xref: /linux/ipc/mqueue.c (revision 7820b0715b6fb1378fab41b27fb7aa3950852cb7)
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 
53 struct posix_msg_tree_node {
54 	struct rb_node		rb_node;
55 	struct list_head	msg_list;
56 	int			priority;
57 };
58 
59 struct 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 
66 struct 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 mq_attr attr;
73 
74 	struct sigevent notify;
75 	struct pid* notify_owner;
76 	struct user_namespace *notify_user_ns;
77 	struct user_struct *user;	/* user who created, for accounting */
78 	struct sock *notify_sock;
79 	struct sk_buff *notify_cookie;
80 
81 	/* for tasks waiting for free space and messages, respectively */
82 	struct ext_wait_queue e_wait_q[2];
83 
84 	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
85 };
86 
87 static const struct inode_operations mqueue_dir_inode_operations;
88 static const struct file_operations mqueue_file_operations;
89 static const struct super_operations mqueue_super_ops;
90 static void remove_notification(struct mqueue_inode_info *info);
91 
92 static struct kmem_cache *mqueue_inode_cachep;
93 
94 static struct ctl_table_header * mq_sysctl_table;
95 
96 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
97 {
98 	return container_of(inode, struct mqueue_inode_info, vfs_inode);
99 }
100 
101 /*
102  * This routine should be called with the mq_lock held.
103  */
104 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
105 {
106 	return get_ipc_ns(inode->i_sb->s_fs_info);
107 }
108 
109 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
110 {
111 	struct ipc_namespace *ns;
112 
113 	spin_lock(&mq_lock);
114 	ns = __get_ns_from_inode(inode);
115 	spin_unlock(&mq_lock);
116 	return ns;
117 }
118 
119 /* Auxiliary functions to manipulate messages' list */
120 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
121 {
122 	struct rb_node **p, *parent = NULL;
123 	struct posix_msg_tree_node *leaf;
124 
125 	p = &info->msg_tree.rb_node;
126 	while (*p) {
127 		parent = *p;
128 		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
129 
130 		if (likely(leaf->priority == msg->m_type))
131 			goto insert_msg;
132 		else if (msg->m_type < leaf->priority)
133 			p = &(*p)->rb_left;
134 		else
135 			p = &(*p)->rb_right;
136 	}
137 	leaf = kzalloc(sizeof(*leaf), GFP_ATOMIC);
138 	if (!leaf)
139 		return -ENOMEM;
140 	rb_init_node(&leaf->rb_node);
141 	INIT_LIST_HEAD(&leaf->msg_list);
142 	leaf->priority = msg->m_type;
143 	rb_link_node(&leaf->rb_node, parent, p);
144 	rb_insert_color(&leaf->rb_node, &info->msg_tree);
145 	info->qsize += sizeof(struct posix_msg_tree_node);
146 insert_msg:
147 	info->attr.mq_curmsgs++;
148 	info->qsize += msg->m_ts;
149 	list_add_tail(&msg->m_list, &leaf->msg_list);
150 	return 0;
151 }
152 
153 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
154 {
155 	struct rb_node **p, *parent = NULL;
156 	struct posix_msg_tree_node *leaf;
157 	struct msg_msg *msg;
158 
159 try_again:
160 	p = &info->msg_tree.rb_node;
161 	while (*p) {
162 		parent = *p;
163 		/*
164 		 * During insert, low priorities go to the left and high to the
165 		 * right.  On receive, we want the highest priorities first, so
166 		 * walk all the way to the right.
167 		 */
168 		p = &(*p)->rb_right;
169 	}
170 	if (!parent) {
171 		if (info->attr.mq_curmsgs) {
172 			pr_warn_once("Inconsistency in POSIX message queue, "
173 				     "no tree element, but supposedly messages "
174 				     "should exist!\n");
175 			info->attr.mq_curmsgs = 0;
176 		}
177 		return NULL;
178 	}
179 	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
180 	if (list_empty(&leaf->msg_list)) {
181 		pr_warn_once("Inconsistency in POSIX message queue, "
182 			     "empty leaf node but we haven't implemented "
183 			     "lazy leaf delete!\n");
184 		rb_erase(&leaf->rb_node, &info->msg_tree);
185 		info->qsize -= sizeof(struct posix_msg_tree_node);
186 		kfree(leaf);
187 		goto try_again;
188 	} else {
189 		msg = list_first_entry(&leaf->msg_list,
190 				       struct msg_msg, m_list);
191 		list_del(&msg->m_list);
192 		if (list_empty(&leaf->msg_list)) {
193 			rb_erase(&leaf->rb_node, &info->msg_tree);
194 			info->qsize -= sizeof(struct posix_msg_tree_node);
195 			kfree(leaf);
196 		}
197 	}
198 	info->attr.mq_curmsgs--;
199 	info->qsize -= msg->m_ts;
200 	return msg;
201 }
202 
203 static struct inode *mqueue_get_inode(struct super_block *sb,
204 		struct ipc_namespace *ipc_ns, umode_t mode,
205 		struct mq_attr *attr)
206 {
207 	struct user_struct *u = current_user();
208 	struct inode *inode;
209 	int ret = -ENOMEM;
210 
211 	inode = new_inode(sb);
212 	if (!inode)
213 		goto err;
214 
215 	inode->i_ino = get_next_ino();
216 	inode->i_mode = mode;
217 	inode->i_uid = current_fsuid();
218 	inode->i_gid = current_fsgid();
219 	inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
220 
221 	if (S_ISREG(mode)) {
222 		struct mqueue_inode_info *info;
223 		unsigned long mq_bytes, mq_treesize;
224 
225 		inode->i_fop = &mqueue_file_operations;
226 		inode->i_size = FILENT_SIZE;
227 		/* mqueue specific info */
228 		info = MQUEUE_I(inode);
229 		spin_lock_init(&info->lock);
230 		init_waitqueue_head(&info->wait_q);
231 		INIT_LIST_HEAD(&info->e_wait_q[0].list);
232 		INIT_LIST_HEAD(&info->e_wait_q[1].list);
233 		info->notify_owner = NULL;
234 		info->notify_user_ns = NULL;
235 		info->qsize = 0;
236 		info->user = NULL;	/* set when all is ok */
237 		info->msg_tree = RB_ROOT;
238 		memset(&info->attr, 0, sizeof(info->attr));
239 		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
240 					   ipc_ns->mq_msg_default);
241 		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
242 					    ipc_ns->mq_msgsize_default);
243 		if (attr) {
244 			info->attr.mq_maxmsg = attr->mq_maxmsg;
245 			info->attr.mq_msgsize = attr->mq_msgsize;
246 		}
247 		/*
248 		 * We used to allocate a static array of pointers and account
249 		 * the size of that array as well as one msg_msg struct per
250 		 * possible message into the queue size. That's no longer
251 		 * accurate as the queue is now an rbtree and will grow and
252 		 * shrink depending on usage patterns.  We can, however, still
253 		 * account one msg_msg struct per message, but the nodes are
254 		 * allocated depending on priority usage, and most programs
255 		 * only use one, or a handful, of priorities.  However, since
256 		 * this is pinned memory, we need to assume worst case, so
257 		 * that means the min(mq_maxmsg, max_priorities) * struct
258 		 * posix_msg_tree_node.
259 		 */
260 		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
261 			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
262 			sizeof(struct posix_msg_tree_node);
263 
264 		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
265 					  info->attr.mq_msgsize);
266 
267 		spin_lock(&mq_lock);
268 		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
269 		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
270 			spin_unlock(&mq_lock);
271 			/* mqueue_evict_inode() releases info->messages */
272 			ret = -EMFILE;
273 			goto out_inode;
274 		}
275 		u->mq_bytes += mq_bytes;
276 		spin_unlock(&mq_lock);
277 
278 		/* all is ok */
279 		info->user = get_uid(u);
280 	} else if (S_ISDIR(mode)) {
281 		inc_nlink(inode);
282 		/* Some things misbehave if size == 0 on a directory */
283 		inode->i_size = 2 * DIRENT_SIZE;
284 		inode->i_op = &mqueue_dir_inode_operations;
285 		inode->i_fop = &simple_dir_operations;
286 	}
287 
288 	return inode;
289 out_inode:
290 	iput(inode);
291 err:
292 	return ERR_PTR(ret);
293 }
294 
295 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
296 {
297 	struct inode *inode;
298 	struct ipc_namespace *ns = data;
299 
300 	sb->s_blocksize = PAGE_CACHE_SIZE;
301 	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
302 	sb->s_magic = MQUEUE_MAGIC;
303 	sb->s_op = &mqueue_super_ops;
304 
305 	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
306 	if (IS_ERR(inode))
307 		return PTR_ERR(inode);
308 
309 	sb->s_root = d_make_root(inode);
310 	if (!sb->s_root)
311 		return -ENOMEM;
312 	return 0;
313 }
314 
315 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
316 			 int flags, const char *dev_name,
317 			 void *data)
318 {
319 	if (!(flags & MS_KERNMOUNT))
320 		data = current->nsproxy->ipc_ns;
321 	return mount_ns(fs_type, flags, data, mqueue_fill_super);
322 }
323 
324 static void init_once(void *foo)
325 {
326 	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
327 
328 	inode_init_once(&p->vfs_inode);
329 }
330 
331 static struct inode *mqueue_alloc_inode(struct super_block *sb)
332 {
333 	struct mqueue_inode_info *ei;
334 
335 	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
336 	if (!ei)
337 		return NULL;
338 	return &ei->vfs_inode;
339 }
340 
341 static void mqueue_i_callback(struct rcu_head *head)
342 {
343 	struct inode *inode = container_of(head, struct inode, i_rcu);
344 	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
345 }
346 
347 static void mqueue_destroy_inode(struct inode *inode)
348 {
349 	call_rcu(&inode->i_rcu, mqueue_i_callback);
350 }
351 
352 static void mqueue_evict_inode(struct inode *inode)
353 {
354 	struct mqueue_inode_info *info;
355 	struct user_struct *user;
356 	unsigned long mq_bytes, mq_treesize;
357 	struct ipc_namespace *ipc_ns;
358 	struct msg_msg *msg;
359 
360 	clear_inode(inode);
361 
362 	if (S_ISDIR(inode->i_mode))
363 		return;
364 
365 	ipc_ns = get_ns_from_inode(inode);
366 	info = MQUEUE_I(inode);
367 	spin_lock(&info->lock);
368 	while ((msg = msg_get(info)) != NULL)
369 		free_msg(msg);
370 	spin_unlock(&info->lock);
371 
372 	/* Total amount of bytes accounted for the mqueue */
373 	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
374 		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
375 		sizeof(struct posix_msg_tree_node);
376 
377 	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
378 				  info->attr.mq_msgsize);
379 
380 	user = info->user;
381 	if (user) {
382 		spin_lock(&mq_lock);
383 		user->mq_bytes -= mq_bytes;
384 		/*
385 		 * get_ns_from_inode() ensures that the
386 		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
387 		 * to which we now hold a reference, or it is NULL.
388 		 * We can't put it here under mq_lock, though.
389 		 */
390 		if (ipc_ns)
391 			ipc_ns->mq_queues_count--;
392 		spin_unlock(&mq_lock);
393 		free_uid(user);
394 	}
395 	if (ipc_ns)
396 		put_ipc_ns(ipc_ns);
397 }
398 
399 static int mqueue_create(struct inode *dir, struct dentry *dentry,
400 				umode_t mode, struct nameidata *nd)
401 {
402 	struct inode *inode;
403 	struct mq_attr *attr = dentry->d_fsdata;
404 	int error;
405 	struct ipc_namespace *ipc_ns;
406 
407 	spin_lock(&mq_lock);
408 	ipc_ns = __get_ns_from_inode(dir);
409 	if (!ipc_ns) {
410 		error = -EACCES;
411 		goto out_unlock;
412 	}
413 	if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX ||
414 	    (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
415 	     !capable(CAP_SYS_RESOURCE))) {
416 		error = -ENOSPC;
417 		goto out_unlock;
418 	}
419 	ipc_ns->mq_queues_count++;
420 	spin_unlock(&mq_lock);
421 
422 	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
423 	if (IS_ERR(inode)) {
424 		error = PTR_ERR(inode);
425 		spin_lock(&mq_lock);
426 		ipc_ns->mq_queues_count--;
427 		goto out_unlock;
428 	}
429 
430 	put_ipc_ns(ipc_ns);
431 	dir->i_size += DIRENT_SIZE;
432 	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
433 
434 	d_instantiate(dentry, inode);
435 	dget(dentry);
436 	return 0;
437 out_unlock:
438 	spin_unlock(&mq_lock);
439 	if (ipc_ns)
440 		put_ipc_ns(ipc_ns);
441 	return error;
442 }
443 
444 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
445 {
446   	struct inode *inode = dentry->d_inode;
447 
448 	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
449 	dir->i_size -= DIRENT_SIZE;
450   	drop_nlink(inode);
451   	dput(dentry);
452   	return 0;
453 }
454 
455 /*
456 *	This is routine for system read from queue file.
457 *	To avoid mess with doing here some sort of mq_receive we allow
458 *	to read only queue size & notification info (the only values
459 *	that are interesting from user point of view and aren't accessible
460 *	through std routines)
461 */
462 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
463 				size_t count, loff_t *off)
464 {
465 	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
466 	char buffer[FILENT_SIZE];
467 	ssize_t ret;
468 
469 	spin_lock(&info->lock);
470 	snprintf(buffer, sizeof(buffer),
471 			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
472 			info->qsize,
473 			info->notify_owner ? info->notify.sigev_notify : 0,
474 			(info->notify_owner &&
475 			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
476 				info->notify.sigev_signo : 0,
477 			pid_vnr(info->notify_owner));
478 	spin_unlock(&info->lock);
479 	buffer[sizeof(buffer)-1] = '\0';
480 
481 	ret = simple_read_from_buffer(u_data, count, off, buffer,
482 				strlen(buffer));
483 	if (ret <= 0)
484 		return ret;
485 
486 	filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
487 	return ret;
488 }
489 
490 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
491 {
492 	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
493 
494 	spin_lock(&info->lock);
495 	if (task_tgid(current) == info->notify_owner)
496 		remove_notification(info);
497 
498 	spin_unlock(&info->lock);
499 	return 0;
500 }
501 
502 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
503 {
504 	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
505 	int retval = 0;
506 
507 	poll_wait(filp, &info->wait_q, poll_tab);
508 
509 	spin_lock(&info->lock);
510 	if (info->attr.mq_curmsgs)
511 		retval = POLLIN | POLLRDNORM;
512 
513 	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
514 		retval |= POLLOUT | POLLWRNORM;
515 	spin_unlock(&info->lock);
516 
517 	return retval;
518 }
519 
520 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
521 static void wq_add(struct mqueue_inode_info *info, int sr,
522 			struct ext_wait_queue *ewp)
523 {
524 	struct ext_wait_queue *walk;
525 
526 	ewp->task = current;
527 
528 	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
529 		if (walk->task->static_prio <= current->static_prio) {
530 			list_add_tail(&ewp->list, &walk->list);
531 			return;
532 		}
533 	}
534 	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
535 }
536 
537 /*
538  * Puts current task to sleep. Caller must hold queue lock. After return
539  * lock isn't held.
540  * sr: SEND or RECV
541  */
542 static int wq_sleep(struct mqueue_inode_info *info, int sr,
543 		    ktime_t *timeout, struct ext_wait_queue *ewp)
544 {
545 	int retval;
546 	signed long time;
547 
548 	wq_add(info, sr, ewp);
549 
550 	for (;;) {
551 		set_current_state(TASK_INTERRUPTIBLE);
552 
553 		spin_unlock(&info->lock);
554 		time = schedule_hrtimeout_range_clock(timeout, 0,
555 			HRTIMER_MODE_ABS, CLOCK_REALTIME);
556 
557 		while (ewp->state == STATE_PENDING)
558 			cpu_relax();
559 
560 		if (ewp->state == STATE_READY) {
561 			retval = 0;
562 			goto out;
563 		}
564 		spin_lock(&info->lock);
565 		if (ewp->state == STATE_READY) {
566 			retval = 0;
567 			goto out_unlock;
568 		}
569 		if (signal_pending(current)) {
570 			retval = -ERESTARTSYS;
571 			break;
572 		}
573 		if (time == 0) {
574 			retval = -ETIMEDOUT;
575 			break;
576 		}
577 	}
578 	list_del(&ewp->list);
579 out_unlock:
580 	spin_unlock(&info->lock);
581 out:
582 	return retval;
583 }
584 
585 /*
586  * Returns waiting task that should be serviced first or NULL if none exists
587  */
588 static struct ext_wait_queue *wq_get_first_waiter(
589 		struct mqueue_inode_info *info, int sr)
590 {
591 	struct list_head *ptr;
592 
593 	ptr = info->e_wait_q[sr].list.prev;
594 	if (ptr == &info->e_wait_q[sr].list)
595 		return NULL;
596 	return list_entry(ptr, struct ext_wait_queue, list);
597 }
598 
599 
600 static inline void set_cookie(struct sk_buff *skb, char code)
601 {
602 	((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
603 }
604 
605 /*
606  * The next function is only to split too long sys_mq_timedsend
607  */
608 static void __do_notify(struct mqueue_inode_info *info)
609 {
610 	/* notification
611 	 * invoked when there is registered process and there isn't process
612 	 * waiting synchronously for message AND state of queue changed from
613 	 * empty to not empty. Here we are sure that no one is waiting
614 	 * synchronously. */
615 	if (info->notify_owner &&
616 	    info->attr.mq_curmsgs == 1) {
617 		struct siginfo sig_i;
618 		switch (info->notify.sigev_notify) {
619 		case SIGEV_NONE:
620 			break;
621 		case SIGEV_SIGNAL:
622 			/* sends signal */
623 
624 			sig_i.si_signo = info->notify.sigev_signo;
625 			sig_i.si_errno = 0;
626 			sig_i.si_code = SI_MESGQ;
627 			sig_i.si_value = info->notify.sigev_value;
628 			/* map current pid/uid into info->owner's namespaces */
629 			rcu_read_lock();
630 			sig_i.si_pid = task_tgid_nr_ns(current,
631 						ns_of_pid(info->notify_owner));
632 			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
633 			rcu_read_unlock();
634 
635 			kill_pid_info(info->notify.sigev_signo,
636 				      &sig_i, info->notify_owner);
637 			break;
638 		case SIGEV_THREAD:
639 			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
640 			netlink_sendskb(info->notify_sock, info->notify_cookie);
641 			break;
642 		}
643 		/* after notification unregisters process */
644 		put_pid(info->notify_owner);
645 		put_user_ns(info->notify_user_ns);
646 		info->notify_owner = NULL;
647 		info->notify_user_ns = NULL;
648 	}
649 	wake_up(&info->wait_q);
650 }
651 
652 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
653 			   ktime_t *expires, struct timespec *ts)
654 {
655 	if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
656 		return -EFAULT;
657 	if (!timespec_valid(ts))
658 		return -EINVAL;
659 
660 	*expires = timespec_to_ktime(*ts);
661 	return 0;
662 }
663 
664 static void remove_notification(struct mqueue_inode_info *info)
665 {
666 	if (info->notify_owner != NULL &&
667 	    info->notify.sigev_notify == SIGEV_THREAD) {
668 		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
669 		netlink_sendskb(info->notify_sock, info->notify_cookie);
670 	}
671 	put_pid(info->notify_owner);
672 	put_user_ns(info->notify_user_ns);
673 	info->notify_owner = NULL;
674 	info->notify_user_ns = NULL;
675 }
676 
677 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
678 {
679 	int mq_treesize;
680 	unsigned long total_size;
681 
682 	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
683 		return -EINVAL;
684 	if (capable(CAP_SYS_RESOURCE)) {
685 		if (attr->mq_maxmsg > HARD_MSGMAX ||
686 		    attr->mq_msgsize > HARD_MSGSIZEMAX)
687 			return -EINVAL;
688 	} else {
689 		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
690 				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
691 			return -EINVAL;
692 	}
693 	/* check for overflow */
694 	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
695 		return -EOVERFLOW;
696 	mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
697 		min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
698 		sizeof(struct posix_msg_tree_node);
699 	total_size = attr->mq_maxmsg * attr->mq_msgsize;
700 	if (total_size + mq_treesize < total_size)
701 		return -EOVERFLOW;
702 	return 0;
703 }
704 
705 /*
706  * Invoked when creating a new queue via sys_mq_open
707  */
708 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
709 			struct dentry *dentry, int oflag, umode_t mode,
710 			struct mq_attr *attr)
711 {
712 	const struct cred *cred = current_cred();
713 	struct file *result;
714 	int ret;
715 
716 	if (attr) {
717 		ret = mq_attr_ok(ipc_ns, attr);
718 		if (ret)
719 			goto out;
720 		/* store for use during create */
721 		dentry->d_fsdata = attr;
722 	} else {
723 		struct mq_attr def_attr;
724 
725 		def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
726 					 ipc_ns->mq_msg_default);
727 		def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
728 					  ipc_ns->mq_msgsize_default);
729 		ret = mq_attr_ok(ipc_ns, &def_attr);
730 		if (ret)
731 			goto out;
732 	}
733 
734 	mode &= ~current_umask();
735 	ret = mnt_want_write(ipc_ns->mq_mnt);
736 	if (ret)
737 		goto out;
738 	ret = vfs_create(dir->d_inode, dentry, mode, NULL);
739 	dentry->d_fsdata = NULL;
740 	if (ret)
741 		goto out_drop_write;
742 
743 	result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
744 	/*
745 	 * dentry_open() took a persistent mnt_want_write(),
746 	 * so we can now drop this one.
747 	 */
748 	mnt_drop_write(ipc_ns->mq_mnt);
749 	return result;
750 
751 out_drop_write:
752 	mnt_drop_write(ipc_ns->mq_mnt);
753 out:
754 	dput(dentry);
755 	mntput(ipc_ns->mq_mnt);
756 	return ERR_PTR(ret);
757 }
758 
759 /* Opens existing queue */
760 static struct file *do_open(struct ipc_namespace *ipc_ns,
761 				struct dentry *dentry, int oflag)
762 {
763 	int ret;
764 	const struct cred *cred = current_cred();
765 
766 	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
767 						  MAY_READ | MAY_WRITE };
768 
769 	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
770 		ret = -EINVAL;
771 		goto err;
772 	}
773 
774 	if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
775 		ret = -EACCES;
776 		goto err;
777 	}
778 
779 	return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
780 
781 err:
782 	dput(dentry);
783 	mntput(ipc_ns->mq_mnt);
784 	return ERR_PTR(ret);
785 }
786 
787 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
788 		struct mq_attr __user *, u_attr)
789 {
790 	struct dentry *dentry;
791 	struct file *filp;
792 	char *name;
793 	struct mq_attr attr;
794 	int fd, error;
795 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
796 
797 	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
798 		return -EFAULT;
799 
800 	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
801 
802 	if (IS_ERR(name = getname(u_name)))
803 		return PTR_ERR(name);
804 
805 	fd = get_unused_fd_flags(O_CLOEXEC);
806 	if (fd < 0)
807 		goto out_putname;
808 
809 	mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
810 	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
811 	if (IS_ERR(dentry)) {
812 		error = PTR_ERR(dentry);
813 		goto out_putfd;
814 	}
815 	mntget(ipc_ns->mq_mnt);
816 
817 	if (oflag & O_CREAT) {
818 		if (dentry->d_inode) {	/* entry already exists */
819 			audit_inode(name, dentry);
820 			if (oflag & O_EXCL) {
821 				error = -EEXIST;
822 				goto out;
823 			}
824 			filp = do_open(ipc_ns, dentry, oflag);
825 		} else {
826 			filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
827 						dentry, oflag, mode,
828 						u_attr ? &attr : NULL);
829 		}
830 	} else {
831 		if (!dentry->d_inode) {
832 			error = -ENOENT;
833 			goto out;
834 		}
835 		audit_inode(name, dentry);
836 		filp = do_open(ipc_ns, dentry, oflag);
837 	}
838 
839 	if (IS_ERR(filp)) {
840 		error = PTR_ERR(filp);
841 		goto out_putfd;
842 	}
843 
844 	fd_install(fd, filp);
845 	goto out_upsem;
846 
847 out:
848 	dput(dentry);
849 	mntput(ipc_ns->mq_mnt);
850 out_putfd:
851 	put_unused_fd(fd);
852 	fd = error;
853 out_upsem:
854 	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
855 out_putname:
856 	putname(name);
857 	return fd;
858 }
859 
860 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
861 {
862 	int err;
863 	char *name;
864 	struct dentry *dentry;
865 	struct inode *inode = NULL;
866 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
867 
868 	name = getname(u_name);
869 	if (IS_ERR(name))
870 		return PTR_ERR(name);
871 
872 	mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
873 			I_MUTEX_PARENT);
874 	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
875 	if (IS_ERR(dentry)) {
876 		err = PTR_ERR(dentry);
877 		goto out_unlock;
878 	}
879 
880 	if (!dentry->d_inode) {
881 		err = -ENOENT;
882 		goto out_err;
883 	}
884 
885 	inode = dentry->d_inode;
886 	if (inode)
887 		ihold(inode);
888 	err = mnt_want_write(ipc_ns->mq_mnt);
889 	if (err)
890 		goto out_err;
891 	err = vfs_unlink(dentry->d_parent->d_inode, dentry);
892 	mnt_drop_write(ipc_ns->mq_mnt);
893 out_err:
894 	dput(dentry);
895 
896 out_unlock:
897 	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
898 	putname(name);
899 	if (inode)
900 		iput(inode);
901 
902 	return err;
903 }
904 
905 /* Pipelined send and receive functions.
906  *
907  * If a receiver finds no waiting message, then it registers itself in the
908  * list of waiting receivers. A sender checks that list before adding the new
909  * message into the message array. If there is a waiting receiver, then it
910  * bypasses the message array and directly hands the message over to the
911  * receiver.
912  * The receiver accepts the message and returns without grabbing the queue
913  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
914  * are necessary. The same algorithm is used for sysv semaphores, see
915  * ipc/sem.c for more details.
916  *
917  * The same algorithm is used for senders.
918  */
919 
920 /* pipelined_send() - send a message directly to the task waiting in
921  * sys_mq_timedreceive() (without inserting message into a queue).
922  */
923 static inline void pipelined_send(struct mqueue_inode_info *info,
924 				  struct msg_msg *message,
925 				  struct ext_wait_queue *receiver)
926 {
927 	receiver->msg = message;
928 	list_del(&receiver->list);
929 	receiver->state = STATE_PENDING;
930 	wake_up_process(receiver->task);
931 	smp_wmb();
932 	receiver->state = STATE_READY;
933 }
934 
935 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
936  * gets its message and put to the queue (we have one free place for sure). */
937 static inline void pipelined_receive(struct mqueue_inode_info *info)
938 {
939 	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
940 
941 	if (!sender) {
942 		/* for poll */
943 		wake_up_interruptible(&info->wait_q);
944 		return;
945 	}
946 	if (msg_insert(sender->msg, info))
947 		return;
948 	list_del(&sender->list);
949 	sender->state = STATE_PENDING;
950 	wake_up_process(sender->task);
951 	smp_wmb();
952 	sender->state = STATE_READY;
953 }
954 
955 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
956 		size_t, msg_len, unsigned int, msg_prio,
957 		const struct timespec __user *, u_abs_timeout)
958 {
959 	struct file *filp;
960 	struct inode *inode;
961 	struct ext_wait_queue wait;
962 	struct ext_wait_queue *receiver;
963 	struct msg_msg *msg_ptr;
964 	struct mqueue_inode_info *info;
965 	ktime_t expires, *timeout = NULL;
966 	struct timespec ts;
967 	int ret;
968 
969 	if (u_abs_timeout) {
970 		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
971 		if (res)
972 			return res;
973 		timeout = &expires;
974 	}
975 
976 	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
977 		return -EINVAL;
978 
979 	audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
980 
981 	filp = fget(mqdes);
982 	if (unlikely(!filp)) {
983 		ret = -EBADF;
984 		goto out;
985 	}
986 
987 	inode = filp->f_path.dentry->d_inode;
988 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
989 		ret = -EBADF;
990 		goto out_fput;
991 	}
992 	info = MQUEUE_I(inode);
993 	audit_inode(NULL, filp->f_path.dentry);
994 
995 	if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
996 		ret = -EBADF;
997 		goto out_fput;
998 	}
999 
1000 	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1001 		ret = -EMSGSIZE;
1002 		goto out_fput;
1003 	}
1004 
1005 	/* First try to allocate memory, before doing anything with
1006 	 * existing queues. */
1007 	msg_ptr = load_msg(u_msg_ptr, msg_len);
1008 	if (IS_ERR(msg_ptr)) {
1009 		ret = PTR_ERR(msg_ptr);
1010 		goto out_fput;
1011 	}
1012 	msg_ptr->m_ts = msg_len;
1013 	msg_ptr->m_type = msg_prio;
1014 
1015 	spin_lock(&info->lock);
1016 
1017 	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1018 		if (filp->f_flags & O_NONBLOCK) {
1019 			spin_unlock(&info->lock);
1020 			ret = -EAGAIN;
1021 		} else {
1022 			wait.task = current;
1023 			wait.msg = (void *) msg_ptr;
1024 			wait.state = STATE_NONE;
1025 			ret = wq_sleep(info, SEND, timeout, &wait);
1026 		}
1027 		if (ret < 0)
1028 			free_msg(msg_ptr);
1029 	} else {
1030 		receiver = wq_get_first_waiter(info, RECV);
1031 		if (receiver) {
1032 			pipelined_send(info, msg_ptr, receiver);
1033 		} else {
1034 			/* adds message to the queue */
1035 			if (msg_insert(msg_ptr, info)) {
1036 				free_msg(msg_ptr);
1037 				ret = -ENOMEM;
1038 				spin_unlock(&info->lock);
1039 				goto out_fput;
1040 			}
1041 			__do_notify(info);
1042 		}
1043 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1044 				CURRENT_TIME;
1045 		spin_unlock(&info->lock);
1046 		ret = 0;
1047 	}
1048 out_fput:
1049 	fput(filp);
1050 out:
1051 	return ret;
1052 }
1053 
1054 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1055 		size_t, msg_len, unsigned int __user *, u_msg_prio,
1056 		const struct timespec __user *, u_abs_timeout)
1057 {
1058 	ssize_t ret;
1059 	struct msg_msg *msg_ptr;
1060 	struct file *filp;
1061 	struct inode *inode;
1062 	struct mqueue_inode_info *info;
1063 	struct ext_wait_queue wait;
1064 	ktime_t expires, *timeout = NULL;
1065 	struct timespec ts;
1066 
1067 	if (u_abs_timeout) {
1068 		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1069 		if (res)
1070 			return res;
1071 		timeout = &expires;
1072 	}
1073 
1074 	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1075 
1076 	filp = fget(mqdes);
1077 	if (unlikely(!filp)) {
1078 		ret = -EBADF;
1079 		goto out;
1080 	}
1081 
1082 	inode = filp->f_path.dentry->d_inode;
1083 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1084 		ret = -EBADF;
1085 		goto out_fput;
1086 	}
1087 	info = MQUEUE_I(inode);
1088 	audit_inode(NULL, filp->f_path.dentry);
1089 
1090 	if (unlikely(!(filp->f_mode & FMODE_READ))) {
1091 		ret = -EBADF;
1092 		goto out_fput;
1093 	}
1094 
1095 	/* checks if buffer is big enough */
1096 	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1097 		ret = -EMSGSIZE;
1098 		goto out_fput;
1099 	}
1100 
1101 	spin_lock(&info->lock);
1102 	if (info->attr.mq_curmsgs == 0) {
1103 		if (filp->f_flags & O_NONBLOCK) {
1104 			spin_unlock(&info->lock);
1105 			ret = -EAGAIN;
1106 		} else {
1107 			wait.task = current;
1108 			wait.state = STATE_NONE;
1109 			ret = wq_sleep(info, RECV, timeout, &wait);
1110 			msg_ptr = wait.msg;
1111 		}
1112 	} else {
1113 		msg_ptr = msg_get(info);
1114 
1115 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1116 				CURRENT_TIME;
1117 
1118 		/* There is now free space in queue. */
1119 		pipelined_receive(info);
1120 		spin_unlock(&info->lock);
1121 		ret = 0;
1122 	}
1123 	if (ret == 0) {
1124 		ret = msg_ptr->m_ts;
1125 
1126 		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1127 			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1128 			ret = -EFAULT;
1129 		}
1130 		free_msg(msg_ptr);
1131 	}
1132 out_fput:
1133 	fput(filp);
1134 out:
1135 	return ret;
1136 }
1137 
1138 /*
1139  * Notes: the case when user wants us to deregister (with NULL as pointer)
1140  * and he isn't currently owner of notification, will be silently discarded.
1141  * It isn't explicitly defined in the POSIX.
1142  */
1143 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1144 		const struct sigevent __user *, u_notification)
1145 {
1146 	int ret;
1147 	struct file *filp;
1148 	struct sock *sock;
1149 	struct inode *inode;
1150 	struct sigevent notification;
1151 	struct mqueue_inode_info *info;
1152 	struct sk_buff *nc;
1153 
1154 	if (u_notification) {
1155 		if (copy_from_user(&notification, u_notification,
1156 					sizeof(struct sigevent)))
1157 			return -EFAULT;
1158 	}
1159 
1160 	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1161 
1162 	nc = NULL;
1163 	sock = NULL;
1164 	if (u_notification != NULL) {
1165 		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1166 			     notification.sigev_notify != SIGEV_SIGNAL &&
1167 			     notification.sigev_notify != SIGEV_THREAD))
1168 			return -EINVAL;
1169 		if (notification.sigev_notify == SIGEV_SIGNAL &&
1170 			!valid_signal(notification.sigev_signo)) {
1171 			return -EINVAL;
1172 		}
1173 		if (notification.sigev_notify == SIGEV_THREAD) {
1174 			long timeo;
1175 
1176 			/* create the notify skb */
1177 			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1178 			if (!nc) {
1179 				ret = -ENOMEM;
1180 				goto out;
1181 			}
1182 			if (copy_from_user(nc->data,
1183 					notification.sigev_value.sival_ptr,
1184 					NOTIFY_COOKIE_LEN)) {
1185 				ret = -EFAULT;
1186 				goto out;
1187 			}
1188 
1189 			/* TODO: add a header? */
1190 			skb_put(nc, NOTIFY_COOKIE_LEN);
1191 			/* and attach it to the socket */
1192 retry:
1193 			filp = fget(notification.sigev_signo);
1194 			if (!filp) {
1195 				ret = -EBADF;
1196 				goto out;
1197 			}
1198 			sock = netlink_getsockbyfilp(filp);
1199 			fput(filp);
1200 			if (IS_ERR(sock)) {
1201 				ret = PTR_ERR(sock);
1202 				sock = NULL;
1203 				goto out;
1204 			}
1205 
1206 			timeo = MAX_SCHEDULE_TIMEOUT;
1207 			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1208 			if (ret == 1)
1209 				goto retry;
1210 			if (ret) {
1211 				sock = NULL;
1212 				nc = NULL;
1213 				goto out;
1214 			}
1215 		}
1216 	}
1217 
1218 	filp = fget(mqdes);
1219 	if (!filp) {
1220 		ret = -EBADF;
1221 		goto out;
1222 	}
1223 
1224 	inode = filp->f_path.dentry->d_inode;
1225 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1226 		ret = -EBADF;
1227 		goto out_fput;
1228 	}
1229 	info = MQUEUE_I(inode);
1230 
1231 	ret = 0;
1232 	spin_lock(&info->lock);
1233 	if (u_notification == NULL) {
1234 		if (info->notify_owner == task_tgid(current)) {
1235 			remove_notification(info);
1236 			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1237 		}
1238 	} else if (info->notify_owner != NULL) {
1239 		ret = -EBUSY;
1240 	} else {
1241 		switch (notification.sigev_notify) {
1242 		case SIGEV_NONE:
1243 			info->notify.sigev_notify = SIGEV_NONE;
1244 			break;
1245 		case SIGEV_THREAD:
1246 			info->notify_sock = sock;
1247 			info->notify_cookie = nc;
1248 			sock = NULL;
1249 			nc = NULL;
1250 			info->notify.sigev_notify = SIGEV_THREAD;
1251 			break;
1252 		case SIGEV_SIGNAL:
1253 			info->notify.sigev_signo = notification.sigev_signo;
1254 			info->notify.sigev_value = notification.sigev_value;
1255 			info->notify.sigev_notify = SIGEV_SIGNAL;
1256 			break;
1257 		}
1258 
1259 		info->notify_owner = get_pid(task_tgid(current));
1260 		info->notify_user_ns = get_user_ns(current_user_ns());
1261 		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1262 	}
1263 	spin_unlock(&info->lock);
1264 out_fput:
1265 	fput(filp);
1266 out:
1267 	if (sock) {
1268 		netlink_detachskb(sock, nc);
1269 	} else if (nc) {
1270 		dev_kfree_skb(nc);
1271 	}
1272 	return ret;
1273 }
1274 
1275 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1276 		const struct mq_attr __user *, u_mqstat,
1277 		struct mq_attr __user *, u_omqstat)
1278 {
1279 	int ret;
1280 	struct mq_attr mqstat, omqstat;
1281 	struct file *filp;
1282 	struct inode *inode;
1283 	struct mqueue_inode_info *info;
1284 
1285 	if (u_mqstat != NULL) {
1286 		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1287 			return -EFAULT;
1288 		if (mqstat.mq_flags & (~O_NONBLOCK))
1289 			return -EINVAL;
1290 	}
1291 
1292 	filp = fget(mqdes);
1293 	if (!filp) {
1294 		ret = -EBADF;
1295 		goto out;
1296 	}
1297 
1298 	inode = filp->f_path.dentry->d_inode;
1299 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1300 		ret = -EBADF;
1301 		goto out_fput;
1302 	}
1303 	info = MQUEUE_I(inode);
1304 
1305 	spin_lock(&info->lock);
1306 
1307 	omqstat = info->attr;
1308 	omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1309 	if (u_mqstat) {
1310 		audit_mq_getsetattr(mqdes, &mqstat);
1311 		spin_lock(&filp->f_lock);
1312 		if (mqstat.mq_flags & O_NONBLOCK)
1313 			filp->f_flags |= O_NONBLOCK;
1314 		else
1315 			filp->f_flags &= ~O_NONBLOCK;
1316 		spin_unlock(&filp->f_lock);
1317 
1318 		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1319 	}
1320 
1321 	spin_unlock(&info->lock);
1322 
1323 	ret = 0;
1324 	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1325 						sizeof(struct mq_attr)))
1326 		ret = -EFAULT;
1327 
1328 out_fput:
1329 	fput(filp);
1330 out:
1331 	return ret;
1332 }
1333 
1334 static const struct inode_operations mqueue_dir_inode_operations = {
1335 	.lookup = simple_lookup,
1336 	.create = mqueue_create,
1337 	.unlink = mqueue_unlink,
1338 };
1339 
1340 static const struct file_operations mqueue_file_operations = {
1341 	.flush = mqueue_flush_file,
1342 	.poll = mqueue_poll_file,
1343 	.read = mqueue_read_file,
1344 	.llseek = default_llseek,
1345 };
1346 
1347 static const struct super_operations mqueue_super_ops = {
1348 	.alloc_inode = mqueue_alloc_inode,
1349 	.destroy_inode = mqueue_destroy_inode,
1350 	.evict_inode = mqueue_evict_inode,
1351 	.statfs = simple_statfs,
1352 };
1353 
1354 static struct file_system_type mqueue_fs_type = {
1355 	.name = "mqueue",
1356 	.mount = mqueue_mount,
1357 	.kill_sb = kill_litter_super,
1358 };
1359 
1360 int mq_init_ns(struct ipc_namespace *ns)
1361 {
1362 	ns->mq_queues_count  = 0;
1363 	ns->mq_queues_max    = DFLT_QUEUESMAX;
1364 	ns->mq_msg_max       = DFLT_MSGMAX;
1365 	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1366 	ns->mq_msg_default   = DFLT_MSG;
1367 	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1368 
1369 	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1370 	if (IS_ERR(ns->mq_mnt)) {
1371 		int err = PTR_ERR(ns->mq_mnt);
1372 		ns->mq_mnt = NULL;
1373 		return err;
1374 	}
1375 	return 0;
1376 }
1377 
1378 void mq_clear_sbinfo(struct ipc_namespace *ns)
1379 {
1380 	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1381 }
1382 
1383 void mq_put_mnt(struct ipc_namespace *ns)
1384 {
1385 	kern_unmount(ns->mq_mnt);
1386 }
1387 
1388 static int __init init_mqueue_fs(void)
1389 {
1390 	int error;
1391 
1392 	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1393 				sizeof(struct mqueue_inode_info), 0,
1394 				SLAB_HWCACHE_ALIGN, init_once);
1395 	if (mqueue_inode_cachep == NULL)
1396 		return -ENOMEM;
1397 
1398 	/* ignore failures - they are not fatal */
1399 	mq_sysctl_table = mq_register_sysctl_table();
1400 
1401 	error = register_filesystem(&mqueue_fs_type);
1402 	if (error)
1403 		goto out_sysctl;
1404 
1405 	spin_lock_init(&mq_lock);
1406 
1407 	error = mq_init_ns(&init_ipc_ns);
1408 	if (error)
1409 		goto out_filesystem;
1410 
1411 	return 0;
1412 
1413 out_filesystem:
1414 	unregister_filesystem(&mqueue_fs_type);
1415 out_sysctl:
1416 	if (mq_sysctl_table)
1417 		unregister_sysctl_table(mq_sysctl_table);
1418 	kmem_cache_destroy(mqueue_inode_cachep);
1419 	return error;
1420 }
1421 
1422 __initcall(init_mqueue_fs);
1423