xref: /openbmc/linux/kernel/fork.c (revision 6e736be7f282fff705db7c34a15313281b372a76)
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13 
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/futex.h>
41 #include <linux/compat.h>
42 #include <linux/kthread.h>
43 #include <linux/task_io_accounting_ops.h>
44 #include <linux/rcupdate.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/audit.h>
48 #include <linux/memcontrol.h>
49 #include <linux/ftrace.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/ksm.h>
53 #include <linux/acct.h>
54 #include <linux/tsacct_kern.h>
55 #include <linux/cn_proc.h>
56 #include <linux/freezer.h>
57 #include <linux/delayacct.h>
58 #include <linux/taskstats_kern.h>
59 #include <linux/random.h>
60 #include <linux/tty.h>
61 #include <linux/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_event.h>
65 #include <linux/posix-timers.h>
66 #include <linux/user-return-notifier.h>
67 #include <linux/oom.h>
68 #include <linux/khugepaged.h>
69 
70 #include <asm/pgtable.h>
71 #include <asm/pgalloc.h>
72 #include <asm/uaccess.h>
73 #include <asm/mmu_context.h>
74 #include <asm/cacheflush.h>
75 #include <asm/tlbflush.h>
76 
77 #include <trace/events/sched.h>
78 
79 /*
80  * Protected counters by write_lock_irq(&tasklist_lock)
81  */
82 unsigned long total_forks;	/* Handle normal Linux uptimes. */
83 int nr_threads;			/* The idle threads do not count.. */
84 
85 int max_threads;		/* tunable limit on nr_threads */
86 
87 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
88 
89 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
90 
91 #ifdef CONFIG_PROVE_RCU
92 int lockdep_tasklist_lock_is_held(void)
93 {
94 	return lockdep_is_held(&tasklist_lock);
95 }
96 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
97 #endif /* #ifdef CONFIG_PROVE_RCU */
98 
99 int nr_processes(void)
100 {
101 	int cpu;
102 	int total = 0;
103 
104 	for_each_possible_cpu(cpu)
105 		total += per_cpu(process_counts, cpu);
106 
107 	return total;
108 }
109 
110 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
111 # define alloc_task_struct_node(node)		\
112 		kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
113 # define free_task_struct(tsk)			\
114 		kmem_cache_free(task_struct_cachep, (tsk))
115 static struct kmem_cache *task_struct_cachep;
116 #endif
117 
118 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
119 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
120 						  int node)
121 {
122 #ifdef CONFIG_DEBUG_STACK_USAGE
123 	gfp_t mask = GFP_KERNEL | __GFP_ZERO;
124 #else
125 	gfp_t mask = GFP_KERNEL;
126 #endif
127 	struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
128 
129 	return page ? page_address(page) : NULL;
130 }
131 
132 static inline void free_thread_info(struct thread_info *ti)
133 {
134 	free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
135 }
136 #endif
137 
138 /* SLAB cache for signal_struct structures (tsk->signal) */
139 static struct kmem_cache *signal_cachep;
140 
141 /* SLAB cache for sighand_struct structures (tsk->sighand) */
142 struct kmem_cache *sighand_cachep;
143 
144 /* SLAB cache for files_struct structures (tsk->files) */
145 struct kmem_cache *files_cachep;
146 
147 /* SLAB cache for fs_struct structures (tsk->fs) */
148 struct kmem_cache *fs_cachep;
149 
150 /* SLAB cache for vm_area_struct structures */
151 struct kmem_cache *vm_area_cachep;
152 
153 /* SLAB cache for mm_struct structures (tsk->mm) */
154 static struct kmem_cache *mm_cachep;
155 
156 static void account_kernel_stack(struct thread_info *ti, int account)
157 {
158 	struct zone *zone = page_zone(virt_to_page(ti));
159 
160 	mod_zone_page_state(zone, NR_KERNEL_STACK, account);
161 }
162 
163 void free_task(struct task_struct *tsk)
164 {
165 	account_kernel_stack(tsk->stack, -1);
166 	free_thread_info(tsk->stack);
167 	rt_mutex_debug_task_free(tsk);
168 	ftrace_graph_exit_task(tsk);
169 	free_task_struct(tsk);
170 }
171 EXPORT_SYMBOL(free_task);
172 
173 static inline void free_signal_struct(struct signal_struct *sig)
174 {
175 	taskstats_tgid_free(sig);
176 	sched_autogroup_exit(sig);
177 	kmem_cache_free(signal_cachep, sig);
178 }
179 
180 static inline void put_signal_struct(struct signal_struct *sig)
181 {
182 	if (atomic_dec_and_test(&sig->sigcnt))
183 		free_signal_struct(sig);
184 }
185 
186 void __put_task_struct(struct task_struct *tsk)
187 {
188 	WARN_ON(!tsk->exit_state);
189 	WARN_ON(atomic_read(&tsk->usage));
190 	WARN_ON(tsk == current);
191 
192 	exit_creds(tsk);
193 	delayacct_tsk_free(tsk);
194 	put_signal_struct(tsk->signal);
195 
196 	if (!profile_handoff_task(tsk))
197 		free_task(tsk);
198 }
199 EXPORT_SYMBOL_GPL(__put_task_struct);
200 
201 /*
202  * macro override instead of weak attribute alias, to workaround
203  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
204  */
205 #ifndef arch_task_cache_init
206 #define arch_task_cache_init()
207 #endif
208 
209 void __init fork_init(unsigned long mempages)
210 {
211 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
212 #ifndef ARCH_MIN_TASKALIGN
213 #define ARCH_MIN_TASKALIGN	L1_CACHE_BYTES
214 #endif
215 	/* create a slab on which task_structs can be allocated */
216 	task_struct_cachep =
217 		kmem_cache_create("task_struct", sizeof(struct task_struct),
218 			ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
219 #endif
220 
221 	/* do the arch specific task caches init */
222 	arch_task_cache_init();
223 
224 	/*
225 	 * The default maximum number of threads is set to a safe
226 	 * value: the thread structures can take up at most half
227 	 * of memory.
228 	 */
229 	max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
230 
231 	/*
232 	 * we need to allow at least 20 threads to boot a system
233 	 */
234 	if (max_threads < 20)
235 		max_threads = 20;
236 
237 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
238 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
239 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
240 		init_task.signal->rlim[RLIMIT_NPROC];
241 }
242 
243 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
244 					       struct task_struct *src)
245 {
246 	*dst = *src;
247 	return 0;
248 }
249 
250 static struct task_struct *dup_task_struct(struct task_struct *orig)
251 {
252 	struct task_struct *tsk;
253 	struct thread_info *ti;
254 	unsigned long *stackend;
255 	int node = tsk_fork_get_node(orig);
256 	int err;
257 
258 	prepare_to_copy(orig);
259 
260 	tsk = alloc_task_struct_node(node);
261 	if (!tsk)
262 		return NULL;
263 
264 	ti = alloc_thread_info_node(tsk, node);
265 	if (!ti) {
266 		free_task_struct(tsk);
267 		return NULL;
268 	}
269 
270 	err = arch_dup_task_struct(tsk, orig);
271 	if (err)
272 		goto out;
273 
274 	tsk->stack = ti;
275 
276 	setup_thread_stack(tsk, orig);
277 	clear_user_return_notifier(tsk);
278 	clear_tsk_need_resched(tsk);
279 	stackend = end_of_stack(tsk);
280 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
281 
282 #ifdef CONFIG_CC_STACKPROTECTOR
283 	tsk->stack_canary = get_random_int();
284 #endif
285 
286 	/*
287 	 * One for us, one for whoever does the "release_task()" (usually
288 	 * parent)
289 	 */
290 	atomic_set(&tsk->usage, 2);
291 #ifdef CONFIG_BLK_DEV_IO_TRACE
292 	tsk->btrace_seq = 0;
293 #endif
294 	tsk->splice_pipe = NULL;
295 
296 	account_kernel_stack(ti, 1);
297 
298 	return tsk;
299 
300 out:
301 	free_thread_info(ti);
302 	free_task_struct(tsk);
303 	return NULL;
304 }
305 
306 #ifdef CONFIG_MMU
307 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
308 {
309 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
310 	struct rb_node **rb_link, *rb_parent;
311 	int retval;
312 	unsigned long charge;
313 	struct mempolicy *pol;
314 
315 	down_write(&oldmm->mmap_sem);
316 	flush_cache_dup_mm(oldmm);
317 	/*
318 	 * Not linked in yet - no deadlock potential:
319 	 */
320 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
321 
322 	mm->locked_vm = 0;
323 	mm->mmap = NULL;
324 	mm->mmap_cache = NULL;
325 	mm->free_area_cache = oldmm->mmap_base;
326 	mm->cached_hole_size = ~0UL;
327 	mm->map_count = 0;
328 	cpumask_clear(mm_cpumask(mm));
329 	mm->mm_rb = RB_ROOT;
330 	rb_link = &mm->mm_rb.rb_node;
331 	rb_parent = NULL;
332 	pprev = &mm->mmap;
333 	retval = ksm_fork(mm, oldmm);
334 	if (retval)
335 		goto out;
336 	retval = khugepaged_fork(mm, oldmm);
337 	if (retval)
338 		goto out;
339 
340 	prev = NULL;
341 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
342 		struct file *file;
343 
344 		if (mpnt->vm_flags & VM_DONTCOPY) {
345 			long pages = vma_pages(mpnt);
346 			mm->total_vm -= pages;
347 			vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
348 								-pages);
349 			continue;
350 		}
351 		charge = 0;
352 		if (mpnt->vm_flags & VM_ACCOUNT) {
353 			unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
354 			if (security_vm_enough_memory(len))
355 				goto fail_nomem;
356 			charge = len;
357 		}
358 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
359 		if (!tmp)
360 			goto fail_nomem;
361 		*tmp = *mpnt;
362 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
363 		pol = mpol_dup(vma_policy(mpnt));
364 		retval = PTR_ERR(pol);
365 		if (IS_ERR(pol))
366 			goto fail_nomem_policy;
367 		vma_set_policy(tmp, pol);
368 		tmp->vm_mm = mm;
369 		if (anon_vma_fork(tmp, mpnt))
370 			goto fail_nomem_anon_vma_fork;
371 		tmp->vm_flags &= ~VM_LOCKED;
372 		tmp->vm_next = tmp->vm_prev = NULL;
373 		file = tmp->vm_file;
374 		if (file) {
375 			struct inode *inode = file->f_path.dentry->d_inode;
376 			struct address_space *mapping = file->f_mapping;
377 
378 			get_file(file);
379 			if (tmp->vm_flags & VM_DENYWRITE)
380 				atomic_dec(&inode->i_writecount);
381 			mutex_lock(&mapping->i_mmap_mutex);
382 			if (tmp->vm_flags & VM_SHARED)
383 				mapping->i_mmap_writable++;
384 			flush_dcache_mmap_lock(mapping);
385 			/* insert tmp into the share list, just after mpnt */
386 			vma_prio_tree_add(tmp, mpnt);
387 			flush_dcache_mmap_unlock(mapping);
388 			mutex_unlock(&mapping->i_mmap_mutex);
389 		}
390 
391 		/*
392 		 * Clear hugetlb-related page reserves for children. This only
393 		 * affects MAP_PRIVATE mappings. Faults generated by the child
394 		 * are not guaranteed to succeed, even if read-only
395 		 */
396 		if (is_vm_hugetlb_page(tmp))
397 			reset_vma_resv_huge_pages(tmp);
398 
399 		/*
400 		 * Link in the new vma and copy the page table entries.
401 		 */
402 		*pprev = tmp;
403 		pprev = &tmp->vm_next;
404 		tmp->vm_prev = prev;
405 		prev = tmp;
406 
407 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
408 		rb_link = &tmp->vm_rb.rb_right;
409 		rb_parent = &tmp->vm_rb;
410 
411 		mm->map_count++;
412 		retval = copy_page_range(mm, oldmm, mpnt);
413 
414 		if (tmp->vm_ops && tmp->vm_ops->open)
415 			tmp->vm_ops->open(tmp);
416 
417 		if (retval)
418 			goto out;
419 	}
420 	/* a new mm has just been created */
421 	arch_dup_mmap(oldmm, mm);
422 	retval = 0;
423 out:
424 	up_write(&mm->mmap_sem);
425 	flush_tlb_mm(oldmm);
426 	up_write(&oldmm->mmap_sem);
427 	return retval;
428 fail_nomem_anon_vma_fork:
429 	mpol_put(pol);
430 fail_nomem_policy:
431 	kmem_cache_free(vm_area_cachep, tmp);
432 fail_nomem:
433 	retval = -ENOMEM;
434 	vm_unacct_memory(charge);
435 	goto out;
436 }
437 
438 static inline int mm_alloc_pgd(struct mm_struct *mm)
439 {
440 	mm->pgd = pgd_alloc(mm);
441 	if (unlikely(!mm->pgd))
442 		return -ENOMEM;
443 	return 0;
444 }
445 
446 static inline void mm_free_pgd(struct mm_struct *mm)
447 {
448 	pgd_free(mm, mm->pgd);
449 }
450 #else
451 #define dup_mmap(mm, oldmm)	(0)
452 #define mm_alloc_pgd(mm)	(0)
453 #define mm_free_pgd(mm)
454 #endif /* CONFIG_MMU */
455 
456 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
457 
458 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
459 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
460 
461 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
462 
463 static int __init coredump_filter_setup(char *s)
464 {
465 	default_dump_filter =
466 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
467 		MMF_DUMP_FILTER_MASK;
468 	return 1;
469 }
470 
471 __setup("coredump_filter=", coredump_filter_setup);
472 
473 #include <linux/init_task.h>
474 
475 static void mm_init_aio(struct mm_struct *mm)
476 {
477 #ifdef CONFIG_AIO
478 	spin_lock_init(&mm->ioctx_lock);
479 	INIT_HLIST_HEAD(&mm->ioctx_list);
480 #endif
481 }
482 
483 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
484 {
485 	atomic_set(&mm->mm_users, 1);
486 	atomic_set(&mm->mm_count, 1);
487 	init_rwsem(&mm->mmap_sem);
488 	INIT_LIST_HEAD(&mm->mmlist);
489 	mm->flags = (current->mm) ?
490 		(current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
491 	mm->core_state = NULL;
492 	mm->nr_ptes = 0;
493 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
494 	spin_lock_init(&mm->page_table_lock);
495 	mm->free_area_cache = TASK_UNMAPPED_BASE;
496 	mm->cached_hole_size = ~0UL;
497 	mm_init_aio(mm);
498 	mm_init_owner(mm, p);
499 
500 	if (likely(!mm_alloc_pgd(mm))) {
501 		mm->def_flags = 0;
502 		mmu_notifier_mm_init(mm);
503 		return mm;
504 	}
505 
506 	free_mm(mm);
507 	return NULL;
508 }
509 
510 /*
511  * Allocate and initialize an mm_struct.
512  */
513 struct mm_struct *mm_alloc(void)
514 {
515 	struct mm_struct *mm;
516 
517 	mm = allocate_mm();
518 	if (!mm)
519 		return NULL;
520 
521 	memset(mm, 0, sizeof(*mm));
522 	mm_init_cpumask(mm);
523 	return mm_init(mm, current);
524 }
525 
526 /*
527  * Called when the last reference to the mm
528  * is dropped: either by a lazy thread or by
529  * mmput. Free the page directory and the mm.
530  */
531 void __mmdrop(struct mm_struct *mm)
532 {
533 	BUG_ON(mm == &init_mm);
534 	mm_free_pgd(mm);
535 	destroy_context(mm);
536 	mmu_notifier_mm_destroy(mm);
537 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
538 	VM_BUG_ON(mm->pmd_huge_pte);
539 #endif
540 	free_mm(mm);
541 }
542 EXPORT_SYMBOL_GPL(__mmdrop);
543 
544 /*
545  * Decrement the use count and release all resources for an mm.
546  */
547 void mmput(struct mm_struct *mm)
548 {
549 	might_sleep();
550 
551 	if (atomic_dec_and_test(&mm->mm_users)) {
552 		exit_aio(mm);
553 		ksm_exit(mm);
554 		khugepaged_exit(mm); /* must run before exit_mmap */
555 		exit_mmap(mm);
556 		set_mm_exe_file(mm, NULL);
557 		if (!list_empty(&mm->mmlist)) {
558 			spin_lock(&mmlist_lock);
559 			list_del(&mm->mmlist);
560 			spin_unlock(&mmlist_lock);
561 		}
562 		put_swap_token(mm);
563 		if (mm->binfmt)
564 			module_put(mm->binfmt->module);
565 		mmdrop(mm);
566 	}
567 }
568 EXPORT_SYMBOL_GPL(mmput);
569 
570 /*
571  * We added or removed a vma mapping the executable. The vmas are only mapped
572  * during exec and are not mapped with the mmap system call.
573  * Callers must hold down_write() on the mm's mmap_sem for these
574  */
575 void added_exe_file_vma(struct mm_struct *mm)
576 {
577 	mm->num_exe_file_vmas++;
578 }
579 
580 void removed_exe_file_vma(struct mm_struct *mm)
581 {
582 	mm->num_exe_file_vmas--;
583 	if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
584 		fput(mm->exe_file);
585 		mm->exe_file = NULL;
586 	}
587 
588 }
589 
590 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
591 {
592 	if (new_exe_file)
593 		get_file(new_exe_file);
594 	if (mm->exe_file)
595 		fput(mm->exe_file);
596 	mm->exe_file = new_exe_file;
597 	mm->num_exe_file_vmas = 0;
598 }
599 
600 struct file *get_mm_exe_file(struct mm_struct *mm)
601 {
602 	struct file *exe_file;
603 
604 	/* We need mmap_sem to protect against races with removal of
605 	 * VM_EXECUTABLE vmas */
606 	down_read(&mm->mmap_sem);
607 	exe_file = mm->exe_file;
608 	if (exe_file)
609 		get_file(exe_file);
610 	up_read(&mm->mmap_sem);
611 	return exe_file;
612 }
613 
614 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
615 {
616 	/* It's safe to write the exe_file pointer without exe_file_lock because
617 	 * this is called during fork when the task is not yet in /proc */
618 	newmm->exe_file = get_mm_exe_file(oldmm);
619 }
620 
621 /**
622  * get_task_mm - acquire a reference to the task's mm
623  *
624  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
625  * this kernel workthread has transiently adopted a user mm with use_mm,
626  * to do its AIO) is not set and if so returns a reference to it, after
627  * bumping up the use count.  User must release the mm via mmput()
628  * after use.  Typically used by /proc and ptrace.
629  */
630 struct mm_struct *get_task_mm(struct task_struct *task)
631 {
632 	struct mm_struct *mm;
633 
634 	task_lock(task);
635 	mm = task->mm;
636 	if (mm) {
637 		if (task->flags & PF_KTHREAD)
638 			mm = NULL;
639 		else
640 			atomic_inc(&mm->mm_users);
641 	}
642 	task_unlock(task);
643 	return mm;
644 }
645 EXPORT_SYMBOL_GPL(get_task_mm);
646 
647 /* Please note the differences between mmput and mm_release.
648  * mmput is called whenever we stop holding onto a mm_struct,
649  * error success whatever.
650  *
651  * mm_release is called after a mm_struct has been removed
652  * from the current process.
653  *
654  * This difference is important for error handling, when we
655  * only half set up a mm_struct for a new process and need to restore
656  * the old one.  Because we mmput the new mm_struct before
657  * restoring the old one. . .
658  * Eric Biederman 10 January 1998
659  */
660 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
661 {
662 	struct completion *vfork_done = tsk->vfork_done;
663 
664 	/* Get rid of any futexes when releasing the mm */
665 #ifdef CONFIG_FUTEX
666 	if (unlikely(tsk->robust_list)) {
667 		exit_robust_list(tsk);
668 		tsk->robust_list = NULL;
669 	}
670 #ifdef CONFIG_COMPAT
671 	if (unlikely(tsk->compat_robust_list)) {
672 		compat_exit_robust_list(tsk);
673 		tsk->compat_robust_list = NULL;
674 	}
675 #endif
676 	if (unlikely(!list_empty(&tsk->pi_state_list)))
677 		exit_pi_state_list(tsk);
678 #endif
679 
680 	/* Get rid of any cached register state */
681 	deactivate_mm(tsk, mm);
682 
683 	/* notify parent sleeping on vfork() */
684 	if (vfork_done) {
685 		tsk->vfork_done = NULL;
686 		complete(vfork_done);
687 	}
688 
689 	/*
690 	 * If we're exiting normally, clear a user-space tid field if
691 	 * requested.  We leave this alone when dying by signal, to leave
692 	 * the value intact in a core dump, and to save the unnecessary
693 	 * trouble otherwise.  Userland only wants this done for a sys_exit.
694 	 */
695 	if (tsk->clear_child_tid) {
696 		if (!(tsk->flags & PF_SIGNALED) &&
697 		    atomic_read(&mm->mm_users) > 1) {
698 			/*
699 			 * We don't check the error code - if userspace has
700 			 * not set up a proper pointer then tough luck.
701 			 */
702 			put_user(0, tsk->clear_child_tid);
703 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
704 					1, NULL, NULL, 0);
705 		}
706 		tsk->clear_child_tid = NULL;
707 	}
708 }
709 
710 /*
711  * Allocate a new mm structure and copy contents from the
712  * mm structure of the passed in task structure.
713  */
714 struct mm_struct *dup_mm(struct task_struct *tsk)
715 {
716 	struct mm_struct *mm, *oldmm = current->mm;
717 	int err;
718 
719 	if (!oldmm)
720 		return NULL;
721 
722 	mm = allocate_mm();
723 	if (!mm)
724 		goto fail_nomem;
725 
726 	memcpy(mm, oldmm, sizeof(*mm));
727 	mm_init_cpumask(mm);
728 
729 	/* Initializing for Swap token stuff */
730 	mm->token_priority = 0;
731 	mm->last_interval = 0;
732 
733 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
734 	mm->pmd_huge_pte = NULL;
735 #endif
736 
737 	if (!mm_init(mm, tsk))
738 		goto fail_nomem;
739 
740 	if (init_new_context(tsk, mm))
741 		goto fail_nocontext;
742 
743 	dup_mm_exe_file(oldmm, mm);
744 
745 	err = dup_mmap(mm, oldmm);
746 	if (err)
747 		goto free_pt;
748 
749 	mm->hiwater_rss = get_mm_rss(mm);
750 	mm->hiwater_vm = mm->total_vm;
751 
752 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
753 		goto free_pt;
754 
755 	return mm;
756 
757 free_pt:
758 	/* don't put binfmt in mmput, we haven't got module yet */
759 	mm->binfmt = NULL;
760 	mmput(mm);
761 
762 fail_nomem:
763 	return NULL;
764 
765 fail_nocontext:
766 	/*
767 	 * If init_new_context() failed, we cannot use mmput() to free the mm
768 	 * because it calls destroy_context()
769 	 */
770 	mm_free_pgd(mm);
771 	free_mm(mm);
772 	return NULL;
773 }
774 
775 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
776 {
777 	struct mm_struct *mm, *oldmm;
778 	int retval;
779 
780 	tsk->min_flt = tsk->maj_flt = 0;
781 	tsk->nvcsw = tsk->nivcsw = 0;
782 #ifdef CONFIG_DETECT_HUNG_TASK
783 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
784 #endif
785 
786 	tsk->mm = NULL;
787 	tsk->active_mm = NULL;
788 
789 	/*
790 	 * Are we cloning a kernel thread?
791 	 *
792 	 * We need to steal a active VM for that..
793 	 */
794 	oldmm = current->mm;
795 	if (!oldmm)
796 		return 0;
797 
798 	if (clone_flags & CLONE_VM) {
799 		atomic_inc(&oldmm->mm_users);
800 		mm = oldmm;
801 		goto good_mm;
802 	}
803 
804 	retval = -ENOMEM;
805 	mm = dup_mm(tsk);
806 	if (!mm)
807 		goto fail_nomem;
808 
809 good_mm:
810 	/* Initializing for Swap token stuff */
811 	mm->token_priority = 0;
812 	mm->last_interval = 0;
813 
814 	tsk->mm = mm;
815 	tsk->active_mm = mm;
816 	return 0;
817 
818 fail_nomem:
819 	return retval;
820 }
821 
822 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
823 {
824 	struct fs_struct *fs = current->fs;
825 	if (clone_flags & CLONE_FS) {
826 		/* tsk->fs is already what we want */
827 		spin_lock(&fs->lock);
828 		if (fs->in_exec) {
829 			spin_unlock(&fs->lock);
830 			return -EAGAIN;
831 		}
832 		fs->users++;
833 		spin_unlock(&fs->lock);
834 		return 0;
835 	}
836 	tsk->fs = copy_fs_struct(fs);
837 	if (!tsk->fs)
838 		return -ENOMEM;
839 	return 0;
840 }
841 
842 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
843 {
844 	struct files_struct *oldf, *newf;
845 	int error = 0;
846 
847 	/*
848 	 * A background process may not have any files ...
849 	 */
850 	oldf = current->files;
851 	if (!oldf)
852 		goto out;
853 
854 	if (clone_flags & CLONE_FILES) {
855 		atomic_inc(&oldf->count);
856 		goto out;
857 	}
858 
859 	newf = dup_fd(oldf, &error);
860 	if (!newf)
861 		goto out;
862 
863 	tsk->files = newf;
864 	error = 0;
865 out:
866 	return error;
867 }
868 
869 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
870 {
871 #ifdef CONFIG_BLOCK
872 	struct io_context *ioc = current->io_context;
873 	struct io_context *new_ioc;
874 
875 	if (!ioc)
876 		return 0;
877 	/*
878 	 * Share io context with parent, if CLONE_IO is set
879 	 */
880 	if (clone_flags & CLONE_IO) {
881 		tsk->io_context = ioc_task_link(ioc);
882 		if (unlikely(!tsk->io_context))
883 			return -ENOMEM;
884 	} else if (ioprio_valid(ioc->ioprio)) {
885 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
886 		if (unlikely(!new_ioc))
887 			return -ENOMEM;
888 
889 		new_ioc->ioprio = ioc->ioprio;
890 		put_io_context(new_ioc);
891 	}
892 #endif
893 	return 0;
894 }
895 
896 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
897 {
898 	struct sighand_struct *sig;
899 
900 	if (clone_flags & CLONE_SIGHAND) {
901 		atomic_inc(&current->sighand->count);
902 		return 0;
903 	}
904 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
905 	rcu_assign_pointer(tsk->sighand, sig);
906 	if (!sig)
907 		return -ENOMEM;
908 	atomic_set(&sig->count, 1);
909 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
910 	return 0;
911 }
912 
913 void __cleanup_sighand(struct sighand_struct *sighand)
914 {
915 	if (atomic_dec_and_test(&sighand->count))
916 		kmem_cache_free(sighand_cachep, sighand);
917 }
918 
919 
920 /*
921  * Initialize POSIX timer handling for a thread group.
922  */
923 static void posix_cpu_timers_init_group(struct signal_struct *sig)
924 {
925 	unsigned long cpu_limit;
926 
927 	/* Thread group counters. */
928 	thread_group_cputime_init(sig);
929 
930 	cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
931 	if (cpu_limit != RLIM_INFINITY) {
932 		sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
933 		sig->cputimer.running = 1;
934 	}
935 
936 	/* The timer lists. */
937 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
938 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
939 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
940 }
941 
942 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
943 {
944 	struct signal_struct *sig;
945 
946 	if (clone_flags & CLONE_THREAD)
947 		return 0;
948 
949 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
950 	tsk->signal = sig;
951 	if (!sig)
952 		return -ENOMEM;
953 
954 	sig->nr_threads = 1;
955 	atomic_set(&sig->live, 1);
956 	atomic_set(&sig->sigcnt, 1);
957 	init_waitqueue_head(&sig->wait_chldexit);
958 	if (clone_flags & CLONE_NEWPID)
959 		sig->flags |= SIGNAL_UNKILLABLE;
960 	sig->curr_target = tsk;
961 	init_sigpending(&sig->shared_pending);
962 	INIT_LIST_HEAD(&sig->posix_timers);
963 
964 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
965 	sig->real_timer.function = it_real_fn;
966 
967 	task_lock(current->group_leader);
968 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
969 	task_unlock(current->group_leader);
970 
971 	posix_cpu_timers_init_group(sig);
972 
973 	tty_audit_fork(sig);
974 	sched_autogroup_fork(sig);
975 
976 #ifdef CONFIG_CGROUPS
977 	init_rwsem(&sig->threadgroup_fork_lock);
978 #endif
979 
980 	sig->oom_adj = current->signal->oom_adj;
981 	sig->oom_score_adj = current->signal->oom_score_adj;
982 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
983 
984 	mutex_init(&sig->cred_guard_mutex);
985 
986 	return 0;
987 }
988 
989 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
990 {
991 	unsigned long new_flags = p->flags;
992 
993 	new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
994 	new_flags |= PF_FORKNOEXEC;
995 	new_flags |= PF_STARTING;
996 	p->flags = new_flags;
997 	clear_freeze_flag(p);
998 }
999 
1000 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1001 {
1002 	current->clear_child_tid = tidptr;
1003 
1004 	return task_pid_vnr(current);
1005 }
1006 
1007 static void rt_mutex_init_task(struct task_struct *p)
1008 {
1009 	raw_spin_lock_init(&p->pi_lock);
1010 #ifdef CONFIG_RT_MUTEXES
1011 	plist_head_init(&p->pi_waiters);
1012 	p->pi_blocked_on = NULL;
1013 #endif
1014 }
1015 
1016 #ifdef CONFIG_MM_OWNER
1017 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1018 {
1019 	mm->owner = p;
1020 }
1021 #endif /* CONFIG_MM_OWNER */
1022 
1023 /*
1024  * Initialize POSIX timer handling for a single task.
1025  */
1026 static void posix_cpu_timers_init(struct task_struct *tsk)
1027 {
1028 	tsk->cputime_expires.prof_exp = cputime_zero;
1029 	tsk->cputime_expires.virt_exp = cputime_zero;
1030 	tsk->cputime_expires.sched_exp = 0;
1031 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1032 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1033 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1034 }
1035 
1036 /*
1037  * This creates a new process as a copy of the old one,
1038  * but does not actually start it yet.
1039  *
1040  * It copies the registers, and all the appropriate
1041  * parts of the process environment (as per the clone
1042  * flags). The actual kick-off is left to the caller.
1043  */
1044 static struct task_struct *copy_process(unsigned long clone_flags,
1045 					unsigned long stack_start,
1046 					struct pt_regs *regs,
1047 					unsigned long stack_size,
1048 					int __user *child_tidptr,
1049 					struct pid *pid,
1050 					int trace)
1051 {
1052 	int retval;
1053 	struct task_struct *p;
1054 	int cgroup_callbacks_done = 0;
1055 
1056 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1057 		return ERR_PTR(-EINVAL);
1058 
1059 	/*
1060 	 * Thread groups must share signals as well, and detached threads
1061 	 * can only be started up within the thread group.
1062 	 */
1063 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1064 		return ERR_PTR(-EINVAL);
1065 
1066 	/*
1067 	 * Shared signal handlers imply shared VM. By way of the above,
1068 	 * thread groups also imply shared VM. Blocking this case allows
1069 	 * for various simplifications in other code.
1070 	 */
1071 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1072 		return ERR_PTR(-EINVAL);
1073 
1074 	/*
1075 	 * Siblings of global init remain as zombies on exit since they are
1076 	 * not reaped by their parent (swapper). To solve this and to avoid
1077 	 * multi-rooted process trees, prevent global and container-inits
1078 	 * from creating siblings.
1079 	 */
1080 	if ((clone_flags & CLONE_PARENT) &&
1081 				current->signal->flags & SIGNAL_UNKILLABLE)
1082 		return ERR_PTR(-EINVAL);
1083 
1084 	retval = security_task_create(clone_flags);
1085 	if (retval)
1086 		goto fork_out;
1087 
1088 	retval = -ENOMEM;
1089 	p = dup_task_struct(current);
1090 	if (!p)
1091 		goto fork_out;
1092 
1093 	ftrace_graph_init_task(p);
1094 
1095 	rt_mutex_init_task(p);
1096 
1097 #ifdef CONFIG_PROVE_LOCKING
1098 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1099 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1100 #endif
1101 	retval = -EAGAIN;
1102 	if (atomic_read(&p->real_cred->user->processes) >=
1103 			task_rlimit(p, RLIMIT_NPROC)) {
1104 		if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1105 		    p->real_cred->user != INIT_USER)
1106 			goto bad_fork_free;
1107 	}
1108 	current->flags &= ~PF_NPROC_EXCEEDED;
1109 
1110 	retval = copy_creds(p, clone_flags);
1111 	if (retval < 0)
1112 		goto bad_fork_free;
1113 
1114 	/*
1115 	 * If multiple threads are within copy_process(), then this check
1116 	 * triggers too late. This doesn't hurt, the check is only there
1117 	 * to stop root fork bombs.
1118 	 */
1119 	retval = -EAGAIN;
1120 	if (nr_threads >= max_threads)
1121 		goto bad_fork_cleanup_count;
1122 
1123 	if (!try_module_get(task_thread_info(p)->exec_domain->module))
1124 		goto bad_fork_cleanup_count;
1125 
1126 	p->did_exec = 0;
1127 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1128 	copy_flags(clone_flags, p);
1129 	INIT_LIST_HEAD(&p->children);
1130 	INIT_LIST_HEAD(&p->sibling);
1131 	rcu_copy_process(p);
1132 	p->vfork_done = NULL;
1133 	spin_lock_init(&p->alloc_lock);
1134 
1135 	init_sigpending(&p->pending);
1136 
1137 	p->utime = cputime_zero;
1138 	p->stime = cputime_zero;
1139 	p->gtime = cputime_zero;
1140 	p->utimescaled = cputime_zero;
1141 	p->stimescaled = cputime_zero;
1142 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1143 	p->prev_utime = cputime_zero;
1144 	p->prev_stime = cputime_zero;
1145 #endif
1146 #if defined(SPLIT_RSS_COUNTING)
1147 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1148 #endif
1149 
1150 	p->default_timer_slack_ns = current->timer_slack_ns;
1151 
1152 	task_io_accounting_init(&p->ioac);
1153 	acct_clear_integrals(p);
1154 
1155 	posix_cpu_timers_init(p);
1156 
1157 	do_posix_clock_monotonic_gettime(&p->start_time);
1158 	p->real_start_time = p->start_time;
1159 	monotonic_to_bootbased(&p->real_start_time);
1160 	p->io_context = NULL;
1161 	p->audit_context = NULL;
1162 	if (clone_flags & CLONE_THREAD)
1163 		threadgroup_fork_read_lock(current);
1164 	cgroup_fork(p);
1165 #ifdef CONFIG_NUMA
1166 	p->mempolicy = mpol_dup(p->mempolicy);
1167 	if (IS_ERR(p->mempolicy)) {
1168 		retval = PTR_ERR(p->mempolicy);
1169 		p->mempolicy = NULL;
1170 		goto bad_fork_cleanup_cgroup;
1171 	}
1172 	mpol_fix_fork_child_flag(p);
1173 #endif
1174 #ifdef CONFIG_CPUSETS
1175 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1176 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1177 #endif
1178 #ifdef CONFIG_TRACE_IRQFLAGS
1179 	p->irq_events = 0;
1180 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1181 	p->hardirqs_enabled = 1;
1182 #else
1183 	p->hardirqs_enabled = 0;
1184 #endif
1185 	p->hardirq_enable_ip = 0;
1186 	p->hardirq_enable_event = 0;
1187 	p->hardirq_disable_ip = _THIS_IP_;
1188 	p->hardirq_disable_event = 0;
1189 	p->softirqs_enabled = 1;
1190 	p->softirq_enable_ip = _THIS_IP_;
1191 	p->softirq_enable_event = 0;
1192 	p->softirq_disable_ip = 0;
1193 	p->softirq_disable_event = 0;
1194 	p->hardirq_context = 0;
1195 	p->softirq_context = 0;
1196 #endif
1197 #ifdef CONFIG_LOCKDEP
1198 	p->lockdep_depth = 0; /* no locks held yet */
1199 	p->curr_chain_key = 0;
1200 	p->lockdep_recursion = 0;
1201 #endif
1202 
1203 #ifdef CONFIG_DEBUG_MUTEXES
1204 	p->blocked_on = NULL; /* not blocked yet */
1205 #endif
1206 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1207 	p->memcg_batch.do_batch = 0;
1208 	p->memcg_batch.memcg = NULL;
1209 #endif
1210 
1211 	/* Perform scheduler related setup. Assign this task to a CPU. */
1212 	sched_fork(p);
1213 
1214 	retval = perf_event_init_task(p);
1215 	if (retval)
1216 		goto bad_fork_cleanup_policy;
1217 	retval = audit_alloc(p);
1218 	if (retval)
1219 		goto bad_fork_cleanup_policy;
1220 	/* copy all the process information */
1221 	retval = copy_semundo(clone_flags, p);
1222 	if (retval)
1223 		goto bad_fork_cleanup_audit;
1224 	retval = copy_files(clone_flags, p);
1225 	if (retval)
1226 		goto bad_fork_cleanup_semundo;
1227 	retval = copy_fs(clone_flags, p);
1228 	if (retval)
1229 		goto bad_fork_cleanup_files;
1230 	retval = copy_sighand(clone_flags, p);
1231 	if (retval)
1232 		goto bad_fork_cleanup_fs;
1233 	retval = copy_signal(clone_flags, p);
1234 	if (retval)
1235 		goto bad_fork_cleanup_sighand;
1236 	retval = copy_mm(clone_flags, p);
1237 	if (retval)
1238 		goto bad_fork_cleanup_signal;
1239 	retval = copy_namespaces(clone_flags, p);
1240 	if (retval)
1241 		goto bad_fork_cleanup_mm;
1242 	retval = copy_io(clone_flags, p);
1243 	if (retval)
1244 		goto bad_fork_cleanup_namespaces;
1245 	retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1246 	if (retval)
1247 		goto bad_fork_cleanup_io;
1248 
1249 	if (pid != &init_struct_pid) {
1250 		retval = -ENOMEM;
1251 		pid = alloc_pid(p->nsproxy->pid_ns);
1252 		if (!pid)
1253 			goto bad_fork_cleanup_io;
1254 	}
1255 
1256 	p->pid = pid_nr(pid);
1257 	p->tgid = p->pid;
1258 	if (clone_flags & CLONE_THREAD)
1259 		p->tgid = current->tgid;
1260 
1261 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1262 	/*
1263 	 * Clear TID on mm_release()?
1264 	 */
1265 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1266 #ifdef CONFIG_BLOCK
1267 	p->plug = NULL;
1268 #endif
1269 #ifdef CONFIG_FUTEX
1270 	p->robust_list = NULL;
1271 #ifdef CONFIG_COMPAT
1272 	p->compat_robust_list = NULL;
1273 #endif
1274 	INIT_LIST_HEAD(&p->pi_state_list);
1275 	p->pi_state_cache = NULL;
1276 #endif
1277 	/*
1278 	 * sigaltstack should be cleared when sharing the same VM
1279 	 */
1280 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1281 		p->sas_ss_sp = p->sas_ss_size = 0;
1282 
1283 	/*
1284 	 * Syscall tracing and stepping should be turned off in the
1285 	 * child regardless of CLONE_PTRACE.
1286 	 */
1287 	user_disable_single_step(p);
1288 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1289 #ifdef TIF_SYSCALL_EMU
1290 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1291 #endif
1292 	clear_all_latency_tracing(p);
1293 
1294 	/* ok, now we should be set up.. */
1295 	p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1296 	p->pdeath_signal = 0;
1297 	p->exit_state = 0;
1298 
1299 	p->nr_dirtied = 0;
1300 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1301 
1302 	/*
1303 	 * Ok, make it visible to the rest of the system.
1304 	 * We dont wake it up yet.
1305 	 */
1306 	p->group_leader = p;
1307 	INIT_LIST_HEAD(&p->thread_group);
1308 
1309 	/* Now that the task is set up, run cgroup callbacks if
1310 	 * necessary. We need to run them before the task is visible
1311 	 * on the tasklist. */
1312 	cgroup_fork_callbacks(p);
1313 	cgroup_callbacks_done = 1;
1314 
1315 	/* Need tasklist lock for parent etc handling! */
1316 	write_lock_irq(&tasklist_lock);
1317 
1318 	/* CLONE_PARENT re-uses the old parent */
1319 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1320 		p->real_parent = current->real_parent;
1321 		p->parent_exec_id = current->parent_exec_id;
1322 	} else {
1323 		p->real_parent = current;
1324 		p->parent_exec_id = current->self_exec_id;
1325 	}
1326 
1327 	spin_lock(&current->sighand->siglock);
1328 
1329 	/*
1330 	 * Process group and session signals need to be delivered to just the
1331 	 * parent before the fork or both the parent and the child after the
1332 	 * fork. Restart if a signal comes in before we add the new process to
1333 	 * it's process group.
1334 	 * A fatal signal pending means that current will exit, so the new
1335 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1336 	*/
1337 	recalc_sigpending();
1338 	if (signal_pending(current)) {
1339 		spin_unlock(&current->sighand->siglock);
1340 		write_unlock_irq(&tasklist_lock);
1341 		retval = -ERESTARTNOINTR;
1342 		goto bad_fork_free_pid;
1343 	}
1344 
1345 	if (clone_flags & CLONE_THREAD) {
1346 		current->signal->nr_threads++;
1347 		atomic_inc(&current->signal->live);
1348 		atomic_inc(&current->signal->sigcnt);
1349 		p->group_leader = current->group_leader;
1350 		list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1351 	}
1352 
1353 	if (likely(p->pid)) {
1354 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1355 
1356 		if (thread_group_leader(p)) {
1357 			if (is_child_reaper(pid))
1358 				p->nsproxy->pid_ns->child_reaper = p;
1359 
1360 			p->signal->leader_pid = pid;
1361 			p->signal->tty = tty_kref_get(current->signal->tty);
1362 			attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1363 			attach_pid(p, PIDTYPE_SID, task_session(current));
1364 			list_add_tail(&p->sibling, &p->real_parent->children);
1365 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1366 			__this_cpu_inc(process_counts);
1367 		}
1368 		attach_pid(p, PIDTYPE_PID, pid);
1369 		nr_threads++;
1370 	}
1371 
1372 	total_forks++;
1373 	spin_unlock(&current->sighand->siglock);
1374 	write_unlock_irq(&tasklist_lock);
1375 	proc_fork_connector(p);
1376 	cgroup_post_fork(p);
1377 	if (clone_flags & CLONE_THREAD)
1378 		threadgroup_fork_read_unlock(current);
1379 	perf_event_fork(p);
1380 	return p;
1381 
1382 bad_fork_free_pid:
1383 	if (pid != &init_struct_pid)
1384 		free_pid(pid);
1385 bad_fork_cleanup_io:
1386 	if (p->io_context)
1387 		exit_io_context(p);
1388 bad_fork_cleanup_namespaces:
1389 	exit_task_namespaces(p);
1390 bad_fork_cleanup_mm:
1391 	if (p->mm)
1392 		mmput(p->mm);
1393 bad_fork_cleanup_signal:
1394 	if (!(clone_flags & CLONE_THREAD))
1395 		free_signal_struct(p->signal);
1396 bad_fork_cleanup_sighand:
1397 	__cleanup_sighand(p->sighand);
1398 bad_fork_cleanup_fs:
1399 	exit_fs(p); /* blocking */
1400 bad_fork_cleanup_files:
1401 	exit_files(p); /* blocking */
1402 bad_fork_cleanup_semundo:
1403 	exit_sem(p);
1404 bad_fork_cleanup_audit:
1405 	audit_free(p);
1406 bad_fork_cleanup_policy:
1407 	perf_event_free_task(p);
1408 #ifdef CONFIG_NUMA
1409 	mpol_put(p->mempolicy);
1410 bad_fork_cleanup_cgroup:
1411 #endif
1412 	if (clone_flags & CLONE_THREAD)
1413 		threadgroup_fork_read_unlock(current);
1414 	cgroup_exit(p, cgroup_callbacks_done);
1415 	delayacct_tsk_free(p);
1416 	module_put(task_thread_info(p)->exec_domain->module);
1417 bad_fork_cleanup_count:
1418 	atomic_dec(&p->cred->user->processes);
1419 	exit_creds(p);
1420 bad_fork_free:
1421 	free_task(p);
1422 fork_out:
1423 	return ERR_PTR(retval);
1424 }
1425 
1426 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1427 {
1428 	memset(regs, 0, sizeof(struct pt_regs));
1429 	return regs;
1430 }
1431 
1432 static inline void init_idle_pids(struct pid_link *links)
1433 {
1434 	enum pid_type type;
1435 
1436 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1437 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1438 		links[type].pid = &init_struct_pid;
1439 	}
1440 }
1441 
1442 struct task_struct * __cpuinit fork_idle(int cpu)
1443 {
1444 	struct task_struct *task;
1445 	struct pt_regs regs;
1446 
1447 	task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1448 			    &init_struct_pid, 0);
1449 	if (!IS_ERR(task)) {
1450 		init_idle_pids(task->pids);
1451 		init_idle(task, cpu);
1452 	}
1453 
1454 	return task;
1455 }
1456 
1457 /*
1458  *  Ok, this is the main fork-routine.
1459  *
1460  * It copies the process, and if successful kick-starts
1461  * it and waits for it to finish using the VM if required.
1462  */
1463 long do_fork(unsigned long clone_flags,
1464 	      unsigned long stack_start,
1465 	      struct pt_regs *regs,
1466 	      unsigned long stack_size,
1467 	      int __user *parent_tidptr,
1468 	      int __user *child_tidptr)
1469 {
1470 	struct task_struct *p;
1471 	int trace = 0;
1472 	long nr;
1473 
1474 	/*
1475 	 * Do some preliminary argument and permissions checking before we
1476 	 * actually start allocating stuff
1477 	 */
1478 	if (clone_flags & CLONE_NEWUSER) {
1479 		if (clone_flags & CLONE_THREAD)
1480 			return -EINVAL;
1481 		/* hopefully this check will go away when userns support is
1482 		 * complete
1483 		 */
1484 		if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1485 				!capable(CAP_SETGID))
1486 			return -EPERM;
1487 	}
1488 
1489 	/*
1490 	 * Determine whether and which event to report to ptracer.  When
1491 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
1492 	 * requested, no event is reported; otherwise, report if the event
1493 	 * for the type of forking is enabled.
1494 	 */
1495 	if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1496 		if (clone_flags & CLONE_VFORK)
1497 			trace = PTRACE_EVENT_VFORK;
1498 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
1499 			trace = PTRACE_EVENT_CLONE;
1500 		else
1501 			trace = PTRACE_EVENT_FORK;
1502 
1503 		if (likely(!ptrace_event_enabled(current, trace)))
1504 			trace = 0;
1505 	}
1506 
1507 	p = copy_process(clone_flags, stack_start, regs, stack_size,
1508 			 child_tidptr, NULL, trace);
1509 	/*
1510 	 * Do this prior waking up the new thread - the thread pointer
1511 	 * might get invalid after that point, if the thread exits quickly.
1512 	 */
1513 	if (!IS_ERR(p)) {
1514 		struct completion vfork;
1515 
1516 		trace_sched_process_fork(current, p);
1517 
1518 		nr = task_pid_vnr(p);
1519 
1520 		if (clone_flags & CLONE_PARENT_SETTID)
1521 			put_user(nr, parent_tidptr);
1522 
1523 		if (clone_flags & CLONE_VFORK) {
1524 			p->vfork_done = &vfork;
1525 			init_completion(&vfork);
1526 		}
1527 
1528 		audit_finish_fork(p);
1529 
1530 		/*
1531 		 * We set PF_STARTING at creation in case tracing wants to
1532 		 * use this to distinguish a fully live task from one that
1533 		 * hasn't finished SIGSTOP raising yet.  Now we clear it
1534 		 * and set the child going.
1535 		 */
1536 		p->flags &= ~PF_STARTING;
1537 
1538 		wake_up_new_task(p);
1539 
1540 		/* forking complete and child started to run, tell ptracer */
1541 		if (unlikely(trace))
1542 			ptrace_event(trace, nr);
1543 
1544 		if (clone_flags & CLONE_VFORK) {
1545 			freezer_do_not_count();
1546 			wait_for_completion(&vfork);
1547 			freezer_count();
1548 			ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1549 		}
1550 	} else {
1551 		nr = PTR_ERR(p);
1552 	}
1553 	return nr;
1554 }
1555 
1556 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1557 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1558 #endif
1559 
1560 static void sighand_ctor(void *data)
1561 {
1562 	struct sighand_struct *sighand = data;
1563 
1564 	spin_lock_init(&sighand->siglock);
1565 	init_waitqueue_head(&sighand->signalfd_wqh);
1566 }
1567 
1568 void __init proc_caches_init(void)
1569 {
1570 	sighand_cachep = kmem_cache_create("sighand_cache",
1571 			sizeof(struct sighand_struct), 0,
1572 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1573 			SLAB_NOTRACK, sighand_ctor);
1574 	signal_cachep = kmem_cache_create("signal_cache",
1575 			sizeof(struct signal_struct), 0,
1576 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1577 	files_cachep = kmem_cache_create("files_cache",
1578 			sizeof(struct files_struct), 0,
1579 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1580 	fs_cachep = kmem_cache_create("fs_cache",
1581 			sizeof(struct fs_struct), 0,
1582 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1583 	/*
1584 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1585 	 * whole struct cpumask for the OFFSTACK case. We could change
1586 	 * this to *only* allocate as much of it as required by the
1587 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
1588 	 * is at the end of the structure, exactly for that reason.
1589 	 */
1590 	mm_cachep = kmem_cache_create("mm_struct",
1591 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1592 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1593 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1594 	mmap_init();
1595 	nsproxy_cache_init();
1596 }
1597 
1598 /*
1599  * Check constraints on flags passed to the unshare system call.
1600  */
1601 static int check_unshare_flags(unsigned long unshare_flags)
1602 {
1603 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1604 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1605 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1606 		return -EINVAL;
1607 	/*
1608 	 * Not implemented, but pretend it works if there is nothing to
1609 	 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1610 	 * needs to unshare vm.
1611 	 */
1612 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1613 		/* FIXME: get_task_mm() increments ->mm_users */
1614 		if (atomic_read(&current->mm->mm_users) > 1)
1615 			return -EINVAL;
1616 	}
1617 
1618 	return 0;
1619 }
1620 
1621 /*
1622  * Unshare the filesystem structure if it is being shared
1623  */
1624 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1625 {
1626 	struct fs_struct *fs = current->fs;
1627 
1628 	if (!(unshare_flags & CLONE_FS) || !fs)
1629 		return 0;
1630 
1631 	/* don't need lock here; in the worst case we'll do useless copy */
1632 	if (fs->users == 1)
1633 		return 0;
1634 
1635 	*new_fsp = copy_fs_struct(fs);
1636 	if (!*new_fsp)
1637 		return -ENOMEM;
1638 
1639 	return 0;
1640 }
1641 
1642 /*
1643  * Unshare file descriptor table if it is being shared
1644  */
1645 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1646 {
1647 	struct files_struct *fd = current->files;
1648 	int error = 0;
1649 
1650 	if ((unshare_flags & CLONE_FILES) &&
1651 	    (fd && atomic_read(&fd->count) > 1)) {
1652 		*new_fdp = dup_fd(fd, &error);
1653 		if (!*new_fdp)
1654 			return error;
1655 	}
1656 
1657 	return 0;
1658 }
1659 
1660 /*
1661  * unshare allows a process to 'unshare' part of the process
1662  * context which was originally shared using clone.  copy_*
1663  * functions used by do_fork() cannot be used here directly
1664  * because they modify an inactive task_struct that is being
1665  * constructed. Here we are modifying the current, active,
1666  * task_struct.
1667  */
1668 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1669 {
1670 	struct fs_struct *fs, *new_fs = NULL;
1671 	struct files_struct *fd, *new_fd = NULL;
1672 	struct nsproxy *new_nsproxy = NULL;
1673 	int do_sysvsem = 0;
1674 	int err;
1675 
1676 	err = check_unshare_flags(unshare_flags);
1677 	if (err)
1678 		goto bad_unshare_out;
1679 
1680 	/*
1681 	 * If unsharing namespace, must also unshare filesystem information.
1682 	 */
1683 	if (unshare_flags & CLONE_NEWNS)
1684 		unshare_flags |= CLONE_FS;
1685 	/*
1686 	 * CLONE_NEWIPC must also detach from the undolist: after switching
1687 	 * to a new ipc namespace, the semaphore arrays from the old
1688 	 * namespace are unreachable.
1689 	 */
1690 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1691 		do_sysvsem = 1;
1692 	err = unshare_fs(unshare_flags, &new_fs);
1693 	if (err)
1694 		goto bad_unshare_out;
1695 	err = unshare_fd(unshare_flags, &new_fd);
1696 	if (err)
1697 		goto bad_unshare_cleanup_fs;
1698 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1699 	if (err)
1700 		goto bad_unshare_cleanup_fd;
1701 
1702 	if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1703 		if (do_sysvsem) {
1704 			/*
1705 			 * CLONE_SYSVSEM is equivalent to sys_exit().
1706 			 */
1707 			exit_sem(current);
1708 		}
1709 
1710 		if (new_nsproxy) {
1711 			switch_task_namespaces(current, new_nsproxy);
1712 			new_nsproxy = NULL;
1713 		}
1714 
1715 		task_lock(current);
1716 
1717 		if (new_fs) {
1718 			fs = current->fs;
1719 			spin_lock(&fs->lock);
1720 			current->fs = new_fs;
1721 			if (--fs->users)
1722 				new_fs = NULL;
1723 			else
1724 				new_fs = fs;
1725 			spin_unlock(&fs->lock);
1726 		}
1727 
1728 		if (new_fd) {
1729 			fd = current->files;
1730 			current->files = new_fd;
1731 			new_fd = fd;
1732 		}
1733 
1734 		task_unlock(current);
1735 	}
1736 
1737 	if (new_nsproxy)
1738 		put_nsproxy(new_nsproxy);
1739 
1740 bad_unshare_cleanup_fd:
1741 	if (new_fd)
1742 		put_files_struct(new_fd);
1743 
1744 bad_unshare_cleanup_fs:
1745 	if (new_fs)
1746 		free_fs_struct(new_fs);
1747 
1748 bad_unshare_out:
1749 	return err;
1750 }
1751 
1752 /*
1753  *	Helper to unshare the files of the current task.
1754  *	We don't want to expose copy_files internals to
1755  *	the exec layer of the kernel.
1756  */
1757 
1758 int unshare_files(struct files_struct **displaced)
1759 {
1760 	struct task_struct *task = current;
1761 	struct files_struct *copy = NULL;
1762 	int error;
1763 
1764 	error = unshare_fd(CLONE_FILES, &copy);
1765 	if (error || !copy) {
1766 		*displaced = NULL;
1767 		return error;
1768 	}
1769 	*displaced = task->files;
1770 	task_lock(task);
1771 	task->files = copy;
1772 	task_unlock(task);
1773 	return 0;
1774 }
1775