xref: /linux-6.15/kernel/fork.c (revision 95813b8f)
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 	if (page)
168 		memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
169 					    1 << THREAD_SIZE_ORDER);
170 
171 	return page ? page_address(page) : NULL;
172 }
173 
174 static inline void free_thread_info(struct thread_info *ti)
175 {
176 	struct page *page = virt_to_page(ti);
177 
178 	memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
179 				    -(1 << THREAD_SIZE_ORDER));
180 	__free_kmem_pages(page, THREAD_SIZE_ORDER);
181 }
182 # else
183 static struct kmem_cache *thread_info_cache;
184 
185 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
186 						  int node)
187 {
188 	return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
189 }
190 
191 static void free_thread_info(struct thread_info *ti)
192 {
193 	kmem_cache_free(thread_info_cache, ti);
194 }
195 
196 void thread_info_cache_init(void)
197 {
198 	thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
199 					      THREAD_SIZE, 0, NULL);
200 	BUG_ON(thread_info_cache == NULL);
201 }
202 # endif
203 #endif
204 
205 /* SLAB cache for signal_struct structures (tsk->signal) */
206 static struct kmem_cache *signal_cachep;
207 
208 /* SLAB cache for sighand_struct structures (tsk->sighand) */
209 struct kmem_cache *sighand_cachep;
210 
211 /* SLAB cache for files_struct structures (tsk->files) */
212 struct kmem_cache *files_cachep;
213 
214 /* SLAB cache for fs_struct structures (tsk->fs) */
215 struct kmem_cache *fs_cachep;
216 
217 /* SLAB cache for vm_area_struct structures */
218 struct kmem_cache *vm_area_cachep;
219 
220 /* SLAB cache for mm_struct structures (tsk->mm) */
221 static struct kmem_cache *mm_cachep;
222 
223 static void account_kernel_stack(struct thread_info *ti, int account)
224 {
225 	struct zone *zone = page_zone(virt_to_page(ti));
226 
227 	mod_zone_page_state(zone, NR_KERNEL_STACK, account);
228 }
229 
230 void free_task(struct task_struct *tsk)
231 {
232 	account_kernel_stack(tsk->stack, -1);
233 	arch_release_thread_info(tsk->stack);
234 	free_thread_info(tsk->stack);
235 	rt_mutex_debug_task_free(tsk);
236 	ftrace_graph_exit_task(tsk);
237 	put_seccomp_filter(tsk);
238 	arch_release_task_struct(tsk);
239 	free_task_struct(tsk);
240 }
241 EXPORT_SYMBOL(free_task);
242 
243 static inline void free_signal_struct(struct signal_struct *sig)
244 {
245 	taskstats_tgid_free(sig);
246 	sched_autogroup_exit(sig);
247 	kmem_cache_free(signal_cachep, sig);
248 }
249 
250 static inline void put_signal_struct(struct signal_struct *sig)
251 {
252 	if (atomic_dec_and_test(&sig->sigcnt))
253 		free_signal_struct(sig);
254 }
255 
256 void __put_task_struct(struct task_struct *tsk)
257 {
258 	WARN_ON(!tsk->exit_state);
259 	WARN_ON(atomic_read(&tsk->usage));
260 	WARN_ON(tsk == current);
261 
262 	cgroup_free(tsk);
263 	task_numa_free(tsk);
264 	security_task_free(tsk);
265 	exit_creds(tsk);
266 	delayacct_tsk_free(tsk);
267 	put_signal_struct(tsk->signal);
268 
269 	if (!profile_handoff_task(tsk))
270 		free_task(tsk);
271 }
272 EXPORT_SYMBOL_GPL(__put_task_struct);
273 
274 void __init __weak arch_task_cache_init(void) { }
275 
276 /*
277  * set_max_threads
278  */
279 static void set_max_threads(unsigned int max_threads_suggested)
280 {
281 	u64 threads;
282 
283 	/*
284 	 * The number of threads shall be limited such that the thread
285 	 * structures may only consume a small part of the available memory.
286 	 */
287 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
288 		threads = MAX_THREADS;
289 	else
290 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
291 				    (u64) THREAD_SIZE * 8UL);
292 
293 	if (threads > max_threads_suggested)
294 		threads = max_threads_suggested;
295 
296 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
297 }
298 
299 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
300 /* Initialized by the architecture: */
301 int arch_task_struct_size __read_mostly;
302 #endif
303 
304 void __init fork_init(void)
305 {
306 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
307 #ifndef ARCH_MIN_TASKALIGN
308 #define ARCH_MIN_TASKALIGN	L1_CACHE_BYTES
309 #endif
310 	/* create a slab on which task_structs can be allocated */
311 	task_struct_cachep = kmem_cache_create("task_struct",
312 			arch_task_struct_size, ARCH_MIN_TASKALIGN,
313 			SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
314 #endif
315 
316 	/* do the arch specific task caches init */
317 	arch_task_cache_init();
318 
319 	set_max_threads(MAX_THREADS);
320 
321 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
322 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
323 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
324 		init_task.signal->rlim[RLIMIT_NPROC];
325 }
326 
327 int __weak arch_dup_task_struct(struct task_struct *dst,
328 					       struct task_struct *src)
329 {
330 	*dst = *src;
331 	return 0;
332 }
333 
334 void set_task_stack_end_magic(struct task_struct *tsk)
335 {
336 	unsigned long *stackend;
337 
338 	stackend = end_of_stack(tsk);
339 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
340 }
341 
342 static struct task_struct *dup_task_struct(struct task_struct *orig)
343 {
344 	struct task_struct *tsk;
345 	struct thread_info *ti;
346 	int node = tsk_fork_get_node(orig);
347 	int err;
348 
349 	tsk = alloc_task_struct_node(node);
350 	if (!tsk)
351 		return NULL;
352 
353 	ti = alloc_thread_info_node(tsk, node);
354 	if (!ti)
355 		goto free_tsk;
356 
357 	err = arch_dup_task_struct(tsk, orig);
358 	if (err)
359 		goto free_ti;
360 
361 	tsk->stack = ti;
362 #ifdef CONFIG_SECCOMP
363 	/*
364 	 * We must handle setting up seccomp filters once we're under
365 	 * the sighand lock in case orig has changed between now and
366 	 * then. Until then, filter must be NULL to avoid messing up
367 	 * the usage counts on the error path calling free_task.
368 	 */
369 	tsk->seccomp.filter = NULL;
370 #endif
371 
372 	setup_thread_stack(tsk, orig);
373 	clear_user_return_notifier(tsk);
374 	clear_tsk_need_resched(tsk);
375 	set_task_stack_end_magic(tsk);
376 
377 #ifdef CONFIG_CC_STACKPROTECTOR
378 	tsk->stack_canary = get_random_int();
379 #endif
380 
381 	/*
382 	 * One for us, one for whoever does the "release_task()" (usually
383 	 * parent)
384 	 */
385 	atomic_set(&tsk->usage, 2);
386 #ifdef CONFIG_BLK_DEV_IO_TRACE
387 	tsk->btrace_seq = 0;
388 #endif
389 	tsk->splice_pipe = NULL;
390 	tsk->task_frag.page = NULL;
391 	tsk->wake_q.next = NULL;
392 
393 	account_kernel_stack(ti, 1);
394 
395 	return tsk;
396 
397 free_ti:
398 	free_thread_info(ti);
399 free_tsk:
400 	free_task_struct(tsk);
401 	return NULL;
402 }
403 
404 #ifdef CONFIG_MMU
405 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
406 {
407 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
408 	struct rb_node **rb_link, *rb_parent;
409 	int retval;
410 	unsigned long charge;
411 
412 	uprobe_start_dup_mmap();
413 	down_write(&oldmm->mmap_sem);
414 	flush_cache_dup_mm(oldmm);
415 	uprobe_dup_mmap(oldmm, mm);
416 	/*
417 	 * Not linked in yet - no deadlock potential:
418 	 */
419 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
420 
421 	/* No ordering required: file already has been exposed. */
422 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
423 
424 	mm->total_vm = oldmm->total_vm;
425 	mm->data_vm = oldmm->data_vm;
426 	mm->exec_vm = oldmm->exec_vm;
427 	mm->stack_vm = oldmm->stack_vm;
428 
429 	rb_link = &mm->mm_rb.rb_node;
430 	rb_parent = NULL;
431 	pprev = &mm->mmap;
432 	retval = ksm_fork(mm, oldmm);
433 	if (retval)
434 		goto out;
435 	retval = khugepaged_fork(mm, oldmm);
436 	if (retval)
437 		goto out;
438 
439 	prev = NULL;
440 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
441 		struct file *file;
442 
443 		if (mpnt->vm_flags & VM_DONTCOPY) {
444 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
445 			continue;
446 		}
447 		charge = 0;
448 		if (mpnt->vm_flags & VM_ACCOUNT) {
449 			unsigned long len = vma_pages(mpnt);
450 
451 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
452 				goto fail_nomem;
453 			charge = len;
454 		}
455 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
456 		if (!tmp)
457 			goto fail_nomem;
458 		*tmp = *mpnt;
459 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
460 		retval = vma_dup_policy(mpnt, tmp);
461 		if (retval)
462 			goto fail_nomem_policy;
463 		tmp->vm_mm = mm;
464 		if (anon_vma_fork(tmp, mpnt))
465 			goto fail_nomem_anon_vma_fork;
466 		tmp->vm_flags &=
467 			~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
468 		tmp->vm_next = tmp->vm_prev = NULL;
469 		tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
470 		file = tmp->vm_file;
471 		if (file) {
472 			struct inode *inode = file_inode(file);
473 			struct address_space *mapping = file->f_mapping;
474 
475 			get_file(file);
476 			if (tmp->vm_flags & VM_DENYWRITE)
477 				atomic_dec(&inode->i_writecount);
478 			i_mmap_lock_write(mapping);
479 			if (tmp->vm_flags & VM_SHARED)
480 				atomic_inc(&mapping->i_mmap_writable);
481 			flush_dcache_mmap_lock(mapping);
482 			/* insert tmp into the share list, just after mpnt */
483 			vma_interval_tree_insert_after(tmp, mpnt,
484 					&mapping->i_mmap);
485 			flush_dcache_mmap_unlock(mapping);
486 			i_mmap_unlock_write(mapping);
487 		}
488 
489 		/*
490 		 * Clear hugetlb-related page reserves for children. This only
491 		 * affects MAP_PRIVATE mappings. Faults generated by the child
492 		 * are not guaranteed to succeed, even if read-only
493 		 */
494 		if (is_vm_hugetlb_page(tmp))
495 			reset_vma_resv_huge_pages(tmp);
496 
497 		/*
498 		 * Link in the new vma and copy the page table entries.
499 		 */
500 		*pprev = tmp;
501 		pprev = &tmp->vm_next;
502 		tmp->vm_prev = prev;
503 		prev = tmp;
504 
505 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
506 		rb_link = &tmp->vm_rb.rb_right;
507 		rb_parent = &tmp->vm_rb;
508 
509 		mm->map_count++;
510 		retval = copy_page_range(mm, oldmm, mpnt);
511 
512 		if (tmp->vm_ops && tmp->vm_ops->open)
513 			tmp->vm_ops->open(tmp);
514 
515 		if (retval)
516 			goto out;
517 	}
518 	/* a new mm has just been created */
519 	arch_dup_mmap(oldmm, mm);
520 	retval = 0;
521 out:
522 	up_write(&mm->mmap_sem);
523 	flush_tlb_mm(oldmm);
524 	up_write(&oldmm->mmap_sem);
525 	uprobe_end_dup_mmap();
526 	return retval;
527 fail_nomem_anon_vma_fork:
528 	mpol_put(vma_policy(tmp));
529 fail_nomem_policy:
530 	kmem_cache_free(vm_area_cachep, tmp);
531 fail_nomem:
532 	retval = -ENOMEM;
533 	vm_unacct_memory(charge);
534 	goto out;
535 }
536 
537 static inline int mm_alloc_pgd(struct mm_struct *mm)
538 {
539 	mm->pgd = pgd_alloc(mm);
540 	if (unlikely(!mm->pgd))
541 		return -ENOMEM;
542 	return 0;
543 }
544 
545 static inline void mm_free_pgd(struct mm_struct *mm)
546 {
547 	pgd_free(mm, mm->pgd);
548 }
549 #else
550 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
551 {
552 	down_write(&oldmm->mmap_sem);
553 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
554 	up_write(&oldmm->mmap_sem);
555 	return 0;
556 }
557 #define mm_alloc_pgd(mm)	(0)
558 #define mm_free_pgd(mm)
559 #endif /* CONFIG_MMU */
560 
561 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
562 
563 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
564 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
565 
566 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
567 
568 static int __init coredump_filter_setup(char *s)
569 {
570 	default_dump_filter =
571 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
572 		MMF_DUMP_FILTER_MASK;
573 	return 1;
574 }
575 
576 __setup("coredump_filter=", coredump_filter_setup);
577 
578 #include <linux/init_task.h>
579 
580 static void mm_init_aio(struct mm_struct *mm)
581 {
582 #ifdef CONFIG_AIO
583 	spin_lock_init(&mm->ioctx_lock);
584 	mm->ioctx_table = NULL;
585 #endif
586 }
587 
588 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
589 {
590 #ifdef CONFIG_MEMCG
591 	mm->owner = p;
592 #endif
593 }
594 
595 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
596 {
597 	mm->mmap = NULL;
598 	mm->mm_rb = RB_ROOT;
599 	mm->vmacache_seqnum = 0;
600 	atomic_set(&mm->mm_users, 1);
601 	atomic_set(&mm->mm_count, 1);
602 	init_rwsem(&mm->mmap_sem);
603 	INIT_LIST_HEAD(&mm->mmlist);
604 	mm->core_state = NULL;
605 	atomic_long_set(&mm->nr_ptes, 0);
606 	mm_nr_pmds_init(mm);
607 	mm->map_count = 0;
608 	mm->locked_vm = 0;
609 	mm->pinned_vm = 0;
610 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
611 	spin_lock_init(&mm->page_table_lock);
612 	mm_init_cpumask(mm);
613 	mm_init_aio(mm);
614 	mm_init_owner(mm, p);
615 	mmu_notifier_mm_init(mm);
616 	clear_tlb_flush_pending(mm);
617 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
618 	mm->pmd_huge_pte = NULL;
619 #endif
620 
621 	if (current->mm) {
622 		mm->flags = current->mm->flags & MMF_INIT_MASK;
623 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
624 	} else {
625 		mm->flags = default_dump_filter;
626 		mm->def_flags = 0;
627 	}
628 
629 	if (mm_alloc_pgd(mm))
630 		goto fail_nopgd;
631 
632 	if (init_new_context(p, mm))
633 		goto fail_nocontext;
634 
635 	return mm;
636 
637 fail_nocontext:
638 	mm_free_pgd(mm);
639 fail_nopgd:
640 	free_mm(mm);
641 	return NULL;
642 }
643 
644 static void check_mm(struct mm_struct *mm)
645 {
646 	int i;
647 
648 	for (i = 0; i < NR_MM_COUNTERS; i++) {
649 		long x = atomic_long_read(&mm->rss_stat.count[i]);
650 
651 		if (unlikely(x))
652 			printk(KERN_ALERT "BUG: Bad rss-counter state "
653 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
654 	}
655 
656 	if (atomic_long_read(&mm->nr_ptes))
657 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
658 				atomic_long_read(&mm->nr_ptes));
659 	if (mm_nr_pmds(mm))
660 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
661 				mm_nr_pmds(mm));
662 
663 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
664 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
665 #endif
666 }
667 
668 /*
669  * Allocate and initialize an mm_struct.
670  */
671 struct mm_struct *mm_alloc(void)
672 {
673 	struct mm_struct *mm;
674 
675 	mm = allocate_mm();
676 	if (!mm)
677 		return NULL;
678 
679 	memset(mm, 0, sizeof(*mm));
680 	return mm_init(mm, current);
681 }
682 
683 /*
684  * Called when the last reference to the mm
685  * is dropped: either by a lazy thread or by
686  * mmput. Free the page directory and the mm.
687  */
688 void __mmdrop(struct mm_struct *mm)
689 {
690 	BUG_ON(mm == &init_mm);
691 	mm_free_pgd(mm);
692 	destroy_context(mm);
693 	mmu_notifier_mm_destroy(mm);
694 	check_mm(mm);
695 	free_mm(mm);
696 }
697 EXPORT_SYMBOL_GPL(__mmdrop);
698 
699 /*
700  * Decrement the use count and release all resources for an mm.
701  */
702 void mmput(struct mm_struct *mm)
703 {
704 	might_sleep();
705 
706 	if (atomic_dec_and_test(&mm->mm_users)) {
707 		uprobe_clear_state(mm);
708 		exit_aio(mm);
709 		ksm_exit(mm);
710 		khugepaged_exit(mm); /* must run before exit_mmap */
711 		exit_mmap(mm);
712 		set_mm_exe_file(mm, NULL);
713 		if (!list_empty(&mm->mmlist)) {
714 			spin_lock(&mmlist_lock);
715 			list_del(&mm->mmlist);
716 			spin_unlock(&mmlist_lock);
717 		}
718 		if (mm->binfmt)
719 			module_put(mm->binfmt->module);
720 		mmdrop(mm);
721 	}
722 }
723 EXPORT_SYMBOL_GPL(mmput);
724 
725 /**
726  * set_mm_exe_file - change a reference to the mm's executable file
727  *
728  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
729  *
730  * Main users are mmput() and sys_execve(). Callers prevent concurrent
731  * invocations: in mmput() nobody alive left, in execve task is single
732  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
733  * mm->exe_file, but does so without using set_mm_exe_file() in order
734  * to do avoid the need for any locks.
735  */
736 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
737 {
738 	struct file *old_exe_file;
739 
740 	/*
741 	 * It is safe to dereference the exe_file without RCU as
742 	 * this function is only called if nobody else can access
743 	 * this mm -- see comment above for justification.
744 	 */
745 	old_exe_file = rcu_dereference_raw(mm->exe_file);
746 
747 	if (new_exe_file)
748 		get_file(new_exe_file);
749 	rcu_assign_pointer(mm->exe_file, new_exe_file);
750 	if (old_exe_file)
751 		fput(old_exe_file);
752 }
753 
754 /**
755  * get_mm_exe_file - acquire a reference to the mm's executable file
756  *
757  * Returns %NULL if mm has no associated executable file.
758  * User must release file via fput().
759  */
760 struct file *get_mm_exe_file(struct mm_struct *mm)
761 {
762 	struct file *exe_file;
763 
764 	rcu_read_lock();
765 	exe_file = rcu_dereference(mm->exe_file);
766 	if (exe_file && !get_file_rcu(exe_file))
767 		exe_file = NULL;
768 	rcu_read_unlock();
769 	return exe_file;
770 }
771 EXPORT_SYMBOL(get_mm_exe_file);
772 
773 /**
774  * get_task_mm - acquire a reference to the task's mm
775  *
776  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
777  * this kernel workthread has transiently adopted a user mm with use_mm,
778  * to do its AIO) is not set and if so returns a reference to it, after
779  * bumping up the use count.  User must release the mm via mmput()
780  * after use.  Typically used by /proc and ptrace.
781  */
782 struct mm_struct *get_task_mm(struct task_struct *task)
783 {
784 	struct mm_struct *mm;
785 
786 	task_lock(task);
787 	mm = task->mm;
788 	if (mm) {
789 		if (task->flags & PF_KTHREAD)
790 			mm = NULL;
791 		else
792 			atomic_inc(&mm->mm_users);
793 	}
794 	task_unlock(task);
795 	return mm;
796 }
797 EXPORT_SYMBOL_GPL(get_task_mm);
798 
799 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
800 {
801 	struct mm_struct *mm;
802 	int err;
803 
804 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
805 	if (err)
806 		return ERR_PTR(err);
807 
808 	mm = get_task_mm(task);
809 	if (mm && mm != current->mm &&
810 			!ptrace_may_access(task, mode)) {
811 		mmput(mm);
812 		mm = ERR_PTR(-EACCES);
813 	}
814 	mutex_unlock(&task->signal->cred_guard_mutex);
815 
816 	return mm;
817 }
818 
819 static void complete_vfork_done(struct task_struct *tsk)
820 {
821 	struct completion *vfork;
822 
823 	task_lock(tsk);
824 	vfork = tsk->vfork_done;
825 	if (likely(vfork)) {
826 		tsk->vfork_done = NULL;
827 		complete(vfork);
828 	}
829 	task_unlock(tsk);
830 }
831 
832 static int wait_for_vfork_done(struct task_struct *child,
833 				struct completion *vfork)
834 {
835 	int killed;
836 
837 	freezer_do_not_count();
838 	killed = wait_for_completion_killable(vfork);
839 	freezer_count();
840 
841 	if (killed) {
842 		task_lock(child);
843 		child->vfork_done = NULL;
844 		task_unlock(child);
845 	}
846 
847 	put_task_struct(child);
848 	return killed;
849 }
850 
851 /* Please note the differences between mmput and mm_release.
852  * mmput is called whenever we stop holding onto a mm_struct,
853  * error success whatever.
854  *
855  * mm_release is called after a mm_struct has been removed
856  * from the current process.
857  *
858  * This difference is important for error handling, when we
859  * only half set up a mm_struct for a new process and need to restore
860  * the old one.  Because we mmput the new mm_struct before
861  * restoring the old one. . .
862  * Eric Biederman 10 January 1998
863  */
864 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
865 {
866 	/* Get rid of any futexes when releasing the mm */
867 #ifdef CONFIG_FUTEX
868 	if (unlikely(tsk->robust_list)) {
869 		exit_robust_list(tsk);
870 		tsk->robust_list = NULL;
871 	}
872 #ifdef CONFIG_COMPAT
873 	if (unlikely(tsk->compat_robust_list)) {
874 		compat_exit_robust_list(tsk);
875 		tsk->compat_robust_list = NULL;
876 	}
877 #endif
878 	if (unlikely(!list_empty(&tsk->pi_state_list)))
879 		exit_pi_state_list(tsk);
880 #endif
881 
882 	uprobe_free_utask(tsk);
883 
884 	/* Get rid of any cached register state */
885 	deactivate_mm(tsk, mm);
886 
887 	/*
888 	 * If we're exiting normally, clear a user-space tid field if
889 	 * requested.  We leave this alone when dying by signal, to leave
890 	 * the value intact in a core dump, and to save the unnecessary
891 	 * trouble, say, a killed vfork parent shouldn't touch this mm.
892 	 * Userland only wants this done for a sys_exit.
893 	 */
894 	if (tsk->clear_child_tid) {
895 		if (!(tsk->flags & PF_SIGNALED) &&
896 		    atomic_read(&mm->mm_users) > 1) {
897 			/*
898 			 * We don't check the error code - if userspace has
899 			 * not set up a proper pointer then tough luck.
900 			 */
901 			put_user(0, tsk->clear_child_tid);
902 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
903 					1, NULL, NULL, 0);
904 		}
905 		tsk->clear_child_tid = NULL;
906 	}
907 
908 	/*
909 	 * All done, finally we can wake up parent and return this mm to him.
910 	 * Also kthread_stop() uses this completion for synchronization.
911 	 */
912 	if (tsk->vfork_done)
913 		complete_vfork_done(tsk);
914 }
915 
916 /*
917  * Allocate a new mm structure and copy contents from the
918  * mm structure of the passed in task structure.
919  */
920 static struct mm_struct *dup_mm(struct task_struct *tsk)
921 {
922 	struct mm_struct *mm, *oldmm = current->mm;
923 	int err;
924 
925 	mm = allocate_mm();
926 	if (!mm)
927 		goto fail_nomem;
928 
929 	memcpy(mm, oldmm, sizeof(*mm));
930 
931 	if (!mm_init(mm, tsk))
932 		goto fail_nomem;
933 
934 	err = dup_mmap(mm, oldmm);
935 	if (err)
936 		goto free_pt;
937 
938 	mm->hiwater_rss = get_mm_rss(mm);
939 	mm->hiwater_vm = mm->total_vm;
940 
941 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
942 		goto free_pt;
943 
944 	return mm;
945 
946 free_pt:
947 	/* don't put binfmt in mmput, we haven't got module yet */
948 	mm->binfmt = NULL;
949 	mmput(mm);
950 
951 fail_nomem:
952 	return NULL;
953 }
954 
955 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
956 {
957 	struct mm_struct *mm, *oldmm;
958 	int retval;
959 
960 	tsk->min_flt = tsk->maj_flt = 0;
961 	tsk->nvcsw = tsk->nivcsw = 0;
962 #ifdef CONFIG_DETECT_HUNG_TASK
963 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
964 #endif
965 
966 	tsk->mm = NULL;
967 	tsk->active_mm = NULL;
968 
969 	/*
970 	 * Are we cloning a kernel thread?
971 	 *
972 	 * We need to steal a active VM for that..
973 	 */
974 	oldmm = current->mm;
975 	if (!oldmm)
976 		return 0;
977 
978 	/* initialize the new vmacache entries */
979 	vmacache_flush(tsk);
980 
981 	if (clone_flags & CLONE_VM) {
982 		atomic_inc(&oldmm->mm_users);
983 		mm = oldmm;
984 		goto good_mm;
985 	}
986 
987 	retval = -ENOMEM;
988 	mm = dup_mm(tsk);
989 	if (!mm)
990 		goto fail_nomem;
991 
992 good_mm:
993 	tsk->mm = mm;
994 	tsk->active_mm = mm;
995 	return 0;
996 
997 fail_nomem:
998 	return retval;
999 }
1000 
1001 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1002 {
1003 	struct fs_struct *fs = current->fs;
1004 	if (clone_flags & CLONE_FS) {
1005 		/* tsk->fs is already what we want */
1006 		spin_lock(&fs->lock);
1007 		if (fs->in_exec) {
1008 			spin_unlock(&fs->lock);
1009 			return -EAGAIN;
1010 		}
1011 		fs->users++;
1012 		spin_unlock(&fs->lock);
1013 		return 0;
1014 	}
1015 	tsk->fs = copy_fs_struct(fs);
1016 	if (!tsk->fs)
1017 		return -ENOMEM;
1018 	return 0;
1019 }
1020 
1021 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1022 {
1023 	struct files_struct *oldf, *newf;
1024 	int error = 0;
1025 
1026 	/*
1027 	 * A background process may not have any files ...
1028 	 */
1029 	oldf = current->files;
1030 	if (!oldf)
1031 		goto out;
1032 
1033 	if (clone_flags & CLONE_FILES) {
1034 		atomic_inc(&oldf->count);
1035 		goto out;
1036 	}
1037 
1038 	newf = dup_fd(oldf, &error);
1039 	if (!newf)
1040 		goto out;
1041 
1042 	tsk->files = newf;
1043 	error = 0;
1044 out:
1045 	return error;
1046 }
1047 
1048 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1049 {
1050 #ifdef CONFIG_BLOCK
1051 	struct io_context *ioc = current->io_context;
1052 	struct io_context *new_ioc;
1053 
1054 	if (!ioc)
1055 		return 0;
1056 	/*
1057 	 * Share io context with parent, if CLONE_IO is set
1058 	 */
1059 	if (clone_flags & CLONE_IO) {
1060 		ioc_task_link(ioc);
1061 		tsk->io_context = ioc;
1062 	} else if (ioprio_valid(ioc->ioprio)) {
1063 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1064 		if (unlikely(!new_ioc))
1065 			return -ENOMEM;
1066 
1067 		new_ioc->ioprio = ioc->ioprio;
1068 		put_io_context(new_ioc);
1069 	}
1070 #endif
1071 	return 0;
1072 }
1073 
1074 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1075 {
1076 	struct sighand_struct *sig;
1077 
1078 	if (clone_flags & CLONE_SIGHAND) {
1079 		atomic_inc(&current->sighand->count);
1080 		return 0;
1081 	}
1082 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1083 	rcu_assign_pointer(tsk->sighand, sig);
1084 	if (!sig)
1085 		return -ENOMEM;
1086 
1087 	atomic_set(&sig->count, 1);
1088 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1089 	return 0;
1090 }
1091 
1092 void __cleanup_sighand(struct sighand_struct *sighand)
1093 {
1094 	if (atomic_dec_and_test(&sighand->count)) {
1095 		signalfd_cleanup(sighand);
1096 		/*
1097 		 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1098 		 * without an RCU grace period, see __lock_task_sighand().
1099 		 */
1100 		kmem_cache_free(sighand_cachep, sighand);
1101 	}
1102 }
1103 
1104 /*
1105  * Initialize POSIX timer handling for a thread group.
1106  */
1107 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1108 {
1109 	unsigned long cpu_limit;
1110 
1111 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1112 	if (cpu_limit != RLIM_INFINITY) {
1113 		sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1114 		sig->cputimer.running = true;
1115 	}
1116 
1117 	/* The timer lists. */
1118 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1119 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1120 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1121 }
1122 
1123 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1124 {
1125 	struct signal_struct *sig;
1126 
1127 	if (clone_flags & CLONE_THREAD)
1128 		return 0;
1129 
1130 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1131 	tsk->signal = sig;
1132 	if (!sig)
1133 		return -ENOMEM;
1134 
1135 	sig->nr_threads = 1;
1136 	atomic_set(&sig->live, 1);
1137 	atomic_set(&sig->sigcnt, 1);
1138 
1139 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1140 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1141 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1142 
1143 	init_waitqueue_head(&sig->wait_chldexit);
1144 	sig->curr_target = tsk;
1145 	init_sigpending(&sig->shared_pending);
1146 	INIT_LIST_HEAD(&sig->posix_timers);
1147 	seqlock_init(&sig->stats_lock);
1148 	prev_cputime_init(&sig->prev_cputime);
1149 
1150 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1151 	sig->real_timer.function = it_real_fn;
1152 
1153 	task_lock(current->group_leader);
1154 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1155 	task_unlock(current->group_leader);
1156 
1157 	posix_cpu_timers_init_group(sig);
1158 
1159 	tty_audit_fork(sig);
1160 	sched_autogroup_fork(sig);
1161 
1162 	sig->oom_score_adj = current->signal->oom_score_adj;
1163 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1164 
1165 	sig->has_child_subreaper = current->signal->has_child_subreaper ||
1166 				   current->signal->is_child_subreaper;
1167 
1168 	mutex_init(&sig->cred_guard_mutex);
1169 
1170 	return 0;
1171 }
1172 
1173 static void copy_seccomp(struct task_struct *p)
1174 {
1175 #ifdef CONFIG_SECCOMP
1176 	/*
1177 	 * Must be called with sighand->lock held, which is common to
1178 	 * all threads in the group. Holding cred_guard_mutex is not
1179 	 * needed because this new task is not yet running and cannot
1180 	 * be racing exec.
1181 	 */
1182 	assert_spin_locked(&current->sighand->siglock);
1183 
1184 	/* Ref-count the new filter user, and assign it. */
1185 	get_seccomp_filter(current);
1186 	p->seccomp = current->seccomp;
1187 
1188 	/*
1189 	 * Explicitly enable no_new_privs here in case it got set
1190 	 * between the task_struct being duplicated and holding the
1191 	 * sighand lock. The seccomp state and nnp must be in sync.
1192 	 */
1193 	if (task_no_new_privs(current))
1194 		task_set_no_new_privs(p);
1195 
1196 	/*
1197 	 * If the parent gained a seccomp mode after copying thread
1198 	 * flags and between before we held the sighand lock, we have
1199 	 * to manually enable the seccomp thread flag here.
1200 	 */
1201 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1202 		set_tsk_thread_flag(p, TIF_SECCOMP);
1203 #endif
1204 }
1205 
1206 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1207 {
1208 	current->clear_child_tid = tidptr;
1209 
1210 	return task_pid_vnr(current);
1211 }
1212 
1213 static void rt_mutex_init_task(struct task_struct *p)
1214 {
1215 	raw_spin_lock_init(&p->pi_lock);
1216 #ifdef CONFIG_RT_MUTEXES
1217 	p->pi_waiters = RB_ROOT;
1218 	p->pi_waiters_leftmost = NULL;
1219 	p->pi_blocked_on = NULL;
1220 #endif
1221 }
1222 
1223 /*
1224  * Initialize POSIX timer handling for a single task.
1225  */
1226 static void posix_cpu_timers_init(struct task_struct *tsk)
1227 {
1228 	tsk->cputime_expires.prof_exp = 0;
1229 	tsk->cputime_expires.virt_exp = 0;
1230 	tsk->cputime_expires.sched_exp = 0;
1231 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1232 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1233 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1234 }
1235 
1236 static inline void
1237 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1238 {
1239 	 task->pids[type].pid = pid;
1240 }
1241 
1242 /*
1243  * This creates a new process as a copy of the old one,
1244  * but does not actually start it yet.
1245  *
1246  * It copies the registers, and all the appropriate
1247  * parts of the process environment (as per the clone
1248  * flags). The actual kick-off is left to the caller.
1249  */
1250 static struct task_struct *copy_process(unsigned long clone_flags,
1251 					unsigned long stack_start,
1252 					unsigned long stack_size,
1253 					int __user *child_tidptr,
1254 					struct pid *pid,
1255 					int trace,
1256 					unsigned long tls)
1257 {
1258 	int retval;
1259 	struct task_struct *p;
1260 
1261 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1262 		return ERR_PTR(-EINVAL);
1263 
1264 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1265 		return ERR_PTR(-EINVAL);
1266 
1267 	/*
1268 	 * Thread groups must share signals as well, and detached threads
1269 	 * can only be started up within the thread group.
1270 	 */
1271 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1272 		return ERR_PTR(-EINVAL);
1273 
1274 	/*
1275 	 * Shared signal handlers imply shared VM. By way of the above,
1276 	 * thread groups also imply shared VM. Blocking this case allows
1277 	 * for various simplifications in other code.
1278 	 */
1279 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1280 		return ERR_PTR(-EINVAL);
1281 
1282 	/*
1283 	 * Siblings of global init remain as zombies on exit since they are
1284 	 * not reaped by their parent (swapper). To solve this and to avoid
1285 	 * multi-rooted process trees, prevent global and container-inits
1286 	 * from creating siblings.
1287 	 */
1288 	if ((clone_flags & CLONE_PARENT) &&
1289 				current->signal->flags & SIGNAL_UNKILLABLE)
1290 		return ERR_PTR(-EINVAL);
1291 
1292 	/*
1293 	 * If the new process will be in a different pid or user namespace
1294 	 * do not allow it to share a thread group with the forking task.
1295 	 */
1296 	if (clone_flags & CLONE_THREAD) {
1297 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1298 		    (task_active_pid_ns(current) !=
1299 				current->nsproxy->pid_ns_for_children))
1300 			return ERR_PTR(-EINVAL);
1301 	}
1302 
1303 	retval = security_task_create(clone_flags);
1304 	if (retval)
1305 		goto fork_out;
1306 
1307 	retval = -ENOMEM;
1308 	p = dup_task_struct(current);
1309 	if (!p)
1310 		goto fork_out;
1311 
1312 	ftrace_graph_init_task(p);
1313 
1314 	rt_mutex_init_task(p);
1315 
1316 #ifdef CONFIG_PROVE_LOCKING
1317 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1318 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1319 #endif
1320 	retval = -EAGAIN;
1321 	if (atomic_read(&p->real_cred->user->processes) >=
1322 			task_rlimit(p, RLIMIT_NPROC)) {
1323 		if (p->real_cred->user != INIT_USER &&
1324 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1325 			goto bad_fork_free;
1326 	}
1327 	current->flags &= ~PF_NPROC_EXCEEDED;
1328 
1329 	retval = copy_creds(p, clone_flags);
1330 	if (retval < 0)
1331 		goto bad_fork_free;
1332 
1333 	/*
1334 	 * If multiple threads are within copy_process(), then this check
1335 	 * triggers too late. This doesn't hurt, the check is only there
1336 	 * to stop root fork bombs.
1337 	 */
1338 	retval = -EAGAIN;
1339 	if (nr_threads >= max_threads)
1340 		goto bad_fork_cleanup_count;
1341 
1342 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1343 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1344 	p->flags |= PF_FORKNOEXEC;
1345 	INIT_LIST_HEAD(&p->children);
1346 	INIT_LIST_HEAD(&p->sibling);
1347 	rcu_copy_process(p);
1348 	p->vfork_done = NULL;
1349 	spin_lock_init(&p->alloc_lock);
1350 
1351 	init_sigpending(&p->pending);
1352 
1353 	p->utime = p->stime = p->gtime = 0;
1354 	p->utimescaled = p->stimescaled = 0;
1355 	prev_cputime_init(&p->prev_cputime);
1356 
1357 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1358 	seqcount_init(&p->vtime_seqcount);
1359 	p->vtime_snap = 0;
1360 	p->vtime_snap_whence = VTIME_INACTIVE;
1361 #endif
1362 
1363 #if defined(SPLIT_RSS_COUNTING)
1364 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1365 #endif
1366 
1367 	p->default_timer_slack_ns = current->timer_slack_ns;
1368 
1369 	task_io_accounting_init(&p->ioac);
1370 	acct_clear_integrals(p);
1371 
1372 	posix_cpu_timers_init(p);
1373 
1374 	p->start_time = ktime_get_ns();
1375 	p->real_start_time = ktime_get_boot_ns();
1376 	p->io_context = NULL;
1377 	p->audit_context = NULL;
1378 	threadgroup_change_begin(current);
1379 	cgroup_fork(p);
1380 #ifdef CONFIG_NUMA
1381 	p->mempolicy = mpol_dup(p->mempolicy);
1382 	if (IS_ERR(p->mempolicy)) {
1383 		retval = PTR_ERR(p->mempolicy);
1384 		p->mempolicy = NULL;
1385 		goto bad_fork_cleanup_threadgroup_lock;
1386 	}
1387 #endif
1388 #ifdef CONFIG_CPUSETS
1389 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1390 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1391 	seqcount_init(&p->mems_allowed_seq);
1392 #endif
1393 #ifdef CONFIG_TRACE_IRQFLAGS
1394 	p->irq_events = 0;
1395 	p->hardirqs_enabled = 0;
1396 	p->hardirq_enable_ip = 0;
1397 	p->hardirq_enable_event = 0;
1398 	p->hardirq_disable_ip = _THIS_IP_;
1399 	p->hardirq_disable_event = 0;
1400 	p->softirqs_enabled = 1;
1401 	p->softirq_enable_ip = _THIS_IP_;
1402 	p->softirq_enable_event = 0;
1403 	p->softirq_disable_ip = 0;
1404 	p->softirq_disable_event = 0;
1405 	p->hardirq_context = 0;
1406 	p->softirq_context = 0;
1407 #endif
1408 
1409 	p->pagefault_disabled = 0;
1410 
1411 #ifdef CONFIG_LOCKDEP
1412 	p->lockdep_depth = 0; /* no locks held yet */
1413 	p->curr_chain_key = 0;
1414 	p->lockdep_recursion = 0;
1415 #endif
1416 
1417 #ifdef CONFIG_DEBUG_MUTEXES
1418 	p->blocked_on = NULL; /* not blocked yet */
1419 #endif
1420 #ifdef CONFIG_BCACHE
1421 	p->sequential_io	= 0;
1422 	p->sequential_io_avg	= 0;
1423 #endif
1424 
1425 	/* Perform scheduler related setup. Assign this task to a CPU. */
1426 	retval = sched_fork(clone_flags, p);
1427 	if (retval)
1428 		goto bad_fork_cleanup_policy;
1429 
1430 	retval = perf_event_init_task(p);
1431 	if (retval)
1432 		goto bad_fork_cleanup_policy;
1433 	retval = audit_alloc(p);
1434 	if (retval)
1435 		goto bad_fork_cleanup_perf;
1436 	/* copy all the process information */
1437 	shm_init_task(p);
1438 	retval = copy_semundo(clone_flags, p);
1439 	if (retval)
1440 		goto bad_fork_cleanup_audit;
1441 	retval = copy_files(clone_flags, p);
1442 	if (retval)
1443 		goto bad_fork_cleanup_semundo;
1444 	retval = copy_fs(clone_flags, p);
1445 	if (retval)
1446 		goto bad_fork_cleanup_files;
1447 	retval = copy_sighand(clone_flags, p);
1448 	if (retval)
1449 		goto bad_fork_cleanup_fs;
1450 	retval = copy_signal(clone_flags, p);
1451 	if (retval)
1452 		goto bad_fork_cleanup_sighand;
1453 	retval = copy_mm(clone_flags, p);
1454 	if (retval)
1455 		goto bad_fork_cleanup_signal;
1456 	retval = copy_namespaces(clone_flags, p);
1457 	if (retval)
1458 		goto bad_fork_cleanup_mm;
1459 	retval = copy_io(clone_flags, p);
1460 	if (retval)
1461 		goto bad_fork_cleanup_namespaces;
1462 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1463 	if (retval)
1464 		goto bad_fork_cleanup_io;
1465 
1466 	if (pid != &init_struct_pid) {
1467 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1468 		if (IS_ERR(pid)) {
1469 			retval = PTR_ERR(pid);
1470 			goto bad_fork_cleanup_io;
1471 		}
1472 	}
1473 
1474 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1475 	/*
1476 	 * Clear TID on mm_release()?
1477 	 */
1478 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1479 #ifdef CONFIG_BLOCK
1480 	p->plug = NULL;
1481 #endif
1482 #ifdef CONFIG_FUTEX
1483 	p->robust_list = NULL;
1484 #ifdef CONFIG_COMPAT
1485 	p->compat_robust_list = NULL;
1486 #endif
1487 	INIT_LIST_HEAD(&p->pi_state_list);
1488 	p->pi_state_cache = NULL;
1489 #endif
1490 	/*
1491 	 * sigaltstack should be cleared when sharing the same VM
1492 	 */
1493 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1494 		p->sas_ss_sp = p->sas_ss_size = 0;
1495 
1496 	/*
1497 	 * Syscall tracing and stepping should be turned off in the
1498 	 * child regardless of CLONE_PTRACE.
1499 	 */
1500 	user_disable_single_step(p);
1501 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1502 #ifdef TIF_SYSCALL_EMU
1503 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1504 #endif
1505 	clear_all_latency_tracing(p);
1506 
1507 	/* ok, now we should be set up.. */
1508 	p->pid = pid_nr(pid);
1509 	if (clone_flags & CLONE_THREAD) {
1510 		p->exit_signal = -1;
1511 		p->group_leader = current->group_leader;
1512 		p->tgid = current->tgid;
1513 	} else {
1514 		if (clone_flags & CLONE_PARENT)
1515 			p->exit_signal = current->group_leader->exit_signal;
1516 		else
1517 			p->exit_signal = (clone_flags & CSIGNAL);
1518 		p->group_leader = p;
1519 		p->tgid = p->pid;
1520 	}
1521 
1522 	p->nr_dirtied = 0;
1523 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1524 	p->dirty_paused_when = 0;
1525 
1526 	p->pdeath_signal = 0;
1527 	INIT_LIST_HEAD(&p->thread_group);
1528 	p->task_works = NULL;
1529 
1530 	/*
1531 	 * Ensure that the cgroup subsystem policies allow the new process to be
1532 	 * forked. It should be noted the the new process's css_set can be changed
1533 	 * between here and cgroup_post_fork() if an organisation operation is in
1534 	 * progress.
1535 	 */
1536 	retval = cgroup_can_fork(p);
1537 	if (retval)
1538 		goto bad_fork_free_pid;
1539 
1540 	/*
1541 	 * Make it visible to the rest of the system, but dont wake it up yet.
1542 	 * Need tasklist lock for parent etc handling!
1543 	 */
1544 	write_lock_irq(&tasklist_lock);
1545 
1546 	/* CLONE_PARENT re-uses the old parent */
1547 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1548 		p->real_parent = current->real_parent;
1549 		p->parent_exec_id = current->parent_exec_id;
1550 	} else {
1551 		p->real_parent = current;
1552 		p->parent_exec_id = current->self_exec_id;
1553 	}
1554 
1555 	spin_lock(&current->sighand->siglock);
1556 
1557 	/*
1558 	 * Copy seccomp details explicitly here, in case they were changed
1559 	 * before holding sighand lock.
1560 	 */
1561 	copy_seccomp(p);
1562 
1563 	/*
1564 	 * Process group and session signals need to be delivered to just the
1565 	 * parent before the fork or both the parent and the child after the
1566 	 * fork. Restart if a signal comes in before we add the new process to
1567 	 * it's process group.
1568 	 * A fatal signal pending means that current will exit, so the new
1569 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1570 	*/
1571 	recalc_sigpending();
1572 	if (signal_pending(current)) {
1573 		spin_unlock(&current->sighand->siglock);
1574 		write_unlock_irq(&tasklist_lock);
1575 		retval = -ERESTARTNOINTR;
1576 		goto bad_fork_cancel_cgroup;
1577 	}
1578 
1579 	if (likely(p->pid)) {
1580 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1581 
1582 		init_task_pid(p, PIDTYPE_PID, pid);
1583 		if (thread_group_leader(p)) {
1584 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1585 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1586 
1587 			if (is_child_reaper(pid)) {
1588 				ns_of_pid(pid)->child_reaper = p;
1589 				p->signal->flags |= SIGNAL_UNKILLABLE;
1590 			}
1591 
1592 			p->signal->leader_pid = pid;
1593 			p->signal->tty = tty_kref_get(current->signal->tty);
1594 			list_add_tail(&p->sibling, &p->real_parent->children);
1595 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1596 			attach_pid(p, PIDTYPE_PGID);
1597 			attach_pid(p, PIDTYPE_SID);
1598 			__this_cpu_inc(process_counts);
1599 		} else {
1600 			current->signal->nr_threads++;
1601 			atomic_inc(&current->signal->live);
1602 			atomic_inc(&current->signal->sigcnt);
1603 			list_add_tail_rcu(&p->thread_group,
1604 					  &p->group_leader->thread_group);
1605 			list_add_tail_rcu(&p->thread_node,
1606 					  &p->signal->thread_head);
1607 		}
1608 		attach_pid(p, PIDTYPE_PID);
1609 		nr_threads++;
1610 	}
1611 
1612 	total_forks++;
1613 	spin_unlock(&current->sighand->siglock);
1614 	syscall_tracepoint_update(p);
1615 	write_unlock_irq(&tasklist_lock);
1616 
1617 	proc_fork_connector(p);
1618 	cgroup_post_fork(p);
1619 	threadgroup_change_end(current);
1620 	perf_event_fork(p);
1621 
1622 	trace_task_newtask(p, clone_flags);
1623 	uprobe_copy_process(p, clone_flags);
1624 
1625 	return p;
1626 
1627 bad_fork_cancel_cgroup:
1628 	cgroup_cancel_fork(p);
1629 bad_fork_free_pid:
1630 	if (pid != &init_struct_pid)
1631 		free_pid(pid);
1632 bad_fork_cleanup_io:
1633 	if (p->io_context)
1634 		exit_io_context(p);
1635 bad_fork_cleanup_namespaces:
1636 	exit_task_namespaces(p);
1637 bad_fork_cleanup_mm:
1638 	if (p->mm)
1639 		mmput(p->mm);
1640 bad_fork_cleanup_signal:
1641 	if (!(clone_flags & CLONE_THREAD))
1642 		free_signal_struct(p->signal);
1643 bad_fork_cleanup_sighand:
1644 	__cleanup_sighand(p->sighand);
1645 bad_fork_cleanup_fs:
1646 	exit_fs(p); /* blocking */
1647 bad_fork_cleanup_files:
1648 	exit_files(p); /* blocking */
1649 bad_fork_cleanup_semundo:
1650 	exit_sem(p);
1651 bad_fork_cleanup_audit:
1652 	audit_free(p);
1653 bad_fork_cleanup_perf:
1654 	perf_event_free_task(p);
1655 bad_fork_cleanup_policy:
1656 #ifdef CONFIG_NUMA
1657 	mpol_put(p->mempolicy);
1658 bad_fork_cleanup_threadgroup_lock:
1659 #endif
1660 	threadgroup_change_end(current);
1661 	delayacct_tsk_free(p);
1662 bad_fork_cleanup_count:
1663 	atomic_dec(&p->cred->user->processes);
1664 	exit_creds(p);
1665 bad_fork_free:
1666 	free_task(p);
1667 fork_out:
1668 	return ERR_PTR(retval);
1669 }
1670 
1671 static inline void init_idle_pids(struct pid_link *links)
1672 {
1673 	enum pid_type type;
1674 
1675 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1676 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1677 		links[type].pid = &init_struct_pid;
1678 	}
1679 }
1680 
1681 struct task_struct *fork_idle(int cpu)
1682 {
1683 	struct task_struct *task;
1684 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1685 	if (!IS_ERR(task)) {
1686 		init_idle_pids(task->pids);
1687 		init_idle(task, cpu);
1688 	}
1689 
1690 	return task;
1691 }
1692 
1693 /*
1694  *  Ok, this is the main fork-routine.
1695  *
1696  * It copies the process, and if successful kick-starts
1697  * it and waits for it to finish using the VM if required.
1698  */
1699 long _do_fork(unsigned long clone_flags,
1700 	      unsigned long stack_start,
1701 	      unsigned long stack_size,
1702 	      int __user *parent_tidptr,
1703 	      int __user *child_tidptr,
1704 	      unsigned long tls)
1705 {
1706 	struct task_struct *p;
1707 	int trace = 0;
1708 	long nr;
1709 
1710 	/*
1711 	 * Determine whether and which event to report to ptracer.  When
1712 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
1713 	 * requested, no event is reported; otherwise, report if the event
1714 	 * for the type of forking is enabled.
1715 	 */
1716 	if (!(clone_flags & CLONE_UNTRACED)) {
1717 		if (clone_flags & CLONE_VFORK)
1718 			trace = PTRACE_EVENT_VFORK;
1719 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
1720 			trace = PTRACE_EVENT_CLONE;
1721 		else
1722 			trace = PTRACE_EVENT_FORK;
1723 
1724 		if (likely(!ptrace_event_enabled(current, trace)))
1725 			trace = 0;
1726 	}
1727 
1728 	p = copy_process(clone_flags, stack_start, stack_size,
1729 			 child_tidptr, NULL, trace, tls);
1730 	/*
1731 	 * Do this prior waking up the new thread - the thread pointer
1732 	 * might get invalid after that point, if the thread exits quickly.
1733 	 */
1734 	if (!IS_ERR(p)) {
1735 		struct completion vfork;
1736 		struct pid *pid;
1737 
1738 		trace_sched_process_fork(current, p);
1739 
1740 		pid = get_task_pid(p, PIDTYPE_PID);
1741 		nr = pid_vnr(pid);
1742 
1743 		if (clone_flags & CLONE_PARENT_SETTID)
1744 			put_user(nr, parent_tidptr);
1745 
1746 		if (clone_flags & CLONE_VFORK) {
1747 			p->vfork_done = &vfork;
1748 			init_completion(&vfork);
1749 			get_task_struct(p);
1750 		}
1751 
1752 		wake_up_new_task(p);
1753 
1754 		/* forking complete and child started to run, tell ptracer */
1755 		if (unlikely(trace))
1756 			ptrace_event_pid(trace, pid);
1757 
1758 		if (clone_flags & CLONE_VFORK) {
1759 			if (!wait_for_vfork_done(p, &vfork))
1760 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1761 		}
1762 
1763 		put_pid(pid);
1764 	} else {
1765 		nr = PTR_ERR(p);
1766 	}
1767 	return nr;
1768 }
1769 
1770 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1771 /* For compatibility with architectures that call do_fork directly rather than
1772  * using the syscall entry points below. */
1773 long do_fork(unsigned long clone_flags,
1774 	      unsigned long stack_start,
1775 	      unsigned long stack_size,
1776 	      int __user *parent_tidptr,
1777 	      int __user *child_tidptr)
1778 {
1779 	return _do_fork(clone_flags, stack_start, stack_size,
1780 			parent_tidptr, child_tidptr, 0);
1781 }
1782 #endif
1783 
1784 /*
1785  * Create a kernel thread.
1786  */
1787 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1788 {
1789 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1790 		(unsigned long)arg, NULL, NULL, 0);
1791 }
1792 
1793 #ifdef __ARCH_WANT_SYS_FORK
1794 SYSCALL_DEFINE0(fork)
1795 {
1796 #ifdef CONFIG_MMU
1797 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1798 #else
1799 	/* can not support in nommu mode */
1800 	return -EINVAL;
1801 #endif
1802 }
1803 #endif
1804 
1805 #ifdef __ARCH_WANT_SYS_VFORK
1806 SYSCALL_DEFINE0(vfork)
1807 {
1808 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1809 			0, NULL, NULL, 0);
1810 }
1811 #endif
1812 
1813 #ifdef __ARCH_WANT_SYS_CLONE
1814 #ifdef CONFIG_CLONE_BACKWARDS
1815 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1816 		 int __user *, parent_tidptr,
1817 		 unsigned long, tls,
1818 		 int __user *, child_tidptr)
1819 #elif defined(CONFIG_CLONE_BACKWARDS2)
1820 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1821 		 int __user *, parent_tidptr,
1822 		 int __user *, child_tidptr,
1823 		 unsigned long, tls)
1824 #elif defined(CONFIG_CLONE_BACKWARDS3)
1825 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1826 		int, stack_size,
1827 		int __user *, parent_tidptr,
1828 		int __user *, child_tidptr,
1829 		unsigned long, tls)
1830 #else
1831 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1832 		 int __user *, parent_tidptr,
1833 		 int __user *, child_tidptr,
1834 		 unsigned long, tls)
1835 #endif
1836 {
1837 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1838 }
1839 #endif
1840 
1841 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1842 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1843 #endif
1844 
1845 static void sighand_ctor(void *data)
1846 {
1847 	struct sighand_struct *sighand = data;
1848 
1849 	spin_lock_init(&sighand->siglock);
1850 	init_waitqueue_head(&sighand->signalfd_wqh);
1851 }
1852 
1853 void __init proc_caches_init(void)
1854 {
1855 	sighand_cachep = kmem_cache_create("sighand_cache",
1856 			sizeof(struct sighand_struct), 0,
1857 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1858 			SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
1859 	signal_cachep = kmem_cache_create("signal_cache",
1860 			sizeof(struct signal_struct), 0,
1861 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1862 			NULL);
1863 	files_cachep = kmem_cache_create("files_cache",
1864 			sizeof(struct files_struct), 0,
1865 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1866 			NULL);
1867 	fs_cachep = kmem_cache_create("fs_cache",
1868 			sizeof(struct fs_struct), 0,
1869 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1870 			NULL);
1871 	/*
1872 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1873 	 * whole struct cpumask for the OFFSTACK case. We could change
1874 	 * this to *only* allocate as much of it as required by the
1875 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
1876 	 * is at the end of the structure, exactly for that reason.
1877 	 */
1878 	mm_cachep = kmem_cache_create("mm_struct",
1879 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1880 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1881 			NULL);
1882 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
1883 	mmap_init();
1884 	nsproxy_cache_init();
1885 }
1886 
1887 /*
1888  * Check constraints on flags passed to the unshare system call.
1889  */
1890 static int check_unshare_flags(unsigned long unshare_flags)
1891 {
1892 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1893 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1894 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1895 				CLONE_NEWUSER|CLONE_NEWPID))
1896 		return -EINVAL;
1897 	/*
1898 	 * Not implemented, but pretend it works if there is nothing
1899 	 * to unshare.  Note that unsharing the address space or the
1900 	 * signal handlers also need to unshare the signal queues (aka
1901 	 * CLONE_THREAD).
1902 	 */
1903 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1904 		if (!thread_group_empty(current))
1905 			return -EINVAL;
1906 	}
1907 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1908 		if (atomic_read(&current->sighand->count) > 1)
1909 			return -EINVAL;
1910 	}
1911 	if (unshare_flags & CLONE_VM) {
1912 		if (!current_is_single_threaded())
1913 			return -EINVAL;
1914 	}
1915 
1916 	return 0;
1917 }
1918 
1919 /*
1920  * Unshare the filesystem structure if it is being shared
1921  */
1922 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1923 {
1924 	struct fs_struct *fs = current->fs;
1925 
1926 	if (!(unshare_flags & CLONE_FS) || !fs)
1927 		return 0;
1928 
1929 	/* don't need lock here; in the worst case we'll do useless copy */
1930 	if (fs->users == 1)
1931 		return 0;
1932 
1933 	*new_fsp = copy_fs_struct(fs);
1934 	if (!*new_fsp)
1935 		return -ENOMEM;
1936 
1937 	return 0;
1938 }
1939 
1940 /*
1941  * Unshare file descriptor table if it is being shared
1942  */
1943 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1944 {
1945 	struct files_struct *fd = current->files;
1946 	int error = 0;
1947 
1948 	if ((unshare_flags & CLONE_FILES) &&
1949 	    (fd && atomic_read(&fd->count) > 1)) {
1950 		*new_fdp = dup_fd(fd, &error);
1951 		if (!*new_fdp)
1952 			return error;
1953 	}
1954 
1955 	return 0;
1956 }
1957 
1958 /*
1959  * unshare allows a process to 'unshare' part of the process
1960  * context which was originally shared using clone.  copy_*
1961  * functions used by do_fork() cannot be used here directly
1962  * because they modify an inactive task_struct that is being
1963  * constructed. Here we are modifying the current, active,
1964  * task_struct.
1965  */
1966 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1967 {
1968 	struct fs_struct *fs, *new_fs = NULL;
1969 	struct files_struct *fd, *new_fd = NULL;
1970 	struct cred *new_cred = NULL;
1971 	struct nsproxy *new_nsproxy = NULL;
1972 	int do_sysvsem = 0;
1973 	int err;
1974 
1975 	/*
1976 	 * If unsharing a user namespace must also unshare the thread group
1977 	 * and unshare the filesystem root and working directories.
1978 	 */
1979 	if (unshare_flags & CLONE_NEWUSER)
1980 		unshare_flags |= CLONE_THREAD | CLONE_FS;
1981 	/*
1982 	 * If unsharing vm, must also unshare signal handlers.
1983 	 */
1984 	if (unshare_flags & CLONE_VM)
1985 		unshare_flags |= CLONE_SIGHAND;
1986 	/*
1987 	 * If unsharing a signal handlers, must also unshare the signal queues.
1988 	 */
1989 	if (unshare_flags & CLONE_SIGHAND)
1990 		unshare_flags |= CLONE_THREAD;
1991 	/*
1992 	 * If unsharing namespace, must also unshare filesystem information.
1993 	 */
1994 	if (unshare_flags & CLONE_NEWNS)
1995 		unshare_flags |= CLONE_FS;
1996 
1997 	err = check_unshare_flags(unshare_flags);
1998 	if (err)
1999 		goto bad_unshare_out;
2000 	/*
2001 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2002 	 * to a new ipc namespace, the semaphore arrays from the old
2003 	 * namespace are unreachable.
2004 	 */
2005 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2006 		do_sysvsem = 1;
2007 	err = unshare_fs(unshare_flags, &new_fs);
2008 	if (err)
2009 		goto bad_unshare_out;
2010 	err = unshare_fd(unshare_flags, &new_fd);
2011 	if (err)
2012 		goto bad_unshare_cleanup_fs;
2013 	err = unshare_userns(unshare_flags, &new_cred);
2014 	if (err)
2015 		goto bad_unshare_cleanup_fd;
2016 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2017 					 new_cred, new_fs);
2018 	if (err)
2019 		goto bad_unshare_cleanup_cred;
2020 
2021 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2022 		if (do_sysvsem) {
2023 			/*
2024 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2025 			 */
2026 			exit_sem(current);
2027 		}
2028 		if (unshare_flags & CLONE_NEWIPC) {
2029 			/* Orphan segments in old ns (see sem above). */
2030 			exit_shm(current);
2031 			shm_init_task(current);
2032 		}
2033 
2034 		if (new_nsproxy)
2035 			switch_task_namespaces(current, new_nsproxy);
2036 
2037 		task_lock(current);
2038 
2039 		if (new_fs) {
2040 			fs = current->fs;
2041 			spin_lock(&fs->lock);
2042 			current->fs = new_fs;
2043 			if (--fs->users)
2044 				new_fs = NULL;
2045 			else
2046 				new_fs = fs;
2047 			spin_unlock(&fs->lock);
2048 		}
2049 
2050 		if (new_fd) {
2051 			fd = current->files;
2052 			current->files = new_fd;
2053 			new_fd = fd;
2054 		}
2055 
2056 		task_unlock(current);
2057 
2058 		if (new_cred) {
2059 			/* Install the new user namespace */
2060 			commit_creds(new_cred);
2061 			new_cred = NULL;
2062 		}
2063 	}
2064 
2065 bad_unshare_cleanup_cred:
2066 	if (new_cred)
2067 		put_cred(new_cred);
2068 bad_unshare_cleanup_fd:
2069 	if (new_fd)
2070 		put_files_struct(new_fd);
2071 
2072 bad_unshare_cleanup_fs:
2073 	if (new_fs)
2074 		free_fs_struct(new_fs);
2075 
2076 bad_unshare_out:
2077 	return err;
2078 }
2079 
2080 /*
2081  *	Helper to unshare the files of the current task.
2082  *	We don't want to expose copy_files internals to
2083  *	the exec layer of the kernel.
2084  */
2085 
2086 int unshare_files(struct files_struct **displaced)
2087 {
2088 	struct task_struct *task = current;
2089 	struct files_struct *copy = NULL;
2090 	int error;
2091 
2092 	error = unshare_fd(CLONE_FILES, &copy);
2093 	if (error || !copy) {
2094 		*displaced = NULL;
2095 		return error;
2096 	}
2097 	*displaced = task->files;
2098 	task_lock(task);
2099 	task->files = copy;
2100 	task_unlock(task);
2101 	return 0;
2102 }
2103 
2104 int sysctl_max_threads(struct ctl_table *table, int write,
2105 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2106 {
2107 	struct ctl_table t;
2108 	int ret;
2109 	int threads = max_threads;
2110 	int min = MIN_THREADS;
2111 	int max = MAX_THREADS;
2112 
2113 	t = *table;
2114 	t.data = &threads;
2115 	t.extra1 = &min;
2116 	t.extra2 = &max;
2117 
2118 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2119 	if (ret || !write)
2120 		return ret;
2121 
2122 	set_max_threads(threads);
2123 
2124 	return 0;
2125 }
2126