xref: /linux-6.15/kernel/fork.c (revision c9d05cfc)
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/tracehook.h>
41 #include <linux/futex.h>
42 #include <linux/compat.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/proc_fs.h>
62 #include <linux/blkdev.h>
63 #include <linux/fs_struct.h>
64 #include <linux/magic.h>
65 #include <linux/perf_event.h>
66 
67 #include <asm/pgtable.h>
68 #include <asm/pgalloc.h>
69 #include <asm/uaccess.h>
70 #include <asm/mmu_context.h>
71 #include <asm/cacheflush.h>
72 #include <asm/tlbflush.h>
73 
74 #include <trace/events/sched.h>
75 
76 /*
77  * Protected counters by write_lock_irq(&tasklist_lock)
78  */
79 unsigned long total_forks;	/* Handle normal Linux uptimes. */
80 int nr_threads; 		/* The idle threads do not count.. */
81 
82 int max_threads;		/* tunable limit on nr_threads */
83 
84 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
85 
86 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
87 
88 int nr_processes(void)
89 {
90 	int cpu;
91 	int total = 0;
92 
93 	for_each_online_cpu(cpu)
94 		total += per_cpu(process_counts, cpu);
95 
96 	return total;
97 }
98 
99 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
100 # define alloc_task_struct()	kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
101 # define free_task_struct(tsk)	kmem_cache_free(task_struct_cachep, (tsk))
102 static struct kmem_cache *task_struct_cachep;
103 #endif
104 
105 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
106 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
107 {
108 #ifdef CONFIG_DEBUG_STACK_USAGE
109 	gfp_t mask = GFP_KERNEL | __GFP_ZERO;
110 #else
111 	gfp_t mask = GFP_KERNEL;
112 #endif
113 	return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
114 }
115 
116 static inline void free_thread_info(struct thread_info *ti)
117 {
118 	free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
119 }
120 #endif
121 
122 /* SLAB cache for signal_struct structures (tsk->signal) */
123 static struct kmem_cache *signal_cachep;
124 
125 /* SLAB cache for sighand_struct structures (tsk->sighand) */
126 struct kmem_cache *sighand_cachep;
127 
128 /* SLAB cache for files_struct structures (tsk->files) */
129 struct kmem_cache *files_cachep;
130 
131 /* SLAB cache for fs_struct structures (tsk->fs) */
132 struct kmem_cache *fs_cachep;
133 
134 /* SLAB cache for vm_area_struct structures */
135 struct kmem_cache *vm_area_cachep;
136 
137 /* SLAB cache for mm_struct structures (tsk->mm) */
138 static struct kmem_cache *mm_cachep;
139 
140 static void account_kernel_stack(struct thread_info *ti, int account)
141 {
142 	struct zone *zone = page_zone(virt_to_page(ti));
143 
144 	mod_zone_page_state(zone, NR_KERNEL_STACK, account);
145 }
146 
147 void free_task(struct task_struct *tsk)
148 {
149 	prop_local_destroy_single(&tsk->dirties);
150 	account_kernel_stack(tsk->stack, -1);
151 	free_thread_info(tsk->stack);
152 	rt_mutex_debug_task_free(tsk);
153 	ftrace_graph_exit_task(tsk);
154 	free_task_struct(tsk);
155 }
156 EXPORT_SYMBOL(free_task);
157 
158 void __put_task_struct(struct task_struct *tsk)
159 {
160 	WARN_ON(!tsk->exit_state);
161 	WARN_ON(atomic_read(&tsk->usage));
162 	WARN_ON(tsk == current);
163 
164 	exit_creds(tsk);
165 	delayacct_tsk_free(tsk);
166 
167 	if (!profile_handoff_task(tsk))
168 		free_task(tsk);
169 }
170 
171 /*
172  * macro override instead of weak attribute alias, to workaround
173  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
174  */
175 #ifndef arch_task_cache_init
176 #define arch_task_cache_init()
177 #endif
178 
179 void __init fork_init(unsigned long mempages)
180 {
181 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
182 #ifndef ARCH_MIN_TASKALIGN
183 #define ARCH_MIN_TASKALIGN	L1_CACHE_BYTES
184 #endif
185 	/* create a slab on which task_structs can be allocated */
186 	task_struct_cachep =
187 		kmem_cache_create("task_struct", sizeof(struct task_struct),
188 			ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
189 #endif
190 
191 	/* do the arch specific task caches init */
192 	arch_task_cache_init();
193 
194 	/*
195 	 * The default maximum number of threads is set to a safe
196 	 * value: the thread structures can take up at most half
197 	 * of memory.
198 	 */
199 	max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
200 
201 	/*
202 	 * we need to allow at least 20 threads to boot a system
203 	 */
204 	if(max_threads < 20)
205 		max_threads = 20;
206 
207 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
208 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
209 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
210 		init_task.signal->rlim[RLIMIT_NPROC];
211 }
212 
213 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
214 					       struct task_struct *src)
215 {
216 	*dst = *src;
217 	return 0;
218 }
219 
220 static struct task_struct *dup_task_struct(struct task_struct *orig)
221 {
222 	struct task_struct *tsk;
223 	struct thread_info *ti;
224 	unsigned long *stackend;
225 
226 	int err;
227 
228 	prepare_to_copy(orig);
229 
230 	tsk = alloc_task_struct();
231 	if (!tsk)
232 		return NULL;
233 
234 	ti = alloc_thread_info(tsk);
235 	if (!ti) {
236 		free_task_struct(tsk);
237 		return NULL;
238 	}
239 
240  	err = arch_dup_task_struct(tsk, orig);
241 	if (err)
242 		goto out;
243 
244 	tsk->stack = ti;
245 
246 	err = prop_local_init_single(&tsk->dirties);
247 	if (err)
248 		goto out;
249 
250 	setup_thread_stack(tsk, orig);
251 	stackend = end_of_stack(tsk);
252 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
253 
254 #ifdef CONFIG_CC_STACKPROTECTOR
255 	tsk->stack_canary = get_random_int();
256 #endif
257 
258 	/* One for us, one for whoever does the "release_task()" (usually parent) */
259 	atomic_set(&tsk->usage,2);
260 	atomic_set(&tsk->fs_excl, 0);
261 #ifdef CONFIG_BLK_DEV_IO_TRACE
262 	tsk->btrace_seq = 0;
263 #endif
264 	tsk->splice_pipe = NULL;
265 
266 	account_kernel_stack(ti, 1);
267 
268 	return tsk;
269 
270 out:
271 	free_thread_info(ti);
272 	free_task_struct(tsk);
273 	return NULL;
274 }
275 
276 #ifdef CONFIG_MMU
277 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
278 {
279 	struct vm_area_struct *mpnt, *tmp, **pprev;
280 	struct rb_node **rb_link, *rb_parent;
281 	int retval;
282 	unsigned long charge;
283 	struct mempolicy *pol;
284 
285 	down_write(&oldmm->mmap_sem);
286 	flush_cache_dup_mm(oldmm);
287 	/*
288 	 * Not linked in yet - no deadlock potential:
289 	 */
290 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
291 
292 	mm->locked_vm = 0;
293 	mm->mmap = NULL;
294 	mm->mmap_cache = NULL;
295 	mm->free_area_cache = oldmm->mmap_base;
296 	mm->cached_hole_size = ~0UL;
297 	mm->map_count = 0;
298 	cpumask_clear(mm_cpumask(mm));
299 	mm->mm_rb = RB_ROOT;
300 	rb_link = &mm->mm_rb.rb_node;
301 	rb_parent = NULL;
302 	pprev = &mm->mmap;
303 	retval = ksm_fork(mm, oldmm);
304 	if (retval)
305 		goto out;
306 
307 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
308 		struct file *file;
309 
310 		if (mpnt->vm_flags & VM_DONTCOPY) {
311 			long pages = vma_pages(mpnt);
312 			mm->total_vm -= pages;
313 			vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
314 								-pages);
315 			continue;
316 		}
317 		charge = 0;
318 		if (mpnt->vm_flags & VM_ACCOUNT) {
319 			unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
320 			if (security_vm_enough_memory(len))
321 				goto fail_nomem;
322 			charge = len;
323 		}
324 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
325 		if (!tmp)
326 			goto fail_nomem;
327 		*tmp = *mpnt;
328 		pol = mpol_dup(vma_policy(mpnt));
329 		retval = PTR_ERR(pol);
330 		if (IS_ERR(pol))
331 			goto fail_nomem_policy;
332 		vma_set_policy(tmp, pol);
333 		tmp->vm_flags &= ~VM_LOCKED;
334 		tmp->vm_mm = mm;
335 		tmp->vm_next = NULL;
336 		anon_vma_link(tmp);
337 		file = tmp->vm_file;
338 		if (file) {
339 			struct inode *inode = file->f_path.dentry->d_inode;
340 			struct address_space *mapping = file->f_mapping;
341 
342 			get_file(file);
343 			if (tmp->vm_flags & VM_DENYWRITE)
344 				atomic_dec(&inode->i_writecount);
345 			spin_lock(&mapping->i_mmap_lock);
346 			if (tmp->vm_flags & VM_SHARED)
347 				mapping->i_mmap_writable++;
348 			tmp->vm_truncate_count = mpnt->vm_truncate_count;
349 			flush_dcache_mmap_lock(mapping);
350 			/* insert tmp into the share list, just after mpnt */
351 			vma_prio_tree_add(tmp, mpnt);
352 			flush_dcache_mmap_unlock(mapping);
353 			spin_unlock(&mapping->i_mmap_lock);
354 		}
355 
356 		/*
357 		 * Clear hugetlb-related page reserves for children. This only
358 		 * affects MAP_PRIVATE mappings. Faults generated by the child
359 		 * are not guaranteed to succeed, even if read-only
360 		 */
361 		if (is_vm_hugetlb_page(tmp))
362 			reset_vma_resv_huge_pages(tmp);
363 
364 		/*
365 		 * Link in the new vma and copy the page table entries.
366 		 */
367 		*pprev = tmp;
368 		pprev = &tmp->vm_next;
369 
370 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
371 		rb_link = &tmp->vm_rb.rb_right;
372 		rb_parent = &tmp->vm_rb;
373 
374 		mm->map_count++;
375 		retval = copy_page_range(mm, oldmm, mpnt);
376 
377 		if (tmp->vm_ops && tmp->vm_ops->open)
378 			tmp->vm_ops->open(tmp);
379 
380 		if (retval)
381 			goto out;
382 	}
383 	/* a new mm has just been created */
384 	arch_dup_mmap(oldmm, mm);
385 	retval = 0;
386 out:
387 	up_write(&mm->mmap_sem);
388 	flush_tlb_mm(oldmm);
389 	up_write(&oldmm->mmap_sem);
390 	return retval;
391 fail_nomem_policy:
392 	kmem_cache_free(vm_area_cachep, tmp);
393 fail_nomem:
394 	retval = -ENOMEM;
395 	vm_unacct_memory(charge);
396 	goto out;
397 }
398 
399 static inline int mm_alloc_pgd(struct mm_struct * mm)
400 {
401 	mm->pgd = pgd_alloc(mm);
402 	if (unlikely(!mm->pgd))
403 		return -ENOMEM;
404 	return 0;
405 }
406 
407 static inline void mm_free_pgd(struct mm_struct * mm)
408 {
409 	pgd_free(mm, mm->pgd);
410 }
411 #else
412 #define dup_mmap(mm, oldmm)	(0)
413 #define mm_alloc_pgd(mm)	(0)
414 #define mm_free_pgd(mm)
415 #endif /* CONFIG_MMU */
416 
417 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
418 
419 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
420 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
421 
422 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
423 
424 static int __init coredump_filter_setup(char *s)
425 {
426 	default_dump_filter =
427 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
428 		MMF_DUMP_FILTER_MASK;
429 	return 1;
430 }
431 
432 __setup("coredump_filter=", coredump_filter_setup);
433 
434 #include <linux/init_task.h>
435 
436 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
437 {
438 	atomic_set(&mm->mm_users, 1);
439 	atomic_set(&mm->mm_count, 1);
440 	init_rwsem(&mm->mmap_sem);
441 	INIT_LIST_HEAD(&mm->mmlist);
442 	mm->flags = (current->mm) ?
443 		(current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
444 	mm->core_state = NULL;
445 	mm->nr_ptes = 0;
446 	set_mm_counter(mm, file_rss, 0);
447 	set_mm_counter(mm, anon_rss, 0);
448 	spin_lock_init(&mm->page_table_lock);
449 	spin_lock_init(&mm->ioctx_lock);
450 	INIT_HLIST_HEAD(&mm->ioctx_list);
451 	mm->free_area_cache = TASK_UNMAPPED_BASE;
452 	mm->cached_hole_size = ~0UL;
453 	mm_init_owner(mm, p);
454 
455 	if (likely(!mm_alloc_pgd(mm))) {
456 		mm->def_flags = 0;
457 		mmu_notifier_mm_init(mm);
458 		return mm;
459 	}
460 
461 	free_mm(mm);
462 	return NULL;
463 }
464 
465 /*
466  * Allocate and initialize an mm_struct.
467  */
468 struct mm_struct * mm_alloc(void)
469 {
470 	struct mm_struct * mm;
471 
472 	mm = allocate_mm();
473 	if (mm) {
474 		memset(mm, 0, sizeof(*mm));
475 		mm = mm_init(mm, current);
476 	}
477 	return mm;
478 }
479 
480 /*
481  * Called when the last reference to the mm
482  * is dropped: either by a lazy thread or by
483  * mmput. Free the page directory and the mm.
484  */
485 void __mmdrop(struct mm_struct *mm)
486 {
487 	BUG_ON(mm == &init_mm);
488 	mm_free_pgd(mm);
489 	destroy_context(mm);
490 	mmu_notifier_mm_destroy(mm);
491 	free_mm(mm);
492 }
493 EXPORT_SYMBOL_GPL(__mmdrop);
494 
495 /*
496  * Decrement the use count and release all resources for an mm.
497  */
498 void mmput(struct mm_struct *mm)
499 {
500 	might_sleep();
501 
502 	if (atomic_dec_and_test(&mm->mm_users)) {
503 		exit_aio(mm);
504 		ksm_exit(mm);
505 		exit_mmap(mm);
506 		set_mm_exe_file(mm, NULL);
507 		if (!list_empty(&mm->mmlist)) {
508 			spin_lock(&mmlist_lock);
509 			list_del(&mm->mmlist);
510 			spin_unlock(&mmlist_lock);
511 		}
512 		put_swap_token(mm);
513 		mmdrop(mm);
514 	}
515 }
516 EXPORT_SYMBOL_GPL(mmput);
517 
518 /**
519  * get_task_mm - acquire a reference to the task's mm
520  *
521  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
522  * this kernel workthread has transiently adopted a user mm with use_mm,
523  * to do its AIO) is not set and if so returns a reference to it, after
524  * bumping up the use count.  User must release the mm via mmput()
525  * after use.  Typically used by /proc and ptrace.
526  */
527 struct mm_struct *get_task_mm(struct task_struct *task)
528 {
529 	struct mm_struct *mm;
530 
531 	task_lock(task);
532 	mm = task->mm;
533 	if (mm) {
534 		if (task->flags & PF_KTHREAD)
535 			mm = NULL;
536 		else
537 			atomic_inc(&mm->mm_users);
538 	}
539 	task_unlock(task);
540 	return mm;
541 }
542 EXPORT_SYMBOL_GPL(get_task_mm);
543 
544 /* Please note the differences between mmput and mm_release.
545  * mmput is called whenever we stop holding onto a mm_struct,
546  * error success whatever.
547  *
548  * mm_release is called after a mm_struct has been removed
549  * from the current process.
550  *
551  * This difference is important for error handling, when we
552  * only half set up a mm_struct for a new process and need to restore
553  * the old one.  Because we mmput the new mm_struct before
554  * restoring the old one. . .
555  * Eric Biederman 10 January 1998
556  */
557 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
558 {
559 	struct completion *vfork_done = tsk->vfork_done;
560 
561 	/* Get rid of any futexes when releasing the mm */
562 #ifdef CONFIG_FUTEX
563 	if (unlikely(tsk->robust_list))
564 		exit_robust_list(tsk);
565 #ifdef CONFIG_COMPAT
566 	if (unlikely(tsk->compat_robust_list))
567 		compat_exit_robust_list(tsk);
568 #endif
569 #endif
570 
571 	/* Get rid of any cached register state */
572 	deactivate_mm(tsk, mm);
573 
574 	/* notify parent sleeping on vfork() */
575 	if (vfork_done) {
576 		tsk->vfork_done = NULL;
577 		complete(vfork_done);
578 	}
579 
580 	/*
581 	 * If we're exiting normally, clear a user-space tid field if
582 	 * requested.  We leave this alone when dying by signal, to leave
583 	 * the value intact in a core dump, and to save the unnecessary
584 	 * trouble otherwise.  Userland only wants this done for a sys_exit.
585 	 */
586 	if (tsk->clear_child_tid) {
587 		if (!(tsk->flags & PF_SIGNALED) &&
588 		    atomic_read(&mm->mm_users) > 1) {
589 			/*
590 			 * We don't check the error code - if userspace has
591 			 * not set up a proper pointer then tough luck.
592 			 */
593 			put_user(0, tsk->clear_child_tid);
594 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
595 					1, NULL, NULL, 0);
596 		}
597 		tsk->clear_child_tid = NULL;
598 	}
599 }
600 
601 /*
602  * Allocate a new mm structure and copy contents from the
603  * mm structure of the passed in task structure.
604  */
605 struct mm_struct *dup_mm(struct task_struct *tsk)
606 {
607 	struct mm_struct *mm, *oldmm = current->mm;
608 	int err;
609 
610 	if (!oldmm)
611 		return NULL;
612 
613 	mm = allocate_mm();
614 	if (!mm)
615 		goto fail_nomem;
616 
617 	memcpy(mm, oldmm, sizeof(*mm));
618 
619 	/* Initializing for Swap token stuff */
620 	mm->token_priority = 0;
621 	mm->last_interval = 0;
622 
623 	if (!mm_init(mm, tsk))
624 		goto fail_nomem;
625 
626 	if (init_new_context(tsk, mm))
627 		goto fail_nocontext;
628 
629 	dup_mm_exe_file(oldmm, mm);
630 
631 	err = dup_mmap(mm, oldmm);
632 	if (err)
633 		goto free_pt;
634 
635 	mm->hiwater_rss = get_mm_rss(mm);
636 	mm->hiwater_vm = mm->total_vm;
637 
638 	return mm;
639 
640 free_pt:
641 	mmput(mm);
642 
643 fail_nomem:
644 	return NULL;
645 
646 fail_nocontext:
647 	/*
648 	 * If init_new_context() failed, we cannot use mmput() to free the mm
649 	 * because it calls destroy_context()
650 	 */
651 	mm_free_pgd(mm);
652 	free_mm(mm);
653 	return NULL;
654 }
655 
656 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
657 {
658 	struct mm_struct * mm, *oldmm;
659 	int retval;
660 
661 	tsk->min_flt = tsk->maj_flt = 0;
662 	tsk->nvcsw = tsk->nivcsw = 0;
663 #ifdef CONFIG_DETECT_HUNG_TASK
664 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
665 #endif
666 
667 	tsk->mm = NULL;
668 	tsk->active_mm = NULL;
669 
670 	/*
671 	 * Are we cloning a kernel thread?
672 	 *
673 	 * We need to steal a active VM for that..
674 	 */
675 	oldmm = current->mm;
676 	if (!oldmm)
677 		return 0;
678 
679 	if (clone_flags & CLONE_VM) {
680 		atomic_inc(&oldmm->mm_users);
681 		mm = oldmm;
682 		goto good_mm;
683 	}
684 
685 	retval = -ENOMEM;
686 	mm = dup_mm(tsk);
687 	if (!mm)
688 		goto fail_nomem;
689 
690 good_mm:
691 	/* Initializing for Swap token stuff */
692 	mm->token_priority = 0;
693 	mm->last_interval = 0;
694 
695 	tsk->mm = mm;
696 	tsk->active_mm = mm;
697 	return 0;
698 
699 fail_nomem:
700 	return retval;
701 }
702 
703 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
704 {
705 	struct fs_struct *fs = current->fs;
706 	if (clone_flags & CLONE_FS) {
707 		/* tsk->fs is already what we want */
708 		write_lock(&fs->lock);
709 		if (fs->in_exec) {
710 			write_unlock(&fs->lock);
711 			return -EAGAIN;
712 		}
713 		fs->users++;
714 		write_unlock(&fs->lock);
715 		return 0;
716 	}
717 	tsk->fs = copy_fs_struct(fs);
718 	if (!tsk->fs)
719 		return -ENOMEM;
720 	return 0;
721 }
722 
723 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
724 {
725 	struct files_struct *oldf, *newf;
726 	int error = 0;
727 
728 	/*
729 	 * A background process may not have any files ...
730 	 */
731 	oldf = current->files;
732 	if (!oldf)
733 		goto out;
734 
735 	if (clone_flags & CLONE_FILES) {
736 		atomic_inc(&oldf->count);
737 		goto out;
738 	}
739 
740 	newf = dup_fd(oldf, &error);
741 	if (!newf)
742 		goto out;
743 
744 	tsk->files = newf;
745 	error = 0;
746 out:
747 	return error;
748 }
749 
750 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
751 {
752 #ifdef CONFIG_BLOCK
753 	struct io_context *ioc = current->io_context;
754 
755 	if (!ioc)
756 		return 0;
757 	/*
758 	 * Share io context with parent, if CLONE_IO is set
759 	 */
760 	if (clone_flags & CLONE_IO) {
761 		tsk->io_context = ioc_task_link(ioc);
762 		if (unlikely(!tsk->io_context))
763 			return -ENOMEM;
764 	} else if (ioprio_valid(ioc->ioprio)) {
765 		tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
766 		if (unlikely(!tsk->io_context))
767 			return -ENOMEM;
768 
769 		tsk->io_context->ioprio = ioc->ioprio;
770 	}
771 #endif
772 	return 0;
773 }
774 
775 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
776 {
777 	struct sighand_struct *sig;
778 
779 	if (clone_flags & CLONE_SIGHAND) {
780 		atomic_inc(&current->sighand->count);
781 		return 0;
782 	}
783 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
784 	rcu_assign_pointer(tsk->sighand, sig);
785 	if (!sig)
786 		return -ENOMEM;
787 	atomic_set(&sig->count, 1);
788 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
789 	return 0;
790 }
791 
792 void __cleanup_sighand(struct sighand_struct *sighand)
793 {
794 	if (atomic_dec_and_test(&sighand->count))
795 		kmem_cache_free(sighand_cachep, sighand);
796 }
797 
798 
799 /*
800  * Initialize POSIX timer handling for a thread group.
801  */
802 static void posix_cpu_timers_init_group(struct signal_struct *sig)
803 {
804 	/* Thread group counters. */
805 	thread_group_cputime_init(sig);
806 
807 	/* Expiration times and increments. */
808 	sig->it_virt_expires = cputime_zero;
809 	sig->it_virt_incr = cputime_zero;
810 	sig->it_prof_expires = cputime_zero;
811 	sig->it_prof_incr = cputime_zero;
812 
813 	/* Cached expiration times. */
814 	sig->cputime_expires.prof_exp = cputime_zero;
815 	sig->cputime_expires.virt_exp = cputime_zero;
816 	sig->cputime_expires.sched_exp = 0;
817 
818 	if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
819 		sig->cputime_expires.prof_exp =
820 			secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
821 		sig->cputimer.running = 1;
822 	}
823 
824 	/* The timer lists. */
825 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
826 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
827 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
828 }
829 
830 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
831 {
832 	struct signal_struct *sig;
833 
834 	if (clone_flags & CLONE_THREAD)
835 		return 0;
836 
837 	sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
838 	tsk->signal = sig;
839 	if (!sig)
840 		return -ENOMEM;
841 
842 	atomic_set(&sig->count, 1);
843 	atomic_set(&sig->live, 1);
844 	init_waitqueue_head(&sig->wait_chldexit);
845 	sig->flags = 0;
846 	if (clone_flags & CLONE_NEWPID)
847 		sig->flags |= SIGNAL_UNKILLABLE;
848 	sig->group_exit_code = 0;
849 	sig->group_exit_task = NULL;
850 	sig->group_stop_count = 0;
851 	sig->curr_target = tsk;
852 	init_sigpending(&sig->shared_pending);
853 	INIT_LIST_HEAD(&sig->posix_timers);
854 
855 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
856 	sig->it_real_incr.tv64 = 0;
857 	sig->real_timer.function = it_real_fn;
858 
859 	sig->leader = 0;	/* session leadership doesn't inherit */
860 	sig->tty_old_pgrp = NULL;
861 	sig->tty = NULL;
862 
863 	sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
864 	sig->gtime = cputime_zero;
865 	sig->cgtime = cputime_zero;
866 	sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
867 	sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
868 	sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
869 	task_io_accounting_init(&sig->ioac);
870 	sig->sum_sched_runtime = 0;
871 	taskstats_tgid_init(sig);
872 
873 	task_lock(current->group_leader);
874 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
875 	task_unlock(current->group_leader);
876 
877 	posix_cpu_timers_init_group(sig);
878 
879 	acct_init_pacct(&sig->pacct);
880 
881 	tty_audit_fork(sig);
882 
883 	return 0;
884 }
885 
886 void __cleanup_signal(struct signal_struct *sig)
887 {
888 	thread_group_cputime_free(sig);
889 	tty_kref_put(sig->tty);
890 	kmem_cache_free(signal_cachep, sig);
891 }
892 
893 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
894 {
895 	unsigned long new_flags = p->flags;
896 
897 	new_flags &= ~PF_SUPERPRIV;
898 	new_flags |= PF_FORKNOEXEC;
899 	new_flags |= PF_STARTING;
900 	p->flags = new_flags;
901 	clear_freeze_flag(p);
902 }
903 
904 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
905 {
906 	current->clear_child_tid = tidptr;
907 
908 	return task_pid_vnr(current);
909 }
910 
911 static void rt_mutex_init_task(struct task_struct *p)
912 {
913 	spin_lock_init(&p->pi_lock);
914 #ifdef CONFIG_RT_MUTEXES
915 	plist_head_init(&p->pi_waiters, &p->pi_lock);
916 	p->pi_blocked_on = NULL;
917 #endif
918 }
919 
920 #ifdef CONFIG_MM_OWNER
921 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
922 {
923 	mm->owner = p;
924 }
925 #endif /* CONFIG_MM_OWNER */
926 
927 /*
928  * Initialize POSIX timer handling for a single task.
929  */
930 static void posix_cpu_timers_init(struct task_struct *tsk)
931 {
932 	tsk->cputime_expires.prof_exp = cputime_zero;
933 	tsk->cputime_expires.virt_exp = cputime_zero;
934 	tsk->cputime_expires.sched_exp = 0;
935 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
936 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
937 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
938 }
939 
940 /*
941  * This creates a new process as a copy of the old one,
942  * but does not actually start it yet.
943  *
944  * It copies the registers, and all the appropriate
945  * parts of the process environment (as per the clone
946  * flags). The actual kick-off is left to the caller.
947  */
948 static struct task_struct *copy_process(unsigned long clone_flags,
949 					unsigned long stack_start,
950 					struct pt_regs *regs,
951 					unsigned long stack_size,
952 					int __user *child_tidptr,
953 					struct pid *pid,
954 					int trace)
955 {
956 	int retval;
957 	struct task_struct *p;
958 	int cgroup_callbacks_done = 0;
959 
960 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
961 		return ERR_PTR(-EINVAL);
962 
963 	/*
964 	 * Thread groups must share signals as well, and detached threads
965 	 * can only be started up within the thread group.
966 	 */
967 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
968 		return ERR_PTR(-EINVAL);
969 
970 	/*
971 	 * Shared signal handlers imply shared VM. By way of the above,
972 	 * thread groups also imply shared VM. Blocking this case allows
973 	 * for various simplifications in other code.
974 	 */
975 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
976 		return ERR_PTR(-EINVAL);
977 
978 	retval = security_task_create(clone_flags);
979 	if (retval)
980 		goto fork_out;
981 
982 	retval = -ENOMEM;
983 	p = dup_task_struct(current);
984 	if (!p)
985 		goto fork_out;
986 
987 	ftrace_graph_init_task(p);
988 
989 	rt_mutex_init_task(p);
990 
991 #ifdef CONFIG_PROVE_LOCKING
992 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
993 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
994 #endif
995 	retval = -EAGAIN;
996 	if (atomic_read(&p->real_cred->user->processes) >=
997 			p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
998 		if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
999 		    p->real_cred->user != INIT_USER)
1000 			goto bad_fork_free;
1001 	}
1002 
1003 	retval = copy_creds(p, clone_flags);
1004 	if (retval < 0)
1005 		goto bad_fork_free;
1006 
1007 	/*
1008 	 * If multiple threads are within copy_process(), then this check
1009 	 * triggers too late. This doesn't hurt, the check is only there
1010 	 * to stop root fork bombs.
1011 	 */
1012 	retval = -EAGAIN;
1013 	if (nr_threads >= max_threads)
1014 		goto bad_fork_cleanup_count;
1015 
1016 	if (!try_module_get(task_thread_info(p)->exec_domain->module))
1017 		goto bad_fork_cleanup_count;
1018 
1019 	if (p->binfmt && !try_module_get(p->binfmt->module))
1020 		goto bad_fork_cleanup_put_domain;
1021 
1022 	p->did_exec = 0;
1023 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1024 	copy_flags(clone_flags, p);
1025 	INIT_LIST_HEAD(&p->children);
1026 	INIT_LIST_HEAD(&p->sibling);
1027 	rcu_copy_process(p);
1028 	p->vfork_done = NULL;
1029 	spin_lock_init(&p->alloc_lock);
1030 
1031 	init_sigpending(&p->pending);
1032 
1033 	p->utime = cputime_zero;
1034 	p->stime = cputime_zero;
1035 	p->gtime = cputime_zero;
1036 	p->utimescaled = cputime_zero;
1037 	p->stimescaled = cputime_zero;
1038 	p->prev_utime = cputime_zero;
1039 	p->prev_stime = cputime_zero;
1040 
1041 	p->default_timer_slack_ns = current->timer_slack_ns;
1042 
1043 	task_io_accounting_init(&p->ioac);
1044 	acct_clear_integrals(p);
1045 
1046 	posix_cpu_timers_init(p);
1047 
1048 	p->lock_depth = -1;		/* -1 = no lock */
1049 	do_posix_clock_monotonic_gettime(&p->start_time);
1050 	p->real_start_time = p->start_time;
1051 	monotonic_to_bootbased(&p->real_start_time);
1052 	p->io_context = NULL;
1053 	p->audit_context = NULL;
1054 	cgroup_fork(p);
1055 #ifdef CONFIG_NUMA
1056 	p->mempolicy = mpol_dup(p->mempolicy);
1057  	if (IS_ERR(p->mempolicy)) {
1058  		retval = PTR_ERR(p->mempolicy);
1059  		p->mempolicy = NULL;
1060  		goto bad_fork_cleanup_cgroup;
1061  	}
1062 	mpol_fix_fork_child_flag(p);
1063 #endif
1064 #ifdef CONFIG_TRACE_IRQFLAGS
1065 	p->irq_events = 0;
1066 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1067 	p->hardirqs_enabled = 1;
1068 #else
1069 	p->hardirqs_enabled = 0;
1070 #endif
1071 	p->hardirq_enable_ip = 0;
1072 	p->hardirq_enable_event = 0;
1073 	p->hardirq_disable_ip = _THIS_IP_;
1074 	p->hardirq_disable_event = 0;
1075 	p->softirqs_enabled = 1;
1076 	p->softirq_enable_ip = _THIS_IP_;
1077 	p->softirq_enable_event = 0;
1078 	p->softirq_disable_ip = 0;
1079 	p->softirq_disable_event = 0;
1080 	p->hardirq_context = 0;
1081 	p->softirq_context = 0;
1082 #endif
1083 #ifdef CONFIG_LOCKDEP
1084 	p->lockdep_depth = 0; /* no locks held yet */
1085 	p->curr_chain_key = 0;
1086 	p->lockdep_recursion = 0;
1087 #endif
1088 
1089 #ifdef CONFIG_DEBUG_MUTEXES
1090 	p->blocked_on = NULL; /* not blocked yet */
1091 #endif
1092 
1093 	p->bts = NULL;
1094 
1095 	/* Perform scheduler related setup. Assign this task to a CPU. */
1096 	sched_fork(p, clone_flags);
1097 
1098 	retval = perf_event_init_task(p);
1099 	if (retval)
1100 		goto bad_fork_cleanup_policy;
1101 
1102 	if ((retval = audit_alloc(p)))
1103 		goto bad_fork_cleanup_policy;
1104 	/* copy all the process information */
1105 	if ((retval = copy_semundo(clone_flags, p)))
1106 		goto bad_fork_cleanup_audit;
1107 	if ((retval = copy_files(clone_flags, p)))
1108 		goto bad_fork_cleanup_semundo;
1109 	if ((retval = copy_fs(clone_flags, p)))
1110 		goto bad_fork_cleanup_files;
1111 	if ((retval = copy_sighand(clone_flags, p)))
1112 		goto bad_fork_cleanup_fs;
1113 	if ((retval = copy_signal(clone_flags, p)))
1114 		goto bad_fork_cleanup_sighand;
1115 	if ((retval = copy_mm(clone_flags, p)))
1116 		goto bad_fork_cleanup_signal;
1117 	if ((retval = copy_namespaces(clone_flags, p)))
1118 		goto bad_fork_cleanup_mm;
1119 	if ((retval = copy_io(clone_flags, p)))
1120 		goto bad_fork_cleanup_namespaces;
1121 	retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1122 	if (retval)
1123 		goto bad_fork_cleanup_io;
1124 
1125 	if (pid != &init_struct_pid) {
1126 		retval = -ENOMEM;
1127 		pid = alloc_pid(p->nsproxy->pid_ns);
1128 		if (!pid)
1129 			goto bad_fork_cleanup_io;
1130 
1131 		if (clone_flags & CLONE_NEWPID) {
1132 			retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1133 			if (retval < 0)
1134 				goto bad_fork_free_pid;
1135 		}
1136 	}
1137 
1138 	p->pid = pid_nr(pid);
1139 	p->tgid = p->pid;
1140 	if (clone_flags & CLONE_THREAD)
1141 		p->tgid = current->tgid;
1142 
1143 	if (current->nsproxy != p->nsproxy) {
1144 		retval = ns_cgroup_clone(p, pid);
1145 		if (retval)
1146 			goto bad_fork_free_pid;
1147 	}
1148 
1149 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1150 	/*
1151 	 * Clear TID on mm_release()?
1152 	 */
1153 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1154 #ifdef CONFIG_FUTEX
1155 	p->robust_list = NULL;
1156 #ifdef CONFIG_COMPAT
1157 	p->compat_robust_list = NULL;
1158 #endif
1159 	INIT_LIST_HEAD(&p->pi_state_list);
1160 	p->pi_state_cache = NULL;
1161 #endif
1162 	/*
1163 	 * sigaltstack should be cleared when sharing the same VM
1164 	 */
1165 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1166 		p->sas_ss_sp = p->sas_ss_size = 0;
1167 
1168 	/*
1169 	 * Syscall tracing should be turned off in the child regardless
1170 	 * of CLONE_PTRACE.
1171 	 */
1172 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1173 #ifdef TIF_SYSCALL_EMU
1174 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1175 #endif
1176 	clear_all_latency_tracing(p);
1177 
1178 	/* ok, now we should be set up.. */
1179 	p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1180 	p->pdeath_signal = 0;
1181 	p->exit_state = 0;
1182 
1183 	/*
1184 	 * Ok, make it visible to the rest of the system.
1185 	 * We dont wake it up yet.
1186 	 */
1187 	p->group_leader = p;
1188 	INIT_LIST_HEAD(&p->thread_group);
1189 
1190 	/* Now that the task is set up, run cgroup callbacks if
1191 	 * necessary. We need to run them before the task is visible
1192 	 * on the tasklist. */
1193 	cgroup_fork_callbacks(p);
1194 	cgroup_callbacks_done = 1;
1195 
1196 	/* Need tasklist lock for parent etc handling! */
1197 	write_lock_irq(&tasklist_lock);
1198 
1199 	/*
1200 	 * The task hasn't been attached yet, so its cpus_allowed mask will
1201 	 * not be changed, nor will its assigned CPU.
1202 	 *
1203 	 * The cpus_allowed mask of the parent may have changed after it was
1204 	 * copied first time - so re-copy it here, then check the child's CPU
1205 	 * to ensure it is on a valid CPU (and if not, just force it back to
1206 	 * parent's CPU). This avoids alot of nasty races.
1207 	 */
1208 	p->cpus_allowed = current->cpus_allowed;
1209 	p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1210 	if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1211 			!cpu_online(task_cpu(p))))
1212 		set_task_cpu(p, smp_processor_id());
1213 
1214 	/* CLONE_PARENT re-uses the old parent */
1215 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1216 		p->real_parent = current->real_parent;
1217 		p->parent_exec_id = current->parent_exec_id;
1218 	} else {
1219 		p->real_parent = current;
1220 		p->parent_exec_id = current->self_exec_id;
1221 	}
1222 
1223 	spin_lock(&current->sighand->siglock);
1224 
1225 	/*
1226 	 * Process group and session signals need to be delivered to just the
1227 	 * parent before the fork or both the parent and the child after the
1228 	 * fork. Restart if a signal comes in before we add the new process to
1229 	 * it's process group.
1230 	 * A fatal signal pending means that current will exit, so the new
1231 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1232  	 */
1233 	recalc_sigpending();
1234 	if (signal_pending(current)) {
1235 		spin_unlock(&current->sighand->siglock);
1236 		write_unlock_irq(&tasklist_lock);
1237 		retval = -ERESTARTNOINTR;
1238 		goto bad_fork_free_pid;
1239 	}
1240 
1241 	if (clone_flags & CLONE_THREAD) {
1242 		atomic_inc(&current->signal->count);
1243 		atomic_inc(&current->signal->live);
1244 		p->group_leader = current->group_leader;
1245 		list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1246 	}
1247 
1248 	if (likely(p->pid)) {
1249 		list_add_tail(&p->sibling, &p->real_parent->children);
1250 		tracehook_finish_clone(p, clone_flags, trace);
1251 
1252 		if (thread_group_leader(p)) {
1253 			if (clone_flags & CLONE_NEWPID)
1254 				p->nsproxy->pid_ns->child_reaper = p;
1255 
1256 			p->signal->leader_pid = pid;
1257 			tty_kref_put(p->signal->tty);
1258 			p->signal->tty = tty_kref_get(current->signal->tty);
1259 			attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1260 			attach_pid(p, PIDTYPE_SID, task_session(current));
1261 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1262 			__get_cpu_var(process_counts)++;
1263 		}
1264 		attach_pid(p, PIDTYPE_PID, pid);
1265 		nr_threads++;
1266 	}
1267 
1268 	total_forks++;
1269 	spin_unlock(&current->sighand->siglock);
1270 	write_unlock_irq(&tasklist_lock);
1271 	proc_fork_connector(p);
1272 	cgroup_post_fork(p);
1273 	perf_event_fork(p);
1274 	return p;
1275 
1276 bad_fork_free_pid:
1277 	if (pid != &init_struct_pid)
1278 		free_pid(pid);
1279 bad_fork_cleanup_io:
1280 	put_io_context(p->io_context);
1281 bad_fork_cleanup_namespaces:
1282 	exit_task_namespaces(p);
1283 bad_fork_cleanup_mm:
1284 	if (p->mm)
1285 		mmput(p->mm);
1286 bad_fork_cleanup_signal:
1287 	if (!(clone_flags & CLONE_THREAD))
1288 		__cleanup_signal(p->signal);
1289 bad_fork_cleanup_sighand:
1290 	__cleanup_sighand(p->sighand);
1291 bad_fork_cleanup_fs:
1292 	exit_fs(p); /* blocking */
1293 bad_fork_cleanup_files:
1294 	exit_files(p); /* blocking */
1295 bad_fork_cleanup_semundo:
1296 	exit_sem(p);
1297 bad_fork_cleanup_audit:
1298 	audit_free(p);
1299 bad_fork_cleanup_policy:
1300 	perf_event_free_task(p);
1301 #ifdef CONFIG_NUMA
1302 	mpol_put(p->mempolicy);
1303 bad_fork_cleanup_cgroup:
1304 #endif
1305 	cgroup_exit(p, cgroup_callbacks_done);
1306 	delayacct_tsk_free(p);
1307 	if (p->binfmt)
1308 		module_put(p->binfmt->module);
1309 bad_fork_cleanup_put_domain:
1310 	module_put(task_thread_info(p)->exec_domain->module);
1311 bad_fork_cleanup_count:
1312 	atomic_dec(&p->cred->user->processes);
1313 	exit_creds(p);
1314 bad_fork_free:
1315 	free_task(p);
1316 fork_out:
1317 	return ERR_PTR(retval);
1318 }
1319 
1320 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1321 {
1322 	memset(regs, 0, sizeof(struct pt_regs));
1323 	return regs;
1324 }
1325 
1326 struct task_struct * __cpuinit fork_idle(int cpu)
1327 {
1328 	struct task_struct *task;
1329 	struct pt_regs regs;
1330 
1331 	task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1332 			    &init_struct_pid, 0);
1333 	if (!IS_ERR(task))
1334 		init_idle(task, cpu);
1335 
1336 	return task;
1337 }
1338 
1339 /*
1340  *  Ok, this is the main fork-routine.
1341  *
1342  * It copies the process, and if successful kick-starts
1343  * it and waits for it to finish using the VM if required.
1344  */
1345 long do_fork(unsigned long clone_flags,
1346 	      unsigned long stack_start,
1347 	      struct pt_regs *regs,
1348 	      unsigned long stack_size,
1349 	      int __user *parent_tidptr,
1350 	      int __user *child_tidptr)
1351 {
1352 	struct task_struct *p;
1353 	int trace = 0;
1354 	long nr;
1355 
1356 	/*
1357 	 * Do some preliminary argument and permissions checking before we
1358 	 * actually start allocating stuff
1359 	 */
1360 	if (clone_flags & CLONE_NEWUSER) {
1361 		if (clone_flags & CLONE_THREAD)
1362 			return -EINVAL;
1363 		/* hopefully this check will go away when userns support is
1364 		 * complete
1365 		 */
1366 		if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1367 				!capable(CAP_SETGID))
1368 			return -EPERM;
1369 	}
1370 
1371 	/*
1372 	 * We hope to recycle these flags after 2.6.26
1373 	 */
1374 	if (unlikely(clone_flags & CLONE_STOPPED)) {
1375 		static int __read_mostly count = 100;
1376 
1377 		if (count > 0 && printk_ratelimit()) {
1378 			char comm[TASK_COMM_LEN];
1379 
1380 			count--;
1381 			printk(KERN_INFO "fork(): process `%s' used deprecated "
1382 					"clone flags 0x%lx\n",
1383 				get_task_comm(comm, current),
1384 				clone_flags & CLONE_STOPPED);
1385 		}
1386 	}
1387 
1388 	/*
1389 	 * When called from kernel_thread, don't do user tracing stuff.
1390 	 */
1391 	if (likely(user_mode(regs)))
1392 		trace = tracehook_prepare_clone(clone_flags);
1393 
1394 	p = copy_process(clone_flags, stack_start, regs, stack_size,
1395 			 child_tidptr, NULL, trace);
1396 	/*
1397 	 * Do this prior waking up the new thread - the thread pointer
1398 	 * might get invalid after that point, if the thread exits quickly.
1399 	 */
1400 	if (!IS_ERR(p)) {
1401 		struct completion vfork;
1402 
1403 		trace_sched_process_fork(current, p);
1404 
1405 		nr = task_pid_vnr(p);
1406 
1407 		if (clone_flags & CLONE_PARENT_SETTID)
1408 			put_user(nr, parent_tidptr);
1409 
1410 		if (clone_flags & CLONE_VFORK) {
1411 			p->vfork_done = &vfork;
1412 			init_completion(&vfork);
1413 		}
1414 
1415 		audit_finish_fork(p);
1416 		tracehook_report_clone(regs, clone_flags, nr, p);
1417 
1418 		/*
1419 		 * We set PF_STARTING at creation in case tracing wants to
1420 		 * use this to distinguish a fully live task from one that
1421 		 * hasn't gotten to tracehook_report_clone() yet.  Now we
1422 		 * clear it and set the child going.
1423 		 */
1424 		p->flags &= ~PF_STARTING;
1425 
1426 		if (unlikely(clone_flags & CLONE_STOPPED)) {
1427 			/*
1428 			 * We'll start up with an immediate SIGSTOP.
1429 			 */
1430 			sigaddset(&p->pending.signal, SIGSTOP);
1431 			set_tsk_thread_flag(p, TIF_SIGPENDING);
1432 			__set_task_state(p, TASK_STOPPED);
1433 		} else {
1434 			wake_up_new_task(p, clone_flags);
1435 		}
1436 
1437 		tracehook_report_clone_complete(trace, regs,
1438 						clone_flags, nr, p);
1439 
1440 		if (clone_flags & CLONE_VFORK) {
1441 			freezer_do_not_count();
1442 			wait_for_completion(&vfork);
1443 			freezer_count();
1444 			tracehook_report_vfork_done(p, nr);
1445 		}
1446 	} else {
1447 		nr = PTR_ERR(p);
1448 	}
1449 	return nr;
1450 }
1451 
1452 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1453 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1454 #endif
1455 
1456 static void sighand_ctor(void *data)
1457 {
1458 	struct sighand_struct *sighand = data;
1459 
1460 	spin_lock_init(&sighand->siglock);
1461 	init_waitqueue_head(&sighand->signalfd_wqh);
1462 }
1463 
1464 void __init proc_caches_init(void)
1465 {
1466 	sighand_cachep = kmem_cache_create("sighand_cache",
1467 			sizeof(struct sighand_struct), 0,
1468 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1469 			SLAB_NOTRACK, sighand_ctor);
1470 	signal_cachep = kmem_cache_create("signal_cache",
1471 			sizeof(struct signal_struct), 0,
1472 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1473 	files_cachep = kmem_cache_create("files_cache",
1474 			sizeof(struct files_struct), 0,
1475 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1476 	fs_cachep = kmem_cache_create("fs_cache",
1477 			sizeof(struct fs_struct), 0,
1478 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1479 	mm_cachep = kmem_cache_create("mm_struct",
1480 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1481 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1482 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1483 	mmap_init();
1484 }
1485 
1486 /*
1487  * Check constraints on flags passed to the unshare system call and
1488  * force unsharing of additional process context as appropriate.
1489  */
1490 static void check_unshare_flags(unsigned long *flags_ptr)
1491 {
1492 	/*
1493 	 * If unsharing a thread from a thread group, must also
1494 	 * unshare vm.
1495 	 */
1496 	if (*flags_ptr & CLONE_THREAD)
1497 		*flags_ptr |= CLONE_VM;
1498 
1499 	/*
1500 	 * If unsharing vm, must also unshare signal handlers.
1501 	 */
1502 	if (*flags_ptr & CLONE_VM)
1503 		*flags_ptr |= CLONE_SIGHAND;
1504 
1505 	/*
1506 	 * If unsharing signal handlers and the task was created
1507 	 * using CLONE_THREAD, then must unshare the thread
1508 	 */
1509 	if ((*flags_ptr & CLONE_SIGHAND) &&
1510 	    (atomic_read(&current->signal->count) > 1))
1511 		*flags_ptr |= CLONE_THREAD;
1512 
1513 	/*
1514 	 * If unsharing namespace, must also unshare filesystem information.
1515 	 */
1516 	if (*flags_ptr & CLONE_NEWNS)
1517 		*flags_ptr |= CLONE_FS;
1518 }
1519 
1520 /*
1521  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1522  */
1523 static int unshare_thread(unsigned long unshare_flags)
1524 {
1525 	if (unshare_flags & CLONE_THREAD)
1526 		return -EINVAL;
1527 
1528 	return 0;
1529 }
1530 
1531 /*
1532  * Unshare the filesystem structure if it is being shared
1533  */
1534 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1535 {
1536 	struct fs_struct *fs = current->fs;
1537 
1538 	if (!(unshare_flags & CLONE_FS) || !fs)
1539 		return 0;
1540 
1541 	/* don't need lock here; in the worst case we'll do useless copy */
1542 	if (fs->users == 1)
1543 		return 0;
1544 
1545 	*new_fsp = copy_fs_struct(fs);
1546 	if (!*new_fsp)
1547 		return -ENOMEM;
1548 
1549 	return 0;
1550 }
1551 
1552 /*
1553  * Unsharing of sighand is not supported yet
1554  */
1555 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1556 {
1557 	struct sighand_struct *sigh = current->sighand;
1558 
1559 	if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1560 		return -EINVAL;
1561 	else
1562 		return 0;
1563 }
1564 
1565 /*
1566  * Unshare vm if it is being shared
1567  */
1568 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1569 {
1570 	struct mm_struct *mm = current->mm;
1571 
1572 	if ((unshare_flags & CLONE_VM) &&
1573 	    (mm && atomic_read(&mm->mm_users) > 1)) {
1574 		return -EINVAL;
1575 	}
1576 
1577 	return 0;
1578 }
1579 
1580 /*
1581  * Unshare file descriptor table if it is being shared
1582  */
1583 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1584 {
1585 	struct files_struct *fd = current->files;
1586 	int error = 0;
1587 
1588 	if ((unshare_flags & CLONE_FILES) &&
1589 	    (fd && atomic_read(&fd->count) > 1)) {
1590 		*new_fdp = dup_fd(fd, &error);
1591 		if (!*new_fdp)
1592 			return error;
1593 	}
1594 
1595 	return 0;
1596 }
1597 
1598 /*
1599  * unshare allows a process to 'unshare' part of the process
1600  * context which was originally shared using clone.  copy_*
1601  * functions used by do_fork() cannot be used here directly
1602  * because they modify an inactive task_struct that is being
1603  * constructed. Here we are modifying the current, active,
1604  * task_struct.
1605  */
1606 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1607 {
1608 	int err = 0;
1609 	struct fs_struct *fs, *new_fs = NULL;
1610 	struct sighand_struct *new_sigh = NULL;
1611 	struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1612 	struct files_struct *fd, *new_fd = NULL;
1613 	struct nsproxy *new_nsproxy = NULL;
1614 	int do_sysvsem = 0;
1615 
1616 	check_unshare_flags(&unshare_flags);
1617 
1618 	/* Return -EINVAL for all unsupported flags */
1619 	err = -EINVAL;
1620 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1621 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1622 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1623 		goto bad_unshare_out;
1624 
1625 	/*
1626 	 * CLONE_NEWIPC must also detach from the undolist: after switching
1627 	 * to a new ipc namespace, the semaphore arrays from the old
1628 	 * namespace are unreachable.
1629 	 */
1630 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1631 		do_sysvsem = 1;
1632 	if ((err = unshare_thread(unshare_flags)))
1633 		goto bad_unshare_out;
1634 	if ((err = unshare_fs(unshare_flags, &new_fs)))
1635 		goto bad_unshare_cleanup_thread;
1636 	if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1637 		goto bad_unshare_cleanup_fs;
1638 	if ((err = unshare_vm(unshare_flags, &new_mm)))
1639 		goto bad_unshare_cleanup_sigh;
1640 	if ((err = unshare_fd(unshare_flags, &new_fd)))
1641 		goto bad_unshare_cleanup_vm;
1642 	if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1643 			new_fs)))
1644 		goto bad_unshare_cleanup_fd;
1645 
1646 	if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1647 		if (do_sysvsem) {
1648 			/*
1649 			 * CLONE_SYSVSEM is equivalent to sys_exit().
1650 			 */
1651 			exit_sem(current);
1652 		}
1653 
1654 		if (new_nsproxy) {
1655 			switch_task_namespaces(current, new_nsproxy);
1656 			new_nsproxy = NULL;
1657 		}
1658 
1659 		task_lock(current);
1660 
1661 		if (new_fs) {
1662 			fs = current->fs;
1663 			write_lock(&fs->lock);
1664 			current->fs = new_fs;
1665 			if (--fs->users)
1666 				new_fs = NULL;
1667 			else
1668 				new_fs = fs;
1669 			write_unlock(&fs->lock);
1670 		}
1671 
1672 		if (new_mm) {
1673 			mm = current->mm;
1674 			active_mm = current->active_mm;
1675 			current->mm = new_mm;
1676 			current->active_mm = new_mm;
1677 			activate_mm(active_mm, new_mm);
1678 			new_mm = mm;
1679 		}
1680 
1681 		if (new_fd) {
1682 			fd = current->files;
1683 			current->files = new_fd;
1684 			new_fd = fd;
1685 		}
1686 
1687 		task_unlock(current);
1688 	}
1689 
1690 	if (new_nsproxy)
1691 		put_nsproxy(new_nsproxy);
1692 
1693 bad_unshare_cleanup_fd:
1694 	if (new_fd)
1695 		put_files_struct(new_fd);
1696 
1697 bad_unshare_cleanup_vm:
1698 	if (new_mm)
1699 		mmput(new_mm);
1700 
1701 bad_unshare_cleanup_sigh:
1702 	if (new_sigh)
1703 		if (atomic_dec_and_test(&new_sigh->count))
1704 			kmem_cache_free(sighand_cachep, new_sigh);
1705 
1706 bad_unshare_cleanup_fs:
1707 	if (new_fs)
1708 		free_fs_struct(new_fs);
1709 
1710 bad_unshare_cleanup_thread:
1711 bad_unshare_out:
1712 	return err;
1713 }
1714 
1715 /*
1716  *	Helper to unshare the files of the current task.
1717  *	We don't want to expose copy_files internals to
1718  *	the exec layer of the kernel.
1719  */
1720 
1721 int unshare_files(struct files_struct **displaced)
1722 {
1723 	struct task_struct *task = current;
1724 	struct files_struct *copy = NULL;
1725 	int error;
1726 
1727 	error = unshare_fd(CLONE_FILES, &copy);
1728 	if (error || !copy) {
1729 		*displaced = NULL;
1730 		return error;
1731 	}
1732 	*displaced = task->files;
1733 	task_lock(task);
1734 	task->files = copy;
1735 	task_unlock(task);
1736 	return 0;
1737 }
1738