xref: /openbmc/linux/ipc/mqueue.c (revision 77d84ff8)
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 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 
88 static const struct inode_operations mqueue_dir_inode_operations;
89 static const struct file_operations mqueue_file_operations;
90 static const struct super_operations mqueue_super_ops;
91 static void remove_notification(struct mqueue_inode_info *info);
92 
93 static struct kmem_cache *mqueue_inode_cachep;
94 
95 static struct ctl_table_header * mq_sysctl_table;
96 
97 static 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  */
105 static 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 
110 static 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 */
121 static 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 		info->qsize += sizeof(*leaf);
147 	}
148 	leaf->priority = msg->m_type;
149 	rb_link_node(&leaf->rb_node, parent, p);
150 	rb_insert_color(&leaf->rb_node, &info->msg_tree);
151 insert_msg:
152 	info->attr.mq_curmsgs++;
153 	info->qsize += msg->m_ts;
154 	list_add_tail(&msg->m_list, &leaf->msg_list);
155 	return 0;
156 }
157 
158 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
159 {
160 	struct rb_node **p, *parent = NULL;
161 	struct posix_msg_tree_node *leaf;
162 	struct msg_msg *msg;
163 
164 try_again:
165 	p = &info->msg_tree.rb_node;
166 	while (*p) {
167 		parent = *p;
168 		/*
169 		 * During insert, low priorities go to the left and high to the
170 		 * right.  On receive, we want the highest priorities first, so
171 		 * walk all the way to the right.
172 		 */
173 		p = &(*p)->rb_right;
174 	}
175 	if (!parent) {
176 		if (info->attr.mq_curmsgs) {
177 			pr_warn_once("Inconsistency in POSIX message queue, "
178 				     "no tree element, but supposedly messages "
179 				     "should exist!\n");
180 			info->attr.mq_curmsgs = 0;
181 		}
182 		return NULL;
183 	}
184 	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
185 	if (unlikely(list_empty(&leaf->msg_list))) {
186 		pr_warn_once("Inconsistency in POSIX message queue, "
187 			     "empty leaf node but we haven't implemented "
188 			     "lazy leaf delete!\n");
189 		rb_erase(&leaf->rb_node, &info->msg_tree);
190 		if (info->node_cache) {
191 			info->qsize -= sizeof(*leaf);
192 			kfree(leaf);
193 		} else {
194 			info->node_cache = leaf;
195 		}
196 		goto try_again;
197 	} else {
198 		msg = list_first_entry(&leaf->msg_list,
199 				       struct msg_msg, m_list);
200 		list_del(&msg->m_list);
201 		if (list_empty(&leaf->msg_list)) {
202 			rb_erase(&leaf->rb_node, &info->msg_tree);
203 			if (info->node_cache) {
204 				info->qsize -= sizeof(*leaf);
205 				kfree(leaf);
206 			} else {
207 				info->node_cache = leaf;
208 			}
209 		}
210 	}
211 	info->attr.mq_curmsgs--;
212 	info->qsize -= msg->m_ts;
213 	return msg;
214 }
215 
216 static struct inode *mqueue_get_inode(struct super_block *sb,
217 		struct ipc_namespace *ipc_ns, umode_t mode,
218 		struct mq_attr *attr)
219 {
220 	struct user_struct *u = current_user();
221 	struct inode *inode;
222 	int ret = -ENOMEM;
223 
224 	inode = new_inode(sb);
225 	if (!inode)
226 		goto err;
227 
228 	inode->i_ino = get_next_ino();
229 	inode->i_mode = mode;
230 	inode->i_uid = current_fsuid();
231 	inode->i_gid = current_fsgid();
232 	inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
233 
234 	if (S_ISREG(mode)) {
235 		struct mqueue_inode_info *info;
236 		unsigned long mq_bytes, mq_treesize;
237 
238 		inode->i_fop = &mqueue_file_operations;
239 		inode->i_size = FILENT_SIZE;
240 		/* mqueue specific info */
241 		info = MQUEUE_I(inode);
242 		spin_lock_init(&info->lock);
243 		init_waitqueue_head(&info->wait_q);
244 		INIT_LIST_HEAD(&info->e_wait_q[0].list);
245 		INIT_LIST_HEAD(&info->e_wait_q[1].list);
246 		info->notify_owner = NULL;
247 		info->notify_user_ns = NULL;
248 		info->qsize = 0;
249 		info->user = NULL;	/* set when all is ok */
250 		info->msg_tree = RB_ROOT;
251 		info->node_cache = NULL;
252 		memset(&info->attr, 0, sizeof(info->attr));
253 		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
254 					   ipc_ns->mq_msg_default);
255 		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
256 					    ipc_ns->mq_msgsize_default);
257 		if (attr) {
258 			info->attr.mq_maxmsg = attr->mq_maxmsg;
259 			info->attr.mq_msgsize = attr->mq_msgsize;
260 		}
261 		/*
262 		 * We used to allocate a static array of pointers and account
263 		 * the size of that array as well as one msg_msg struct per
264 		 * possible message into the queue size. That's no longer
265 		 * accurate as the queue is now an rbtree and will grow and
266 		 * shrink depending on usage patterns.  We can, however, still
267 		 * account one msg_msg struct per message, but the nodes are
268 		 * allocated depending on priority usage, and most programs
269 		 * only use one, or a handful, of priorities.  However, since
270 		 * this is pinned memory, we need to assume worst case, so
271 		 * that means the min(mq_maxmsg, max_priorities) * struct
272 		 * posix_msg_tree_node.
273 		 */
274 		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
275 			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
276 			sizeof(struct posix_msg_tree_node);
277 
278 		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
279 					  info->attr.mq_msgsize);
280 
281 		spin_lock(&mq_lock);
282 		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
283 		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
284 			spin_unlock(&mq_lock);
285 			/* mqueue_evict_inode() releases info->messages */
286 			ret = -EMFILE;
287 			goto out_inode;
288 		}
289 		u->mq_bytes += mq_bytes;
290 		spin_unlock(&mq_lock);
291 
292 		/* all is ok */
293 		info->user = get_uid(u);
294 	} else if (S_ISDIR(mode)) {
295 		inc_nlink(inode);
296 		/* Some things misbehave if size == 0 on a directory */
297 		inode->i_size = 2 * DIRENT_SIZE;
298 		inode->i_op = &mqueue_dir_inode_operations;
299 		inode->i_fop = &simple_dir_operations;
300 	}
301 
302 	return inode;
303 out_inode:
304 	iput(inode);
305 err:
306 	return ERR_PTR(ret);
307 }
308 
309 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
310 {
311 	struct inode *inode;
312 	struct ipc_namespace *ns = data;
313 
314 	sb->s_blocksize = PAGE_CACHE_SIZE;
315 	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
316 	sb->s_magic = MQUEUE_MAGIC;
317 	sb->s_op = &mqueue_super_ops;
318 
319 	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
320 	if (IS_ERR(inode))
321 		return PTR_ERR(inode);
322 
323 	sb->s_root = d_make_root(inode);
324 	if (!sb->s_root)
325 		return -ENOMEM;
326 	return 0;
327 }
328 
329 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
330 			 int flags, const char *dev_name,
331 			 void *data)
332 {
333 	if (!(flags & MS_KERNMOUNT)) {
334 		struct ipc_namespace *ns = current->nsproxy->ipc_ns;
335 		/* Don't allow mounting unless the caller has CAP_SYS_ADMIN
336 		 * over the ipc namespace.
337 		 */
338 		if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
339 			return ERR_PTR(-EPERM);
340 
341 		data = ns;
342 	}
343 	return mount_ns(fs_type, flags, data, mqueue_fill_super);
344 }
345 
346 static void init_once(void *foo)
347 {
348 	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
349 
350 	inode_init_once(&p->vfs_inode);
351 }
352 
353 static struct inode *mqueue_alloc_inode(struct super_block *sb)
354 {
355 	struct mqueue_inode_info *ei;
356 
357 	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
358 	if (!ei)
359 		return NULL;
360 	return &ei->vfs_inode;
361 }
362 
363 static void mqueue_i_callback(struct rcu_head *head)
364 {
365 	struct inode *inode = container_of(head, struct inode, i_rcu);
366 	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
367 }
368 
369 static void mqueue_destroy_inode(struct inode *inode)
370 {
371 	call_rcu(&inode->i_rcu, mqueue_i_callback);
372 }
373 
374 static void mqueue_evict_inode(struct inode *inode)
375 {
376 	struct mqueue_inode_info *info;
377 	struct user_struct *user;
378 	unsigned long mq_bytes, mq_treesize;
379 	struct ipc_namespace *ipc_ns;
380 	struct msg_msg *msg;
381 
382 	clear_inode(inode);
383 
384 	if (S_ISDIR(inode->i_mode))
385 		return;
386 
387 	ipc_ns = get_ns_from_inode(inode);
388 	info = MQUEUE_I(inode);
389 	spin_lock(&info->lock);
390 	while ((msg = msg_get(info)) != NULL)
391 		free_msg(msg);
392 	kfree(info->node_cache);
393 	spin_unlock(&info->lock);
394 
395 	/* Total amount of bytes accounted for the mqueue */
396 	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
397 		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
398 		sizeof(struct posix_msg_tree_node);
399 
400 	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
401 				  info->attr.mq_msgsize);
402 
403 	user = info->user;
404 	if (user) {
405 		spin_lock(&mq_lock);
406 		user->mq_bytes -= mq_bytes;
407 		/*
408 		 * get_ns_from_inode() ensures that the
409 		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
410 		 * to which we now hold a reference, or it is NULL.
411 		 * We can't put it here under mq_lock, though.
412 		 */
413 		if (ipc_ns)
414 			ipc_ns->mq_queues_count--;
415 		spin_unlock(&mq_lock);
416 		free_uid(user);
417 	}
418 	if (ipc_ns)
419 		put_ipc_ns(ipc_ns);
420 }
421 
422 static int mqueue_create(struct inode *dir, struct dentry *dentry,
423 				umode_t mode, bool excl)
424 {
425 	struct inode *inode;
426 	struct mq_attr *attr = dentry->d_fsdata;
427 	int error;
428 	struct ipc_namespace *ipc_ns;
429 
430 	spin_lock(&mq_lock);
431 	ipc_ns = __get_ns_from_inode(dir);
432 	if (!ipc_ns) {
433 		error = -EACCES;
434 		goto out_unlock;
435 	}
436 	if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX ||
437 	    (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
438 	     !capable(CAP_SYS_RESOURCE))) {
439 		error = -ENOSPC;
440 		goto out_unlock;
441 	}
442 	ipc_ns->mq_queues_count++;
443 	spin_unlock(&mq_lock);
444 
445 	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
446 	if (IS_ERR(inode)) {
447 		error = PTR_ERR(inode);
448 		spin_lock(&mq_lock);
449 		ipc_ns->mq_queues_count--;
450 		goto out_unlock;
451 	}
452 
453 	put_ipc_ns(ipc_ns);
454 	dir->i_size += DIRENT_SIZE;
455 	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
456 
457 	d_instantiate(dentry, inode);
458 	dget(dentry);
459 	return 0;
460 out_unlock:
461 	spin_unlock(&mq_lock);
462 	if (ipc_ns)
463 		put_ipc_ns(ipc_ns);
464 	return error;
465 }
466 
467 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
468 {
469   	struct inode *inode = dentry->d_inode;
470 
471 	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
472 	dir->i_size -= DIRENT_SIZE;
473   	drop_nlink(inode);
474   	dput(dentry);
475   	return 0;
476 }
477 
478 /*
479 *	This is routine for system read from queue file.
480 *	To avoid mess with doing here some sort of mq_receive we allow
481 *	to read only queue size & notification info (the only values
482 *	that are interesting from user point of view and aren't accessible
483 *	through std routines)
484 */
485 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
486 				size_t count, loff_t *off)
487 {
488 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
489 	char buffer[FILENT_SIZE];
490 	ssize_t ret;
491 
492 	spin_lock(&info->lock);
493 	snprintf(buffer, sizeof(buffer),
494 			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
495 			info->qsize,
496 			info->notify_owner ? info->notify.sigev_notify : 0,
497 			(info->notify_owner &&
498 			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
499 				info->notify.sigev_signo : 0,
500 			pid_vnr(info->notify_owner));
501 	spin_unlock(&info->lock);
502 	buffer[sizeof(buffer)-1] = '\0';
503 
504 	ret = simple_read_from_buffer(u_data, count, off, buffer,
505 				strlen(buffer));
506 	if (ret <= 0)
507 		return ret;
508 
509 	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
510 	return ret;
511 }
512 
513 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
514 {
515 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
516 
517 	spin_lock(&info->lock);
518 	if (task_tgid(current) == info->notify_owner)
519 		remove_notification(info);
520 
521 	spin_unlock(&info->lock);
522 	return 0;
523 }
524 
525 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
526 {
527 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
528 	int retval = 0;
529 
530 	poll_wait(filp, &info->wait_q, poll_tab);
531 
532 	spin_lock(&info->lock);
533 	if (info->attr.mq_curmsgs)
534 		retval = POLLIN | POLLRDNORM;
535 
536 	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
537 		retval |= POLLOUT | POLLWRNORM;
538 	spin_unlock(&info->lock);
539 
540 	return retval;
541 }
542 
543 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
544 static void wq_add(struct mqueue_inode_info *info, int sr,
545 			struct ext_wait_queue *ewp)
546 {
547 	struct ext_wait_queue *walk;
548 
549 	ewp->task = current;
550 
551 	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
552 		if (walk->task->static_prio <= current->static_prio) {
553 			list_add_tail(&ewp->list, &walk->list);
554 			return;
555 		}
556 	}
557 	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
558 }
559 
560 /*
561  * Puts current task to sleep. Caller must hold queue lock. After return
562  * lock isn't held.
563  * sr: SEND or RECV
564  */
565 static int wq_sleep(struct mqueue_inode_info *info, int sr,
566 		    ktime_t *timeout, struct ext_wait_queue *ewp)
567 {
568 	int retval;
569 	signed long time;
570 
571 	wq_add(info, sr, ewp);
572 
573 	for (;;) {
574 		set_current_state(TASK_INTERRUPTIBLE);
575 
576 		spin_unlock(&info->lock);
577 		time = schedule_hrtimeout_range_clock(timeout, 0,
578 			HRTIMER_MODE_ABS, CLOCK_REALTIME);
579 
580 		while (ewp->state == STATE_PENDING)
581 			cpu_relax();
582 
583 		if (ewp->state == STATE_READY) {
584 			retval = 0;
585 			goto out;
586 		}
587 		spin_lock(&info->lock);
588 		if (ewp->state == STATE_READY) {
589 			retval = 0;
590 			goto out_unlock;
591 		}
592 		if (signal_pending(current)) {
593 			retval = -ERESTARTSYS;
594 			break;
595 		}
596 		if (time == 0) {
597 			retval = -ETIMEDOUT;
598 			break;
599 		}
600 	}
601 	list_del(&ewp->list);
602 out_unlock:
603 	spin_unlock(&info->lock);
604 out:
605 	return retval;
606 }
607 
608 /*
609  * Returns waiting task that should be serviced first or NULL if none exists
610  */
611 static struct ext_wait_queue *wq_get_first_waiter(
612 		struct mqueue_inode_info *info, int sr)
613 {
614 	struct list_head *ptr;
615 
616 	ptr = info->e_wait_q[sr].list.prev;
617 	if (ptr == &info->e_wait_q[sr].list)
618 		return NULL;
619 	return list_entry(ptr, struct ext_wait_queue, list);
620 }
621 
622 
623 static inline void set_cookie(struct sk_buff *skb, char code)
624 {
625 	((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
626 }
627 
628 /*
629  * The next function is only to split too long sys_mq_timedsend
630  */
631 static void __do_notify(struct mqueue_inode_info *info)
632 {
633 	/* notification
634 	 * invoked when there is registered process and there isn't process
635 	 * waiting synchronously for message AND state of queue changed from
636 	 * empty to not empty. Here we are sure that no one is waiting
637 	 * synchronously. */
638 	if (info->notify_owner &&
639 	    info->attr.mq_curmsgs == 1) {
640 		struct siginfo sig_i;
641 		switch (info->notify.sigev_notify) {
642 		case SIGEV_NONE:
643 			break;
644 		case SIGEV_SIGNAL:
645 			/* sends signal */
646 
647 			sig_i.si_signo = info->notify.sigev_signo;
648 			sig_i.si_errno = 0;
649 			sig_i.si_code = SI_MESGQ;
650 			sig_i.si_value = info->notify.sigev_value;
651 			/* map current pid/uid into info->owner's namespaces */
652 			rcu_read_lock();
653 			sig_i.si_pid = task_tgid_nr_ns(current,
654 						ns_of_pid(info->notify_owner));
655 			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
656 			rcu_read_unlock();
657 
658 			kill_pid_info(info->notify.sigev_signo,
659 				      &sig_i, info->notify_owner);
660 			break;
661 		case SIGEV_THREAD:
662 			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
663 			netlink_sendskb(info->notify_sock, info->notify_cookie);
664 			break;
665 		}
666 		/* after notification unregisters process */
667 		put_pid(info->notify_owner);
668 		put_user_ns(info->notify_user_ns);
669 		info->notify_owner = NULL;
670 		info->notify_user_ns = NULL;
671 	}
672 	wake_up(&info->wait_q);
673 }
674 
675 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
676 			   ktime_t *expires, struct timespec *ts)
677 {
678 	if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
679 		return -EFAULT;
680 	if (!timespec_valid(ts))
681 		return -EINVAL;
682 
683 	*expires = timespec_to_ktime(*ts);
684 	return 0;
685 }
686 
687 static void remove_notification(struct mqueue_inode_info *info)
688 {
689 	if (info->notify_owner != NULL &&
690 	    info->notify.sigev_notify == SIGEV_THREAD) {
691 		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
692 		netlink_sendskb(info->notify_sock, info->notify_cookie);
693 	}
694 	put_pid(info->notify_owner);
695 	put_user_ns(info->notify_user_ns);
696 	info->notify_owner = NULL;
697 	info->notify_user_ns = NULL;
698 }
699 
700 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
701 {
702 	int mq_treesize;
703 	unsigned long total_size;
704 
705 	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
706 		return -EINVAL;
707 	if (capable(CAP_SYS_RESOURCE)) {
708 		if (attr->mq_maxmsg > HARD_MSGMAX ||
709 		    attr->mq_msgsize > HARD_MSGSIZEMAX)
710 			return -EINVAL;
711 	} else {
712 		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
713 				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
714 			return -EINVAL;
715 	}
716 	/* check for overflow */
717 	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
718 		return -EOVERFLOW;
719 	mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
720 		min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
721 		sizeof(struct posix_msg_tree_node);
722 	total_size = attr->mq_maxmsg * attr->mq_msgsize;
723 	if (total_size + mq_treesize < total_size)
724 		return -EOVERFLOW;
725 	return 0;
726 }
727 
728 /*
729  * Invoked when creating a new queue via sys_mq_open
730  */
731 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
732 			struct path *path, int oflag, umode_t mode,
733 			struct mq_attr *attr)
734 {
735 	const struct cred *cred = current_cred();
736 	int ret;
737 
738 	if (attr) {
739 		ret = mq_attr_ok(ipc_ns, attr);
740 		if (ret)
741 			return ERR_PTR(ret);
742 		/* store for use during create */
743 		path->dentry->d_fsdata = attr;
744 	} else {
745 		struct mq_attr def_attr;
746 
747 		def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
748 					 ipc_ns->mq_msg_default);
749 		def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
750 					  ipc_ns->mq_msgsize_default);
751 		ret = mq_attr_ok(ipc_ns, &def_attr);
752 		if (ret)
753 			return ERR_PTR(ret);
754 	}
755 
756 	mode &= ~current_umask();
757 	ret = vfs_create(dir, path->dentry, mode, true);
758 	path->dentry->d_fsdata = NULL;
759 	if (ret)
760 		return ERR_PTR(ret);
761 	return dentry_open(path, oflag, cred);
762 }
763 
764 /* Opens existing queue */
765 static struct file *do_open(struct path *path, int oflag)
766 {
767 	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
768 						  MAY_READ | MAY_WRITE };
769 	int acc;
770 	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
771 		return ERR_PTR(-EINVAL);
772 	acc = oflag2acc[oflag & O_ACCMODE];
773 	if (inode_permission(path->dentry->d_inode, acc))
774 		return ERR_PTR(-EACCES);
775 	return dentry_open(path, oflag, current_cred());
776 }
777 
778 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
779 		struct mq_attr __user *, u_attr)
780 {
781 	struct path path;
782 	struct file *filp;
783 	struct filename *name;
784 	struct mq_attr attr;
785 	int fd, error;
786 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
787 	struct vfsmount *mnt = ipc_ns->mq_mnt;
788 	struct dentry *root = mnt->mnt_root;
789 	int ro;
790 
791 	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
792 		return -EFAULT;
793 
794 	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
795 
796 	if (IS_ERR(name = getname(u_name)))
797 		return PTR_ERR(name);
798 
799 	fd = get_unused_fd_flags(O_CLOEXEC);
800 	if (fd < 0)
801 		goto out_putname;
802 
803 	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
804 	error = 0;
805 	mutex_lock(&root->d_inode->i_mutex);
806 	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
807 	if (IS_ERR(path.dentry)) {
808 		error = PTR_ERR(path.dentry);
809 		goto out_putfd;
810 	}
811 	path.mnt = mntget(mnt);
812 
813 	if (oflag & O_CREAT) {
814 		if (path.dentry->d_inode) {	/* entry already exists */
815 			audit_inode(name, path.dentry, 0);
816 			if (oflag & O_EXCL) {
817 				error = -EEXIST;
818 				goto out;
819 			}
820 			filp = do_open(&path, oflag);
821 		} else {
822 			if (ro) {
823 				error = ro;
824 				goto out;
825 			}
826 			audit_inode_parent_hidden(name, root);
827 			filp = do_create(ipc_ns, root->d_inode,
828 						&path, oflag, mode,
829 						u_attr ? &attr : NULL);
830 		}
831 	} else {
832 		if (!path.dentry->d_inode) {
833 			error = -ENOENT;
834 			goto out;
835 		}
836 		audit_inode(name, path.dentry, 0);
837 		filp = do_open(&path, oflag);
838 	}
839 
840 	if (!IS_ERR(filp))
841 		fd_install(fd, filp);
842 	else
843 		error = PTR_ERR(filp);
844 out:
845 	path_put(&path);
846 out_putfd:
847 	if (error) {
848 		put_unused_fd(fd);
849 		fd = error;
850 	}
851 	mutex_unlock(&root->d_inode->i_mutex);
852 	if (!ro)
853 		mnt_drop_write(mnt);
854 out_putname:
855 	putname(name);
856 	return fd;
857 }
858 
859 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
860 {
861 	int err;
862 	struct filename *name;
863 	struct dentry *dentry;
864 	struct inode *inode = NULL;
865 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
866 	struct vfsmount *mnt = ipc_ns->mq_mnt;
867 
868 	name = getname(u_name);
869 	if (IS_ERR(name))
870 		return PTR_ERR(name);
871 
872 	audit_inode_parent_hidden(name, mnt->mnt_root);
873 	err = mnt_want_write(mnt);
874 	if (err)
875 		goto out_name;
876 	mutex_lock_nested(&mnt->mnt_root->d_inode->i_mutex, I_MUTEX_PARENT);
877 	dentry = lookup_one_len(name->name, mnt->mnt_root,
878 				strlen(name->name));
879 	if (IS_ERR(dentry)) {
880 		err = PTR_ERR(dentry);
881 		goto out_unlock;
882 	}
883 
884 	inode = dentry->d_inode;
885 	if (!inode) {
886 		err = -ENOENT;
887 	} else {
888 		ihold(inode);
889 		err = vfs_unlink(dentry->d_parent->d_inode, dentry, NULL);
890 	}
891 	dput(dentry);
892 
893 out_unlock:
894 	mutex_unlock(&mnt->mnt_root->d_inode->i_mutex);
895 	if (inode)
896 		iput(inode);
897 	mnt_drop_write(mnt);
898 out_name:
899 	putname(name);
900 
901 	return err;
902 }
903 
904 /* Pipelined send and receive functions.
905  *
906  * If a receiver finds no waiting message, then it registers itself in the
907  * list of waiting receivers. A sender checks that list before adding the new
908  * message into the message array. If there is a waiting receiver, then it
909  * bypasses the message array and directly hands the message over to the
910  * receiver.
911  * The receiver accepts the message and returns without grabbing the queue
912  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
913  * are necessary. The same algorithm is used for sysv semaphores, see
914  * ipc/sem.c for more details.
915  *
916  * The same algorithm is used for senders.
917  */
918 
919 /* pipelined_send() - send a message directly to the task waiting in
920  * sys_mq_timedreceive() (without inserting message into a queue).
921  */
922 static inline void pipelined_send(struct mqueue_inode_info *info,
923 				  struct msg_msg *message,
924 				  struct ext_wait_queue *receiver)
925 {
926 	receiver->msg = message;
927 	list_del(&receiver->list);
928 	receiver->state = STATE_PENDING;
929 	wake_up_process(receiver->task);
930 	smp_wmb();
931 	receiver->state = STATE_READY;
932 }
933 
934 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
935  * gets its message and put to the queue (we have one free place for sure). */
936 static inline void pipelined_receive(struct mqueue_inode_info *info)
937 {
938 	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
939 
940 	if (!sender) {
941 		/* for poll */
942 		wake_up_interruptible(&info->wait_q);
943 		return;
944 	}
945 	if (msg_insert(sender->msg, info))
946 		return;
947 	list_del(&sender->list);
948 	sender->state = STATE_PENDING;
949 	wake_up_process(sender->task);
950 	smp_wmb();
951 	sender->state = STATE_READY;
952 }
953 
954 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
955 		size_t, msg_len, unsigned int, msg_prio,
956 		const struct timespec __user *, u_abs_timeout)
957 {
958 	struct fd f;
959 	struct inode *inode;
960 	struct ext_wait_queue wait;
961 	struct ext_wait_queue *receiver;
962 	struct msg_msg *msg_ptr;
963 	struct mqueue_inode_info *info;
964 	ktime_t expires, *timeout = NULL;
965 	struct timespec ts;
966 	struct posix_msg_tree_node *new_leaf = NULL;
967 	int ret = 0;
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 	f = fdget(mqdes);
982 	if (unlikely(!f.file)) {
983 		ret = -EBADF;
984 		goto out;
985 	}
986 
987 	inode = file_inode(f.file);
988 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
989 		ret = -EBADF;
990 		goto out_fput;
991 	}
992 	info = MQUEUE_I(inode);
993 	audit_inode(NULL, f.file->f_path.dentry, 0);
994 
995 	if (unlikely(!(f.file->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 	/*
1016 	 * msg_insert really wants us to have a valid, spare node struct so
1017 	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1018 	 * fall back to that if necessary.
1019 	 */
1020 	if (!info->node_cache)
1021 		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1022 
1023 	spin_lock(&info->lock);
1024 
1025 	if (!info->node_cache && new_leaf) {
1026 		/* Save our speculative allocation into the cache */
1027 		INIT_LIST_HEAD(&new_leaf->msg_list);
1028 		info->node_cache = new_leaf;
1029 		info->qsize += sizeof(*new_leaf);
1030 		new_leaf = NULL;
1031 	} else {
1032 		kfree(new_leaf);
1033 	}
1034 
1035 	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1036 		if (f.file->f_flags & O_NONBLOCK) {
1037 			ret = -EAGAIN;
1038 		} else {
1039 			wait.task = current;
1040 			wait.msg = (void *) msg_ptr;
1041 			wait.state = STATE_NONE;
1042 			ret = wq_sleep(info, SEND, timeout, &wait);
1043 			/*
1044 			 * wq_sleep must be called with info->lock held, and
1045 			 * returns with the lock released
1046 			 */
1047 			goto out_free;
1048 		}
1049 	} else {
1050 		receiver = wq_get_first_waiter(info, RECV);
1051 		if (receiver) {
1052 			pipelined_send(info, msg_ptr, receiver);
1053 		} else {
1054 			/* adds message to the queue */
1055 			ret = msg_insert(msg_ptr, info);
1056 			if (ret)
1057 				goto out_unlock;
1058 			__do_notify(info);
1059 		}
1060 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1061 				CURRENT_TIME;
1062 	}
1063 out_unlock:
1064 	spin_unlock(&info->lock);
1065 out_free:
1066 	if (ret)
1067 		free_msg(msg_ptr);
1068 out_fput:
1069 	fdput(f);
1070 out:
1071 	return ret;
1072 }
1073 
1074 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1075 		size_t, msg_len, unsigned int __user *, u_msg_prio,
1076 		const struct timespec __user *, u_abs_timeout)
1077 {
1078 	ssize_t ret;
1079 	struct msg_msg *msg_ptr;
1080 	struct fd f;
1081 	struct inode *inode;
1082 	struct mqueue_inode_info *info;
1083 	struct ext_wait_queue wait;
1084 	ktime_t expires, *timeout = NULL;
1085 	struct timespec ts;
1086 	struct posix_msg_tree_node *new_leaf = NULL;
1087 
1088 	if (u_abs_timeout) {
1089 		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1090 		if (res)
1091 			return res;
1092 		timeout = &expires;
1093 	}
1094 
1095 	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1096 
1097 	f = fdget(mqdes);
1098 	if (unlikely(!f.file)) {
1099 		ret = -EBADF;
1100 		goto out;
1101 	}
1102 
1103 	inode = file_inode(f.file);
1104 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1105 		ret = -EBADF;
1106 		goto out_fput;
1107 	}
1108 	info = MQUEUE_I(inode);
1109 	audit_inode(NULL, f.file->f_path.dentry, 0);
1110 
1111 	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1112 		ret = -EBADF;
1113 		goto out_fput;
1114 	}
1115 
1116 	/* checks if buffer is big enough */
1117 	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1118 		ret = -EMSGSIZE;
1119 		goto out_fput;
1120 	}
1121 
1122 	/*
1123 	 * msg_insert really wants us to have a valid, spare node struct so
1124 	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1125 	 * fall back to that if necessary.
1126 	 */
1127 	if (!info->node_cache)
1128 		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1129 
1130 	spin_lock(&info->lock);
1131 
1132 	if (!info->node_cache && new_leaf) {
1133 		/* Save our speculative allocation into the cache */
1134 		INIT_LIST_HEAD(&new_leaf->msg_list);
1135 		info->node_cache = new_leaf;
1136 		info->qsize += sizeof(*new_leaf);
1137 	} else {
1138 		kfree(new_leaf);
1139 	}
1140 
1141 	if (info->attr.mq_curmsgs == 0) {
1142 		if (f.file->f_flags & O_NONBLOCK) {
1143 			spin_unlock(&info->lock);
1144 			ret = -EAGAIN;
1145 		} else {
1146 			wait.task = current;
1147 			wait.state = STATE_NONE;
1148 			ret = wq_sleep(info, RECV, timeout, &wait);
1149 			msg_ptr = wait.msg;
1150 		}
1151 	} else {
1152 		msg_ptr = msg_get(info);
1153 
1154 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1155 				CURRENT_TIME;
1156 
1157 		/* There is now free space in queue. */
1158 		pipelined_receive(info);
1159 		spin_unlock(&info->lock);
1160 		ret = 0;
1161 	}
1162 	if (ret == 0) {
1163 		ret = msg_ptr->m_ts;
1164 
1165 		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1166 			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1167 			ret = -EFAULT;
1168 		}
1169 		free_msg(msg_ptr);
1170 	}
1171 out_fput:
1172 	fdput(f);
1173 out:
1174 	return ret;
1175 }
1176 
1177 /*
1178  * Notes: the case when user wants us to deregister (with NULL as pointer)
1179  * and he isn't currently owner of notification, will be silently discarded.
1180  * It isn't explicitly defined in the POSIX.
1181  */
1182 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1183 		const struct sigevent __user *, u_notification)
1184 {
1185 	int ret;
1186 	struct fd f;
1187 	struct sock *sock;
1188 	struct inode *inode;
1189 	struct sigevent notification;
1190 	struct mqueue_inode_info *info;
1191 	struct sk_buff *nc;
1192 
1193 	if (u_notification) {
1194 		if (copy_from_user(&notification, u_notification,
1195 					sizeof(struct sigevent)))
1196 			return -EFAULT;
1197 	}
1198 
1199 	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1200 
1201 	nc = NULL;
1202 	sock = NULL;
1203 	if (u_notification != NULL) {
1204 		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1205 			     notification.sigev_notify != SIGEV_SIGNAL &&
1206 			     notification.sigev_notify != SIGEV_THREAD))
1207 			return -EINVAL;
1208 		if (notification.sigev_notify == SIGEV_SIGNAL &&
1209 			!valid_signal(notification.sigev_signo)) {
1210 			return -EINVAL;
1211 		}
1212 		if (notification.sigev_notify == SIGEV_THREAD) {
1213 			long timeo;
1214 
1215 			/* create the notify skb */
1216 			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1217 			if (!nc) {
1218 				ret = -ENOMEM;
1219 				goto out;
1220 			}
1221 			if (copy_from_user(nc->data,
1222 					notification.sigev_value.sival_ptr,
1223 					NOTIFY_COOKIE_LEN)) {
1224 				ret = -EFAULT;
1225 				goto out;
1226 			}
1227 
1228 			/* TODO: add a header? */
1229 			skb_put(nc, NOTIFY_COOKIE_LEN);
1230 			/* and attach it to the socket */
1231 retry:
1232 			f = fdget(notification.sigev_signo);
1233 			if (!f.file) {
1234 				ret = -EBADF;
1235 				goto out;
1236 			}
1237 			sock = netlink_getsockbyfilp(f.file);
1238 			fdput(f);
1239 			if (IS_ERR(sock)) {
1240 				ret = PTR_ERR(sock);
1241 				sock = NULL;
1242 				goto out;
1243 			}
1244 
1245 			timeo = MAX_SCHEDULE_TIMEOUT;
1246 			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1247 			if (ret == 1)
1248 				goto retry;
1249 			if (ret) {
1250 				sock = NULL;
1251 				nc = NULL;
1252 				goto out;
1253 			}
1254 		}
1255 	}
1256 
1257 	f = fdget(mqdes);
1258 	if (!f.file) {
1259 		ret = -EBADF;
1260 		goto out;
1261 	}
1262 
1263 	inode = file_inode(f.file);
1264 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1265 		ret = -EBADF;
1266 		goto out_fput;
1267 	}
1268 	info = MQUEUE_I(inode);
1269 
1270 	ret = 0;
1271 	spin_lock(&info->lock);
1272 	if (u_notification == NULL) {
1273 		if (info->notify_owner == task_tgid(current)) {
1274 			remove_notification(info);
1275 			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1276 		}
1277 	} else if (info->notify_owner != NULL) {
1278 		ret = -EBUSY;
1279 	} else {
1280 		switch (notification.sigev_notify) {
1281 		case SIGEV_NONE:
1282 			info->notify.sigev_notify = SIGEV_NONE;
1283 			break;
1284 		case SIGEV_THREAD:
1285 			info->notify_sock = sock;
1286 			info->notify_cookie = nc;
1287 			sock = NULL;
1288 			nc = NULL;
1289 			info->notify.sigev_notify = SIGEV_THREAD;
1290 			break;
1291 		case SIGEV_SIGNAL:
1292 			info->notify.sigev_signo = notification.sigev_signo;
1293 			info->notify.sigev_value = notification.sigev_value;
1294 			info->notify.sigev_notify = SIGEV_SIGNAL;
1295 			break;
1296 		}
1297 
1298 		info->notify_owner = get_pid(task_tgid(current));
1299 		info->notify_user_ns = get_user_ns(current_user_ns());
1300 		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1301 	}
1302 	spin_unlock(&info->lock);
1303 out_fput:
1304 	fdput(f);
1305 out:
1306 	if (sock) {
1307 		netlink_detachskb(sock, nc);
1308 	} else if (nc) {
1309 		dev_kfree_skb(nc);
1310 	}
1311 	return ret;
1312 }
1313 
1314 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1315 		const struct mq_attr __user *, u_mqstat,
1316 		struct mq_attr __user *, u_omqstat)
1317 {
1318 	int ret;
1319 	struct mq_attr mqstat, omqstat;
1320 	struct fd f;
1321 	struct inode *inode;
1322 	struct mqueue_inode_info *info;
1323 
1324 	if (u_mqstat != NULL) {
1325 		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1326 			return -EFAULT;
1327 		if (mqstat.mq_flags & (~O_NONBLOCK))
1328 			return -EINVAL;
1329 	}
1330 
1331 	f = fdget(mqdes);
1332 	if (!f.file) {
1333 		ret = -EBADF;
1334 		goto out;
1335 	}
1336 
1337 	inode = file_inode(f.file);
1338 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1339 		ret = -EBADF;
1340 		goto out_fput;
1341 	}
1342 	info = MQUEUE_I(inode);
1343 
1344 	spin_lock(&info->lock);
1345 
1346 	omqstat = info->attr;
1347 	omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1348 	if (u_mqstat) {
1349 		audit_mq_getsetattr(mqdes, &mqstat);
1350 		spin_lock(&f.file->f_lock);
1351 		if (mqstat.mq_flags & O_NONBLOCK)
1352 			f.file->f_flags |= O_NONBLOCK;
1353 		else
1354 			f.file->f_flags &= ~O_NONBLOCK;
1355 		spin_unlock(&f.file->f_lock);
1356 
1357 		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1358 	}
1359 
1360 	spin_unlock(&info->lock);
1361 
1362 	ret = 0;
1363 	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1364 						sizeof(struct mq_attr)))
1365 		ret = -EFAULT;
1366 
1367 out_fput:
1368 	fdput(f);
1369 out:
1370 	return ret;
1371 }
1372 
1373 static const struct inode_operations mqueue_dir_inode_operations = {
1374 	.lookup = simple_lookup,
1375 	.create = mqueue_create,
1376 	.unlink = mqueue_unlink,
1377 };
1378 
1379 static const struct file_operations mqueue_file_operations = {
1380 	.flush = mqueue_flush_file,
1381 	.poll = mqueue_poll_file,
1382 	.read = mqueue_read_file,
1383 	.llseek = default_llseek,
1384 };
1385 
1386 static const struct super_operations mqueue_super_ops = {
1387 	.alloc_inode = mqueue_alloc_inode,
1388 	.destroy_inode = mqueue_destroy_inode,
1389 	.evict_inode = mqueue_evict_inode,
1390 	.statfs = simple_statfs,
1391 };
1392 
1393 static struct file_system_type mqueue_fs_type = {
1394 	.name = "mqueue",
1395 	.mount = mqueue_mount,
1396 	.kill_sb = kill_litter_super,
1397 	.fs_flags = FS_USERNS_MOUNT,
1398 };
1399 
1400 int mq_init_ns(struct ipc_namespace *ns)
1401 {
1402 	ns->mq_queues_count  = 0;
1403 	ns->mq_queues_max    = DFLT_QUEUESMAX;
1404 	ns->mq_msg_max       = DFLT_MSGMAX;
1405 	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1406 	ns->mq_msg_default   = DFLT_MSG;
1407 	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1408 
1409 	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1410 	if (IS_ERR(ns->mq_mnt)) {
1411 		int err = PTR_ERR(ns->mq_mnt);
1412 		ns->mq_mnt = NULL;
1413 		return err;
1414 	}
1415 	return 0;
1416 }
1417 
1418 void mq_clear_sbinfo(struct ipc_namespace *ns)
1419 {
1420 	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1421 }
1422 
1423 void mq_put_mnt(struct ipc_namespace *ns)
1424 {
1425 	kern_unmount(ns->mq_mnt);
1426 }
1427 
1428 static int __init init_mqueue_fs(void)
1429 {
1430 	int error;
1431 
1432 	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1433 				sizeof(struct mqueue_inode_info), 0,
1434 				SLAB_HWCACHE_ALIGN, init_once);
1435 	if (mqueue_inode_cachep == NULL)
1436 		return -ENOMEM;
1437 
1438 	/* ignore failures - they are not fatal */
1439 	mq_sysctl_table = mq_register_sysctl_table();
1440 
1441 	error = register_filesystem(&mqueue_fs_type);
1442 	if (error)
1443 		goto out_sysctl;
1444 
1445 	spin_lock_init(&mq_lock);
1446 
1447 	error = mq_init_ns(&init_ipc_ns);
1448 	if (error)
1449 		goto out_filesystem;
1450 
1451 	return 0;
1452 
1453 out_filesystem:
1454 	unregister_filesystem(&mqueue_fs_type);
1455 out_sysctl:
1456 	if (mq_sysctl_table)
1457 		unregister_sysctl_table(mq_sysctl_table);
1458 	kmem_cache_destroy(mqueue_inode_cachep);
1459 	return error;
1460 }
1461 
1462 __initcall(init_mqueue_fs);
1463