xref: /linux/ipc/mqueue.c (revision 43afe4d3)
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/fs_context.h>
22 #include <linux/namei.h>
23 #include <linux/sysctl.h>
24 #include <linux/poll.h>
25 #include <linux/mqueue.h>
26 #include <linux/msg.h>
27 #include <linux/skbuff.h>
28 #include <linux/vmalloc.h>
29 #include <linux/netlink.h>
30 #include <linux/syscalls.h>
31 #include <linux/audit.h>
32 #include <linux/signal.h>
33 #include <linux/mutex.h>
34 #include <linux/nsproxy.h>
35 #include <linux/pid.h>
36 #include <linux/ipc_namespace.h>
37 #include <linux/user_namespace.h>
38 #include <linux/slab.h>
39 #include <linux/sched/wake_q.h>
40 #include <linux/sched/signal.h>
41 #include <linux/sched/user.h>
42 
43 #include <net/sock.h>
44 #include "util.h"
45 
46 struct mqueue_fs_context {
47 	struct ipc_namespace	*ipc_ns;
48 };
49 
50 #define MQUEUE_MAGIC	0x19800202
51 #define DIRENT_SIZE	20
52 #define FILENT_SIZE	80
53 
54 #define SEND		0
55 #define RECV		1
56 
57 #define STATE_NONE	0
58 #define STATE_READY	1
59 
60 struct posix_msg_tree_node {
61 	struct rb_node		rb_node;
62 	struct list_head	msg_list;
63 	int			priority;
64 };
65 
66 /*
67  * Locking:
68  *
69  * Accesses to a message queue are synchronized by acquiring info->lock.
70  *
71  * There are two notable exceptions:
72  * - The actual wakeup of a sleeping task is performed using the wake_q
73  *   framework. info->lock is already released when wake_up_q is called.
74  * - The exit codepaths after sleeping check ext_wait_queue->state without
75  *   any locks. If it is STATE_READY, then the syscall is completed without
76  *   acquiring info->lock.
77  *
78  * MQ_BARRIER:
79  * To achieve proper release/acquire memory barrier pairing, the state is set to
80  * STATE_READY with smp_store_release(), and it is read with READ_ONCE followed
81  * by smp_acquire__after_ctrl_dep(). In addition, wake_q_add_safe() is used.
82  *
83  * This prevents the following races:
84  *
85  * 1) With the simple wake_q_add(), the task could be gone already before
86  *    the increase of the reference happens
87  * Thread A
88  *				Thread B
89  * WRITE_ONCE(wait.state, STATE_NONE);
90  * schedule_hrtimeout()
91  *				wake_q_add(A)
92  *				if (cmpxchg()) // success
93  *				   ->state = STATE_READY (reordered)
94  * <timeout returns>
95  * if (wait.state == STATE_READY) return;
96  * sysret to user space
97  * sys_exit()
98  *				get_task_struct() // UaF
99  *
100  * Solution: Use wake_q_add_safe() and perform the get_task_struct() before
101  * the smp_store_release() that does ->state = STATE_READY.
102  *
103  * 2) Without proper _release/_acquire barriers, the woken up task
104  *    could read stale data
105  *
106  * Thread A
107  *				Thread B
108  * do_mq_timedreceive
109  * WRITE_ONCE(wait.state, STATE_NONE);
110  * schedule_hrtimeout()
111  *				state = STATE_READY;
112  * <timeout returns>
113  * if (wait.state == STATE_READY) return;
114  * msg_ptr = wait.msg;		// Access to stale data!
115  *				receiver->msg = message; (reordered)
116  *
117  * Solution: use _release and _acquire barriers.
118  *
119  * 3) There is intentionally no barrier when setting current->state
120  *    to TASK_INTERRUPTIBLE: spin_unlock(&info->lock) provides the
121  *    release memory barrier, and the wakeup is triggered when holding
122  *    info->lock, i.e. spin_lock(&info->lock) provided a pairing
123  *    acquire memory barrier.
124  */
125 
126 struct ext_wait_queue {		/* queue of sleeping tasks */
127 	struct task_struct *task;
128 	struct list_head list;
129 	struct msg_msg *msg;	/* ptr of loaded message */
130 	int state;		/* one of STATE_* values */
131 };
132 
133 struct mqueue_inode_info {
134 	spinlock_t lock;
135 	struct inode vfs_inode;
136 	wait_queue_head_t wait_q;
137 
138 	struct rb_root msg_tree;
139 	struct rb_node *msg_tree_rightmost;
140 	struct posix_msg_tree_node *node_cache;
141 	struct mq_attr attr;
142 
143 	struct sigevent notify;
144 	struct pid *notify_owner;
145 	struct user_namespace *notify_user_ns;
146 	struct user_struct *user;	/* user who created, for accounting */
147 	struct sock *notify_sock;
148 	struct sk_buff *notify_cookie;
149 
150 	/* for tasks waiting for free space and messages, respectively */
151 	struct ext_wait_queue e_wait_q[2];
152 
153 	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
154 };
155 
156 static struct file_system_type mqueue_fs_type;
157 static const struct inode_operations mqueue_dir_inode_operations;
158 static const struct file_operations mqueue_file_operations;
159 static const struct super_operations mqueue_super_ops;
160 static const struct fs_context_operations mqueue_fs_context_ops;
161 static void remove_notification(struct mqueue_inode_info *info);
162 
163 static struct kmem_cache *mqueue_inode_cachep;
164 
165 static struct ctl_table_header *mq_sysctl_table;
166 
167 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
168 {
169 	return container_of(inode, struct mqueue_inode_info, vfs_inode);
170 }
171 
172 /*
173  * This routine should be called with the mq_lock held.
174  */
175 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
176 {
177 	return get_ipc_ns(inode->i_sb->s_fs_info);
178 }
179 
180 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
181 {
182 	struct ipc_namespace *ns;
183 
184 	spin_lock(&mq_lock);
185 	ns = __get_ns_from_inode(inode);
186 	spin_unlock(&mq_lock);
187 	return ns;
188 }
189 
190 /* Auxiliary functions to manipulate messages' list */
191 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
192 {
193 	struct rb_node **p, *parent = NULL;
194 	struct posix_msg_tree_node *leaf;
195 	bool rightmost = true;
196 
197 	p = &info->msg_tree.rb_node;
198 	while (*p) {
199 		parent = *p;
200 		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
201 
202 		if (likely(leaf->priority == msg->m_type))
203 			goto insert_msg;
204 		else if (msg->m_type < leaf->priority) {
205 			p = &(*p)->rb_left;
206 			rightmost = false;
207 		} else
208 			p = &(*p)->rb_right;
209 	}
210 	if (info->node_cache) {
211 		leaf = info->node_cache;
212 		info->node_cache = NULL;
213 	} else {
214 		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
215 		if (!leaf)
216 			return -ENOMEM;
217 		INIT_LIST_HEAD(&leaf->msg_list);
218 	}
219 	leaf->priority = msg->m_type;
220 
221 	if (rightmost)
222 		info->msg_tree_rightmost = &leaf->rb_node;
223 
224 	rb_link_node(&leaf->rb_node, parent, p);
225 	rb_insert_color(&leaf->rb_node, &info->msg_tree);
226 insert_msg:
227 	info->attr.mq_curmsgs++;
228 	info->qsize += msg->m_ts;
229 	list_add_tail(&msg->m_list, &leaf->msg_list);
230 	return 0;
231 }
232 
233 static inline void msg_tree_erase(struct posix_msg_tree_node *leaf,
234 				  struct mqueue_inode_info *info)
235 {
236 	struct rb_node *node = &leaf->rb_node;
237 
238 	if (info->msg_tree_rightmost == node)
239 		info->msg_tree_rightmost = rb_prev(node);
240 
241 	rb_erase(node, &info->msg_tree);
242 	if (info->node_cache)
243 		kfree(leaf);
244 	else
245 		info->node_cache = leaf;
246 }
247 
248 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
249 {
250 	struct rb_node *parent = NULL;
251 	struct posix_msg_tree_node *leaf;
252 	struct msg_msg *msg;
253 
254 try_again:
255 	/*
256 	 * During insert, low priorities go to the left and high to the
257 	 * right.  On receive, we want the highest priorities first, so
258 	 * walk all the way to the right.
259 	 */
260 	parent = info->msg_tree_rightmost;
261 	if (!parent) {
262 		if (info->attr.mq_curmsgs) {
263 			pr_warn_once("Inconsistency in POSIX message queue, "
264 				     "no tree element, but supposedly messages "
265 				     "should exist!\n");
266 			info->attr.mq_curmsgs = 0;
267 		}
268 		return NULL;
269 	}
270 	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
271 	if (unlikely(list_empty(&leaf->msg_list))) {
272 		pr_warn_once("Inconsistency in POSIX message queue, "
273 			     "empty leaf node but we haven't implemented "
274 			     "lazy leaf delete!\n");
275 		msg_tree_erase(leaf, info);
276 		goto try_again;
277 	} else {
278 		msg = list_first_entry(&leaf->msg_list,
279 				       struct msg_msg, m_list);
280 		list_del(&msg->m_list);
281 		if (list_empty(&leaf->msg_list)) {
282 			msg_tree_erase(leaf, info);
283 		}
284 	}
285 	info->attr.mq_curmsgs--;
286 	info->qsize -= msg->m_ts;
287 	return msg;
288 }
289 
290 static struct inode *mqueue_get_inode(struct super_block *sb,
291 		struct ipc_namespace *ipc_ns, umode_t mode,
292 		struct mq_attr *attr)
293 {
294 	struct user_struct *u = current_user();
295 	struct inode *inode;
296 	int ret = -ENOMEM;
297 
298 	inode = new_inode(sb);
299 	if (!inode)
300 		goto err;
301 
302 	inode->i_ino = get_next_ino();
303 	inode->i_mode = mode;
304 	inode->i_uid = current_fsuid();
305 	inode->i_gid = current_fsgid();
306 	inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
307 
308 	if (S_ISREG(mode)) {
309 		struct mqueue_inode_info *info;
310 		unsigned long mq_bytes, mq_treesize;
311 
312 		inode->i_fop = &mqueue_file_operations;
313 		inode->i_size = FILENT_SIZE;
314 		/* mqueue specific info */
315 		info = MQUEUE_I(inode);
316 		spin_lock_init(&info->lock);
317 		init_waitqueue_head(&info->wait_q);
318 		INIT_LIST_HEAD(&info->e_wait_q[0].list);
319 		INIT_LIST_HEAD(&info->e_wait_q[1].list);
320 		info->notify_owner = NULL;
321 		info->notify_user_ns = NULL;
322 		info->qsize = 0;
323 		info->user = NULL;	/* set when all is ok */
324 		info->msg_tree = RB_ROOT;
325 		info->msg_tree_rightmost = NULL;
326 		info->node_cache = NULL;
327 		memset(&info->attr, 0, sizeof(info->attr));
328 		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
329 					   ipc_ns->mq_msg_default);
330 		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
331 					    ipc_ns->mq_msgsize_default);
332 		if (attr) {
333 			info->attr.mq_maxmsg = attr->mq_maxmsg;
334 			info->attr.mq_msgsize = attr->mq_msgsize;
335 		}
336 		/*
337 		 * We used to allocate a static array of pointers and account
338 		 * the size of that array as well as one msg_msg struct per
339 		 * possible message into the queue size. That's no longer
340 		 * accurate as the queue is now an rbtree and will grow and
341 		 * shrink depending on usage patterns.  We can, however, still
342 		 * account one msg_msg struct per message, but the nodes are
343 		 * allocated depending on priority usage, and most programs
344 		 * only use one, or a handful, of priorities.  However, since
345 		 * this is pinned memory, we need to assume worst case, so
346 		 * that means the min(mq_maxmsg, max_priorities) * struct
347 		 * posix_msg_tree_node.
348 		 */
349 
350 		ret = -EINVAL;
351 		if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
352 			goto out_inode;
353 		if (capable(CAP_SYS_RESOURCE)) {
354 			if (info->attr.mq_maxmsg > HARD_MSGMAX ||
355 			    info->attr.mq_msgsize > HARD_MSGSIZEMAX)
356 				goto out_inode;
357 		} else {
358 			if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
359 					info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
360 				goto out_inode;
361 		}
362 		ret = -EOVERFLOW;
363 		/* check for overflow */
364 		if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
365 			goto out_inode;
366 		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
367 			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
368 			sizeof(struct posix_msg_tree_node);
369 		mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
370 		if (mq_bytes + mq_treesize < mq_bytes)
371 			goto out_inode;
372 		mq_bytes += mq_treesize;
373 		spin_lock(&mq_lock);
374 		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
375 		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
376 			spin_unlock(&mq_lock);
377 			/* mqueue_evict_inode() releases info->messages */
378 			ret = -EMFILE;
379 			goto out_inode;
380 		}
381 		u->mq_bytes += mq_bytes;
382 		spin_unlock(&mq_lock);
383 
384 		/* all is ok */
385 		info->user = get_uid(u);
386 	} else if (S_ISDIR(mode)) {
387 		inc_nlink(inode);
388 		/* Some things misbehave if size == 0 on a directory */
389 		inode->i_size = 2 * DIRENT_SIZE;
390 		inode->i_op = &mqueue_dir_inode_operations;
391 		inode->i_fop = &simple_dir_operations;
392 	}
393 
394 	return inode;
395 out_inode:
396 	iput(inode);
397 err:
398 	return ERR_PTR(ret);
399 }
400 
401 static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
402 {
403 	struct inode *inode;
404 	struct ipc_namespace *ns = sb->s_fs_info;
405 
406 	sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
407 	sb->s_blocksize = PAGE_SIZE;
408 	sb->s_blocksize_bits = PAGE_SHIFT;
409 	sb->s_magic = MQUEUE_MAGIC;
410 	sb->s_op = &mqueue_super_ops;
411 
412 	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
413 	if (IS_ERR(inode))
414 		return PTR_ERR(inode);
415 
416 	sb->s_root = d_make_root(inode);
417 	if (!sb->s_root)
418 		return -ENOMEM;
419 	return 0;
420 }
421 
422 static int mqueue_get_tree(struct fs_context *fc)
423 {
424 	struct mqueue_fs_context *ctx = fc->fs_private;
425 
426 	return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
427 }
428 
429 static void mqueue_fs_context_free(struct fs_context *fc)
430 {
431 	struct mqueue_fs_context *ctx = fc->fs_private;
432 
433 	put_ipc_ns(ctx->ipc_ns);
434 	kfree(ctx);
435 }
436 
437 static int mqueue_init_fs_context(struct fs_context *fc)
438 {
439 	struct mqueue_fs_context *ctx;
440 
441 	ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
442 	if (!ctx)
443 		return -ENOMEM;
444 
445 	ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
446 	put_user_ns(fc->user_ns);
447 	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
448 	fc->fs_private = ctx;
449 	fc->ops = &mqueue_fs_context_ops;
450 	return 0;
451 }
452 
453 static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
454 {
455 	struct mqueue_fs_context *ctx;
456 	struct fs_context *fc;
457 	struct vfsmount *mnt;
458 
459 	fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
460 	if (IS_ERR(fc))
461 		return ERR_CAST(fc);
462 
463 	ctx = fc->fs_private;
464 	put_ipc_ns(ctx->ipc_ns);
465 	ctx->ipc_ns = get_ipc_ns(ns);
466 	put_user_ns(fc->user_ns);
467 	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
468 
469 	mnt = fc_mount(fc);
470 	put_fs_context(fc);
471 	return mnt;
472 }
473 
474 static void init_once(void *foo)
475 {
476 	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
477 
478 	inode_init_once(&p->vfs_inode);
479 }
480 
481 static struct inode *mqueue_alloc_inode(struct super_block *sb)
482 {
483 	struct mqueue_inode_info *ei;
484 
485 	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
486 	if (!ei)
487 		return NULL;
488 	return &ei->vfs_inode;
489 }
490 
491 static void mqueue_free_inode(struct inode *inode)
492 {
493 	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
494 }
495 
496 static void mqueue_evict_inode(struct inode *inode)
497 {
498 	struct mqueue_inode_info *info;
499 	struct user_struct *user;
500 	struct ipc_namespace *ipc_ns;
501 	struct msg_msg *msg, *nmsg;
502 	LIST_HEAD(tmp_msg);
503 
504 	clear_inode(inode);
505 
506 	if (S_ISDIR(inode->i_mode))
507 		return;
508 
509 	ipc_ns = get_ns_from_inode(inode);
510 	info = MQUEUE_I(inode);
511 	spin_lock(&info->lock);
512 	while ((msg = msg_get(info)) != NULL)
513 		list_add_tail(&msg->m_list, &tmp_msg);
514 	kfree(info->node_cache);
515 	spin_unlock(&info->lock);
516 
517 	list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
518 		list_del(&msg->m_list);
519 		free_msg(msg);
520 	}
521 
522 	user = info->user;
523 	if (user) {
524 		unsigned long mq_bytes, mq_treesize;
525 
526 		/* Total amount of bytes accounted for the mqueue */
527 		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
528 			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
529 			sizeof(struct posix_msg_tree_node);
530 
531 		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
532 					  info->attr.mq_msgsize);
533 
534 		spin_lock(&mq_lock);
535 		user->mq_bytes -= mq_bytes;
536 		/*
537 		 * get_ns_from_inode() ensures that the
538 		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
539 		 * to which we now hold a reference, or it is NULL.
540 		 * We can't put it here under mq_lock, though.
541 		 */
542 		if (ipc_ns)
543 			ipc_ns->mq_queues_count--;
544 		spin_unlock(&mq_lock);
545 		free_uid(user);
546 	}
547 	if (ipc_ns)
548 		put_ipc_ns(ipc_ns);
549 }
550 
551 static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
552 {
553 	struct inode *dir = dentry->d_parent->d_inode;
554 	struct inode *inode;
555 	struct mq_attr *attr = arg;
556 	int error;
557 	struct ipc_namespace *ipc_ns;
558 
559 	spin_lock(&mq_lock);
560 	ipc_ns = __get_ns_from_inode(dir);
561 	if (!ipc_ns) {
562 		error = -EACCES;
563 		goto out_unlock;
564 	}
565 
566 	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
567 	    !capable(CAP_SYS_RESOURCE)) {
568 		error = -ENOSPC;
569 		goto out_unlock;
570 	}
571 	ipc_ns->mq_queues_count++;
572 	spin_unlock(&mq_lock);
573 
574 	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
575 	if (IS_ERR(inode)) {
576 		error = PTR_ERR(inode);
577 		spin_lock(&mq_lock);
578 		ipc_ns->mq_queues_count--;
579 		goto out_unlock;
580 	}
581 
582 	put_ipc_ns(ipc_ns);
583 	dir->i_size += DIRENT_SIZE;
584 	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
585 
586 	d_instantiate(dentry, inode);
587 	dget(dentry);
588 	return 0;
589 out_unlock:
590 	spin_unlock(&mq_lock);
591 	if (ipc_ns)
592 		put_ipc_ns(ipc_ns);
593 	return error;
594 }
595 
596 static int mqueue_create(struct inode *dir, struct dentry *dentry,
597 				umode_t mode, bool excl)
598 {
599 	return mqueue_create_attr(dentry, mode, NULL);
600 }
601 
602 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
603 {
604 	struct inode *inode = d_inode(dentry);
605 
606 	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
607 	dir->i_size -= DIRENT_SIZE;
608 	drop_nlink(inode);
609 	dput(dentry);
610 	return 0;
611 }
612 
613 /*
614 *	This is routine for system read from queue file.
615 *	To avoid mess with doing here some sort of mq_receive we allow
616 *	to read only queue size & notification info (the only values
617 *	that are interesting from user point of view and aren't accessible
618 *	through std routines)
619 */
620 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
621 				size_t count, loff_t *off)
622 {
623 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
624 	char buffer[FILENT_SIZE];
625 	ssize_t ret;
626 
627 	spin_lock(&info->lock);
628 	snprintf(buffer, sizeof(buffer),
629 			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
630 			info->qsize,
631 			info->notify_owner ? info->notify.sigev_notify : 0,
632 			(info->notify_owner &&
633 			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
634 				info->notify.sigev_signo : 0,
635 			pid_vnr(info->notify_owner));
636 	spin_unlock(&info->lock);
637 	buffer[sizeof(buffer)-1] = '\0';
638 
639 	ret = simple_read_from_buffer(u_data, count, off, buffer,
640 				strlen(buffer));
641 	if (ret <= 0)
642 		return ret;
643 
644 	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
645 	return ret;
646 }
647 
648 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
649 {
650 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
651 
652 	spin_lock(&info->lock);
653 	if (task_tgid(current) == info->notify_owner)
654 		remove_notification(info);
655 
656 	spin_unlock(&info->lock);
657 	return 0;
658 }
659 
660 static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
661 {
662 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
663 	__poll_t retval = 0;
664 
665 	poll_wait(filp, &info->wait_q, poll_tab);
666 
667 	spin_lock(&info->lock);
668 	if (info->attr.mq_curmsgs)
669 		retval = EPOLLIN | EPOLLRDNORM;
670 
671 	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
672 		retval |= EPOLLOUT | EPOLLWRNORM;
673 	spin_unlock(&info->lock);
674 
675 	return retval;
676 }
677 
678 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
679 static void wq_add(struct mqueue_inode_info *info, int sr,
680 			struct ext_wait_queue *ewp)
681 {
682 	struct ext_wait_queue *walk;
683 
684 	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
685 		if (walk->task->prio <= current->prio) {
686 			list_add_tail(&ewp->list, &walk->list);
687 			return;
688 		}
689 	}
690 	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
691 }
692 
693 /*
694  * Puts current task to sleep. Caller must hold queue lock. After return
695  * lock isn't held.
696  * sr: SEND or RECV
697  */
698 static int wq_sleep(struct mqueue_inode_info *info, int sr,
699 		    ktime_t *timeout, struct ext_wait_queue *ewp)
700 	__releases(&info->lock)
701 {
702 	int retval;
703 	signed long time;
704 
705 	wq_add(info, sr, ewp);
706 
707 	for (;;) {
708 		/* memory barrier not required, we hold info->lock */
709 		__set_current_state(TASK_INTERRUPTIBLE);
710 
711 		spin_unlock(&info->lock);
712 		time = schedule_hrtimeout_range_clock(timeout, 0,
713 			HRTIMER_MODE_ABS, CLOCK_REALTIME);
714 
715 		if (READ_ONCE(ewp->state) == STATE_READY) {
716 			/* see MQ_BARRIER for purpose/pairing */
717 			smp_acquire__after_ctrl_dep();
718 			retval = 0;
719 			goto out;
720 		}
721 		spin_lock(&info->lock);
722 
723 		/* we hold info->lock, so no memory barrier required */
724 		if (READ_ONCE(ewp->state) == STATE_READY) {
725 			retval = 0;
726 			goto out_unlock;
727 		}
728 		if (signal_pending(current)) {
729 			retval = -ERESTARTSYS;
730 			break;
731 		}
732 		if (time == 0) {
733 			retval = -ETIMEDOUT;
734 			break;
735 		}
736 	}
737 	list_del(&ewp->list);
738 out_unlock:
739 	spin_unlock(&info->lock);
740 out:
741 	return retval;
742 }
743 
744 /*
745  * Returns waiting task that should be serviced first or NULL if none exists
746  */
747 static struct ext_wait_queue *wq_get_first_waiter(
748 		struct mqueue_inode_info *info, int sr)
749 {
750 	struct list_head *ptr;
751 
752 	ptr = info->e_wait_q[sr].list.prev;
753 	if (ptr == &info->e_wait_q[sr].list)
754 		return NULL;
755 	return list_entry(ptr, struct ext_wait_queue, list);
756 }
757 
758 
759 static inline void set_cookie(struct sk_buff *skb, char code)
760 {
761 	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
762 }
763 
764 /*
765  * The next function is only to split too long sys_mq_timedsend
766  */
767 static void __do_notify(struct mqueue_inode_info *info)
768 {
769 	/* notification
770 	 * invoked when there is registered process and there isn't process
771 	 * waiting synchronously for message AND state of queue changed from
772 	 * empty to not empty. Here we are sure that no one is waiting
773 	 * synchronously. */
774 	if (info->notify_owner &&
775 	    info->attr.mq_curmsgs == 1) {
776 		struct kernel_siginfo sig_i;
777 		switch (info->notify.sigev_notify) {
778 		case SIGEV_NONE:
779 			break;
780 		case SIGEV_SIGNAL:
781 			/* sends signal */
782 
783 			clear_siginfo(&sig_i);
784 			sig_i.si_signo = info->notify.sigev_signo;
785 			sig_i.si_errno = 0;
786 			sig_i.si_code = SI_MESGQ;
787 			sig_i.si_value = info->notify.sigev_value;
788 			/* map current pid/uid into info->owner's namespaces */
789 			rcu_read_lock();
790 			sig_i.si_pid = task_tgid_nr_ns(current,
791 						ns_of_pid(info->notify_owner));
792 			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
793 			rcu_read_unlock();
794 
795 			kill_pid_info(info->notify.sigev_signo,
796 				      &sig_i, info->notify_owner);
797 			break;
798 		case SIGEV_THREAD:
799 			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
800 			netlink_sendskb(info->notify_sock, info->notify_cookie);
801 			break;
802 		}
803 		/* after notification unregisters process */
804 		put_pid(info->notify_owner);
805 		put_user_ns(info->notify_user_ns);
806 		info->notify_owner = NULL;
807 		info->notify_user_ns = NULL;
808 	}
809 	wake_up(&info->wait_q);
810 }
811 
812 static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
813 			   struct timespec64 *ts)
814 {
815 	if (get_timespec64(ts, u_abs_timeout))
816 		return -EFAULT;
817 	if (!timespec64_valid(ts))
818 		return -EINVAL;
819 	return 0;
820 }
821 
822 static void remove_notification(struct mqueue_inode_info *info)
823 {
824 	if (info->notify_owner != NULL &&
825 	    info->notify.sigev_notify == SIGEV_THREAD) {
826 		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
827 		netlink_sendskb(info->notify_sock, info->notify_cookie);
828 	}
829 	put_pid(info->notify_owner);
830 	put_user_ns(info->notify_user_ns);
831 	info->notify_owner = NULL;
832 	info->notify_user_ns = NULL;
833 }
834 
835 static int prepare_open(struct dentry *dentry, int oflag, int ro,
836 			umode_t mode, struct filename *name,
837 			struct mq_attr *attr)
838 {
839 	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
840 						  MAY_READ | MAY_WRITE };
841 	int acc;
842 
843 	if (d_really_is_negative(dentry)) {
844 		if (!(oflag & O_CREAT))
845 			return -ENOENT;
846 		if (ro)
847 			return ro;
848 		audit_inode_parent_hidden(name, dentry->d_parent);
849 		return vfs_mkobj(dentry, mode & ~current_umask(),
850 				  mqueue_create_attr, attr);
851 	}
852 	/* it already existed */
853 	audit_inode(name, dentry, 0);
854 	if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
855 		return -EEXIST;
856 	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
857 		return -EINVAL;
858 	acc = oflag2acc[oflag & O_ACCMODE];
859 	return inode_permission(d_inode(dentry), acc);
860 }
861 
862 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
863 		      struct mq_attr *attr)
864 {
865 	struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
866 	struct dentry *root = mnt->mnt_root;
867 	struct filename *name;
868 	struct path path;
869 	int fd, error;
870 	int ro;
871 
872 	audit_mq_open(oflag, mode, attr);
873 
874 	if (IS_ERR(name = getname(u_name)))
875 		return PTR_ERR(name);
876 
877 	fd = get_unused_fd_flags(O_CLOEXEC);
878 	if (fd < 0)
879 		goto out_putname;
880 
881 	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
882 	inode_lock(d_inode(root));
883 	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
884 	if (IS_ERR(path.dentry)) {
885 		error = PTR_ERR(path.dentry);
886 		goto out_putfd;
887 	}
888 	path.mnt = mntget(mnt);
889 	error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
890 	if (!error) {
891 		struct file *file = dentry_open(&path, oflag, current_cred());
892 		if (!IS_ERR(file))
893 			fd_install(fd, file);
894 		else
895 			error = PTR_ERR(file);
896 	}
897 	path_put(&path);
898 out_putfd:
899 	if (error) {
900 		put_unused_fd(fd);
901 		fd = error;
902 	}
903 	inode_unlock(d_inode(root));
904 	if (!ro)
905 		mnt_drop_write(mnt);
906 out_putname:
907 	putname(name);
908 	return fd;
909 }
910 
911 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
912 		struct mq_attr __user *, u_attr)
913 {
914 	struct mq_attr attr;
915 	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
916 		return -EFAULT;
917 
918 	return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
919 }
920 
921 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
922 {
923 	int err;
924 	struct filename *name;
925 	struct dentry *dentry;
926 	struct inode *inode = NULL;
927 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
928 	struct vfsmount *mnt = ipc_ns->mq_mnt;
929 
930 	name = getname(u_name);
931 	if (IS_ERR(name))
932 		return PTR_ERR(name);
933 
934 	audit_inode_parent_hidden(name, mnt->mnt_root);
935 	err = mnt_want_write(mnt);
936 	if (err)
937 		goto out_name;
938 	inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
939 	dentry = lookup_one_len(name->name, mnt->mnt_root,
940 				strlen(name->name));
941 	if (IS_ERR(dentry)) {
942 		err = PTR_ERR(dentry);
943 		goto out_unlock;
944 	}
945 
946 	inode = d_inode(dentry);
947 	if (!inode) {
948 		err = -ENOENT;
949 	} else {
950 		ihold(inode);
951 		err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
952 	}
953 	dput(dentry);
954 
955 out_unlock:
956 	inode_unlock(d_inode(mnt->mnt_root));
957 	if (inode)
958 		iput(inode);
959 	mnt_drop_write(mnt);
960 out_name:
961 	putname(name);
962 
963 	return err;
964 }
965 
966 /* Pipelined send and receive functions.
967  *
968  * If a receiver finds no waiting message, then it registers itself in the
969  * list of waiting receivers. A sender checks that list before adding the new
970  * message into the message array. If there is a waiting receiver, then it
971  * bypasses the message array and directly hands the message over to the
972  * receiver. The receiver accepts the message and returns without grabbing the
973  * queue spinlock:
974  *
975  * - Set pointer to message.
976  * - Queue the receiver task for later wakeup (without the info->lock).
977  * - Update its state to STATE_READY. Now the receiver can continue.
978  * - Wake up the process after the lock is dropped. Should the process wake up
979  *   before this wakeup (due to a timeout or a signal) it will either see
980  *   STATE_READY and continue or acquire the lock to check the state again.
981  *
982  * The same algorithm is used for senders.
983  */
984 
985 static inline void __pipelined_op(struct wake_q_head *wake_q,
986 				  struct mqueue_inode_info *info,
987 				  struct ext_wait_queue *this)
988 {
989 	list_del(&this->list);
990 	get_task_struct(this->task);
991 
992 	/* see MQ_BARRIER for purpose/pairing */
993 	smp_store_release(&this->state, STATE_READY);
994 	wake_q_add_safe(wake_q, this->task);
995 }
996 
997 /* pipelined_send() - send a message directly to the task waiting in
998  * sys_mq_timedreceive() (without inserting message into a queue).
999  */
1000 static inline void pipelined_send(struct wake_q_head *wake_q,
1001 				  struct mqueue_inode_info *info,
1002 				  struct msg_msg *message,
1003 				  struct ext_wait_queue *receiver)
1004 {
1005 	receiver->msg = message;
1006 	__pipelined_op(wake_q, info, receiver);
1007 }
1008 
1009 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
1010  * gets its message and put to the queue (we have one free place for sure). */
1011 static inline void pipelined_receive(struct wake_q_head *wake_q,
1012 				     struct mqueue_inode_info *info)
1013 {
1014 	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
1015 
1016 	if (!sender) {
1017 		/* for poll */
1018 		wake_up_interruptible(&info->wait_q);
1019 		return;
1020 	}
1021 	if (msg_insert(sender->msg, info))
1022 		return;
1023 
1024 	__pipelined_op(wake_q, info, sender);
1025 }
1026 
1027 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
1028 		size_t msg_len, unsigned int msg_prio,
1029 		struct timespec64 *ts)
1030 {
1031 	struct fd f;
1032 	struct inode *inode;
1033 	struct ext_wait_queue wait;
1034 	struct ext_wait_queue *receiver;
1035 	struct msg_msg *msg_ptr;
1036 	struct mqueue_inode_info *info;
1037 	ktime_t expires, *timeout = NULL;
1038 	struct posix_msg_tree_node *new_leaf = NULL;
1039 	int ret = 0;
1040 	DEFINE_WAKE_Q(wake_q);
1041 
1042 	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
1043 		return -EINVAL;
1044 
1045 	if (ts) {
1046 		expires = timespec64_to_ktime(*ts);
1047 		timeout = &expires;
1048 	}
1049 
1050 	audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
1051 
1052 	f = fdget(mqdes);
1053 	if (unlikely(!f.file)) {
1054 		ret = -EBADF;
1055 		goto out;
1056 	}
1057 
1058 	inode = file_inode(f.file);
1059 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1060 		ret = -EBADF;
1061 		goto out_fput;
1062 	}
1063 	info = MQUEUE_I(inode);
1064 	audit_file(f.file);
1065 
1066 	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1067 		ret = -EBADF;
1068 		goto out_fput;
1069 	}
1070 
1071 	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1072 		ret = -EMSGSIZE;
1073 		goto out_fput;
1074 	}
1075 
1076 	/* First try to allocate memory, before doing anything with
1077 	 * existing queues. */
1078 	msg_ptr = load_msg(u_msg_ptr, msg_len);
1079 	if (IS_ERR(msg_ptr)) {
1080 		ret = PTR_ERR(msg_ptr);
1081 		goto out_fput;
1082 	}
1083 	msg_ptr->m_ts = msg_len;
1084 	msg_ptr->m_type = msg_prio;
1085 
1086 	/*
1087 	 * msg_insert really wants us to have a valid, spare node struct so
1088 	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1089 	 * fall back to that if necessary.
1090 	 */
1091 	if (!info->node_cache)
1092 		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1093 
1094 	spin_lock(&info->lock);
1095 
1096 	if (!info->node_cache && new_leaf) {
1097 		/* Save our speculative allocation into the cache */
1098 		INIT_LIST_HEAD(&new_leaf->msg_list);
1099 		info->node_cache = new_leaf;
1100 		new_leaf = NULL;
1101 	} else {
1102 		kfree(new_leaf);
1103 	}
1104 
1105 	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1106 		if (f.file->f_flags & O_NONBLOCK) {
1107 			ret = -EAGAIN;
1108 		} else {
1109 			wait.task = current;
1110 			wait.msg = (void *) msg_ptr;
1111 
1112 			/* memory barrier not required, we hold info->lock */
1113 			WRITE_ONCE(wait.state, STATE_NONE);
1114 			ret = wq_sleep(info, SEND, timeout, &wait);
1115 			/*
1116 			 * wq_sleep must be called with info->lock held, and
1117 			 * returns with the lock released
1118 			 */
1119 			goto out_free;
1120 		}
1121 	} else {
1122 		receiver = wq_get_first_waiter(info, RECV);
1123 		if (receiver) {
1124 			pipelined_send(&wake_q, info, msg_ptr, receiver);
1125 		} else {
1126 			/* adds message to the queue */
1127 			ret = msg_insert(msg_ptr, info);
1128 			if (ret)
1129 				goto out_unlock;
1130 			__do_notify(info);
1131 		}
1132 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1133 				current_time(inode);
1134 	}
1135 out_unlock:
1136 	spin_unlock(&info->lock);
1137 	wake_up_q(&wake_q);
1138 out_free:
1139 	if (ret)
1140 		free_msg(msg_ptr);
1141 out_fput:
1142 	fdput(f);
1143 out:
1144 	return ret;
1145 }
1146 
1147 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1148 		size_t msg_len, unsigned int __user *u_msg_prio,
1149 		struct timespec64 *ts)
1150 {
1151 	ssize_t ret;
1152 	struct msg_msg *msg_ptr;
1153 	struct fd f;
1154 	struct inode *inode;
1155 	struct mqueue_inode_info *info;
1156 	struct ext_wait_queue wait;
1157 	ktime_t expires, *timeout = NULL;
1158 	struct posix_msg_tree_node *new_leaf = NULL;
1159 
1160 	if (ts) {
1161 		expires = timespec64_to_ktime(*ts);
1162 		timeout = &expires;
1163 	}
1164 
1165 	audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1166 
1167 	f = fdget(mqdes);
1168 	if (unlikely(!f.file)) {
1169 		ret = -EBADF;
1170 		goto out;
1171 	}
1172 
1173 	inode = file_inode(f.file);
1174 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1175 		ret = -EBADF;
1176 		goto out_fput;
1177 	}
1178 	info = MQUEUE_I(inode);
1179 	audit_file(f.file);
1180 
1181 	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1182 		ret = -EBADF;
1183 		goto out_fput;
1184 	}
1185 
1186 	/* checks if buffer is big enough */
1187 	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1188 		ret = -EMSGSIZE;
1189 		goto out_fput;
1190 	}
1191 
1192 	/*
1193 	 * msg_insert really wants us to have a valid, spare node struct so
1194 	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1195 	 * fall back to that if necessary.
1196 	 */
1197 	if (!info->node_cache)
1198 		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1199 
1200 	spin_lock(&info->lock);
1201 
1202 	if (!info->node_cache && new_leaf) {
1203 		/* Save our speculative allocation into the cache */
1204 		INIT_LIST_HEAD(&new_leaf->msg_list);
1205 		info->node_cache = new_leaf;
1206 	} else {
1207 		kfree(new_leaf);
1208 	}
1209 
1210 	if (info->attr.mq_curmsgs == 0) {
1211 		if (f.file->f_flags & O_NONBLOCK) {
1212 			spin_unlock(&info->lock);
1213 			ret = -EAGAIN;
1214 		} else {
1215 			wait.task = current;
1216 
1217 			/* memory barrier not required, we hold info->lock */
1218 			WRITE_ONCE(wait.state, STATE_NONE);
1219 			ret = wq_sleep(info, RECV, timeout, &wait);
1220 			msg_ptr = wait.msg;
1221 		}
1222 	} else {
1223 		DEFINE_WAKE_Q(wake_q);
1224 
1225 		msg_ptr = msg_get(info);
1226 
1227 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1228 				current_time(inode);
1229 
1230 		/* There is now free space in queue. */
1231 		pipelined_receive(&wake_q, info);
1232 		spin_unlock(&info->lock);
1233 		wake_up_q(&wake_q);
1234 		ret = 0;
1235 	}
1236 	if (ret == 0) {
1237 		ret = msg_ptr->m_ts;
1238 
1239 		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1240 			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1241 			ret = -EFAULT;
1242 		}
1243 		free_msg(msg_ptr);
1244 	}
1245 out_fput:
1246 	fdput(f);
1247 out:
1248 	return ret;
1249 }
1250 
1251 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1252 		size_t, msg_len, unsigned int, msg_prio,
1253 		const struct __kernel_timespec __user *, u_abs_timeout)
1254 {
1255 	struct timespec64 ts, *p = NULL;
1256 	if (u_abs_timeout) {
1257 		int res = prepare_timeout(u_abs_timeout, &ts);
1258 		if (res)
1259 			return res;
1260 		p = &ts;
1261 	}
1262 	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1263 }
1264 
1265 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1266 		size_t, msg_len, unsigned int __user *, u_msg_prio,
1267 		const struct __kernel_timespec __user *, u_abs_timeout)
1268 {
1269 	struct timespec64 ts, *p = NULL;
1270 	if (u_abs_timeout) {
1271 		int res = prepare_timeout(u_abs_timeout, &ts);
1272 		if (res)
1273 			return res;
1274 		p = &ts;
1275 	}
1276 	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1277 }
1278 
1279 /*
1280  * Notes: the case when user wants us to deregister (with NULL as pointer)
1281  * and he isn't currently owner of notification, will be silently discarded.
1282  * It isn't explicitly defined in the POSIX.
1283  */
1284 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1285 {
1286 	int ret;
1287 	struct fd f;
1288 	struct sock *sock;
1289 	struct inode *inode;
1290 	struct mqueue_inode_info *info;
1291 	struct sk_buff *nc;
1292 
1293 	audit_mq_notify(mqdes, notification);
1294 
1295 	nc = NULL;
1296 	sock = NULL;
1297 	if (notification != NULL) {
1298 		if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1299 			     notification->sigev_notify != SIGEV_SIGNAL &&
1300 			     notification->sigev_notify != SIGEV_THREAD))
1301 			return -EINVAL;
1302 		if (notification->sigev_notify == SIGEV_SIGNAL &&
1303 			!valid_signal(notification->sigev_signo)) {
1304 			return -EINVAL;
1305 		}
1306 		if (notification->sigev_notify == SIGEV_THREAD) {
1307 			long timeo;
1308 
1309 			/* create the notify skb */
1310 			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1311 			if (!nc)
1312 				return -ENOMEM;
1313 
1314 			if (copy_from_user(nc->data,
1315 					notification->sigev_value.sival_ptr,
1316 					NOTIFY_COOKIE_LEN)) {
1317 				ret = -EFAULT;
1318 				goto free_skb;
1319 			}
1320 
1321 			/* TODO: add a header? */
1322 			skb_put(nc, NOTIFY_COOKIE_LEN);
1323 			/* and attach it to the socket */
1324 retry:
1325 			f = fdget(notification->sigev_signo);
1326 			if (!f.file) {
1327 				ret = -EBADF;
1328 				goto out;
1329 			}
1330 			sock = netlink_getsockbyfilp(f.file);
1331 			fdput(f);
1332 			if (IS_ERR(sock)) {
1333 				ret = PTR_ERR(sock);
1334 				goto free_skb;
1335 			}
1336 
1337 			timeo = MAX_SCHEDULE_TIMEOUT;
1338 			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1339 			if (ret == 1) {
1340 				sock = NULL;
1341 				goto retry;
1342 			}
1343 			if (ret)
1344 				return ret;
1345 		}
1346 	}
1347 
1348 	f = fdget(mqdes);
1349 	if (!f.file) {
1350 		ret = -EBADF;
1351 		goto out;
1352 	}
1353 
1354 	inode = file_inode(f.file);
1355 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1356 		ret = -EBADF;
1357 		goto out_fput;
1358 	}
1359 	info = MQUEUE_I(inode);
1360 
1361 	ret = 0;
1362 	spin_lock(&info->lock);
1363 	if (notification == NULL) {
1364 		if (info->notify_owner == task_tgid(current)) {
1365 			remove_notification(info);
1366 			inode->i_atime = inode->i_ctime = current_time(inode);
1367 		}
1368 	} else if (info->notify_owner != NULL) {
1369 		ret = -EBUSY;
1370 	} else {
1371 		switch (notification->sigev_notify) {
1372 		case SIGEV_NONE:
1373 			info->notify.sigev_notify = SIGEV_NONE;
1374 			break;
1375 		case SIGEV_THREAD:
1376 			info->notify_sock = sock;
1377 			info->notify_cookie = nc;
1378 			sock = NULL;
1379 			nc = NULL;
1380 			info->notify.sigev_notify = SIGEV_THREAD;
1381 			break;
1382 		case SIGEV_SIGNAL:
1383 			info->notify.sigev_signo = notification->sigev_signo;
1384 			info->notify.sigev_value = notification->sigev_value;
1385 			info->notify.sigev_notify = SIGEV_SIGNAL;
1386 			break;
1387 		}
1388 
1389 		info->notify_owner = get_pid(task_tgid(current));
1390 		info->notify_user_ns = get_user_ns(current_user_ns());
1391 		inode->i_atime = inode->i_ctime = current_time(inode);
1392 	}
1393 	spin_unlock(&info->lock);
1394 out_fput:
1395 	fdput(f);
1396 out:
1397 	if (sock)
1398 		netlink_detachskb(sock, nc);
1399 	else
1400 free_skb:
1401 		dev_kfree_skb(nc);
1402 
1403 	return ret;
1404 }
1405 
1406 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1407 		const struct sigevent __user *, u_notification)
1408 {
1409 	struct sigevent n, *p = NULL;
1410 	if (u_notification) {
1411 		if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1412 			return -EFAULT;
1413 		p = &n;
1414 	}
1415 	return do_mq_notify(mqdes, p);
1416 }
1417 
1418 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1419 {
1420 	struct fd f;
1421 	struct inode *inode;
1422 	struct mqueue_inode_info *info;
1423 
1424 	if (new && (new->mq_flags & (~O_NONBLOCK)))
1425 		return -EINVAL;
1426 
1427 	f = fdget(mqdes);
1428 	if (!f.file)
1429 		return -EBADF;
1430 
1431 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1432 		fdput(f);
1433 		return -EBADF;
1434 	}
1435 
1436 	inode = file_inode(f.file);
1437 	info = MQUEUE_I(inode);
1438 
1439 	spin_lock(&info->lock);
1440 
1441 	if (old) {
1442 		*old = info->attr;
1443 		old->mq_flags = f.file->f_flags & O_NONBLOCK;
1444 	}
1445 	if (new) {
1446 		audit_mq_getsetattr(mqdes, new);
1447 		spin_lock(&f.file->f_lock);
1448 		if (new->mq_flags & O_NONBLOCK)
1449 			f.file->f_flags |= O_NONBLOCK;
1450 		else
1451 			f.file->f_flags &= ~O_NONBLOCK;
1452 		spin_unlock(&f.file->f_lock);
1453 
1454 		inode->i_atime = inode->i_ctime = current_time(inode);
1455 	}
1456 
1457 	spin_unlock(&info->lock);
1458 	fdput(f);
1459 	return 0;
1460 }
1461 
1462 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1463 		const struct mq_attr __user *, u_mqstat,
1464 		struct mq_attr __user *, u_omqstat)
1465 {
1466 	int ret;
1467 	struct mq_attr mqstat, omqstat;
1468 	struct mq_attr *new = NULL, *old = NULL;
1469 
1470 	if (u_mqstat) {
1471 		new = &mqstat;
1472 		if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1473 			return -EFAULT;
1474 	}
1475 	if (u_omqstat)
1476 		old = &omqstat;
1477 
1478 	ret = do_mq_getsetattr(mqdes, new, old);
1479 	if (ret || !old)
1480 		return ret;
1481 
1482 	if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1483 		return -EFAULT;
1484 	return 0;
1485 }
1486 
1487 #ifdef CONFIG_COMPAT
1488 
1489 struct compat_mq_attr {
1490 	compat_long_t mq_flags;      /* message queue flags		     */
1491 	compat_long_t mq_maxmsg;     /* maximum number of messages	     */
1492 	compat_long_t mq_msgsize;    /* maximum message size		     */
1493 	compat_long_t mq_curmsgs;    /* number of messages currently queued  */
1494 	compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1495 };
1496 
1497 static inline int get_compat_mq_attr(struct mq_attr *attr,
1498 			const struct compat_mq_attr __user *uattr)
1499 {
1500 	struct compat_mq_attr v;
1501 
1502 	if (copy_from_user(&v, uattr, sizeof(*uattr)))
1503 		return -EFAULT;
1504 
1505 	memset(attr, 0, sizeof(*attr));
1506 	attr->mq_flags = v.mq_flags;
1507 	attr->mq_maxmsg = v.mq_maxmsg;
1508 	attr->mq_msgsize = v.mq_msgsize;
1509 	attr->mq_curmsgs = v.mq_curmsgs;
1510 	return 0;
1511 }
1512 
1513 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1514 			struct compat_mq_attr __user *uattr)
1515 {
1516 	struct compat_mq_attr v;
1517 
1518 	memset(&v, 0, sizeof(v));
1519 	v.mq_flags = attr->mq_flags;
1520 	v.mq_maxmsg = attr->mq_maxmsg;
1521 	v.mq_msgsize = attr->mq_msgsize;
1522 	v.mq_curmsgs = attr->mq_curmsgs;
1523 	if (copy_to_user(uattr, &v, sizeof(*uattr)))
1524 		return -EFAULT;
1525 	return 0;
1526 }
1527 
1528 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1529 		       int, oflag, compat_mode_t, mode,
1530 		       struct compat_mq_attr __user *, u_attr)
1531 {
1532 	struct mq_attr attr, *p = NULL;
1533 	if (u_attr && oflag & O_CREAT) {
1534 		p = &attr;
1535 		if (get_compat_mq_attr(&attr, u_attr))
1536 			return -EFAULT;
1537 	}
1538 	return do_mq_open(u_name, oflag, mode, p);
1539 }
1540 
1541 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1542 		       const struct compat_sigevent __user *, u_notification)
1543 {
1544 	struct sigevent n, *p = NULL;
1545 	if (u_notification) {
1546 		if (get_compat_sigevent(&n, u_notification))
1547 			return -EFAULT;
1548 		if (n.sigev_notify == SIGEV_THREAD)
1549 			n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1550 		p = &n;
1551 	}
1552 	return do_mq_notify(mqdes, p);
1553 }
1554 
1555 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1556 		       const struct compat_mq_attr __user *, u_mqstat,
1557 		       struct compat_mq_attr __user *, u_omqstat)
1558 {
1559 	int ret;
1560 	struct mq_attr mqstat, omqstat;
1561 	struct mq_attr *new = NULL, *old = NULL;
1562 
1563 	if (u_mqstat) {
1564 		new = &mqstat;
1565 		if (get_compat_mq_attr(new, u_mqstat))
1566 			return -EFAULT;
1567 	}
1568 	if (u_omqstat)
1569 		old = &omqstat;
1570 
1571 	ret = do_mq_getsetattr(mqdes, new, old);
1572 	if (ret || !old)
1573 		return ret;
1574 
1575 	if (put_compat_mq_attr(old, u_omqstat))
1576 		return -EFAULT;
1577 	return 0;
1578 }
1579 #endif
1580 
1581 #ifdef CONFIG_COMPAT_32BIT_TIME
1582 static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1583 				   struct timespec64 *ts)
1584 {
1585 	if (get_old_timespec32(ts, p))
1586 		return -EFAULT;
1587 	if (!timespec64_valid(ts))
1588 		return -EINVAL;
1589 	return 0;
1590 }
1591 
1592 SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1593 		const char __user *, u_msg_ptr,
1594 		unsigned int, msg_len, unsigned int, msg_prio,
1595 		const struct old_timespec32 __user *, u_abs_timeout)
1596 {
1597 	struct timespec64 ts, *p = NULL;
1598 	if (u_abs_timeout) {
1599 		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1600 		if (res)
1601 			return res;
1602 		p = &ts;
1603 	}
1604 	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1605 }
1606 
1607 SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1608 		char __user *, u_msg_ptr,
1609 		unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1610 		const struct old_timespec32 __user *, u_abs_timeout)
1611 {
1612 	struct timespec64 ts, *p = NULL;
1613 	if (u_abs_timeout) {
1614 		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1615 		if (res)
1616 			return res;
1617 		p = &ts;
1618 	}
1619 	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1620 }
1621 #endif
1622 
1623 static const struct inode_operations mqueue_dir_inode_operations = {
1624 	.lookup = simple_lookup,
1625 	.create = mqueue_create,
1626 	.unlink = mqueue_unlink,
1627 };
1628 
1629 static const struct file_operations mqueue_file_operations = {
1630 	.flush = mqueue_flush_file,
1631 	.poll = mqueue_poll_file,
1632 	.read = mqueue_read_file,
1633 	.llseek = default_llseek,
1634 };
1635 
1636 static const struct super_operations mqueue_super_ops = {
1637 	.alloc_inode = mqueue_alloc_inode,
1638 	.free_inode = mqueue_free_inode,
1639 	.evict_inode = mqueue_evict_inode,
1640 	.statfs = simple_statfs,
1641 };
1642 
1643 static const struct fs_context_operations mqueue_fs_context_ops = {
1644 	.free		= mqueue_fs_context_free,
1645 	.get_tree	= mqueue_get_tree,
1646 };
1647 
1648 static struct file_system_type mqueue_fs_type = {
1649 	.name			= "mqueue",
1650 	.init_fs_context	= mqueue_init_fs_context,
1651 	.kill_sb		= kill_litter_super,
1652 	.fs_flags		= FS_USERNS_MOUNT,
1653 };
1654 
1655 int mq_init_ns(struct ipc_namespace *ns)
1656 {
1657 	struct vfsmount *m;
1658 
1659 	ns->mq_queues_count  = 0;
1660 	ns->mq_queues_max    = DFLT_QUEUESMAX;
1661 	ns->mq_msg_max       = DFLT_MSGMAX;
1662 	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1663 	ns->mq_msg_default   = DFLT_MSG;
1664 	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1665 
1666 	m = mq_create_mount(ns);
1667 	if (IS_ERR(m))
1668 		return PTR_ERR(m);
1669 	ns->mq_mnt = m;
1670 	return 0;
1671 }
1672 
1673 void mq_clear_sbinfo(struct ipc_namespace *ns)
1674 {
1675 	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1676 }
1677 
1678 void mq_put_mnt(struct ipc_namespace *ns)
1679 {
1680 	kern_unmount(ns->mq_mnt);
1681 }
1682 
1683 static int __init init_mqueue_fs(void)
1684 {
1685 	int error;
1686 
1687 	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1688 				sizeof(struct mqueue_inode_info), 0,
1689 				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1690 	if (mqueue_inode_cachep == NULL)
1691 		return -ENOMEM;
1692 
1693 	/* ignore failures - they are not fatal */
1694 	mq_sysctl_table = mq_register_sysctl_table();
1695 
1696 	error = register_filesystem(&mqueue_fs_type);
1697 	if (error)
1698 		goto out_sysctl;
1699 
1700 	spin_lock_init(&mq_lock);
1701 
1702 	error = mq_init_ns(&init_ipc_ns);
1703 	if (error)
1704 		goto out_filesystem;
1705 
1706 	return 0;
1707 
1708 out_filesystem:
1709 	unregister_filesystem(&mqueue_fs_type);
1710 out_sysctl:
1711 	if (mq_sysctl_table)
1712 		unregister_sysctl_table(mq_sysctl_table);
1713 	kmem_cache_destroy(mqueue_inode_cachep);
1714 	return error;
1715 }
1716 
1717 device_initcall(init_mqueue_fs);
1718