmp->vm_ops->open)
667 			tmp->vm_ops->open(tmp);
668 
669 		if (retval)
670 			goto out;
671 	}
672 	/* a new mm has just been created */
673 	arch_dup_mmap(oldmm, mm);
674 	retval = 0;
675 out:
676 	up_write(&mm->mmap_sem);
677 	flush_tlb_mm(oldmm);
678 	up_write(&oldmm->mmap_sem);
679 fail_uprobe_end:
680 	uprobe_end_dup_mmap();
681 	return retval;
682 fail_nomem_anon_vma_fork:
683 	mpol_put(vma_policy(tmp));
684 fail_nomem_policy:
685 	kmem_cache_free(vm_area_cachep, tmp);
686 fail_nomem:
687 	retval = -ENOMEM;
688 	vm_unacct_memory(charge);
689 	goto out;
690 }
691 
692 static inline int mm_alloc_pgd(struct mm_struct *mm)
693 {
694 	mm->pgd = pgd_alloc(mm);
695 	if (unlikely(!mm->pgd))
696 		return -ENOMEM;
697 	return 0;
698 }
699 
700 static inline void mm_free_pgd(struct mm_struct *mm)
701 {
702 	pgd_free(mm, mm->pgd);
703 }
704 #else
705 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
706 {
707 	down_write(&oldmm->mmap_sem);
708 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
709 	up_write(&oldmm->mmap_sem);
710 	return 0;
711 }
712 #define mm_alloc_pgd(mm)	(0)
713 #define mm_free_pgd(mm)
714 #endif /* CONFIG_MMU */
715 
716 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
717 
718 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
719 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
720 
721 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
722 
723 static int __init coredump_filter_setup(char *s)
724 {
725 	default_dump_filter =
726 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
727 		MMF_DUMP_FILTER_MASK;
728 	return 1;
729 }
730 
731 __setup("coredump_filter=", coredump_filter_setup);
732 
733 #include <linux/init_task.h>
734 
735 static void mm_init_aio(struct mm_struct *mm)
736 {
737 #ifdef CONFIG_AIO
738 	spin_lock_init(&mm->ioctx_lock);
739 	mm->ioctx_table = NULL;
740 #endif
741 }
742 
743 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
744 {
745 #ifdef CONFIG_MEMCG
746 	mm->owner = p;
747 #endif
748 }
749 
750 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
751 {
752 	mm->mmap = NULL;
753 	mm->mm_rb = RB_ROOT;
754 	mm->vmacache_seqnum = 0;
755 	atomic_set(&mm->mm_users, 1);
756 	atomic_set(&mm->mm_count, 1);
757 	init_rwsem(&mm->mmap_sem);
758 	INIT_LIST_HEAD(&mm->mmlist);
759 	mm->core_state = NULL;
760 	atomic_long_set(&mm->nr_ptes, 0);
761 	mm_nr_pmds_init(mm);
762 	mm->map_count = 0;
763 	mm->locked_vm = 0;
764 	mm->pinned_vm = 0;
765 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
766 	spin_lock_init(&mm->page_table_lock);
767 	mm_init_cpumask(mm);
768 	mm_init_aio(mm);
769 	mm_init_owner(mm, p);
770 	mmu_notifier_mm_init(mm);
771 	clear_tlb_flush_pending(mm);
772 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
773 	mm->pmd_huge_pte = NULL;
774 #endif
775 
776 	if (current->mm) {
777 		mm->flags = current->mm->flags & MMF_INIT_MASK;
778 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
779 	} else {
780 		mm->flags = default_dump_filter;
781 		mm->def_flags = 0;
782 	}
783 
784 	if (mm_alloc_pgd(mm))
785 		goto fail_nopgd;
786 
787 	if (init_new_context(p, mm))
788 		goto fail_nocontext;
789 
790 	return mm;
791 
792 fail_nocontext:
793 	mm_free_pgd(mm);
794 fail_nopgd:
795 	free_mm(mm);
796 	return NULL;
797 }
798 
799 static void check_mm(struct mm_struct *mm)
800 {
801 	int i;
802 
803 	for (i = 0; i < NR_MM_COUNTERS; i++) {
804 		long x = atomic_long_read(&mm->rss_stat.count[i]);
805 
806 		if (unlikely(x))
807 			printk(KERN_ALERT "BUG: Bad rss-counter state "
808 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
809 	}
810 
811 	if (atomic_long_read(&mm->nr_ptes))
812 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
813 				atomic_long_read(&mm->nr_ptes));
814 	if (mm_nr_pmds(mm))
815 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
816 				mm_nr_pmds(mm));
817 
818 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
819 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
820 #endif
821 }
822 
823 /*
824  * Allocate and initialize an mm_struct.
825  */
826 struct mm_struct *mm_alloc(void)
827 {
828 	struct mm_struct *mm;
829 
830 	mm = allocate_mm();
831 	if (!mm)
832 		return NULL;
833 
834 	memset(mm, 0, sizeof(*mm));
835 	return mm_init(mm, current);
836 }
837 
838 /*
839  * Called when the last reference to the mm
840  * is dropped: either by a lazy thread or by
841  * mmput. Free the page directory and the mm.
842  */
843 void __mmdrop(struct mm_struct *mm)
844 {
845 	BUG_ON(mm == &init_mm);
846 	mm_free_pgd(mm);
847 	destroy_context(mm);
848 	mmu_notifier_mm_destroy(mm);
849 	check_mm(mm);
850 	free_mm(mm);
851 }
852 EXPORT_SYMBOL_GPL(__mmdrop);
853 
854 static inline void __mmput(struct mm_struct *mm)
855 {
856 	VM_BUG_ON(atomic_read(&mm->mm_users));
857 
858 	uprobe_clear_state(mm);
859 	exit_aio(mm);
860 	ksm_exit(mm);
861 	khugepaged_exit(mm); /* must run before exit_mmap */
862 	exit_mmap(mm);
863 	mm_put_huge_zero_page(mm);
864 	set_mm_exe_file(mm, NULL);
865 	if (!list_empty(&mm->mmlist)) {
866 		spin_lock(&mmlist_lock);
867 		list_del(&mm->mmlist);
868 		spin_unlock(&mmlist_lock);
869 	}
870 	if (mm->binfmt)
871 		module_put(mm->binfmt->module);
872 	set_bit(MMF_OOM_SKIP, &mm->flags);
873 	mmdrop(mm);
874 }
875 
876 /*
877  * Decrement the use count and release all resources for an mm.
878  */
879 void mmput(struct mm_struct *mm)
880 {
881 	might_sleep();
882 
883 	if (atomic_dec_and_test(&mm->mm_users))
884 		__mmput(mm);
885 }
886 EXPORT_SYMBOL_GPL(mmput);
887 
888 #ifdef CONFIG_MMU
889 static void mmput_async_fn(struct work_struct *work)
890 {
891 	struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
892 	__mmput(mm);
893 }
894 
895 void mmput_async(struct mm_struct *mm)
896 {
897 	if (atomic_dec_and_test(&mm->mm_users)) {
898 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
899 		schedule_work(&mm->async_put_work);
900 	}
901 }
902 #endif
903 
904 /**
905  * set_mm_exe_file - change a reference to the mm's executable file
906  *
907  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
908  *
909  * Main users are mmput() and sys_execve(). Callers prevent concurrent
910  * invocations: in mmput() nobody alive left, in execve task is single
911  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
912  * mm->exe_file, but does so without using set_mm_exe_file() in order
913  * to do avoid the need for any locks.
914  */
915 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
916 {
917 	struct file *old_exe_file;
918 
919 	/*
920 	 * It is safe to dereference the exe_file without RCU as
921 	 * this function is only called if nobody else can access
922 	 * this mm -- see comment above for justification.
923 	 */
924 	old_exe_file = rcu_dereference_raw(mm->exe_file);
925 
926 	if (new_exe_file)
927 		get_file(new_exe_file);
928 	rcu_assign_pointer(mm->exe_file, new_exe_file);
929 	if (old_exe_file)
930 		fput(old_exe_file);
931 }
932 
933 /**
934  * get_mm_exe_file - acquire a reference to the mm's executable file
935  *
936  * Returns %NULL if mm has no associated executable file.
937  * User must release file via fput().
938  */
939 struct file *get_mm_exe_file(struct mm_struct *mm)
940 {
941 	struct file *exe_file;
942 
943 	rcu_read_lock();
944 	exe_file = rcu_dereference(mm->exe_file);
945 	if (exe_file && !get_file_rcu(exe_file))
946 		exe_file = NULL;
947 	rcu_read_unlock();
948 	return exe_file;
949 }
950 EXPORT_SYMBOL(get_mm_exe_file);
951 
952 /**
953  * get_task_exe_file - acquire a reference to the task's executable file
954  *
955  * Returns %NULL if task's mm (if any) has no associated executable file or
956  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
957  * User must release file via fput().
958  */
959 struct file *get_task_exe_file(struct task_struct *task)
960 {
961 	struct file *exe_file = NULL;
962 	struct mm_struct *mm;
963 
964 	task_lock(task);
965 	mm = task->mm;
966 	if (mm) {
967 		if (!(task->flags & PF_KTHREAD))
968 			exe_file = get_mm_exe_file(mm);
969 	}
970 	task_unlock(task);
971 	return exe_file;
972 }
973 EXPORT_SYMBOL(get_task_exe_file);
974 
975 /**
976  * get_task_mm - acquire a reference to the task's mm
977  *
978  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
979  * this kernel workthread has transiently adopted a user mm with use_mm,
980  * to do its AIO) is not set and if so returns a reference to it, after
981  * bumping up the use count.  User must release the mm via mmput()
982  * after use.  Typically used by /proc and ptrace.
983  */
984 struct mm_struct *get_task_mm(struct task_struct *task)
985 {
986 	struct mm_struct *mm;
987 
988 	task_lock(task);
989 	mm = task->mm;
990 	if (mm) {
991 		if (task->flags & PF_KTHREAD)
992 			mm = NULL;
993 		else
994 			atomic_inc(&mm->mm_users);
995 	}
996 	task_unlock(task);
997 	return mm;
998 }
999 EXPORT_SYMBOL_GPL(get_task_mm);
1000 
1001 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1002 {
1003 	struct mm_struct *mm;
1004 	int err;
1005 
1006 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1007 	if (err)
1008 		return ERR_PTR(err);
1009 
1010 	mm = get_task_mm(task);
1011 	if (mm && mm != current->mm &&
1012 			!ptrace_may_access(task, mode)) {
1013 		mmput(mm);
1014 		mm = ERR_PTR(-EACCES);
1015 	}
1016 	mutex_unlock(&task->signal->cred_guard_mutex);
1017 
1018 	return mm;
1019 }
1020 
1021 static void complete_vfork_done(struct task_struct *tsk)
1022 {
1023 	struct completion *vfork;
1024 
1025 	task_lock(tsk);
1026 	vfork = tsk->vfork_done;
1027 	if (likely(vfork)) {
1028 		tsk->vfork_done = NULL;
1029 		complete(vfork);
1030 	}
1031 	task_unlock(tsk);
1032 }
1033 
1034 static int wait_for_vfork_done(struct task_struct *child,
1035 				struct completion *vfork)
1036 {
1037 	int killed;
1038 
1039 	freezer_do_not_count();
1040 	killed = wait_for_completion_killable(vfork);
1041 	freezer_count();
1042 
1043 	if (killed) {
1044 		task_lock(child);
1045 		child->vfork_done = NULL;
1046 		task_unlock(child);
1047 	}
1048 
1049 	put_task_struct(child);
1050 	return killed;
1051 }
1052 
1053 /* Please note the differences between mmput and mm_release.
1054  * mmput is called whenever we stop holding onto a mm_struct,
1055  * error success whatever.
1056  *
1057  * mm_release is called after a mm_struct has been removed
1058  * from the current process.
1059  *
1060  * This difference is important for error handling, when we
1061  * only half set up a mm_struct for a new process and need to restore
1062  * the old one.  Because we mmput the new mm_struct before
1063  * restoring the old one. . .
1064  * Eric Biederman 10 January 1998
1065  */
1066 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1067 {
1068 	/* Get rid of any futexes when releasing the mm */
1069 #ifdef CONFIG_FUTEX
1070 	if (unlikely(tsk->robust_list)) {
1071 		exit_robust_list(tsk);
1072 		tsk->robust_list = NULL;
1073 	}
1074 #ifdef CONFIG_COMPAT
1075 	if (unlikely(tsk->compat_robust_list)) {
1076 		compat_exit_robust_list(tsk);
1077 		tsk->compat_robust_list = NULL;
1078 	}
1079 #endif
1080 	if (unlikely(!list_empty(&tsk->pi_state_list)))
1081 		exit_pi_state_list(tsk);
1082 #endif
1083 
1084 	uprobe_free_utask(tsk);
1085 
1086 	/* Get rid of any cached register state */
1087 	deactivate_mm(tsk, mm);
1088 
1089 	/*
1090 	 * Signal userspace if we're not exiting with a core dump
1091 	 * because we want to leave the value intact for debugging
1092 	 * purposes.
1093 	 */
1094 	if (tsk->clear_child_tid) {
1095 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1096 		    atomic_read(&mm->mm_users) > 1) {
1097 			/*
1098 			 * We don't check the error code - if userspace has
1099 			 * not set up a proper pointer then tough luck.
1100 			 */
1101 			put_user(0, tsk->clear_child_tid);
1102 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1103 					1, NULL, NULL, 0);
1104 		}
1105 		tsk->clear_child_tid = NULL;
1106 	}
1107 
1108 	/*
1109 	 * All done, finally we can wake up parent and return this mm to him.
1110 	 * Also kthread_stop() uses this completion for synchronization.
1111 	 */
1112 	if (tsk->vfork_done)
1113 		complete_vfork_done(tsk);
1114 }
1115 
1116 /*
1117  * Allocate a new mm structure and copy contents from the
1118  * mm structure of the passed in task structure.
1119  */
1120 static struct mm_struct *dup_mm(struct task_struct *tsk)
1121 {
1122 	struct mm_struct *mm, *oldmm = current->mm;
1123 	int err;
1124 
1125 	mm = allocate_mm();
1126 	if (!mm)
1127 		goto fail_nomem;
1128 
1129 	memcpy(mm, oldmm, sizeof(*mm));
1130 
1131 	if (!mm_init(mm, tsk))
1132 		goto fail_nomem;
1133 
1134 	err = dup_mmap(mm, oldmm);
1135 	if (err)
1136 		goto free_pt;
1137 
1138 	mm->hiwater_rss = get_mm_rss(mm);
1139 	mm->hiwater_vm = mm->total_vm;
1140 
1141 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1142 		goto free_pt;
1143 
1144 	return mm;
1145 
1146 free_pt:
1147 	/* don't put binfmt in mmput, we haven't got module yet */
1148 	mm->binfmt = NULL;
1149 	mmput(mm);
1150 
1151 fail_nomem:
1152 	return NULL;
1153 }
1154 
1155 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1156 {
1157 	struct mm_struct *mm, *oldmm;
1158 	int retval;
1159 
1160 	tsk->min_flt = tsk->maj_flt = 0;
1161 	tsk->nvcsw = tsk->nivcsw = 0;
1162 #ifdef CONFIG_DETECT_HUNG_TASK
1163 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1164 #endif
1165 
1166 	tsk->mm = NULL;
1167 	tsk->active_mm = NULL;
1168 
1169 	/*
1170 	 * Are we cloning a kernel thread?
1171 	 *
1172 	 * We need to steal a active VM for that..
1173 	 */
1174 	oldmm = current->mm;
1175 	if (!oldmm)
1176 		return 0;
1177 
1178 	/* initialize the new vmacache entries */
1179 	vmacache_flush(tsk);
1180 
1181 	if (clone_flags & CLONE_VM) {
1182 		atomic_inc(&oldmm->mm_users);
1183 		mm = oldmm;
1184 		goto good_mm;
1185 	}
1186 
1187 	retval = -ENOMEM;
1188 	mm = dup_mm(tsk);
1189 	if (!mm)
1190 		goto fail_nomem;
1191 
1192 good_mm:
1193 	tsk->mm = mm;
1194 	tsk->active_mm = mm;
1195 	return 0;
1196 
1197 fail_nomem:
1198 	return retval;
1199 }
1200 
1201 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1202 {
1203 	struct fs_struct *fs = current->fs;
1204 	if (clone_flags & CLONE_FS) {
1205 		/* tsk->fs is already what we want */
1206 		spin_lock(&fs->lock);
1207 		if (fs->in_exec) {
1208 			spin_unlock(&fs->lock);
1209 			return -EAGAIN;
1210 		}
1211 		fs->users++;
1212 		spin_unlock(&fs->lock);
1213 		return 0;
1214 	}
1215 	tsk->fs = copy_fs_struct(fs);
1216 	if (!tsk->fs)
1217 		return -ENOMEM;
1218 	return 0;
1219 }
1220 
1221 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1222 {
1223 	struct files_struct *oldf, *newf;
1224 	int error = 0;
1225 
1226 	/*
1227 	 * A background process may not have any files ...
1228 	 */
1229 	oldf = current->files;
1230 	if (!oldf)
1231 		goto out;
1232 
1233 	if (clone_flags & CLONE_FILES) {
1234 		atomic_inc(&oldf->count);
1235 		goto out;
1236 	}
1237 
1238 	newf = dup_fd(oldf, &error);
1239 	if (!newf)
1240 		goto out;
1241 
1242 	tsk->files = newf;
1243 	error = 0;
1244 out:
1245 	return error;
1246 }
1247 
1248 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1249 {
1250 #ifdef CONFIG_BLOCK
1251 	struct io_context *ioc = current->io_context;
1252 	struct io_context *new_ioc;
1253 
1254 	if (!ioc)
1255 		return 0;
1256 	/*
1257 	 * Share io context with parent, if CLONE_IO is set
1258 	 */
1259 	if (clone_flags & CLONE_IO) {
1260 		ioc_task_link(ioc);
1261 		tsk->io_context = ioc;
1262 	} else if (ioprio_valid(ioc->ioprio)) {
1263 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1264 		if (unlikely(!new_ioc))
1265 			return -ENOMEM;
1266 
1267 		new_ioc->ioprio = ioc->ioprio;
1268 		put_io_context(new_ioc);
1269 	}
1270 #endif
1271 	return 0;
1272 }
1273 
1274 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1275 {
1276 	struct sighand_struct *sig;
1277 
1278 	if (clone_flags & CLONE_SIGHAND) {
1279 		atomic_inc(&current->sighand->count);
1280 		return 0;
1281 	}
1282 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1283 	rcu_assign_pointer(tsk->sighand, sig);
1284 	if (!sig)
1285 		return -ENOMEM;
1286 
1287 	atomic_set(&sig->count, 1);
1288 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1289 	return 0;
1290 }
1291 
1292 void __cleanup_sighand(struct sighand_struct *sighand)
1293 {
1294 	if (atomic_dec_and_test(&sighand->count)) {
1295 		signalfd_cleanup(sighand);
1296 		/*
1297 		 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1298 		 * without an RCU grace period, see __lock_task_sighand().
1299 		 */
1300 		kmem_cache_free(sighand_cachep, sighand);
1301 	}
1302 }
1303 
1304 /*
1305  * Initialize POSIX timer handling for a thread group.
1306  */
1307 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1308 {
1309 	unsigned long cpu_limit;
1310 
1311 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1312 	if (cpu_limit != RLIM_INFINITY) {
1313 		sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1314 		sig->cputimer.running = true;
1315 	}
1316 
1317 	/* The timer lists. */
1318 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1319 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1320 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1321 }
1322 
1323 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1324 {
1325 	struct signal_struct *sig;
1326 
1327 	if (clone_flags & CLONE_THREAD)
1328 		return 0;
1329 
1330 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1331 	tsk->signal = sig;
1332 	if (!sig)
1333 		return -ENOMEM;
1334 
1335 	sig->nr_threads = 1;
1336 	atomic_set(&sig->live, 1);
1337 	atomic_set(&sig->sigcnt, 1);
1338 
1339 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1340 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1341 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1342 
1343 	init_waitqueue_head(&sig->wait_chldexit);
1344 	sig->curr_target = tsk;
1345 	init_sigpending(&sig->shared_pending);
1346 	INIT_LIST_HEAD(&sig->posix_timers);
1347 	seqlock_init(&sig->stats_lock);
1348 	prev_cputime_init(&sig->prev_cputime);
1349 
1350 #ifdef CONFIG_POSIX_TIMERS
1351 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1352 	sig->real_timer.function = it_real_fn;
1353 #endif
1354 
1355 	task_lock(current->group_leader);
1356 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1357 	task_unlock(current->group_leader);
1358 
1359 	posix_cpu_timers_init_group(sig);
1360 
1361 	tty_audit_fork(sig);
1362 	sched_autogroup_fork(sig);
1363 
1364 	sig->oom_score_adj = current->signal->oom_score_adj;
1365 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1366 
1367 	sig->has_child_subreaper = current->signal->has_child_subreaper ||
1368 				   current->signal->is_child_subreaper;
1369 
1370 	mutex_init(&sig->cred_guard_mutex);
1371 
1372 	return 0;
1373 }
1374 
1375 static void copy_seccomp(struct task_struct *p)
1376 {
1377 #ifdef CONFIG_SECCOMP
1378 	/*
1379 	 * Must be called with sighand->lock held, which is common to
1380 	 * all threads in the group. Holding cred_guard_mutex is not
1381 	 * needed because this new task is not yet running and cannot
1382 	 * be racing exec.
1383 	 */
1384 	assert_spin_locked(&current->sighand->siglock);
1385 
1386 	/* Ref-count the new filter user, and assign it. */
1387 	get_seccomp_filter(current);
1388 	p->seccomp = current->seccomp;
1389 
1390 	/*
1391 	 * Explicitly enable no_new_privs here in case it got set
1392 	 * between the task_struct being duplicated and holding the
1393 	 * sighand lock. The seccomp state and nnp must be in sync.
1394 	 */
1395 	if (task_no_new_privs(current))
1396 		task_set_no_new_privs(p);
1397 
1398 	/*
1399 	 * If the parent gained a seccomp mode after copying thread
1400 	 * flags and between before we held the sighand lock, we have
1401 	 * to manually enable the seccomp thread flag here.
1402 	 */
1403 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1404 		set_tsk_thread_flag(p, TIF_SECCOMP);
1405 #endif
1406 }
1407 
1408 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1409 {
1410 	current->clear_child_tid = tidptr;
1411 
1412 	return task_pid_vnr(current);
1413 }
1414 
1415 static void rt_mutex_init_task(struct task_struct *p)
1416 {
1417 	raw_spin_lock_init(&p->pi_lock);
1418 #ifdef CONFIG_RT_MUTEXES
1419 	p->pi_waiters = RB_ROOT;
1420 	p->pi_waiters_leftmost = NULL;
1421 	p->pi_blocked_on = NULL;
1422 #endif
1423 }
1424 
1425 /*
1426  * Initialize POSIX timer handling for a single task.
1427  */
1428 static void posix_cpu_timers_init(struct task_struct *tsk)
1429 {
1430 	tsk->cputime_expires.prof_exp = 0;
1431 	tsk->cputime_expires.virt_exp = 0;
1432 	tsk->cputime_expires.sched_exp = 0;
1433 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1434 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1435 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1436 }
1437 
1438 static inline void
1439 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1440 {
1441 	 task->pids[type].pid = pid;
1442 }
1443 
1444 /*
1445  * This creates a new process as a copy of the old one,
1446  * but does not actually start it yet.
1447  *
1448  * It copies the registers, and all the appropriate
1449  * parts of the process environment (as per the clone
1450  * flags). The actual kick-off is left to the caller.
1451  */
1452 static __latent_entropy struct task_struct *copy_process(
1453 					unsigned long clone_flags,
1454 					unsigned long stack_start,
1455 					unsigned long stack_size,
1456 					int __user *child_tidptr,
1457 					struct pid *pid,
1458 					int trace,
1459 					unsigned long tls,
1460 					int node)
1461 {
1462 	int retval;
1463 	struct task_struct *p;
1464 
1465 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1466 		return ERR_PTR(-EINVAL);
1467 
1468 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1469 		return ERR_PTR(-EINVAL);
1470 
1471 	/*
1472 	 * Thread groups must share signals as well, and detached threads
1473 	 * can only be started up within the thread group.
1474 	 */
1475 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1476 		return ERR_PTR(-EINVAL);
1477 
1478 	/*
1479 	 * Shared signal handlers imply shared VM. By way of the above,
1480 	 * thread groups also imply shared VM. Blocking this case allows
1481 	 * for various simplifications in other code.
1482 	 */
1483 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1484 		return ERR_PTR(-EINVAL);
1485 
1486 	/*
1487 	 * Siblings of global init remain as zombies on exit since they are
1488 	 * not reaped by their parent (swapper). To solve this and to avoid
1489 	 * multi-rooted process trees, prevent global and container-inits
1490 	 * from creating siblings.
1491 	 */
1492 	if ((clone_flags & CLONE_PARENT) &&
1493 				current->signal->flags & SIGNAL_UNKILLABLE)
1494 		return ERR_PTR(-EINVAL);
1495 
1496 	/*
1497 	 * If the new process will be in a different pid or user namespace
1498 	 * do not allow it to share a thread group with the forking task.
1499 	 */
1500 	if (clone_flags & CLONE_THREAD) {
1501 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1502 		    (task_active_pid_ns(current) !=
1503 				current->nsproxy->pid_ns_for_children))
1504 			return ERR_PTR(-EINVAL);
1505 	}
1506 
1507 	retval = security_task_create(clone_flags);
1508 	if (retval)
1509 		goto fork_out;
1510 
1511 	retval = -ENOMEM;
1512 	p = dup_task_struct(current, node);
1513 	if (!p)
1514 		goto fork_out;
1515 
1516 	ftrace_graph_init_task(p);
1517 
1518 	rt_mutex_init_task(p);
1519 
1520 #ifdef CONFIG_PROVE_LOCKING
1521 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1522 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1523 #endif
1524 	retval = -EAGAIN;
1525 	if (atomic_read(&p->real_cred->user->processes) >=
1526 			task_rlimit(p, RLIMIT_NPROC)) {
1527 		if (p->real_cred->user != INIT_USER &&
1528 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1529 			goto bad_fork_free;
1530 	}
1531 	current->flags &= ~PF_NPROC_EXCEEDED;
1532 
1533 	retval = copy_creds(p, clone_flags);
1534 	if (retval < 0)
1535 		goto bad_fork_free;
1536 
1537 	/*
1538 	 * If multiple threads are within copy_process(), then this check
1539 	 * triggers too late. This doesn't hurt, the check is only there
1540 	 * to stop root fork bombs.
1541 	 */
1542 	retval = -EAGAIN;
1543 	if (nr_threads >= max_threads)
1544 		goto bad_fork_cleanup_count;
1545 
1546 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1547 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1548 	p->flags |= PF_FORKNOEXEC;
1549 	INIT_LIST_HEAD(&p->children);
1550 	INIT_LIST_HEAD(&p->sibling);
1551 	rcu_copy_process(p);
1552 	p->vfork_done = NULL;
1553 	spin_lock_init(&p->alloc_lock);
1554 
1555 	init_sigpending(&p->pending);
1556 
1557 	p->utime = p->stime = p->gtime = 0;
1558 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1559 	p->utimescaled = p->stimescaled = 0;
1560 #endif
1561 	prev_cputime_init(&p->prev_cputime);
1562 
1563 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1564 	seqcount_init(&p->vtime_seqcount);
1565 	p->vtime_snap = 0;
1566 	p->vtime_snap_whence = VTIME_INACTIVE;
1567 #endif
1568 
1569 #if defined(SPLIT_RSS_COUNTING)
1570 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1571 #endif
1572 
1573 	p->default_timer_slack_ns = current->timer_slack_ns;
1574 
1575 	task_io_accounting_init(&p->ioac);
1576 	acct_clear_integrals(p);
1577 
1578 	posix_cpu_timers_init(p);
1579 
1580 	p->start_time = ktime_get_ns();
1581 	p->real_start_time = ktime_get_boot_ns();
1582 	p->io_context = NULL;
1583 	p->audit_context = NULL;
1584 	cgroup_fork(p);
1585 #ifdef CONFIG_NUMA
1586 	p->mempolicy = mpol_dup(p->mempolicy);
1587 	if (IS_ERR(p->mempolicy)) {
1588 		retval = PTR_ERR(p->mempolicy);
1589 		p->mempolicy = NULL;
1590 		goto bad_fork_cleanup_threadgroup_lock;
1591 	}
1592 #endif
1593 #ifdef CONFIG_CPUSETS
1594 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1595 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1596 	seqcount_init(&p->mems_allowed_seq);
1597 #endif
1598 #ifdef CONFIG_TRACE_IRQFLAGS
1599 	p->irq_events = 0;
1600 	p->hardirqs_enabled = 0;
1601 	p->hardirq_enable_ip = 0;
1602 	p->hardirq_enable_event = 0;
1603 	p->hardirq_disable_ip = _THIS_IP_;
1604 	p->hardirq_disable_event = 0;
1605 	p->softirqs_enabled = 1;
1606 	p->softirq_enable_ip = _THIS_IP_;
1607 	p->softirq_enable_event = 0;
1608 	p->softirq_disable_ip = 0;
1609 	p->softirq_disable_event = 0;
1610 	p->hardirq_context = 0;
1611 	p->softirq_context = 0;
1612 #endif
1613 
1614 	p->pagefault_disabled = 0;
1615 
1616 #ifdef CONFIG_LOCKDEP
1617 	p->lockdep_depth = 0; /* no locks held yet */
1618 	p->curr_chain_key = 0;
1619 	p->lockdep_recursion = 0;
1620 #endif
1621 
1622 #ifdef CONFIG_DEBUG_MUTEXES
1623 	p->blocked_on = NULL; /* not blocked yet */
1624 #endif
1625 #ifdef CONFIG_BCACHE
1626 	p->sequential_io	= 0;
1627 	p->sequential_io_avg	= 0;
1628 #endif
1629 
1630 	/* Perform scheduler related setup. Assign this task to a CPU. */
1631 	retval = sched_fork(clone_flags, p);
1632 	if (retval)
1633 		goto bad_fork_cleanup_policy;
1634 
1635 	retval = perf_event_init_task(p);
1636 	if (retval)
1637 		goto bad_fork_cleanup_policy;
1638 	retval = audit_alloc(p);
1639 	if (retval)
1640 		goto bad_fork_cleanup_perf;
1641 	/* copy all the process information */
1642 	shm_init_task(p);
1643 	retval = copy_semundo(clone_flags, p);
1644 	if (retval)
1645 		goto bad_fork_cleanup_audit;
1646 	retval = copy_files(clone_flags, p);
1647 	if (retval)
1648 		goto bad_fork_cleanup_semundo;
1649 	retval = copy_fs(clone_flags, p);
1650 	if (retval)
1651 		goto bad_fork_cleanup_files;
1652 	retval = copy_sighand(clone_flags, p);
1653 	if (retval)
1654 		goto bad_fork_cleanup_fs;
1655 	retval = copy_signal(clone_flags, p);
1656 	if (retval)
1657 		goto bad_fork_cleanup_sighand;
1658 	retval = copy_mm(clone_flags, p);
1659 	if (retval)
1660 		goto bad_fork_cleanup_signal;
1661 	retval = copy_namespaces(clone_flags, p);
1662 	if (retval)
1663 		goto bad_fork_cleanup_mm;
1664 	retval = copy_io(clone_flags, p);
1665 	if (retval)
1666 		goto bad_fork_cleanup_namespaces;
1667 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1668 	if (retval)
1669 		goto bad_fork_cleanup_io;
1670 
1671 	if (pid != &init_struct_pid) {
1672 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1673 		if (IS_ERR(pid)) {
1674 			retval = PTR_ERR(pid);
1675 			goto bad_fork_cleanup_thread;
1676 		}
1677 	}
1678 
1679 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1680 	/*
1681 	 * Clear TID on mm_release()?
1682 	 */
1683 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1684 #ifdef CONFIG_BLOCK
1685 	p->plug = NULL;
1686 #endif
1687 #ifdef CONFIG_FUTEX
1688 	p->robust_list = NULL;
1689 #ifdef CONFIG_COMPAT
1690 	p->compat_robust_list = NULL;
1691 #endif
1692 	INIT_LIST_HEAD(&p->pi_state_list);
1693 	p->pi_state_cache = NULL;
1694 #endif
1695 	/*
1696 	 * sigaltstack should be cleared when sharing the same VM
1697 	 */
1698 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1699 		sas_ss_reset(p);
1700 
1701 	/*
1702 	 * Syscall tracing and stepping should be turned off in the
1703 	 * child regardless of CLONE_PTRACE.
1704 	 */
1705 	user_disable_single_step(p);
1706 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1707 #ifdef TIF_SYSCALL_EMU
1708 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1709 #endif
1710 	clear_all_latency_tracing(p);
1711 
1712 	/* ok, now we should be set up.. */
1713 	p->pid = pid_nr(pid);
1714 	if (clone_flags & CLONE_THREAD) {
1715 		p->exit_signal = -1;
1716 		p->group_leader = current->group_leader;
1717 		p->tgid = current->tgid;
1718 	} else {
1719 		if (clone_flags & CLONE_PARENT)
1720 			p->exit_signal = current->group_leader->exit_signal;
1721 		else
1722 			p->exit_signal = (clone_flags & CSIGNAL);
1723 		p->group_leader = p;
1724 		p->tgid = p->pid;
1725 	}
1726 
1727 	p->nr_dirtied = 0;
1728 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1729 	p->dirty_paused_when = 0;
1730 
1731 	p->pdeath_signal = 0;
1732 	INIT_LIST_HEAD(&p->thread_group);
1733 	p->task_works = NULL;
1734 
1735 	threadgroup_change_begin(current);
1736 	/*
1737 	 * Ensure that the cgroup subsystem policies allow the new process to be
1738 	 * forked. It should be noted the the new process's css_set can be changed
1739 	 * between here and cgroup_post_fork() if an organisation operation is in
1740 	 * progress.
1741 	 */
1742 	retval = cgroup_can_fork(p);
1743 	if (retval)
1744 		goto bad_fork_free_pid;
1745 
1746 	/*
1747 	 * Make it visible to the rest of the system, but dont wake it up yet.
1748 	 * Need tasklist lock for parent etc handling!
1749 	 */
1750 	write_lock_irq(&tasklist_lock);
1751 
1752 	/* CLONE_PARENT re-uses the old parent */
1753 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1754 		p->real_parent = current->real_parent;
1755 		p->parent_exec_id = current->parent_exec_id;
1756 	} else {
1757 		p->real_parent = current;
1758 		p->parent_exec_id = current->self_exec_id;
1759 	}
1760 
1761 	spin_lock(&current->sighand->siglock);
1762 
1763 	/*
1764 	 * Copy seccomp details explicitly here, in case they were changed
1765 	 * before holding sighand lock.
1766 	 */
1767 	copy_seccomp(p);
1768 
1769 	/*
1770 	 * Process group and session signals need to be delivered to just the
1771 	 * parent before the fork or both the parent and the child after the
1772 	 * fork. Restart if a signal comes in before we add the new process to
1773 	 * it's process group.
1774 	 * A fatal signal pending means that current will exit, so the new
1775 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1776 	*/
1777 	recalc_sigpending();
1778 	if (signal_pending(current)) {
1779 		spin_unlock(&current->sighand->siglock);
1780 		write_unlock_irq(&tasklist_lock);
1781 		retval = -ERESTARTNOINTR;
1782 		goto bad_fork_cancel_cgroup;
1783 	}
1784 
1785 	if (likely(p->pid)) {
1786 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1787 
1788 		init_task_pid(p, PIDTYPE_PID, pid);
1789 		if (thread_group_leader(p)) {
1790 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1791 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1792 
1793 			if (is_child_reaper(pid)) {
1794 				ns_of_pid(pid)->child_reaper = p;
1795 				p->signal->flags |= SIGNAL_UNKILLABLE;
1796 			}
1797 
1798 			p->signal->leader_pid = pid;
1799 			p->signal->tty = tty_kref_get(current->signal->tty);
1800 			list_add_tail(&p->sibling, &p->real_parent->children);
1801 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1802 			attach_pid(p, PIDTYPE_PGID);
1803 			attach_pid(p, PIDTYPE_SID);
1804 			__this_cpu_inc(process_counts);
1805 		} else {
1806 			current->signal->nr_threads++;
1807 			atomic_inc(&current->signal->live);
1808 			atomic_inc(&current->signal->sigcnt);
1809 			list_add_tail_rcu(&p->thread_group,
1810 					  &p->group_leader->thread_group);
1811 			list_add_tail_rcu(&p->thread_node,
1812 					  &p->signal->thread_head);
1813 		}
1814 		attach_pid(p, PIDTYPE_PID);
1815 		nr_threads++;
1816 	}
1817 
1818 	total_forks++;
1819 	spin_unlock(&current->sighand->siglock);
1820 	syscall_tracepoint_update(p);
1821 	write_unlock_irq(&tasklist_lock);
1822 
1823 	proc_fork_connector(p);
1824 	cgroup_post_fork(p);
1825 	threadgroup_change_end(current);
1826 	perf_event_fork(p);
1827 
1828 	trace_task_newtask(p, clone_flags);
1829 	uprobe_copy_process(p, clone_flags);
1830 
1831 	return p;
1832 
1833 bad_fork_cancel_cgroup:
1834 	cgroup_cancel_fork(p);
1835 bad_fork_free_pid:
1836 	threadgroup_change_end(current);
1837 	if (pid != &init_struct_pid)
1838 		free_pid(pid);
1839 bad_fork_cleanup_thread:
1840 	exit_thread(p);
1841 bad_fork_cleanup_io:
1842 	if (p->io_context)
1843 		exit_io_context(p);
1844 bad_fork_cleanup_namespaces:
1845 	exit_task_namespaces(p);
1846 bad_fork_cleanup_mm:
1847 	if (p->mm)
1848 		mmput(p->mm);
1849 bad_fork_cleanup_signal:
1850 	if (!(clone_flags & CLONE_THREAD))
1851 		free_signal_struct(p->signal);
1852 bad_fork_cleanup_sighand:
1853 	__cleanup_sighand(p->sighand);
1854 bad_fork_cleanup_fs:
1855 	exit_fs(p); /* blocking */
1856 bad_fork_cleanup_files:
1857 	exit_files(p); /* blocking */
1858 bad_fork_cleanup_semundo:
1859 	exit_sem(p);
1860 bad_fork_cleanup_audit:
1861 	audit_free(p);
1862 bad_fork_cleanup_perf:
1863 	perf_event_free_task(p);
1864 bad_fork_cleanup_policy:
1865 #ifdef CONFIG_NUMA
1866 	mpol_put(p->mempolicy);
1867 bad_fork_cleanup_threadgroup_lock:
1868 #endif
1869 	delayacct_tsk_free(p);
1870 bad_fork_cleanup_count:
1871 	atomic_dec(&p->cred->user->processes);
1872 	exit_creds(p);
1873 bad_fork_free:
1874 	p->state = TASK_DEAD;
1875 	put_task_stack(p);
1876 	free_task(p);
1877 fork_out:
1878 	return ERR_PTR(retval);
1879 }
1880 
1881 static inline void init_idle_pids(struct pid_link *links)
1882 {
1883 	enum pid_type type;
1884 
1885 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1886 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1887 		links[type].pid = &init_struct_pid;
1888 	}
1889 }
1890 
1891 struct task_struct *fork_idle(int cpu)
1892 {
1893 	struct task_struct *task;
1894 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1895 			    cpu_to_node(cpu));
1896 	if (!IS_ERR(task)) {
1897 		init_idle_pids(task->pids);
1898 		init_idle(task, cpu);
1899 	}
1900 
1901 	return task;
1902 }
1903 
1904 /*
1905  *  Ok, this is the main fork-routine.
1906  *
1907  * It copies the process, and if successful kick-starts
1908  * it and waits for it to finish using the VM if required.
1909  */
1910 long _do_fork(unsigned long clone_flags,
1911 	      unsigned long stack_start,
1912 	      unsigned long stack_size,
1913 	      int __user *parent_tidptr,
1914 	      int __user *child_tidptr,
1915 	      unsigned long tls)
1916 {
1917 	struct task_struct *p;
1918 	int trace = 0;
1919 	long nr;
1920 
1921 	/*
1922 	 * Determine whether and which event to report to ptracer.  When
1923 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
1924 	 * requested, no event is reported; otherwise, report if the event
1925 	 * for the type of forking is enabled.
1926 	 */
1927 	if (!(clone_flags & CLONE_UNTRACED)) {
1928 		if (clone_flags & CLONE_VFORK)
1929 			trace = PTRACE_EVENT_VFORK;
1930 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
1931 			trace = PTRACE_EVENT_CLONE;
1932 		else
1933 			trace = PTRACE_EVENT_FORK;
1934 
1935 		if (likely(!ptrace_event_enabled(current, trace)))
1936 			trace = 0;
1937 	}
1938 
1939 	p = copy_process(clone_flags, stack_start, stack_size,
1940 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1941 	add_latent_entropy();
1942 	/*
1943 	 * Do this prior waking up the new thread - the thread pointer
1944 	 * might get invalid after that point, if the thread exits quickly.
1945 	 */
1946 	if (!IS_ERR(p)) {
1947 		struct completion vfork;
1948 		struct pid *pid;
1949 
1950 		trace_sched_process_fork(current, p);
1951 
1952 		pid = get_task_pid(p, PIDTYPE_PID);
1953 		nr = pid_vnr(pid);
1954 
1955 		if (clone_flags & CLONE_PARENT_SETTID)
1956 			put_user(nr, parent_tidptr);
1957 
1958 		if (clone_flags & CLONE_VFORK) {
1959 			p->vfork_done = &vfork;
1960 			init_completion(&vfork);
1961 			get_task_struct(p);
1962 		}
1963 
1964 		wake_up_new_task(p);
1965 
1966 		/* forking complete and child started to run, tell ptracer */
1967 		if (unlikely(trace))
1968 			ptrace_event_pid(trace, pid);
1969 
1970 		if (clone_flags & CLONE_VFORK) {
1971 			if (!wait_for_vfork_done(p, &vfork))
1972 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1973 		}
1974 
1975 		put_pid(pid);
1976 	} else {
1977 		nr = PTR_ERR(p);
1978 	}
1979 	return nr;
1980 }
1981 
1982 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1983 /* For compatibility with architectures that call do_fork directly rather than
1984  * using the syscall entry points below. */
1985 long do_fork(unsigned long clone_flags,
1986 	      unsigned long stack_start,
1987 	      unsigned long stack_size,
1988 	      int __user *parent_tidptr,
1989 	      int __user *child_tidptr)
1990 {
1991 	return _do_fork(clone_flags, stack_start, stack_size,
1992 			parent_tidptr, child_tidptr, 0);
1993 }
1994 #endif
1995 
1996 /*
1997  * Create a kernel thread.
1998  */
1999 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2000 {
2001 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2002 		(unsigned long)arg, NULL, NULL, 0);
2003 }
2004 
2005 #ifdef __ARCH_WANT_SYS_FORK
2006 SYSCALL_DEFINE0(fork)
2007 {
2008 #ifdef CONFIG_MMU
2009 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2010 #else
2011 	/* can not support in nommu mode */
2012 	return -EINVAL;
2013 #endif
2014 }
2015 #endif
2016 
2017 #ifdef __ARCH_WANT_SYS_VFORK
2018 SYSCALL_DEFINE0(vfork)
2019 {
2020 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2021 			0, NULL, NULL, 0);
2022 }
2023 #endif
2024 
2025 #ifdef __ARCH_WANT_SYS_CLONE
2026 #ifdef CONFIG_CLONE_BACKWARDS
2027 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2028 		 int __user *, parent_tidptr,
2029 		 unsigned long, tls,
2030 		 int __user *, child_tidptr)
2031 #elif defined(CONFIG_CLONE_BACKWARDS2)
2032 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2033 		 int __user *, parent_tidptr,
2034 		 int __user *, child_tidptr,
2035 		 unsigned long, tls)
2036 #elif defined(CONFIG_CLONE_BACKWARDS3)
2037 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2038 		int, stack_size,
2039 		int __user *, parent_tidptr,
2040 		int __user *, child_tidptr,
2041 		unsigned long, tls)
2042 #else
2043 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2044 		 int __user *, parent_tidptr,
2045 		 int __user *, child_tidptr,
2046 		 unsigned long, tls)
2047 #endif
2048 {
2049 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2050 }
2051 #endif
2052 
2053 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2054 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2055 #endif
2056 
2057 static void sighand_ctor(void *data)
2058 {
2059 	struct sighand_struct *sighand = data;
2060 
2061 	spin_lock_init(&sighand->siglock);
2062 	init_waitqueue_head(&sighand->signalfd_wqh);
2063 }
2064 
2065 void __init proc_caches_init(void)
2066 {
2067 	sighand_cachep = kmem_cache_create("sighand_cache",
2068 			sizeof(struct sighand_struct), 0,
2069 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2070 			SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2071 	signal_cachep = kmem_cache_create("signal_cache",
2072 			sizeof(struct signal_struct), 0,
2073 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2074 			NULL);
2075 	files_cachep = kmem_cache_create("files_cache",
2076 			sizeof(struct files_struct), 0,
2077 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2078 			NULL);
2079 	fs_cachep = kmem_cache_create("fs_cache",
2080 			sizeof(struct fs_struct), 0,
2081 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2082 			NULL);
2083 	/*
2084 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2085 	 * whole struct cpumask for the OFFSTACK case. We could change
2086 	 * this to *only* allocate as much of it as required by the
2087 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
2088 	 * is at the end of the structure, exactly for that reason.
2089 	 */
2090 	mm_cachep = kmem_cache_create("mm_struct",
2091 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2092 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2093 			NULL);
2094 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2095 	mmap_init();
2096 	nsproxy_cache_init();
2097 }
2098 
2099 /*
2100  * Check constraints on flags passed to the unshare system call.
2101  */
2102 static int check_unshare_flags(unsigned long unshare_flags)
2103 {
2104 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2105 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2106 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2107 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2108 		return -EINVAL;
2109 	/*
2110 	 * Not implemented, but pretend it works if there is nothing
2111 	 * to unshare.  Note that unsharing the address space or the
2112 	 * signal handlers also need to unshare the signal queues (aka
2113 	 * CLONE_THREAD).
2114 	 */
2115 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2116 		if (!thread_group_empty(current))
2117 			return -EINVAL;
2118 	}
2119 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2120 		if (atomic_read(&current->sighand->count) > 1)
2121 			return -EINVAL;
2122 	}
2123 	if (unshare_flags & CLONE_VM) {
2124 		if (!current_is_single_threaded())
2125 			return -EINVAL;
2126 	}
2127 
2128 	return 0;
2129 }
2130 
2131 /*
2132  * Unshare the filesystem structure if it is being shared
2133  */
2134 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2135 {
2136 	struct fs_struct *fs = current->fs;
2137 
2138 	if (!(unshare_flags & CLONE_FS) || !fs)
2139 		return 0;
2140 
2141 	/* don't need lock here; in the worst case we'll do useless copy */
2142 	if (fs->users == 1)
2143 		return 0;
2144 
2145 	*new_fsp = copy_fs_struct(fs);
2146 	if (!*new_fsp)
2147 		return -ENOMEM;
2148 
2149 	return 0;
2150 }
2151 
2152 /*
2153  * Unshare file descriptor table if it is being shared
2154  */
2155 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2156 {
2157 	struct files_struct *fd = current->files;
2158 	int error = 0;
2159 
2160 	if ((unshare_flags & CLONE_FILES) &&
2161 	    (fd && atomic_read(&fd->count) > 1)) {
2162 		*new_fdp = dup_fd(fd, &error);
2163 		if (!*new_fdp)
2164 			return error;
2165 	}
2166 
2167 	return 0;
2168 }
2169 
2170 /*
2171  * unshare allows a process to 'unshare' part of the process
2172  * context which was originally shared using clone.  copy_*
2173  * functions used by do_fork() cannot be used here directly
2174  * because they modify an inactive task_struct that is being
2175  * constructed. Here we are modifying the current, active,
2176  * task_struct.
2177  */
2178 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2179 {
2180 	struct fs_struct *fs, *new_fs = NULL;
2181 	struct files_struct *fd, *new_fd = NULL;
2182 	struct cred *new_cred = NULL;
2183 	struct nsproxy *new_nsproxy = NULL;
2184 	int do_sysvsem = 0;
2185 	int err;
2186 
2187 	/*
2188 	 * If unsharing a user namespace must also unshare the thread group
2189 	 * and unshare the filesystem root and working directories.
2190 	 */
2191 	if (unshare_flags & CLONE_NEWUSER)
2192 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2193 	/*
2194 	 * If unsharing vm, must also unshare signal handlers.
2195 	 */
2196 	if (unshare_flags & CLONE_VM)
2197 		unshare_flags |= CLONE_SIGHAND;
2198 	/*
2199 	 * If unsharing a signal handlers, must also unshare the signal queues.
2200 	 */
2201 	if (unshare_flags & CLONE_SIGHAND)
2202 		unshare_flags |= CLONE_THREAD;
2203 	/*
2204 	 * If unsharing namespace, must also unshare filesystem information.
2205 	 */
2206 	if (unshare_flags & CLONE_NEWNS)
2207 		unshare_flags |= CLONE_FS;
2208 
2209 	err = check_unshare_flags(unshare_flags);
2210 	if (err)
2211 		goto bad_unshare_out;
2212 	/*
2213 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2214 	 * to a new ipc namespace, the semaphore arrays from the old
2215 	 * namespace are unreachable.
2216 	 */
2217 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2218 		do_sysvsem = 1;
2219 	err = unshare_fs(unshare_flags, &new_fs);
2220 	if (err)
2221 		goto bad_unshare_out;
2222 	err = unshare_fd(unshare_flags, &new_fd);
2223 	if (err)
2224 		goto bad_unshare_cleanup_fs;
2225 	err = unshare_userns(unshare_flags, &new_cred);
2226 	if (err)
2227 		goto bad_unshare_cleanup_fd;
2228 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2229 					 new_cred, new_fs);
2230 	if (err)
2231 		goto bad_unshare_cleanup_cred;
2232 
2233 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2234 		if (do_sysvsem) {
2235 			/*
2236 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2237 			 */
2238 			exit_sem(current);
2239 		}
2240 		if (unshare_flags & CLONE_NEWIPC) {
2241 			/* Orphan segments in old ns (see sem above). */
2242 			exit_shm(current);
2243 			shm_init_task(current);
2244 		}
2245 
2246 		if (new_nsproxy)
2247 			switch_task_namespaces(current, new_nsproxy);
2248 
2249 		task_lock(current);
2250 
2251 		if (new_fs) {
2252 			fs = current->fs;
2253 			spin_lock(&fs->lock);
2254 			current->fs = new_fs;
2255 			if (--fs->users)
2256 				new_fs = NULL;
2257 			else
2258 				new_fs = fs;
2259 			spin_unlock(&fs->lock);
2260 		}
2261 
2262 		if (new_fd) {
2263 			fd = current->files;
2264 			current->files = new_fd;
2265 			new_fd = fd;
2266 		}
2267 
2268 		task_unlock(current);
2269 
2270 		if (new_cred) {
2271 			/* Install the new user namespace */
2272 			commit_creds(new_cred);
2273 			new_cred = NULL;
2274 		}
2275 	}
2276 
2277 bad_unshare_cleanup_cred:
2278 	if (new_cred)
2279 		put_cred(new_cred);
2280 bad_unshare_cleanup_fd:
2281 	if (new_fd)
2282 		put_files_struct(new_fd);
2283 
2284 bad_unshare_cleanup_fs:
2285 	if (new_fs)
2286 		free_fs_struct(new_fs);
2287 
2288 bad_unshare_out:
2289 	return err;
2290 }
2291 
2292 /*
2293  *	Helper to unshare the files of the current task.
2294  *	We don't want to expose copy_files internals to
2295  *	the exec layer of the kernel.
2296  */
2297 
2298 int unshare_files(struct files_struct **displaced)
2299 {
2300 	struct task_struct *task = current;
2301 	struct files_struct *copy = NULL;
2302 	int error;
2303 
2304 	error = unshare_fd(CLONE_FILES, &copy);
2305 	if (error || !copy) {
2306 		*displaced = NULL;
2307 		return error;
2308 	}
2309 	*displaced = task->files;
2310 	task_lock(task);
2311 	task->files = copy;
2312 	task_unlock(task);
2313 	return 0;
2314 }
2315 
2316 int sysctl_max_threads(struct ctl_table *table, int write,
2317 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2318 {
2319 	struct ctl_table t;
2320 	int ret;
2321 	int threads = max_threads;
2322 	int min = MIN_THREADS;
2323 	int max = MAX_THREADS;
2324 
2325 	t = *table;
2326 	t.data = &threads;
2327 	t.extra1 = &min;
2328 	t.extra2 = &max;
2329 
2330 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2331 	if (ret || !write)
2332 		return ret;
2333 
2334 	set_max_threads(threads);
2335 
2336 	return 0;
2337 }
2338