xref: /openbmc/linux/kernel/fork.c (revision 545e4006)
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/mnt_namespace.h>
21 #include <linux/personality.h>
22 #include <linux/mempolicy.h>
23 #include <linux/sem.h>
24 #include <linux/file.h>
25 #include <linux/fdtable.h>
26 #include <linux/iocontext.h>
27 #include <linux/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.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/task_io_accounting_ops.h>
42 #include <linux/rcupdate.h>
43 #include <linux/ptrace.h>
44 #include <linux/mount.h>
45 #include <linux/audit.h>
46 #include <linux/memcontrol.h>
47 #include <linux/profile.h>
48 #include <linux/rmap.h>
49 #include <linux/acct.h>
50 #include <linux/tsacct_kern.h>
51 #include <linux/cn_proc.h>
52 #include <linux/freezer.h>
53 #include <linux/delayacct.h>
54 #include <linux/taskstats_kern.h>
55 #include <linux/random.h>
56 #include <linux/tty.h>
57 #include <linux/proc_fs.h>
58 #include <linux/blkdev.h>
59 
60 #include <asm/pgtable.h>
61 #include <asm/pgalloc.h>
62 #include <asm/uaccess.h>
63 #include <asm/mmu_context.h>
64 #include <asm/cacheflush.h>
65 #include <asm/tlbflush.h>
66 
67 /*
68  * Protected counters by write_lock_irq(&tasklist_lock)
69  */
70 unsigned long total_forks;	/* Handle normal Linux uptimes. */
71 int nr_threads; 		/* The idle threads do not count.. */
72 
73 int max_threads;		/* tunable limit on nr_threads */
74 
75 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
76 
77 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
78 
79 int nr_processes(void)
80 {
81 	int cpu;
82 	int total = 0;
83 
84 	for_each_online_cpu(cpu)
85 		total += per_cpu(process_counts, cpu);
86 
87 	return total;
88 }
89 
90 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
91 # define alloc_task_struct()	kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
92 # define free_task_struct(tsk)	kmem_cache_free(task_struct_cachep, (tsk))
93 static struct kmem_cache *task_struct_cachep;
94 #endif
95 
96 /* SLAB cache for signal_struct structures (tsk->signal) */
97 static struct kmem_cache *signal_cachep;
98 
99 /* SLAB cache for sighand_struct structures (tsk->sighand) */
100 struct kmem_cache *sighand_cachep;
101 
102 /* SLAB cache for files_struct structures (tsk->files) */
103 struct kmem_cache *files_cachep;
104 
105 /* SLAB cache for fs_struct structures (tsk->fs) */
106 struct kmem_cache *fs_cachep;
107 
108 /* SLAB cache for vm_area_struct structures */
109 struct kmem_cache *vm_area_cachep;
110 
111 /* SLAB cache for mm_struct structures (tsk->mm) */
112 static struct kmem_cache *mm_cachep;
113 
114 void free_task(struct task_struct *tsk)
115 {
116 	prop_local_destroy_single(&tsk->dirties);
117 	free_thread_info(tsk->stack);
118 	rt_mutex_debug_task_free(tsk);
119 	free_task_struct(tsk);
120 }
121 EXPORT_SYMBOL(free_task);
122 
123 void __put_task_struct(struct task_struct *tsk)
124 {
125 	WARN_ON(!tsk->exit_state);
126 	WARN_ON(atomic_read(&tsk->usage));
127 	WARN_ON(tsk == current);
128 
129 	security_task_free(tsk);
130 	free_uid(tsk->user);
131 	put_group_info(tsk->group_info);
132 	delayacct_tsk_free(tsk);
133 
134 	if (!profile_handoff_task(tsk))
135 		free_task(tsk);
136 }
137 
138 /*
139  * macro override instead of weak attribute alias, to workaround
140  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
141  */
142 #ifndef arch_task_cache_init
143 #define arch_task_cache_init()
144 #endif
145 
146 void __init fork_init(unsigned long mempages)
147 {
148 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
149 #ifndef ARCH_MIN_TASKALIGN
150 #define ARCH_MIN_TASKALIGN	L1_CACHE_BYTES
151 #endif
152 	/* create a slab on which task_structs can be allocated */
153 	task_struct_cachep =
154 		kmem_cache_create("task_struct", sizeof(struct task_struct),
155 			ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
156 #endif
157 
158 	/* do the arch specific task caches init */
159 	arch_task_cache_init();
160 
161 	/*
162 	 * The default maximum number of threads is set to a safe
163 	 * value: the thread structures can take up at most half
164 	 * of memory.
165 	 */
166 	max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
167 
168 	/*
169 	 * we need to allow at least 20 threads to boot a system
170 	 */
171 	if(max_threads < 20)
172 		max_threads = 20;
173 
174 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
175 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
176 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
177 		init_task.signal->rlim[RLIMIT_NPROC];
178 }
179 
180 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
181 					       struct task_struct *src)
182 {
183 	*dst = *src;
184 	return 0;
185 }
186 
187 static struct task_struct *dup_task_struct(struct task_struct *orig)
188 {
189 	struct task_struct *tsk;
190 	struct thread_info *ti;
191 	int err;
192 
193 	prepare_to_copy(orig);
194 
195 	tsk = alloc_task_struct();
196 	if (!tsk)
197 		return NULL;
198 
199 	ti = alloc_thread_info(tsk);
200 	if (!ti) {
201 		free_task_struct(tsk);
202 		return NULL;
203 	}
204 
205  	err = arch_dup_task_struct(tsk, orig);
206 	if (err)
207 		goto out;
208 
209 	tsk->stack = ti;
210 
211 	err = prop_local_init_single(&tsk->dirties);
212 	if (err)
213 		goto out;
214 
215 	setup_thread_stack(tsk, orig);
216 
217 #ifdef CONFIG_CC_STACKPROTECTOR
218 	tsk->stack_canary = get_random_int();
219 #endif
220 
221 	/* One for us, one for whoever does the "release_task()" (usually parent) */
222 	atomic_set(&tsk->usage,2);
223 	atomic_set(&tsk->fs_excl, 0);
224 #ifdef CONFIG_BLK_DEV_IO_TRACE
225 	tsk->btrace_seq = 0;
226 #endif
227 	tsk->splice_pipe = NULL;
228 	return tsk;
229 
230 out:
231 	free_thread_info(ti);
232 	free_task_struct(tsk);
233 	return NULL;
234 }
235 
236 #ifdef CONFIG_MMU
237 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
238 {
239 	struct vm_area_struct *mpnt, *tmp, **pprev;
240 	struct rb_node **rb_link, *rb_parent;
241 	int retval;
242 	unsigned long charge;
243 	struct mempolicy *pol;
244 
245 	down_write(&oldmm->mmap_sem);
246 	flush_cache_dup_mm(oldmm);
247 	/*
248 	 * Not linked in yet - no deadlock potential:
249 	 */
250 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
251 
252 	mm->locked_vm = 0;
253 	mm->mmap = NULL;
254 	mm->mmap_cache = NULL;
255 	mm->free_area_cache = oldmm->mmap_base;
256 	mm->cached_hole_size = ~0UL;
257 	mm->map_count = 0;
258 	cpus_clear(mm->cpu_vm_mask);
259 	mm->mm_rb = RB_ROOT;
260 	rb_link = &mm->mm_rb.rb_node;
261 	rb_parent = NULL;
262 	pprev = &mm->mmap;
263 
264 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
265 		struct file *file;
266 
267 		if (mpnt->vm_flags & VM_DONTCOPY) {
268 			long pages = vma_pages(mpnt);
269 			mm->total_vm -= pages;
270 			vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
271 								-pages);
272 			continue;
273 		}
274 		charge = 0;
275 		if (mpnt->vm_flags & VM_ACCOUNT) {
276 			unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
277 			if (security_vm_enough_memory(len))
278 				goto fail_nomem;
279 			charge = len;
280 		}
281 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
282 		if (!tmp)
283 			goto fail_nomem;
284 		*tmp = *mpnt;
285 		pol = mpol_dup(vma_policy(mpnt));
286 		retval = PTR_ERR(pol);
287 		if (IS_ERR(pol))
288 			goto fail_nomem_policy;
289 		vma_set_policy(tmp, pol);
290 		tmp->vm_flags &= ~VM_LOCKED;
291 		tmp->vm_mm = mm;
292 		tmp->vm_next = NULL;
293 		anon_vma_link(tmp);
294 		file = tmp->vm_file;
295 		if (file) {
296 			struct inode *inode = file->f_path.dentry->d_inode;
297 			get_file(file);
298 			if (tmp->vm_flags & VM_DENYWRITE)
299 				atomic_dec(&inode->i_writecount);
300 
301 			/* insert tmp into the share list, just after mpnt */
302 			spin_lock(&file->f_mapping->i_mmap_lock);
303 			tmp->vm_truncate_count = mpnt->vm_truncate_count;
304 			flush_dcache_mmap_lock(file->f_mapping);
305 			vma_prio_tree_add(tmp, mpnt);
306 			flush_dcache_mmap_unlock(file->f_mapping);
307 			spin_unlock(&file->f_mapping->i_mmap_lock);
308 		}
309 
310 		/*
311 		 * Clear hugetlb-related page reserves for children. This only
312 		 * affects MAP_PRIVATE mappings. Faults generated by the child
313 		 * are not guaranteed to succeed, even if read-only
314 		 */
315 		if (is_vm_hugetlb_page(tmp))
316 			reset_vma_resv_huge_pages(tmp);
317 
318 		/*
319 		 * Link in the new vma and copy the page table entries.
320 		 */
321 		*pprev = tmp;
322 		pprev = &tmp->vm_next;
323 
324 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
325 		rb_link = &tmp->vm_rb.rb_right;
326 		rb_parent = &tmp->vm_rb;
327 
328 		mm->map_count++;
329 		retval = copy_page_range(mm, oldmm, mpnt);
330 
331 		if (tmp->vm_ops && tmp->vm_ops->open)
332 			tmp->vm_ops->open(tmp);
333 
334 		if (retval)
335 			goto out;
336 	}
337 	/* a new mm has just been created */
338 	arch_dup_mmap(oldmm, mm);
339 	retval = 0;
340 out:
341 	up_write(&mm->mmap_sem);
342 	flush_tlb_mm(oldmm);
343 	up_write(&oldmm->mmap_sem);
344 	return retval;
345 fail_nomem_policy:
346 	kmem_cache_free(vm_area_cachep, tmp);
347 fail_nomem:
348 	retval = -ENOMEM;
349 	vm_unacct_memory(charge);
350 	goto out;
351 }
352 
353 static inline int mm_alloc_pgd(struct mm_struct * mm)
354 {
355 	mm->pgd = pgd_alloc(mm);
356 	if (unlikely(!mm->pgd))
357 		return -ENOMEM;
358 	return 0;
359 }
360 
361 static inline void mm_free_pgd(struct mm_struct * mm)
362 {
363 	pgd_free(mm, mm->pgd);
364 }
365 #else
366 #define dup_mmap(mm, oldmm)	(0)
367 #define mm_alloc_pgd(mm)	(0)
368 #define mm_free_pgd(mm)
369 #endif /* CONFIG_MMU */
370 
371 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
372 
373 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
374 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
375 
376 #include <linux/init_task.h>
377 
378 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
379 {
380 	atomic_set(&mm->mm_users, 1);
381 	atomic_set(&mm->mm_count, 1);
382 	init_rwsem(&mm->mmap_sem);
383 	INIT_LIST_HEAD(&mm->mmlist);
384 	mm->flags = (current->mm) ? current->mm->flags
385 				  : MMF_DUMP_FILTER_DEFAULT;
386 	mm->core_waiters = 0;
387 	mm->nr_ptes = 0;
388 	set_mm_counter(mm, file_rss, 0);
389 	set_mm_counter(mm, anon_rss, 0);
390 	spin_lock_init(&mm->page_table_lock);
391 	rwlock_init(&mm->ioctx_list_lock);
392 	mm->ioctx_list = NULL;
393 	mm->free_area_cache = TASK_UNMAPPED_BASE;
394 	mm->cached_hole_size = ~0UL;
395 	mm_init_owner(mm, p);
396 
397 	if (likely(!mm_alloc_pgd(mm))) {
398 		mm->def_flags = 0;
399 		return mm;
400 	}
401 
402 	free_mm(mm);
403 	return NULL;
404 }
405 
406 /*
407  * Allocate and initialize an mm_struct.
408  */
409 struct mm_struct * mm_alloc(void)
410 {
411 	struct mm_struct * mm;
412 
413 	mm = allocate_mm();
414 	if (mm) {
415 		memset(mm, 0, sizeof(*mm));
416 		mm = mm_init(mm, current);
417 	}
418 	return mm;
419 }
420 
421 /*
422  * Called when the last reference to the mm
423  * is dropped: either by a lazy thread or by
424  * mmput. Free the page directory and the mm.
425  */
426 void __mmdrop(struct mm_struct *mm)
427 {
428 	BUG_ON(mm == &init_mm);
429 	mm_free_pgd(mm);
430 	destroy_context(mm);
431 	free_mm(mm);
432 }
433 EXPORT_SYMBOL_GPL(__mmdrop);
434 
435 /*
436  * Decrement the use count and release all resources for an mm.
437  */
438 void mmput(struct mm_struct *mm)
439 {
440 	might_sleep();
441 
442 	if (atomic_dec_and_test(&mm->mm_users)) {
443 		exit_aio(mm);
444 		exit_mmap(mm);
445 		set_mm_exe_file(mm, NULL);
446 		if (!list_empty(&mm->mmlist)) {
447 			spin_lock(&mmlist_lock);
448 			list_del(&mm->mmlist);
449 			spin_unlock(&mmlist_lock);
450 		}
451 		put_swap_token(mm);
452 		mmdrop(mm);
453 	}
454 }
455 EXPORT_SYMBOL_GPL(mmput);
456 
457 /**
458  * get_task_mm - acquire a reference to the task's mm
459  *
460  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
461  * this kernel workthread has transiently adopted a user mm with use_mm,
462  * to do its AIO) is not set and if so returns a reference to it, after
463  * bumping up the use count.  User must release the mm via mmput()
464  * after use.  Typically used by /proc and ptrace.
465  */
466 struct mm_struct *get_task_mm(struct task_struct *task)
467 {
468 	struct mm_struct *mm;
469 
470 	task_lock(task);
471 	mm = task->mm;
472 	if (mm) {
473 		if (task->flags & PF_BORROWED_MM)
474 			mm = NULL;
475 		else
476 			atomic_inc(&mm->mm_users);
477 	}
478 	task_unlock(task);
479 	return mm;
480 }
481 EXPORT_SYMBOL_GPL(get_task_mm);
482 
483 /* Please note the differences between mmput and mm_release.
484  * mmput is called whenever we stop holding onto a mm_struct,
485  * error success whatever.
486  *
487  * mm_release is called after a mm_struct has been removed
488  * from the current process.
489  *
490  * This difference is important for error handling, when we
491  * only half set up a mm_struct for a new process and need to restore
492  * the old one.  Because we mmput the new mm_struct before
493  * restoring the old one. . .
494  * Eric Biederman 10 January 1998
495  */
496 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
497 {
498 	struct completion *vfork_done = tsk->vfork_done;
499 
500 	/* Get rid of any cached register state */
501 	deactivate_mm(tsk, mm);
502 
503 	/* notify parent sleeping on vfork() */
504 	if (vfork_done) {
505 		tsk->vfork_done = NULL;
506 		complete(vfork_done);
507 	}
508 
509 	/*
510 	 * If we're exiting normally, clear a user-space tid field if
511 	 * requested.  We leave this alone when dying by signal, to leave
512 	 * the value intact in a core dump, and to save the unnecessary
513 	 * trouble otherwise.  Userland only wants this done for a sys_exit.
514 	 */
515 	if (tsk->clear_child_tid
516 	    && !(tsk->flags & PF_SIGNALED)
517 	    && atomic_read(&mm->mm_users) > 1) {
518 		u32 __user * tidptr = tsk->clear_child_tid;
519 		tsk->clear_child_tid = NULL;
520 
521 		/*
522 		 * We don't check the error code - if userspace has
523 		 * not set up a proper pointer then tough luck.
524 		 */
525 		put_user(0, tidptr);
526 		sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
527 	}
528 }
529 
530 /*
531  * Allocate a new mm structure and copy contents from the
532  * mm structure of the passed in task structure.
533  */
534 struct mm_struct *dup_mm(struct task_struct *tsk)
535 {
536 	struct mm_struct *mm, *oldmm = current->mm;
537 	int err;
538 
539 	if (!oldmm)
540 		return NULL;
541 
542 	mm = allocate_mm();
543 	if (!mm)
544 		goto fail_nomem;
545 
546 	memcpy(mm, oldmm, sizeof(*mm));
547 
548 	/* Initializing for Swap token stuff */
549 	mm->token_priority = 0;
550 	mm->last_interval = 0;
551 
552 	if (!mm_init(mm, tsk))
553 		goto fail_nomem;
554 
555 	if (init_new_context(tsk, mm))
556 		goto fail_nocontext;
557 
558 	dup_mm_exe_file(oldmm, mm);
559 
560 	err = dup_mmap(mm, oldmm);
561 	if (err)
562 		goto free_pt;
563 
564 	mm->hiwater_rss = get_mm_rss(mm);
565 	mm->hiwater_vm = mm->total_vm;
566 
567 	return mm;
568 
569 free_pt:
570 	mmput(mm);
571 
572 fail_nomem:
573 	return NULL;
574 
575 fail_nocontext:
576 	/*
577 	 * If init_new_context() failed, we cannot use mmput() to free the mm
578 	 * because it calls destroy_context()
579 	 */
580 	mm_free_pgd(mm);
581 	free_mm(mm);
582 	return NULL;
583 }
584 
585 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
586 {
587 	struct mm_struct * mm, *oldmm;
588 	int retval;
589 
590 	tsk->min_flt = tsk->maj_flt = 0;
591 	tsk->nvcsw = tsk->nivcsw = 0;
592 
593 	tsk->mm = NULL;
594 	tsk->active_mm = NULL;
595 
596 	/*
597 	 * Are we cloning a kernel thread?
598 	 *
599 	 * We need to steal a active VM for that..
600 	 */
601 	oldmm = current->mm;
602 	if (!oldmm)
603 		return 0;
604 
605 	if (clone_flags & CLONE_VM) {
606 		atomic_inc(&oldmm->mm_users);
607 		mm = oldmm;
608 		goto good_mm;
609 	}
610 
611 	retval = -ENOMEM;
612 	mm = dup_mm(tsk);
613 	if (!mm)
614 		goto fail_nomem;
615 
616 good_mm:
617 	/* Initializing for Swap token stuff */
618 	mm->token_priority = 0;
619 	mm->last_interval = 0;
620 
621 	tsk->mm = mm;
622 	tsk->active_mm = mm;
623 	return 0;
624 
625 fail_nomem:
626 	return retval;
627 }
628 
629 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
630 {
631 	struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
632 	/* We don't need to lock fs - think why ;-) */
633 	if (fs) {
634 		atomic_set(&fs->count, 1);
635 		rwlock_init(&fs->lock);
636 		fs->umask = old->umask;
637 		read_lock(&old->lock);
638 		fs->root = old->root;
639 		path_get(&old->root);
640 		fs->pwd = old->pwd;
641 		path_get(&old->pwd);
642 		if (old->altroot.dentry) {
643 			fs->altroot = old->altroot;
644 			path_get(&old->altroot);
645 		} else {
646 			fs->altroot.mnt = NULL;
647 			fs->altroot.dentry = NULL;
648 		}
649 		read_unlock(&old->lock);
650 	}
651 	return fs;
652 }
653 
654 struct fs_struct *copy_fs_struct(struct fs_struct *old)
655 {
656 	return __copy_fs_struct(old);
657 }
658 
659 EXPORT_SYMBOL_GPL(copy_fs_struct);
660 
661 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
662 {
663 	if (clone_flags & CLONE_FS) {
664 		atomic_inc(&current->fs->count);
665 		return 0;
666 	}
667 	tsk->fs = __copy_fs_struct(current->fs);
668 	if (!tsk->fs)
669 		return -ENOMEM;
670 	return 0;
671 }
672 
673 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
674 {
675 	struct files_struct *oldf, *newf;
676 	int error = 0;
677 
678 	/*
679 	 * A background process may not have any files ...
680 	 */
681 	oldf = current->files;
682 	if (!oldf)
683 		goto out;
684 
685 	if (clone_flags & CLONE_FILES) {
686 		atomic_inc(&oldf->count);
687 		goto out;
688 	}
689 
690 	newf = dup_fd(oldf, &error);
691 	if (!newf)
692 		goto out;
693 
694 	tsk->files = newf;
695 	error = 0;
696 out:
697 	return error;
698 }
699 
700 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
701 {
702 #ifdef CONFIG_BLOCK
703 	struct io_context *ioc = current->io_context;
704 
705 	if (!ioc)
706 		return 0;
707 	/*
708 	 * Share io context with parent, if CLONE_IO is set
709 	 */
710 	if (clone_flags & CLONE_IO) {
711 		tsk->io_context = ioc_task_link(ioc);
712 		if (unlikely(!tsk->io_context))
713 			return -ENOMEM;
714 	} else if (ioprio_valid(ioc->ioprio)) {
715 		tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
716 		if (unlikely(!tsk->io_context))
717 			return -ENOMEM;
718 
719 		tsk->io_context->ioprio = ioc->ioprio;
720 	}
721 #endif
722 	return 0;
723 }
724 
725 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
726 {
727 	struct sighand_struct *sig;
728 
729 	if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
730 		atomic_inc(&current->sighand->count);
731 		return 0;
732 	}
733 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
734 	rcu_assign_pointer(tsk->sighand, sig);
735 	if (!sig)
736 		return -ENOMEM;
737 	atomic_set(&sig->count, 1);
738 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
739 	return 0;
740 }
741 
742 void __cleanup_sighand(struct sighand_struct *sighand)
743 {
744 	if (atomic_dec_and_test(&sighand->count))
745 		kmem_cache_free(sighand_cachep, sighand);
746 }
747 
748 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
749 {
750 	struct signal_struct *sig;
751 	int ret;
752 
753 	if (clone_flags & CLONE_THREAD) {
754 		atomic_inc(&current->signal->count);
755 		atomic_inc(&current->signal->live);
756 		return 0;
757 	}
758 	sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
759 	tsk->signal = sig;
760 	if (!sig)
761 		return -ENOMEM;
762 
763 	ret = copy_thread_group_keys(tsk);
764 	if (ret < 0) {
765 		kmem_cache_free(signal_cachep, sig);
766 		return ret;
767 	}
768 
769 	atomic_set(&sig->count, 1);
770 	atomic_set(&sig->live, 1);
771 	init_waitqueue_head(&sig->wait_chldexit);
772 	sig->flags = 0;
773 	sig->group_exit_code = 0;
774 	sig->group_exit_task = NULL;
775 	sig->group_stop_count = 0;
776 	sig->curr_target = tsk;
777 	init_sigpending(&sig->shared_pending);
778 	INIT_LIST_HEAD(&sig->posix_timers);
779 
780 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
781 	sig->it_real_incr.tv64 = 0;
782 	sig->real_timer.function = it_real_fn;
783 
784 	sig->it_virt_expires = cputime_zero;
785 	sig->it_virt_incr = cputime_zero;
786 	sig->it_prof_expires = cputime_zero;
787 	sig->it_prof_incr = cputime_zero;
788 
789 	sig->leader = 0;	/* session leadership doesn't inherit */
790 	sig->tty_old_pgrp = NULL;
791 
792 	sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
793 	sig->gtime = cputime_zero;
794 	sig->cgtime = cputime_zero;
795 	sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
796 	sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
797 	sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
798 	sig->sum_sched_runtime = 0;
799 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
800 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
801 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
802 	taskstats_tgid_init(sig);
803 
804 	task_lock(current->group_leader);
805 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
806 	task_unlock(current->group_leader);
807 
808 	if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
809 		/*
810 		 * New sole thread in the process gets an expiry time
811 		 * of the whole CPU time limit.
812 		 */
813 		tsk->it_prof_expires =
814 			secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
815 	}
816 	acct_init_pacct(&sig->pacct);
817 
818 	tty_audit_fork(sig);
819 
820 	return 0;
821 }
822 
823 void __cleanup_signal(struct signal_struct *sig)
824 {
825 	exit_thread_group_keys(sig);
826 	kmem_cache_free(signal_cachep, sig);
827 }
828 
829 static void cleanup_signal(struct task_struct *tsk)
830 {
831 	struct signal_struct *sig = tsk->signal;
832 
833 	atomic_dec(&sig->live);
834 
835 	if (atomic_dec_and_test(&sig->count))
836 		__cleanup_signal(sig);
837 }
838 
839 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
840 {
841 	unsigned long new_flags = p->flags;
842 
843 	new_flags &= ~PF_SUPERPRIV;
844 	new_flags |= PF_FORKNOEXEC;
845 	if (!(clone_flags & CLONE_PTRACE))
846 		p->ptrace = 0;
847 	p->flags = new_flags;
848 	clear_freeze_flag(p);
849 }
850 
851 asmlinkage long sys_set_tid_address(int __user *tidptr)
852 {
853 	current->clear_child_tid = tidptr;
854 
855 	return task_pid_vnr(current);
856 }
857 
858 static void rt_mutex_init_task(struct task_struct *p)
859 {
860 	spin_lock_init(&p->pi_lock);
861 #ifdef CONFIG_RT_MUTEXES
862 	plist_head_init(&p->pi_waiters, &p->pi_lock);
863 	p->pi_blocked_on = NULL;
864 #endif
865 }
866 
867 #ifdef CONFIG_MM_OWNER
868 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
869 {
870 	mm->owner = p;
871 }
872 #endif /* CONFIG_MM_OWNER */
873 
874 /*
875  * This creates a new process as a copy of the old one,
876  * but does not actually start it yet.
877  *
878  * It copies the registers, and all the appropriate
879  * parts of the process environment (as per the clone
880  * flags). The actual kick-off is left to the caller.
881  */
882 static struct task_struct *copy_process(unsigned long clone_flags,
883 					unsigned long stack_start,
884 					struct pt_regs *regs,
885 					unsigned long stack_size,
886 					int __user *child_tidptr,
887 					struct pid *pid)
888 {
889 	int retval;
890 	struct task_struct *p;
891 	int cgroup_callbacks_done = 0;
892 
893 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
894 		return ERR_PTR(-EINVAL);
895 
896 	/*
897 	 * Thread groups must share signals as well, and detached threads
898 	 * can only be started up within the thread group.
899 	 */
900 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
901 		return ERR_PTR(-EINVAL);
902 
903 	/*
904 	 * Shared signal handlers imply shared VM. By way of the above,
905 	 * thread groups also imply shared VM. Blocking this case allows
906 	 * for various simplifications in other code.
907 	 */
908 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
909 		return ERR_PTR(-EINVAL);
910 
911 	retval = security_task_create(clone_flags);
912 	if (retval)
913 		goto fork_out;
914 
915 	retval = -ENOMEM;
916 	p = dup_task_struct(current);
917 	if (!p)
918 		goto fork_out;
919 
920 	rt_mutex_init_task(p);
921 
922 #ifdef CONFIG_PROVE_LOCKING
923 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
924 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
925 #endif
926 	retval = -EAGAIN;
927 	if (atomic_read(&p->user->processes) >=
928 			p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
929 		if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
930 		    p->user != current->nsproxy->user_ns->root_user)
931 			goto bad_fork_free;
932 	}
933 
934 	atomic_inc(&p->user->__count);
935 	atomic_inc(&p->user->processes);
936 	get_group_info(p->group_info);
937 
938 	/*
939 	 * If multiple threads are within copy_process(), then this check
940 	 * triggers too late. This doesn't hurt, the check is only there
941 	 * to stop root fork bombs.
942 	 */
943 	if (nr_threads >= max_threads)
944 		goto bad_fork_cleanup_count;
945 
946 	if (!try_module_get(task_thread_info(p)->exec_domain->module))
947 		goto bad_fork_cleanup_count;
948 
949 	if (p->binfmt && !try_module_get(p->binfmt->module))
950 		goto bad_fork_cleanup_put_domain;
951 
952 	p->did_exec = 0;
953 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
954 	copy_flags(clone_flags, p);
955 	INIT_LIST_HEAD(&p->children);
956 	INIT_LIST_HEAD(&p->sibling);
957 #ifdef CONFIG_PREEMPT_RCU
958 	p->rcu_read_lock_nesting = 0;
959 	p->rcu_flipctr_idx = 0;
960 #endif /* #ifdef CONFIG_PREEMPT_RCU */
961 	p->vfork_done = NULL;
962 	spin_lock_init(&p->alloc_lock);
963 
964 	clear_tsk_thread_flag(p, TIF_SIGPENDING);
965 	init_sigpending(&p->pending);
966 
967 	p->utime = cputime_zero;
968 	p->stime = cputime_zero;
969 	p->gtime = cputime_zero;
970 	p->utimescaled = cputime_zero;
971 	p->stimescaled = cputime_zero;
972 	p->prev_utime = cputime_zero;
973 	p->prev_stime = cputime_zero;
974 
975 #ifdef CONFIG_DETECT_SOFTLOCKUP
976 	p->last_switch_count = 0;
977 	p->last_switch_timestamp = 0;
978 #endif
979 
980 #ifdef CONFIG_TASK_XACCT
981 	p->rchar = 0;		/* I/O counter: bytes read */
982 	p->wchar = 0;		/* I/O counter: bytes written */
983 	p->syscr = 0;		/* I/O counter: read syscalls */
984 	p->syscw = 0;		/* I/O counter: write syscalls */
985 #endif
986 	task_io_accounting_init(p);
987 	acct_clear_integrals(p);
988 
989 	p->it_virt_expires = cputime_zero;
990 	p->it_prof_expires = cputime_zero;
991 	p->it_sched_expires = 0;
992 	INIT_LIST_HEAD(&p->cpu_timers[0]);
993 	INIT_LIST_HEAD(&p->cpu_timers[1]);
994 	INIT_LIST_HEAD(&p->cpu_timers[2]);
995 
996 	p->lock_depth = -1;		/* -1 = no lock */
997 	do_posix_clock_monotonic_gettime(&p->start_time);
998 	p->real_start_time = p->start_time;
999 	monotonic_to_bootbased(&p->real_start_time);
1000 #ifdef CONFIG_SECURITY
1001 	p->security = NULL;
1002 #endif
1003 	p->cap_bset = current->cap_bset;
1004 	p->io_context = NULL;
1005 	p->audit_context = NULL;
1006 	cgroup_fork(p);
1007 #ifdef CONFIG_NUMA
1008 	p->mempolicy = mpol_dup(p->mempolicy);
1009  	if (IS_ERR(p->mempolicy)) {
1010  		retval = PTR_ERR(p->mempolicy);
1011  		p->mempolicy = NULL;
1012  		goto bad_fork_cleanup_cgroup;
1013  	}
1014 	mpol_fix_fork_child_flag(p);
1015 #endif
1016 #ifdef CONFIG_TRACE_IRQFLAGS
1017 	p->irq_events = 0;
1018 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1019 	p->hardirqs_enabled = 1;
1020 #else
1021 	p->hardirqs_enabled = 0;
1022 #endif
1023 	p->hardirq_enable_ip = 0;
1024 	p->hardirq_enable_event = 0;
1025 	p->hardirq_disable_ip = _THIS_IP_;
1026 	p->hardirq_disable_event = 0;
1027 	p->softirqs_enabled = 1;
1028 	p->softirq_enable_ip = _THIS_IP_;
1029 	p->softirq_enable_event = 0;
1030 	p->softirq_disable_ip = 0;
1031 	p->softirq_disable_event = 0;
1032 	p->hardirq_context = 0;
1033 	p->softirq_context = 0;
1034 #endif
1035 #ifdef CONFIG_LOCKDEP
1036 	p->lockdep_depth = 0; /* no locks held yet */
1037 	p->curr_chain_key = 0;
1038 	p->lockdep_recursion = 0;
1039 #endif
1040 
1041 #ifdef CONFIG_DEBUG_MUTEXES
1042 	p->blocked_on = NULL; /* not blocked yet */
1043 #endif
1044 
1045 	/* Perform scheduler related setup. Assign this task to a CPU. */
1046 	sched_fork(p, clone_flags);
1047 
1048 	if ((retval = security_task_alloc(p)))
1049 		goto bad_fork_cleanup_policy;
1050 	if ((retval = audit_alloc(p)))
1051 		goto bad_fork_cleanup_security;
1052 	/* copy all the process information */
1053 	if ((retval = copy_semundo(clone_flags, p)))
1054 		goto bad_fork_cleanup_audit;
1055 	if ((retval = copy_files(clone_flags, p)))
1056 		goto bad_fork_cleanup_semundo;
1057 	if ((retval = copy_fs(clone_flags, p)))
1058 		goto bad_fork_cleanup_files;
1059 	if ((retval = copy_sighand(clone_flags, p)))
1060 		goto bad_fork_cleanup_fs;
1061 	if ((retval = copy_signal(clone_flags, p)))
1062 		goto bad_fork_cleanup_sighand;
1063 	if ((retval = copy_mm(clone_flags, p)))
1064 		goto bad_fork_cleanup_signal;
1065 	if ((retval = copy_keys(clone_flags, p)))
1066 		goto bad_fork_cleanup_mm;
1067 	if ((retval = copy_namespaces(clone_flags, p)))
1068 		goto bad_fork_cleanup_keys;
1069 	if ((retval = copy_io(clone_flags, p)))
1070 		goto bad_fork_cleanup_namespaces;
1071 	retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1072 	if (retval)
1073 		goto bad_fork_cleanup_io;
1074 
1075 	if (pid != &init_struct_pid) {
1076 		retval = -ENOMEM;
1077 		pid = alloc_pid(task_active_pid_ns(p));
1078 		if (!pid)
1079 			goto bad_fork_cleanup_io;
1080 
1081 		if (clone_flags & CLONE_NEWPID) {
1082 			retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1083 			if (retval < 0)
1084 				goto bad_fork_free_pid;
1085 		}
1086 	}
1087 
1088 	p->pid = pid_nr(pid);
1089 	p->tgid = p->pid;
1090 	if (clone_flags & CLONE_THREAD)
1091 		p->tgid = current->tgid;
1092 
1093 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1094 	/*
1095 	 * Clear TID on mm_release()?
1096 	 */
1097 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1098 #ifdef CONFIG_FUTEX
1099 	p->robust_list = NULL;
1100 #ifdef CONFIG_COMPAT
1101 	p->compat_robust_list = NULL;
1102 #endif
1103 	INIT_LIST_HEAD(&p->pi_state_list);
1104 	p->pi_state_cache = NULL;
1105 #endif
1106 	/*
1107 	 * sigaltstack should be cleared when sharing the same VM
1108 	 */
1109 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1110 		p->sas_ss_sp = p->sas_ss_size = 0;
1111 
1112 	/*
1113 	 * Syscall tracing should be turned off in the child regardless
1114 	 * of CLONE_PTRACE.
1115 	 */
1116 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1117 #ifdef TIF_SYSCALL_EMU
1118 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1119 #endif
1120 	clear_all_latency_tracing(p);
1121 
1122 	/* Our parent execution domain becomes current domain
1123 	   These must match for thread signalling to apply */
1124 	p->parent_exec_id = p->self_exec_id;
1125 
1126 	/* ok, now we should be set up.. */
1127 	p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1128 	p->pdeath_signal = 0;
1129 	p->exit_state = 0;
1130 
1131 	/*
1132 	 * Ok, make it visible to the rest of the system.
1133 	 * We dont wake it up yet.
1134 	 */
1135 	p->group_leader = p;
1136 	INIT_LIST_HEAD(&p->thread_group);
1137 	INIT_LIST_HEAD(&p->ptrace_entry);
1138 	INIT_LIST_HEAD(&p->ptraced);
1139 
1140 	/* Now that the task is set up, run cgroup callbacks if
1141 	 * necessary. We need to run them before the task is visible
1142 	 * on the tasklist. */
1143 	cgroup_fork_callbacks(p);
1144 	cgroup_callbacks_done = 1;
1145 
1146 	/* Need tasklist lock for parent etc handling! */
1147 	write_lock_irq(&tasklist_lock);
1148 
1149 	/*
1150 	 * The task hasn't been attached yet, so its cpus_allowed mask will
1151 	 * not be changed, nor will its assigned CPU.
1152 	 *
1153 	 * The cpus_allowed mask of the parent may have changed after it was
1154 	 * copied first time - so re-copy it here, then check the child's CPU
1155 	 * to ensure it is on a valid CPU (and if not, just force it back to
1156 	 * parent's CPU). This avoids alot of nasty races.
1157 	 */
1158 	p->cpus_allowed = current->cpus_allowed;
1159 	p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1160 	if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1161 			!cpu_online(task_cpu(p))))
1162 		set_task_cpu(p, smp_processor_id());
1163 
1164 	/* CLONE_PARENT re-uses the old parent */
1165 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1166 		p->real_parent = current->real_parent;
1167 	else
1168 		p->real_parent = current;
1169 	p->parent = p->real_parent;
1170 
1171 	spin_lock(&current->sighand->siglock);
1172 
1173 	/*
1174 	 * Process group and session signals need to be delivered to just the
1175 	 * parent before the fork or both the parent and the child after the
1176 	 * fork. Restart if a signal comes in before we add the new process to
1177 	 * it's process group.
1178 	 * A fatal signal pending means that current will exit, so the new
1179 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1180  	 */
1181 	recalc_sigpending();
1182 	if (signal_pending(current)) {
1183 		spin_unlock(&current->sighand->siglock);
1184 		write_unlock_irq(&tasklist_lock);
1185 		retval = -ERESTARTNOINTR;
1186 		goto bad_fork_free_pid;
1187 	}
1188 
1189 	if (clone_flags & CLONE_THREAD) {
1190 		p->group_leader = current->group_leader;
1191 		list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1192 
1193 		if (!cputime_eq(current->signal->it_virt_expires,
1194 				cputime_zero) ||
1195 		    !cputime_eq(current->signal->it_prof_expires,
1196 				cputime_zero) ||
1197 		    current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1198 		    !list_empty(&current->signal->cpu_timers[0]) ||
1199 		    !list_empty(&current->signal->cpu_timers[1]) ||
1200 		    !list_empty(&current->signal->cpu_timers[2])) {
1201 			/*
1202 			 * Have child wake up on its first tick to check
1203 			 * for process CPU timers.
1204 			 */
1205 			p->it_prof_expires = jiffies_to_cputime(1);
1206 		}
1207 	}
1208 
1209 	if (likely(p->pid)) {
1210 		list_add_tail(&p->sibling, &p->real_parent->children);
1211 		if (unlikely(p->ptrace & PT_PTRACED))
1212 			__ptrace_link(p, current->parent);
1213 
1214 		if (thread_group_leader(p)) {
1215 			if (clone_flags & CLONE_NEWPID)
1216 				p->nsproxy->pid_ns->child_reaper = p;
1217 
1218 			p->signal->leader_pid = pid;
1219 			p->signal->tty = current->signal->tty;
1220 			set_task_pgrp(p, task_pgrp_nr(current));
1221 			set_task_session(p, task_session_nr(current));
1222 			attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1223 			attach_pid(p, PIDTYPE_SID, task_session(current));
1224 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1225 			__get_cpu_var(process_counts)++;
1226 		}
1227 		attach_pid(p, PIDTYPE_PID, pid);
1228 		nr_threads++;
1229 	}
1230 
1231 	total_forks++;
1232 	spin_unlock(&current->sighand->siglock);
1233 	write_unlock_irq(&tasklist_lock);
1234 	proc_fork_connector(p);
1235 	cgroup_post_fork(p);
1236 	return p;
1237 
1238 bad_fork_free_pid:
1239 	if (pid != &init_struct_pid)
1240 		free_pid(pid);
1241 bad_fork_cleanup_io:
1242 	put_io_context(p->io_context);
1243 bad_fork_cleanup_namespaces:
1244 	exit_task_namespaces(p);
1245 bad_fork_cleanup_keys:
1246 	exit_keys(p);
1247 bad_fork_cleanup_mm:
1248 	if (p->mm)
1249 		mmput(p->mm);
1250 bad_fork_cleanup_signal:
1251 	cleanup_signal(p);
1252 bad_fork_cleanup_sighand:
1253 	__cleanup_sighand(p->sighand);
1254 bad_fork_cleanup_fs:
1255 	exit_fs(p); /* blocking */
1256 bad_fork_cleanup_files:
1257 	exit_files(p); /* blocking */
1258 bad_fork_cleanup_semundo:
1259 	exit_sem(p);
1260 bad_fork_cleanup_audit:
1261 	audit_free(p);
1262 bad_fork_cleanup_security:
1263 	security_task_free(p);
1264 bad_fork_cleanup_policy:
1265 #ifdef CONFIG_NUMA
1266 	mpol_put(p->mempolicy);
1267 bad_fork_cleanup_cgroup:
1268 #endif
1269 	cgroup_exit(p, cgroup_callbacks_done);
1270 	delayacct_tsk_free(p);
1271 	if (p->binfmt)
1272 		module_put(p->binfmt->module);
1273 bad_fork_cleanup_put_domain:
1274 	module_put(task_thread_info(p)->exec_domain->module);
1275 bad_fork_cleanup_count:
1276 	put_group_info(p->group_info);
1277 	atomic_dec(&p->user->processes);
1278 	free_uid(p->user);
1279 bad_fork_free:
1280 	free_task(p);
1281 fork_out:
1282 	return ERR_PTR(retval);
1283 }
1284 
1285 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1286 {
1287 	memset(regs, 0, sizeof(struct pt_regs));
1288 	return regs;
1289 }
1290 
1291 struct task_struct * __cpuinit fork_idle(int cpu)
1292 {
1293 	struct task_struct *task;
1294 	struct pt_regs regs;
1295 
1296 	task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1297 				&init_struct_pid);
1298 	if (!IS_ERR(task))
1299 		init_idle(task, cpu);
1300 
1301 	return task;
1302 }
1303 
1304 static int fork_traceflag(unsigned clone_flags)
1305 {
1306 	if (clone_flags & CLONE_UNTRACED)
1307 		return 0;
1308 	else if (clone_flags & CLONE_VFORK) {
1309 		if (current->ptrace & PT_TRACE_VFORK)
1310 			return PTRACE_EVENT_VFORK;
1311 	} else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1312 		if (current->ptrace & PT_TRACE_CLONE)
1313 			return PTRACE_EVENT_CLONE;
1314 	} else if (current->ptrace & PT_TRACE_FORK)
1315 		return PTRACE_EVENT_FORK;
1316 
1317 	return 0;
1318 }
1319 
1320 /*
1321  *  Ok, this is the main fork-routine.
1322  *
1323  * It copies the process, and if successful kick-starts
1324  * it and waits for it to finish using the VM if required.
1325  */
1326 long do_fork(unsigned long clone_flags,
1327 	      unsigned long stack_start,
1328 	      struct pt_regs *regs,
1329 	      unsigned long stack_size,
1330 	      int __user *parent_tidptr,
1331 	      int __user *child_tidptr)
1332 {
1333 	struct task_struct *p;
1334 	int trace = 0;
1335 	long nr;
1336 
1337 	/*
1338 	 * We hope to recycle these flags after 2.6.26
1339 	 */
1340 	if (unlikely(clone_flags & CLONE_STOPPED)) {
1341 		static int __read_mostly count = 100;
1342 
1343 		if (count > 0 && printk_ratelimit()) {
1344 			char comm[TASK_COMM_LEN];
1345 
1346 			count--;
1347 			printk(KERN_INFO "fork(): process `%s' used deprecated "
1348 					"clone flags 0x%lx\n",
1349 				get_task_comm(comm, current),
1350 				clone_flags & CLONE_STOPPED);
1351 		}
1352 	}
1353 
1354 	if (unlikely(current->ptrace)) {
1355 		trace = fork_traceflag (clone_flags);
1356 		if (trace)
1357 			clone_flags |= CLONE_PTRACE;
1358 	}
1359 
1360 	p = copy_process(clone_flags, stack_start, regs, stack_size,
1361 			child_tidptr, NULL);
1362 	/*
1363 	 * Do this prior waking up the new thread - the thread pointer
1364 	 * might get invalid after that point, if the thread exits quickly.
1365 	 */
1366 	if (!IS_ERR(p)) {
1367 		struct completion vfork;
1368 
1369 		nr = task_pid_vnr(p);
1370 
1371 		if (clone_flags & CLONE_PARENT_SETTID)
1372 			put_user(nr, parent_tidptr);
1373 
1374 		if (clone_flags & CLONE_VFORK) {
1375 			p->vfork_done = &vfork;
1376 			init_completion(&vfork);
1377 		}
1378 
1379 		if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1380 			/*
1381 			 * We'll start up with an immediate SIGSTOP.
1382 			 */
1383 			sigaddset(&p->pending.signal, SIGSTOP);
1384 			set_tsk_thread_flag(p, TIF_SIGPENDING);
1385 		}
1386 
1387 		if (!(clone_flags & CLONE_STOPPED))
1388 			wake_up_new_task(p, clone_flags);
1389 		else
1390 			__set_task_state(p, TASK_STOPPED);
1391 
1392 		if (unlikely (trace)) {
1393 			current->ptrace_message = nr;
1394 			ptrace_notify ((trace << 8) | SIGTRAP);
1395 		}
1396 
1397 		if (clone_flags & CLONE_VFORK) {
1398 			freezer_do_not_count();
1399 			wait_for_completion(&vfork);
1400 			freezer_count();
1401 			if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1402 				current->ptrace_message = nr;
1403 				ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1404 			}
1405 		}
1406 	} else {
1407 		nr = PTR_ERR(p);
1408 	}
1409 	return nr;
1410 }
1411 
1412 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1413 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1414 #endif
1415 
1416 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1417 {
1418 	struct sighand_struct *sighand = data;
1419 
1420 	spin_lock_init(&sighand->siglock);
1421 	init_waitqueue_head(&sighand->signalfd_wqh);
1422 }
1423 
1424 void __init proc_caches_init(void)
1425 {
1426 	sighand_cachep = kmem_cache_create("sighand_cache",
1427 			sizeof(struct sighand_struct), 0,
1428 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1429 			sighand_ctor);
1430 	signal_cachep = kmem_cache_create("signal_cache",
1431 			sizeof(struct signal_struct), 0,
1432 			SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1433 	files_cachep = kmem_cache_create("files_cache",
1434 			sizeof(struct files_struct), 0,
1435 			SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1436 	fs_cachep = kmem_cache_create("fs_cache",
1437 			sizeof(struct fs_struct), 0,
1438 			SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1439 	vm_area_cachep = kmem_cache_create("vm_area_struct",
1440 			sizeof(struct vm_area_struct), 0,
1441 			SLAB_PANIC, NULL);
1442 	mm_cachep = kmem_cache_create("mm_struct",
1443 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1444 			SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1445 }
1446 
1447 /*
1448  * Check constraints on flags passed to the unshare system call and
1449  * force unsharing of additional process context as appropriate.
1450  */
1451 static void check_unshare_flags(unsigned long *flags_ptr)
1452 {
1453 	/*
1454 	 * If unsharing a thread from a thread group, must also
1455 	 * unshare vm.
1456 	 */
1457 	if (*flags_ptr & CLONE_THREAD)
1458 		*flags_ptr |= CLONE_VM;
1459 
1460 	/*
1461 	 * If unsharing vm, must also unshare signal handlers.
1462 	 */
1463 	if (*flags_ptr & CLONE_VM)
1464 		*flags_ptr |= CLONE_SIGHAND;
1465 
1466 	/*
1467 	 * If unsharing signal handlers and the task was created
1468 	 * using CLONE_THREAD, then must unshare the thread
1469 	 */
1470 	if ((*flags_ptr & CLONE_SIGHAND) &&
1471 	    (atomic_read(&current->signal->count) > 1))
1472 		*flags_ptr |= CLONE_THREAD;
1473 
1474 	/*
1475 	 * If unsharing namespace, must also unshare filesystem information.
1476 	 */
1477 	if (*flags_ptr & CLONE_NEWNS)
1478 		*flags_ptr |= CLONE_FS;
1479 }
1480 
1481 /*
1482  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1483  */
1484 static int unshare_thread(unsigned long unshare_flags)
1485 {
1486 	if (unshare_flags & CLONE_THREAD)
1487 		return -EINVAL;
1488 
1489 	return 0;
1490 }
1491 
1492 /*
1493  * Unshare the filesystem structure if it is being shared
1494  */
1495 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1496 {
1497 	struct fs_struct *fs = current->fs;
1498 
1499 	if ((unshare_flags & CLONE_FS) &&
1500 	    (fs && atomic_read(&fs->count) > 1)) {
1501 		*new_fsp = __copy_fs_struct(current->fs);
1502 		if (!*new_fsp)
1503 			return -ENOMEM;
1504 	}
1505 
1506 	return 0;
1507 }
1508 
1509 /*
1510  * Unsharing of sighand is not supported yet
1511  */
1512 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1513 {
1514 	struct sighand_struct *sigh = current->sighand;
1515 
1516 	if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1517 		return -EINVAL;
1518 	else
1519 		return 0;
1520 }
1521 
1522 /*
1523  * Unshare vm if it is being shared
1524  */
1525 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1526 {
1527 	struct mm_struct *mm = current->mm;
1528 
1529 	if ((unshare_flags & CLONE_VM) &&
1530 	    (mm && atomic_read(&mm->mm_users) > 1)) {
1531 		return -EINVAL;
1532 	}
1533 
1534 	return 0;
1535 }
1536 
1537 /*
1538  * Unshare file descriptor table if it is being shared
1539  */
1540 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1541 {
1542 	struct files_struct *fd = current->files;
1543 	int error = 0;
1544 
1545 	if ((unshare_flags & CLONE_FILES) &&
1546 	    (fd && atomic_read(&fd->count) > 1)) {
1547 		*new_fdp = dup_fd(fd, &error);
1548 		if (!*new_fdp)
1549 			return error;
1550 	}
1551 
1552 	return 0;
1553 }
1554 
1555 /*
1556  * unshare allows a process to 'unshare' part of the process
1557  * context which was originally shared using clone.  copy_*
1558  * functions used by do_fork() cannot be used here directly
1559  * because they modify an inactive task_struct that is being
1560  * constructed. Here we are modifying the current, active,
1561  * task_struct.
1562  */
1563 asmlinkage long sys_unshare(unsigned long unshare_flags)
1564 {
1565 	int err = 0;
1566 	struct fs_struct *fs, *new_fs = NULL;
1567 	struct sighand_struct *new_sigh = NULL;
1568 	struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1569 	struct files_struct *fd, *new_fd = NULL;
1570 	struct nsproxy *new_nsproxy = NULL;
1571 	int do_sysvsem = 0;
1572 
1573 	check_unshare_flags(&unshare_flags);
1574 
1575 	/* Return -EINVAL for all unsupported flags */
1576 	err = -EINVAL;
1577 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1578 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1579 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1580 				CLONE_NEWNET))
1581 		goto bad_unshare_out;
1582 
1583 	/*
1584 	 * CLONE_NEWIPC must also detach from the undolist: after switching
1585 	 * to a new ipc namespace, the semaphore arrays from the old
1586 	 * namespace are unreachable.
1587 	 */
1588 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1589 		do_sysvsem = 1;
1590 	if ((err = unshare_thread(unshare_flags)))
1591 		goto bad_unshare_out;
1592 	if ((err = unshare_fs(unshare_flags, &new_fs)))
1593 		goto bad_unshare_cleanup_thread;
1594 	if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1595 		goto bad_unshare_cleanup_fs;
1596 	if ((err = unshare_vm(unshare_flags, &new_mm)))
1597 		goto bad_unshare_cleanup_sigh;
1598 	if ((err = unshare_fd(unshare_flags, &new_fd)))
1599 		goto bad_unshare_cleanup_vm;
1600 	if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1601 			new_fs)))
1602 		goto bad_unshare_cleanup_fd;
1603 
1604 	if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1605 		if (do_sysvsem) {
1606 			/*
1607 			 * CLONE_SYSVSEM is equivalent to sys_exit().
1608 			 */
1609 			exit_sem(current);
1610 		}
1611 
1612 		if (new_nsproxy) {
1613 			switch_task_namespaces(current, new_nsproxy);
1614 			new_nsproxy = NULL;
1615 		}
1616 
1617 		task_lock(current);
1618 
1619 		if (new_fs) {
1620 			fs = current->fs;
1621 			current->fs = new_fs;
1622 			new_fs = fs;
1623 		}
1624 
1625 		if (new_mm) {
1626 			mm = current->mm;
1627 			active_mm = current->active_mm;
1628 			current->mm = new_mm;
1629 			current->active_mm = new_mm;
1630 			activate_mm(active_mm, new_mm);
1631 			new_mm = mm;
1632 		}
1633 
1634 		if (new_fd) {
1635 			fd = current->files;
1636 			current->files = new_fd;
1637 			new_fd = fd;
1638 		}
1639 
1640 		task_unlock(current);
1641 	}
1642 
1643 	if (new_nsproxy)
1644 		put_nsproxy(new_nsproxy);
1645 
1646 bad_unshare_cleanup_fd:
1647 	if (new_fd)
1648 		put_files_struct(new_fd);
1649 
1650 bad_unshare_cleanup_vm:
1651 	if (new_mm)
1652 		mmput(new_mm);
1653 
1654 bad_unshare_cleanup_sigh:
1655 	if (new_sigh)
1656 		if (atomic_dec_and_test(&new_sigh->count))
1657 			kmem_cache_free(sighand_cachep, new_sigh);
1658 
1659 bad_unshare_cleanup_fs:
1660 	if (new_fs)
1661 		put_fs_struct(new_fs);
1662 
1663 bad_unshare_cleanup_thread:
1664 bad_unshare_out:
1665 	return err;
1666 }
1667 
1668 /*
1669  *	Helper to unshare the files of the current task.
1670  *	We don't want to expose copy_files internals to
1671  *	the exec layer of the kernel.
1672  */
1673 
1674 int unshare_files(struct files_struct **displaced)
1675 {
1676 	struct task_struct *task = current;
1677 	struct files_struct *copy = NULL;
1678 	int error;
1679 
1680 	error = unshare_fd(CLONE_FILES, &copy);
1681 	if (error || !copy) {
1682 		*displaced = NULL;
1683 		return error;
1684 	}
1685 	*displaced = task->files;
1686 	task_lock(task);
1687 	task->files = copy;
1688 	task_unlock(task);
1689 	return 0;
1690 }
1691