1/*
2 * POSIX message queues filesystem for Linux.
3 *
4 * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5 *                          Michal Wronski          (michal.wronski@gmail.com)
6 *
7 * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
9 *			    Manfred Spraul	    (manfred@colorfullife.com)
10 *
11 * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
12 *
13 * This file is released under the GPL.
14 */
15
16#include <linux/capability.h>
17#include <linux/init.h>
18#include <linux/pagemap.h>
19#include <linux/file.h>
20#include <linux/mount.h>
21#include <linux/namei.h>
22#include <linux/sysctl.h>
23#include <linux/poll.h>
24#include <linux/mqueue.h>
25#include <linux/msg.h>
26#include <linux/skbuff.h>
27#include <linux/vmalloc.h>
28#include <linux/netlink.h>
29#include <linux/syscalls.h>
30#include <linux/audit.h>
31#include <linux/signal.h>
32#include <linux/mutex.h>
33#include <linux/nsproxy.h>
34#include <linux/pid.h>
35#include <linux/ipc_namespace.h>
36#include <linux/user_namespace.h>
37#include <linux/slab.h>
38
39#include <net/sock.h>
40#include "util.h"
41
42#define MQUEUE_MAGIC	0x19800202
43#define DIRENT_SIZE	20
44#define FILENT_SIZE	80
45
46#define SEND		0
47#define RECV		1
48
49#define STATE_NONE	0
50#define STATE_PENDING	1
51#define STATE_READY	2
52
53struct posix_msg_tree_node {
54	struct rb_node		rb_node;
55	struct list_head	msg_list;
56	int			priority;
57};
58
59struct ext_wait_queue {		/* queue of sleeping tasks */
60	struct task_struct *task;
61	struct list_head list;
62	struct msg_msg *msg;	/* ptr of loaded message */
63	int state;		/* one of STATE_* values */
64};
65
66struct mqueue_inode_info {
67	spinlock_t lock;
68	struct inode vfs_inode;
69	wait_queue_head_t wait_q;
70
71	struct rb_root msg_tree;
72	struct posix_msg_tree_node *node_cache;
73	struct mq_attr attr;
74
75	struct sigevent notify;
76	struct pid *notify_owner;
77	struct user_namespace *notify_user_ns;
78	struct user_struct *user;	/* user who created, for accounting */
79	struct sock *notify_sock;
80	struct sk_buff *notify_cookie;
81
82	/* for tasks waiting for free space and messages, respectively */
83	struct ext_wait_queue e_wait_q[2];
84
85	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
86};
87
88static const struct inode_operations mqueue_dir_inode_operations;
89static const struct file_operations mqueue_file_operations;
90static const struct super_operations mqueue_super_ops;
91static void remove_notification(struct mqueue_inode_info *info);
92
93static struct kmem_cache *mqueue_inode_cachep;
94
95static struct ctl_table_header *mq_sysctl_table;
96
97static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
98{
99	return container_of(inode, struct mqueue_inode_info, vfs_inode);
100}
101
102/*
103 * This routine should be called with the mq_lock held.
104 */
105static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
106{
107	return get_ipc_ns(inode->i_sb->s_fs_info);
108}
109
110static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
111{
112	struct ipc_namespace *ns;
113
114	spin_lock(&mq_lock);
115	ns = __get_ns_from_inode(inode);
116	spin_unlock(&mq_lock);
117	return ns;
118}
119
120/* Auxiliary functions to manipulate messages' list */
121static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
122{
123	struct rb_node **p, *parent = NULL;
124	struct posix_msg_tree_node *leaf;
125
126	p = &info->msg_tree.rb_node;
127	while (*p) {
128		parent = *p;
129		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
130
131		if (likely(leaf->priority == msg->m_type))
132			goto insert_msg;
133		else if (msg->m_type < leaf->priority)
134			p = &(*p)->rb_left;
135		else
136			p = &(*p)->rb_right;
137	}
138	if (info->node_cache) {
139		leaf = info->node_cache;
140		info->node_cache = NULL;
141	} else {
142		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
143		if (!leaf)
144			return -ENOMEM;
145		INIT_LIST_HEAD(&leaf->msg_list);
146	}
147	leaf->priority = msg->m_type;
148	rb_link_node(&leaf->rb_node, parent, p);
149	rb_insert_color(&leaf->rb_node, &info->msg_tree);
150insert_msg:
151	info->attr.mq_curmsgs++;
152	info->qsize += msg->m_ts;
153	list_add_tail(&msg->m_list, &leaf->msg_list);
154	return 0;
155}
156
157static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
158{
159	struct rb_node **p, *parent = NULL;
160	struct posix_msg_tree_node *leaf;
161	struct msg_msg *msg;
162
163try_again:
164	p = &info->msg_tree.rb_node;
165	while (*p) {
166		parent = *p;
167		/*
168		 * During insert, low priorities go to the left and high to the
169		 * right.  On receive, we want the highest priorities first, so
170		 * walk all the way to the right.
171		 */
172		p = &(*p)->rb_right;
173	}
174	if (!parent) {
175		if (info->attr.mq_curmsgs) {
176			pr_warn_once("Inconsistency in POSIX message queue, "
177				     "no tree element, but supposedly messages "
178				     "should exist!\n");
179			info->attr.mq_curmsgs = 0;
180		}
181		return NULL;
182	}
183	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
184	if (unlikely(list_empty(&leaf->msg_list))) {
185		pr_warn_once("Inconsistency in POSIX message queue, "
186			     "empty leaf node but we haven't implemented "
187			     "lazy leaf delete!\n");
188		rb_erase(&leaf->rb_node, &info->msg_tree);
189		if (info->node_cache) {
190			kfree(leaf);
191		} else {
192			info->node_cache = leaf;
193		}
194		goto try_again;
195	} else {
196		msg = list_first_entry(&leaf->msg_list,
197				       struct msg_msg, m_list);
198		list_del(&msg->m_list);
199		if (list_empty(&leaf->msg_list)) {
200			rb_erase(&leaf->rb_node, &info->msg_tree);
201			if (info->node_cache) {
202				kfree(leaf);
203			} else {
204				info->node_cache = leaf;
205			}
206		}
207	}
208	info->attr.mq_curmsgs--;
209	info->qsize -= msg->m_ts;
210	return msg;
211}
212
213static struct inode *mqueue_get_inode(struct super_block *sb,
214		struct ipc_namespace *ipc_ns, umode_t mode,
215		struct mq_attr *attr)
216{
217	struct user_struct *u = current_user();
218	struct inode *inode;
219	int ret = -ENOMEM;
220
221	inode = new_inode(sb);
222	if (!inode)
223		goto err;
224
225	inode->i_ino = get_next_ino();
226	inode->i_mode = mode;
227	inode->i_uid = current_fsuid();
228	inode->i_gid = current_fsgid();
229	inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
230
231	if (S_ISREG(mode)) {
232		struct mqueue_inode_info *info;
233		unsigned long mq_bytes, mq_treesize;
234
235		inode->i_fop = &mqueue_file_operations;
236		inode->i_size = FILENT_SIZE;
237		/* mqueue specific info */
238		info = MQUEUE_I(inode);
239		spin_lock_init(&info->lock);
240		init_waitqueue_head(&info->wait_q);
241		INIT_LIST_HEAD(&info->e_wait_q[0].list);
242		INIT_LIST_HEAD(&info->e_wait_q[1].list);
243		info->notify_owner = NULL;
244		info->notify_user_ns = NULL;
245		info->qsize = 0;
246		info->user = NULL;	/* set when all is ok */
247		info->msg_tree = RB_ROOT;
248		info->node_cache = NULL;
249		memset(&info->attr, 0, sizeof(info->attr));
250		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
251					   ipc_ns->mq_msg_default);
252		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
253					    ipc_ns->mq_msgsize_default);
254		if (attr) {
255			info->attr.mq_maxmsg = attr->mq_maxmsg;
256			info->attr.mq_msgsize = attr->mq_msgsize;
257		}
258		/*
259		 * We used to allocate a static array of pointers and account
260		 * the size of that array as well as one msg_msg struct per
261		 * possible message into the queue size. That's no longer
262		 * accurate as the queue is now an rbtree and will grow and
263		 * shrink depending on usage patterns.  We can, however, still
264		 * account one msg_msg struct per message, but the nodes are
265		 * allocated depending on priority usage, and most programs
266		 * only use one, or a handful, of priorities.  However, since
267		 * this is pinned memory, we need to assume worst case, so
268		 * that means the min(mq_maxmsg, max_priorities) * struct
269		 * posix_msg_tree_node.
270		 */
271		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
272			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
273			sizeof(struct posix_msg_tree_node);
274
275		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
276					  info->attr.mq_msgsize);
277
278		spin_lock(&mq_lock);
279		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
280		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
281			spin_unlock(&mq_lock);
282			/* mqueue_evict_inode() releases info->messages */
283			ret = -EMFILE;
284			goto out_inode;
285		}
286		u->mq_bytes += mq_bytes;
287		spin_unlock(&mq_lock);
288
289		/* all is ok */
290		info->user = get_uid(u);
291	} else if (S_ISDIR(mode)) {
292		inc_nlink(inode);
293		/* Some things misbehave if size == 0 on a directory */
294		inode->i_size = 2 * DIRENT_SIZE;
295		inode->i_op = &mqueue_dir_inode_operations;
296		inode->i_fop = &simple_dir_operations;
297	}
298
299	return inode;
300out_inode:
301	iput(inode);
302err:
303	return ERR_PTR(ret);
304}
305
306static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
307{
308	struct inode *inode;
309	struct ipc_namespace *ns = data;
310
311	sb->s_blocksize = PAGE_CACHE_SIZE;
312	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
313	sb->s_magic = MQUEUE_MAGIC;
314	sb->s_op = &mqueue_super_ops;
315
316	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
317	if (IS_ERR(inode))
318		return PTR_ERR(inode);
319
320	sb->s_root = d_make_root(inode);
321	if (!sb->s_root)
322		return -ENOMEM;
323	return 0;
324}
325
326static struct dentry *mqueue_mount(struct file_system_type *fs_type,
327			 int flags, const char *dev_name,
328			 void *data)
329{
330	if (!(flags & MS_KERNMOUNT)) {
331		struct ipc_namespace *ns = current->nsproxy->ipc_ns;
332		/* Don't allow mounting unless the caller has CAP_SYS_ADMIN
333		 * over the ipc namespace.
334		 */
335		if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
336			return ERR_PTR(-EPERM);
337
338		data = ns;
339	}
340	return mount_ns(fs_type, flags, data, mqueue_fill_super);
341}
342
343static void init_once(void *foo)
344{
345	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
346
347	inode_init_once(&p->vfs_inode);
348}
349
350static struct inode *mqueue_alloc_inode(struct super_block *sb)
351{
352	struct mqueue_inode_info *ei;
353
354	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
355	if (!ei)
356		return NULL;
357	return &ei->vfs_inode;
358}
359
360static void mqueue_i_callback(struct rcu_head *head)
361{
362	struct inode *inode = container_of(head, struct inode, i_rcu);
363	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
364}
365
366static void mqueue_destroy_inode(struct inode *inode)
367{
368	call_rcu(&inode->i_rcu, mqueue_i_callback);
369}
370
371static void mqueue_evict_inode(struct inode *inode)
372{
373	struct mqueue_inode_info *info;
374	struct user_struct *user;
375	unsigned long mq_bytes, mq_treesize;
376	struct ipc_namespace *ipc_ns;
377	struct msg_msg *msg;
378
379	clear_inode(inode);
380
381	if (S_ISDIR(inode->i_mode))
382		return;
383
384	ipc_ns = get_ns_from_inode(inode);
385	info = MQUEUE_I(inode);
386	spin_lock(&info->lock);
387	while ((msg = msg_get(info)) != NULL)
388		free_msg(msg);
389	kfree(info->node_cache);
390	spin_unlock(&info->lock);
391
392	/* Total amount of bytes accounted for the mqueue */
393	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
394		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
395		sizeof(struct posix_msg_tree_node);
396
397	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
398				  info->attr.mq_msgsize);
399
400	user = info->user;
401	if (user) {
402		spin_lock(&mq_lock);
403		user->mq_bytes -= mq_bytes;
404		/*
405		 * get_ns_from_inode() ensures that the
406		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
407		 * to which we now hold a reference, or it is NULL.
408		 * We can't put it here under mq_lock, though.
409		 */
410		if (ipc_ns)
411			ipc_ns->mq_queues_count--;
412		spin_unlock(&mq_lock);
413		free_uid(user);
414	}
415	if (ipc_ns)
416		put_ipc_ns(ipc_ns);
417}
418
419static int mqueue_create(struct inode *dir, struct dentry *dentry,
420				umode_t mode, bool excl)
421{
422	struct inode *inode;
423	struct mq_attr *attr = dentry->d_fsdata;
424	int error;
425	struct ipc_namespace *ipc_ns;
426
427	spin_lock(&mq_lock);
428	ipc_ns = __get_ns_from_inode(dir);
429	if (!ipc_ns) {
430		error = -EACCES;
431		goto out_unlock;
432	}
433
434	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
435	    !capable(CAP_SYS_RESOURCE)) {
436		error = -ENOSPC;
437		goto out_unlock;
438	}
439	ipc_ns->mq_queues_count++;
440	spin_unlock(&mq_lock);
441
442	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
443	if (IS_ERR(inode)) {
444		error = PTR_ERR(inode);
445		spin_lock(&mq_lock);
446		ipc_ns->mq_queues_count--;
447		goto out_unlock;
448	}
449
450	put_ipc_ns(ipc_ns);
451	dir->i_size += DIRENT_SIZE;
452	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
453
454	d_instantiate(dentry, inode);
455	dget(dentry);
456	return 0;
457out_unlock:
458	spin_unlock(&mq_lock);
459	if (ipc_ns)
460		put_ipc_ns(ipc_ns);
461	return error;
462}
463
464static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
465{
466	struct inode *inode = d_inode(dentry);
467
468	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
469	dir->i_size -= DIRENT_SIZE;
470	drop_nlink(inode);
471	dput(dentry);
472	return 0;
473}
474
475/*
476*	This is routine for system read from queue file.
477*	To avoid mess with doing here some sort of mq_receive we allow
478*	to read only queue size & notification info (the only values
479*	that are interesting from user point of view and aren't accessible
480*	through std routines)
481*/
482static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
483				size_t count, loff_t *off)
484{
485	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
486	char buffer[FILENT_SIZE];
487	ssize_t ret;
488
489	spin_lock(&info->lock);
490	snprintf(buffer, sizeof(buffer),
491			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
492			info->qsize,
493			info->notify_owner ? info->notify.sigev_notify : 0,
494			(info->notify_owner &&
495			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
496				info->notify.sigev_signo : 0,
497			pid_vnr(info->notify_owner));
498	spin_unlock(&info->lock);
499	buffer[sizeof(buffer)-1] = '\0';
500
501	ret = simple_read_from_buffer(u_data, count, off, buffer,
502				strlen(buffer));
503	if (ret <= 0)
504		return ret;
505
506	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
507	return ret;
508}
509
510static int mqueue_flush_file(struct file *filp, fl_owner_t id)
511{
512	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
513
514	spin_lock(&info->lock);
515	if (task_tgid(current) == info->notify_owner)
516		remove_notification(info);
517
518	spin_unlock(&info->lock);
519	return 0;
520}
521
522static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
523{
524	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
525	int retval = 0;
526
527	poll_wait(filp, &info->wait_q, poll_tab);
528
529	spin_lock(&info->lock);
530	if (info->attr.mq_curmsgs)
531		retval = POLLIN | POLLRDNORM;
532
533	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
534		retval |= POLLOUT | POLLWRNORM;
535	spin_unlock(&info->lock);
536
537	return retval;
538}
539
540/* Adds current to info->e_wait_q[sr] before element with smaller prio */
541static void wq_add(struct mqueue_inode_info *info, int sr,
542			struct ext_wait_queue *ewp)
543{
544	struct ext_wait_queue *walk;
545
546	ewp->task = current;
547
548	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
549		if (walk->task->static_prio <= current->static_prio) {
550			list_add_tail(&ewp->list, &walk->list);
551			return;
552		}
553	}
554	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
555}
556
557/*
558 * Puts current task to sleep. Caller must hold queue lock. After return
559 * lock isn't held.
560 * sr: SEND or RECV
561 */
562static int wq_sleep(struct mqueue_inode_info *info, int sr,
563		    ktime_t *timeout, struct ext_wait_queue *ewp)
564{
565	int retval;
566	signed long time;
567
568	wq_add(info, sr, ewp);
569
570	for (;;) {
571		set_current_state(TASK_INTERRUPTIBLE);
572
573		spin_unlock(&info->lock);
574		time = schedule_hrtimeout_range_clock(timeout, 0,
575			HRTIMER_MODE_ABS, CLOCK_REALTIME);
576
577		while (ewp->state == STATE_PENDING)
578			cpu_relax();
579
580		if (ewp->state == STATE_READY) {
581			retval = 0;
582			goto out;
583		}
584		spin_lock(&info->lock);
585		if (ewp->state == STATE_READY) {
586			retval = 0;
587			goto out_unlock;
588		}
589		if (signal_pending(current)) {
590			retval = -ERESTARTSYS;
591			break;
592		}
593		if (time == 0) {
594			retval = -ETIMEDOUT;
595			break;
596		}
597	}
598	list_del(&ewp->list);
599out_unlock:
600	spin_unlock(&info->lock);
601out:
602	return retval;
603}
604
605/*
606 * Returns waiting task that should be serviced first or NULL if none exists
607 */
608static struct ext_wait_queue *wq_get_first_waiter(
609		struct mqueue_inode_info *info, int sr)
610{
611	struct list_head *ptr;
612
613	ptr = info->e_wait_q[sr].list.prev;
614	if (ptr == &info->e_wait_q[sr].list)
615		return NULL;
616	return list_entry(ptr, struct ext_wait_queue, list);
617}
618
619
620static inline void set_cookie(struct sk_buff *skb, char code)
621{
622	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
623}
624
625/*
626 * The next function is only to split too long sys_mq_timedsend
627 */
628static void __do_notify(struct mqueue_inode_info *info)
629{
630	/* notification
631	 * invoked when there is registered process and there isn't process
632	 * waiting synchronously for message AND state of queue changed from
633	 * empty to not empty. Here we are sure that no one is waiting
634	 * synchronously. */
635	if (info->notify_owner &&
636	    info->attr.mq_curmsgs == 1) {
637		struct siginfo sig_i;
638		switch (info->notify.sigev_notify) {
639		case SIGEV_NONE:
640			break;
641		case SIGEV_SIGNAL:
642			/* sends signal */
643
644			sig_i.si_signo = info->notify.sigev_signo;
645			sig_i.si_errno = 0;
646			sig_i.si_code = SI_MESGQ;
647			sig_i.si_value = info->notify.sigev_value;
648			/* map current pid/uid into info->owner's namespaces */
649			rcu_read_lock();
650			sig_i.si_pid = task_tgid_nr_ns(current,
651						ns_of_pid(info->notify_owner));
652			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
653			rcu_read_unlock();
654
655			kill_pid_info(info->notify.sigev_signo,
656				      &sig_i, info->notify_owner);
657			break;
658		case SIGEV_THREAD:
659			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
660			netlink_sendskb(info->notify_sock, info->notify_cookie);
661			break;
662		}
663		/* after notification unregisters process */
664		put_pid(info->notify_owner);
665		put_user_ns(info->notify_user_ns);
666		info->notify_owner = NULL;
667		info->notify_user_ns = NULL;
668	}
669	wake_up(&info->wait_q);
670}
671
672static int prepare_timeout(const struct timespec __user *u_abs_timeout,
673			   ktime_t *expires, struct timespec *ts)
674{
675	if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
676		return -EFAULT;
677	if (!timespec_valid(ts))
678		return -EINVAL;
679
680	*expires = timespec_to_ktime(*ts);
681	return 0;
682}
683
684static void remove_notification(struct mqueue_inode_info *info)
685{
686	if (info->notify_owner != NULL &&
687	    info->notify.sigev_notify == SIGEV_THREAD) {
688		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
689		netlink_sendskb(info->notify_sock, info->notify_cookie);
690	}
691	put_pid(info->notify_owner);
692	put_user_ns(info->notify_user_ns);
693	info->notify_owner = NULL;
694	info->notify_user_ns = NULL;
695}
696
697static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
698{
699	int mq_treesize;
700	unsigned long total_size;
701
702	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
703		return -EINVAL;
704	if (capable(CAP_SYS_RESOURCE)) {
705		if (attr->mq_maxmsg > HARD_MSGMAX ||
706		    attr->mq_msgsize > HARD_MSGSIZEMAX)
707			return -EINVAL;
708	} else {
709		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
710				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
711			return -EINVAL;
712	}
713	/* check for overflow */
714	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
715		return -EOVERFLOW;
716	mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
717		min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
718		sizeof(struct posix_msg_tree_node);
719	total_size = attr->mq_maxmsg * attr->mq_msgsize;
720	if (total_size + mq_treesize < total_size)
721		return -EOVERFLOW;
722	return 0;
723}
724
725/*
726 * Invoked when creating a new queue via sys_mq_open
727 */
728static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
729			struct path *path, int oflag, umode_t mode,
730			struct mq_attr *attr)
731{
732	const struct cred *cred = current_cred();
733	int ret;
734
735	if (attr) {
736		ret = mq_attr_ok(ipc_ns, attr);
737		if (ret)
738			return ERR_PTR(ret);
739		/* store for use during create */
740		path->dentry->d_fsdata = attr;
741	} else {
742		struct mq_attr def_attr;
743
744		def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
745					 ipc_ns->mq_msg_default);
746		def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
747					  ipc_ns->mq_msgsize_default);
748		ret = mq_attr_ok(ipc_ns, &def_attr);
749		if (ret)
750			return ERR_PTR(ret);
751	}
752
753	mode &= ~current_umask();
754	ret = vfs_create(dir, path->dentry, mode, true);
755	path->dentry->d_fsdata = NULL;
756	if (ret)
757		return ERR_PTR(ret);
758	return dentry_open(path, oflag, cred);
759}
760
761/* Opens existing queue */
762static struct file *do_open(struct path *path, int oflag)
763{
764	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
765						  MAY_READ | MAY_WRITE };
766	int acc;
767	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
768		return ERR_PTR(-EINVAL);
769	acc = oflag2acc[oflag & O_ACCMODE];
770	if (inode_permission(d_inode(path->dentry), acc))
771		return ERR_PTR(-EACCES);
772	return dentry_open(path, oflag, current_cred());
773}
774
775SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
776		struct mq_attr __user *, u_attr)
777{
778	struct path path;
779	struct file *filp;
780	struct filename *name;
781	struct mq_attr attr;
782	int fd, error;
783	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
784	struct vfsmount *mnt = ipc_ns->mq_mnt;
785	struct dentry *root = mnt->mnt_root;
786	int ro;
787
788	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
789		return -EFAULT;
790
791	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
792
793	if (IS_ERR(name = getname(u_name)))
794		return PTR_ERR(name);
795
796	fd = get_unused_fd_flags(O_CLOEXEC);
797	if (fd < 0)
798		goto out_putname;
799
800	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
801	error = 0;
802	mutex_lock(&d_inode(root)->i_mutex);
803	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
804	if (IS_ERR(path.dentry)) {
805		error = PTR_ERR(path.dentry);
806		goto out_putfd;
807	}
808	path.mnt = mntget(mnt);
809
810	if (oflag & O_CREAT) {
811		if (d_really_is_positive(path.dentry)) {	/* entry already exists */
812			audit_inode(name, path.dentry, 0);
813			if (oflag & O_EXCL) {
814				error = -EEXIST;
815				goto out;
816			}
817			filp = do_open(&path, oflag);
818		} else {
819			if (ro) {
820				error = ro;
821				goto out;
822			}
823			audit_inode_parent_hidden(name, root);
824			filp = do_create(ipc_ns, d_inode(root),
825						&path, oflag, mode,
826						u_attr ? &attr : NULL);
827		}
828	} else {
829		if (d_really_is_negative(path.dentry)) {
830			error = -ENOENT;
831			goto out;
832		}
833		audit_inode(name, path.dentry, 0);
834		filp = do_open(&path, oflag);
835	}
836
837	if (!IS_ERR(filp))
838		fd_install(fd, filp);
839	else
840		error = PTR_ERR(filp);
841out:
842	path_put(&path);
843out_putfd:
844	if (error) {
845		put_unused_fd(fd);
846		fd = error;
847	}
848	mutex_unlock(&d_inode(root)->i_mutex);
849	if (!ro)
850		mnt_drop_write(mnt);
851out_putname:
852	putname(name);
853	return fd;
854}
855
856SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
857{
858	int err;
859	struct filename *name;
860	struct dentry *dentry;
861	struct inode *inode = NULL;
862	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
863	struct vfsmount *mnt = ipc_ns->mq_mnt;
864
865	name = getname(u_name);
866	if (IS_ERR(name))
867		return PTR_ERR(name);
868
869	audit_inode_parent_hidden(name, mnt->mnt_root);
870	err = mnt_want_write(mnt);
871	if (err)
872		goto out_name;
873	mutex_lock_nested(&d_inode(mnt->mnt_root)->i_mutex, I_MUTEX_PARENT);
874	dentry = lookup_one_len(name->name, mnt->mnt_root,
875				strlen(name->name));
876	if (IS_ERR(dentry)) {
877		err = PTR_ERR(dentry);
878		goto out_unlock;
879	}
880
881	inode = d_inode(dentry);
882	if (!inode) {
883		err = -ENOENT;
884	} else {
885		ihold(inode);
886		err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
887	}
888	dput(dentry);
889
890out_unlock:
891	mutex_unlock(&d_inode(mnt->mnt_root)->i_mutex);
892	if (inode)
893		iput(inode);
894	mnt_drop_write(mnt);
895out_name:
896	putname(name);
897
898	return err;
899}
900
901/* Pipelined send and receive functions.
902 *
903 * If a receiver finds no waiting message, then it registers itself in the
904 * list of waiting receivers. A sender checks that list before adding the new
905 * message into the message array. If there is a waiting receiver, then it
906 * bypasses the message array and directly hands the message over to the
907 * receiver.
908 * The receiver accepts the message and returns without grabbing the queue
909 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
910 * are necessary. The same algorithm is used for sysv semaphores, see
911 * ipc/sem.c for more details.
912 *
913 * The same algorithm is used for senders.
914 */
915
916/* pipelined_send() - send a message directly to the task waiting in
917 * sys_mq_timedreceive() (without inserting message into a queue).
918 */
919static inline void pipelined_send(struct mqueue_inode_info *info,
920				  struct msg_msg *message,
921				  struct ext_wait_queue *receiver)
922{
923	receiver->msg = message;
924	list_del(&receiver->list);
925	receiver->state = STATE_PENDING;
926	wake_up_process(receiver->task);
927	smp_wmb();
928	receiver->state = STATE_READY;
929}
930
931/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
932 * gets its message and put to the queue (we have one free place for sure). */
933static inline void pipelined_receive(struct mqueue_inode_info *info)
934{
935	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
936
937	if (!sender) {
938		/* for poll */
939		wake_up_interruptible(&info->wait_q);
940		return;
941	}
942	if (msg_insert(sender->msg, info))
943		return;
944	list_del(&sender->list);
945	sender->state = STATE_PENDING;
946	wake_up_process(sender->task);
947	smp_wmb();
948	sender->state = STATE_READY;
949}
950
951SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
952		size_t, msg_len, unsigned int, msg_prio,
953		const struct timespec __user *, u_abs_timeout)
954{
955	struct fd f;
956	struct inode *inode;
957	struct ext_wait_queue wait;
958	struct ext_wait_queue *receiver;
959	struct msg_msg *msg_ptr;
960	struct mqueue_inode_info *info;
961	ktime_t expires, *timeout = NULL;
962	struct timespec ts;
963	struct posix_msg_tree_node *new_leaf = NULL;
964	int ret = 0;
965
966	if (u_abs_timeout) {
967		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
968		if (res)
969			return res;
970		timeout = &expires;
971	}
972
973	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
974		return -EINVAL;
975
976	audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
977
978	f = fdget(mqdes);
979	if (unlikely(!f.file)) {
980		ret = -EBADF;
981		goto out;
982	}
983
984	inode = file_inode(f.file);
985	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
986		ret = -EBADF;
987		goto out_fput;
988	}
989	info = MQUEUE_I(inode);
990	audit_file(f.file);
991
992	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
993		ret = -EBADF;
994		goto out_fput;
995	}
996
997	if (unlikely(msg_len > info->attr.mq_msgsize)) {
998		ret = -EMSGSIZE;
999		goto out_fput;
1000	}
1001
1002	/* First try to allocate memory, before doing anything with
1003	 * existing queues. */
1004	msg_ptr = load_msg(u_msg_ptr, msg_len);
1005	if (IS_ERR(msg_ptr)) {
1006		ret = PTR_ERR(msg_ptr);
1007		goto out_fput;
1008	}
1009	msg_ptr->m_ts = msg_len;
1010	msg_ptr->m_type = msg_prio;
1011
1012	/*
1013	 * msg_insert really wants us to have a valid, spare node struct so
1014	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1015	 * fall back to that if necessary.
1016	 */
1017	if (!info->node_cache)
1018		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1019
1020	spin_lock(&info->lock);
1021
1022	if (!info->node_cache && new_leaf) {
1023		/* Save our speculative allocation into the cache */
1024		INIT_LIST_HEAD(&new_leaf->msg_list);
1025		info->node_cache = new_leaf;
1026		new_leaf = NULL;
1027	} else {
1028		kfree(new_leaf);
1029	}
1030
1031	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1032		if (f.file->f_flags & O_NONBLOCK) {
1033			ret = -EAGAIN;
1034		} else {
1035			wait.task = current;
1036			wait.msg = (void *) msg_ptr;
1037			wait.state = STATE_NONE;
1038			ret = wq_sleep(info, SEND, timeout, &wait);
1039			/*
1040			 * wq_sleep must be called with info->lock held, and
1041			 * returns with the lock released
1042			 */
1043			goto out_free;
1044		}
1045	} else {
1046		receiver = wq_get_first_waiter(info, RECV);
1047		if (receiver) {
1048			pipelined_send(info, msg_ptr, receiver);
1049		} else {
1050			/* adds message to the queue */
1051			ret = msg_insert(msg_ptr, info);
1052			if (ret)
1053				goto out_unlock;
1054			__do_notify(info);
1055		}
1056		inode->i_atime = inode->i_mtime = inode->i_ctime =
1057				CURRENT_TIME;
1058	}
1059out_unlock:
1060	spin_unlock(&info->lock);
1061out_free:
1062	if (ret)
1063		free_msg(msg_ptr);
1064out_fput:
1065	fdput(f);
1066out:
1067	return ret;
1068}
1069
1070SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1071		size_t, msg_len, unsigned int __user *, u_msg_prio,
1072		const struct timespec __user *, u_abs_timeout)
1073{
1074	ssize_t ret;
1075	struct msg_msg *msg_ptr;
1076	struct fd f;
1077	struct inode *inode;
1078	struct mqueue_inode_info *info;
1079	struct ext_wait_queue wait;
1080	ktime_t expires, *timeout = NULL;
1081	struct timespec ts;
1082	struct posix_msg_tree_node *new_leaf = NULL;
1083
1084	if (u_abs_timeout) {
1085		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1086		if (res)
1087			return res;
1088		timeout = &expires;
1089	}
1090
1091	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1092
1093	f = fdget(mqdes);
1094	if (unlikely(!f.file)) {
1095		ret = -EBADF;
1096		goto out;
1097	}
1098
1099	inode = file_inode(f.file);
1100	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1101		ret = -EBADF;
1102		goto out_fput;
1103	}
1104	info = MQUEUE_I(inode);
1105	audit_file(f.file);
1106
1107	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1108		ret = -EBADF;
1109		goto out_fput;
1110	}
1111
1112	/* checks if buffer is big enough */
1113	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1114		ret = -EMSGSIZE;
1115		goto out_fput;
1116	}
1117
1118	/*
1119	 * msg_insert really wants us to have a valid, spare node struct so
1120	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1121	 * fall back to that if necessary.
1122	 */
1123	if (!info->node_cache)
1124		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1125
1126	spin_lock(&info->lock);
1127
1128	if (!info->node_cache && new_leaf) {
1129		/* Save our speculative allocation into the cache */
1130		INIT_LIST_HEAD(&new_leaf->msg_list);
1131		info->node_cache = new_leaf;
1132	} else {
1133		kfree(new_leaf);
1134	}
1135
1136	if (info->attr.mq_curmsgs == 0) {
1137		if (f.file->f_flags & O_NONBLOCK) {
1138			spin_unlock(&info->lock);
1139			ret = -EAGAIN;
1140		} else {
1141			wait.task = current;
1142			wait.state = STATE_NONE;
1143			ret = wq_sleep(info, RECV, timeout, &wait);
1144			msg_ptr = wait.msg;
1145		}
1146	} else {
1147		msg_ptr = msg_get(info);
1148
1149		inode->i_atime = inode->i_mtime = inode->i_ctime =
1150				CURRENT_TIME;
1151
1152		/* There is now free space in queue. */
1153		pipelined_receive(info);
1154		spin_unlock(&info->lock);
1155		ret = 0;
1156	}
1157	if (ret == 0) {
1158		ret = msg_ptr->m_ts;
1159
1160		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1161			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1162			ret = -EFAULT;
1163		}
1164		free_msg(msg_ptr);
1165	}
1166out_fput:
1167	fdput(f);
1168out:
1169	return ret;
1170}
1171
1172/*
1173 * Notes: the case when user wants us to deregister (with NULL as pointer)
1174 * and he isn't currently owner of notification, will be silently discarded.
1175 * It isn't explicitly defined in the POSIX.
1176 */
1177SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1178		const struct sigevent __user *, u_notification)
1179{
1180	int ret;
1181	struct fd f;
1182	struct sock *sock;
1183	struct inode *inode;
1184	struct sigevent notification;
1185	struct mqueue_inode_info *info;
1186	struct sk_buff *nc;
1187
1188	if (u_notification) {
1189		if (copy_from_user(&notification, u_notification,
1190					sizeof(struct sigevent)))
1191			return -EFAULT;
1192	}
1193
1194	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1195
1196	nc = NULL;
1197	sock = NULL;
1198	if (u_notification != NULL) {
1199		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1200			     notification.sigev_notify != SIGEV_SIGNAL &&
1201			     notification.sigev_notify != SIGEV_THREAD))
1202			return -EINVAL;
1203		if (notification.sigev_notify == SIGEV_SIGNAL &&
1204			!valid_signal(notification.sigev_signo)) {
1205			return -EINVAL;
1206		}
1207		if (notification.sigev_notify == SIGEV_THREAD) {
1208			long timeo;
1209
1210			/* create the notify skb */
1211			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1212			if (!nc) {
1213				ret = -ENOMEM;
1214				goto out;
1215			}
1216			if (copy_from_user(nc->data,
1217					notification.sigev_value.sival_ptr,
1218					NOTIFY_COOKIE_LEN)) {
1219				ret = -EFAULT;
1220				goto out;
1221			}
1222
1223			/* TODO: add a header? */
1224			skb_put(nc, NOTIFY_COOKIE_LEN);
1225			/* and attach it to the socket */
1226retry:
1227			f = fdget(notification.sigev_signo);
1228			if (!f.file) {
1229				ret = -EBADF;
1230				goto out;
1231			}
1232			sock = netlink_getsockbyfilp(f.file);
1233			fdput(f);
1234			if (IS_ERR(sock)) {
1235				ret = PTR_ERR(sock);
1236				sock = NULL;
1237				goto out;
1238			}
1239
1240			timeo = MAX_SCHEDULE_TIMEOUT;
1241			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1242			if (ret == 1)
1243				goto retry;
1244			if (ret) {
1245				sock = NULL;
1246				nc = NULL;
1247				goto out;
1248			}
1249		}
1250	}
1251
1252	f = fdget(mqdes);
1253	if (!f.file) {
1254		ret = -EBADF;
1255		goto out;
1256	}
1257
1258	inode = file_inode(f.file);
1259	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1260		ret = -EBADF;
1261		goto out_fput;
1262	}
1263	info = MQUEUE_I(inode);
1264
1265	ret = 0;
1266	spin_lock(&info->lock);
1267	if (u_notification == NULL) {
1268		if (info->notify_owner == task_tgid(current)) {
1269			remove_notification(info);
1270			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1271		}
1272	} else if (info->notify_owner != NULL) {
1273		ret = -EBUSY;
1274	} else {
1275		switch (notification.sigev_notify) {
1276		case SIGEV_NONE:
1277			info->notify.sigev_notify = SIGEV_NONE;
1278			break;
1279		case SIGEV_THREAD:
1280			info->notify_sock = sock;
1281			info->notify_cookie = nc;
1282			sock = NULL;
1283			nc = NULL;
1284			info->notify.sigev_notify = SIGEV_THREAD;
1285			break;
1286		case SIGEV_SIGNAL:
1287			info->notify.sigev_signo = notification.sigev_signo;
1288			info->notify.sigev_value = notification.sigev_value;
1289			info->notify.sigev_notify = SIGEV_SIGNAL;
1290			break;
1291		}
1292
1293		info->notify_owner = get_pid(task_tgid(current));
1294		info->notify_user_ns = get_user_ns(current_user_ns());
1295		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1296	}
1297	spin_unlock(&info->lock);
1298out_fput:
1299	fdput(f);
1300out:
1301	if (sock)
1302		netlink_detachskb(sock, nc);
1303	else if (nc)
1304		dev_kfree_skb(nc);
1305
1306	return ret;
1307}
1308
1309SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1310		const struct mq_attr __user *, u_mqstat,
1311		struct mq_attr __user *, u_omqstat)
1312{
1313	int ret;
1314	struct mq_attr mqstat, omqstat;
1315	struct fd f;
1316	struct inode *inode;
1317	struct mqueue_inode_info *info;
1318
1319	if (u_mqstat != NULL) {
1320		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1321			return -EFAULT;
1322		if (mqstat.mq_flags & (~O_NONBLOCK))
1323			return -EINVAL;
1324	}
1325
1326	f = fdget(mqdes);
1327	if (!f.file) {
1328		ret = -EBADF;
1329		goto out;
1330	}
1331
1332	inode = file_inode(f.file);
1333	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1334		ret = -EBADF;
1335		goto out_fput;
1336	}
1337	info = MQUEUE_I(inode);
1338
1339	spin_lock(&info->lock);
1340
1341	omqstat = info->attr;
1342	omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1343	if (u_mqstat) {
1344		audit_mq_getsetattr(mqdes, &mqstat);
1345		spin_lock(&f.file->f_lock);
1346		if (mqstat.mq_flags & O_NONBLOCK)
1347			f.file->f_flags |= O_NONBLOCK;
1348		else
1349			f.file->f_flags &= ~O_NONBLOCK;
1350		spin_unlock(&f.file->f_lock);
1351
1352		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1353	}
1354
1355	spin_unlock(&info->lock);
1356
1357	ret = 0;
1358	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1359						sizeof(struct mq_attr)))
1360		ret = -EFAULT;
1361
1362out_fput:
1363	fdput(f);
1364out:
1365	return ret;
1366}
1367
1368static const struct inode_operations mqueue_dir_inode_operations = {
1369	.lookup = simple_lookup,
1370	.create = mqueue_create,
1371	.unlink = mqueue_unlink,
1372};
1373
1374static const struct file_operations mqueue_file_operations = {
1375	.flush = mqueue_flush_file,
1376	.poll = mqueue_poll_file,
1377	.read = mqueue_read_file,
1378	.llseek = default_llseek,
1379};
1380
1381static const struct super_operations mqueue_super_ops = {
1382	.alloc_inode = mqueue_alloc_inode,
1383	.destroy_inode = mqueue_destroy_inode,
1384	.evict_inode = mqueue_evict_inode,
1385	.statfs = simple_statfs,
1386};
1387
1388static struct file_system_type mqueue_fs_type = {
1389	.name = "mqueue",
1390	.mount = mqueue_mount,
1391	.kill_sb = kill_litter_super,
1392	.fs_flags = FS_USERNS_MOUNT,
1393};
1394
1395int mq_init_ns(struct ipc_namespace *ns)
1396{
1397	ns->mq_queues_count  = 0;
1398	ns->mq_queues_max    = DFLT_QUEUESMAX;
1399	ns->mq_msg_max       = DFLT_MSGMAX;
1400	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1401	ns->mq_msg_default   = DFLT_MSG;
1402	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1403
1404	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1405	if (IS_ERR(ns->mq_mnt)) {
1406		int err = PTR_ERR(ns->mq_mnt);
1407		ns->mq_mnt = NULL;
1408		return err;
1409	}
1410	return 0;
1411}
1412
1413void mq_clear_sbinfo(struct ipc_namespace *ns)
1414{
1415	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1416}
1417
1418void mq_put_mnt(struct ipc_namespace *ns)
1419{
1420	kern_unmount(ns->mq_mnt);
1421}
1422
1423static int __init init_mqueue_fs(void)
1424{
1425	int error;
1426
1427	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1428				sizeof(struct mqueue_inode_info), 0,
1429				SLAB_HWCACHE_ALIGN, init_once);
1430	if (mqueue_inode_cachep == NULL)
1431		return -ENOMEM;
1432
1433	/* ignore failures - they are not fatal */
1434	mq_sysctl_table = mq_register_sysctl_table();
1435
1436	error = register_filesystem(&mqueue_fs_type);
1437	if (error)
1438		goto out_sysctl;
1439
1440	spin_lock_init(&mq_lock);
1441
1442	error = mq_init_ns(&init_ipc_ns);
1443	if (error)
1444		goto out_filesystem;
1445
1446	return 0;
1447
1448out_filesystem:
1449	unregister_filesystem(&mqueue_fs_type);
1450out_sysctl:
1451	if (mq_sysctl_table)
1452		unregister_sysctl_table(mq_sysctl_table);
1453	kmem_cache_destroy(mqueue_inode_cachep);
1454	return error;
1455}
1456
1457device_initcall(init_mqueue_fs);
1458