xref: /linux/ipc/mqueue.c (revision c5b2cbdb)
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 
249 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
250 {
251 	struct rb_node *parent = NULL;
252 	struct posix_msg_tree_node *leaf;
253 	struct msg_msg *msg;
254 
255 try_again:
256 	/*
257 	 * During insert, low priorities go to the left and high to the
258 	 * right.  On receive, we want the highest priorities first, so
259 	 * walk all the way to the right.
260 	 */
261 	parent = info->msg_tree_rightmost;
262 	if (!parent) {
263 		if (info->attr.mq_curmsgs) {
264 			pr_warn_once("Inconsistency in POSIX message queue, "
265 				     "no tree element, but supposedly messages "
266 				     "should exist!\n");
267 			info->attr.mq_curmsgs = 0;
268 		}
269 		return NULL;
270 	}
271 	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
272 	if (unlikely(list_empty(&leaf->msg_list))) {
273 		pr_warn_once("Inconsistency in POSIX message queue, "
274 			     "empty leaf node but we haven't implemented "
275 			     "lazy leaf delete!\n");
276 		msg_tree_erase(leaf, info);
277 		goto try_again;
278 	} else {
279 		msg = list_first_entry(&leaf->msg_list,
280 				       struct msg_msg, m_list);
281 		list_del(&msg->m_list);
282 		if (list_empty(&leaf->msg_list)) {
283 			msg_tree_erase(leaf, info);
284 		}
285 	}
286 	info->attr.mq_curmsgs--;
287 	info->qsize -= msg->m_ts;
288 	return msg;
289 }
290 
291 static struct inode *mqueue_get_inode(struct super_block *sb,
292 		struct ipc_namespace *ipc_ns, umode_t mode,
293 		struct mq_attr *attr)
294 {
295 	struct user_struct *u = current_user();
296 	struct inode *inode;
297 	int ret = -ENOMEM;
298 
299 	inode = new_inode(sb);
300 	if (!inode)
301 		goto err;
302 
303 	inode->i_ino = get_next_ino();
304 	inode->i_mode = mode;
305 	inode->i_uid = current_fsuid();
306 	inode->i_gid = current_fsgid();
307 	inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
308 
309 	if (S_ISREG(mode)) {
310 		struct mqueue_inode_info *info;
311 		unsigned long mq_bytes, mq_treesize;
312 
313 		inode->i_fop = &mqueue_file_operations;
314 		inode->i_size = FILENT_SIZE;
315 		/* mqueue specific info */
316 		info = MQUEUE_I(inode);
317 		spin_lock_init(&info->lock);
318 		init_waitqueue_head(&info->wait_q);
319 		INIT_LIST_HEAD(&info->e_wait_q[0].list);
320 		INIT_LIST_HEAD(&info->e_wait_q[1].list);
321 		info->notify_owner = NULL;
322 		info->notify_user_ns = NULL;
323 		info->qsize = 0;
324 		info->user = NULL;	/* set when all is ok */
325 		info->msg_tree = RB_ROOT;
326 		info->msg_tree_rightmost = NULL;
327 		info->node_cache = NULL;
328 		memset(&info->attr, 0, sizeof(info->attr));
329 		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
330 					   ipc_ns->mq_msg_default);
331 		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
332 					    ipc_ns->mq_msgsize_default);
333 		if (attr) {
334 			info->attr.mq_maxmsg = attr->mq_maxmsg;
335 			info->attr.mq_msgsize = attr->mq_msgsize;
336 		}
337 		/*
338 		 * We used to allocate a static array of pointers and account
339 		 * the size of that array as well as one msg_msg struct per
340 		 * possible message into the queue size. That's no longer
341 		 * accurate as the queue is now an rbtree and will grow and
342 		 * shrink depending on usage patterns.  We can, however, still
343 		 * account one msg_msg struct per message, but the nodes are
344 		 * allocated depending on priority usage, and most programs
345 		 * only use one, or a handful, of priorities.  However, since
346 		 * this is pinned memory, we need to assume worst case, so
347 		 * that means the min(mq_maxmsg, max_priorities) * struct
348 		 * posix_msg_tree_node.
349 		 */
350 
351 		ret = -EINVAL;
352 		if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
353 			goto out_inode;
354 		if (capable(CAP_SYS_RESOURCE)) {
355 			if (info->attr.mq_maxmsg > HARD_MSGMAX ||
356 			    info->attr.mq_msgsize > HARD_MSGSIZEMAX)
357 				goto out_inode;
358 		} else {
359 			if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
360 					info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
361 				goto out_inode;
362 		}
363 		ret = -EOVERFLOW;
364 		/* check for overflow */
365 		if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
366 			goto out_inode;
367 		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
368 			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
369 			sizeof(struct posix_msg_tree_node);
370 		mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
371 		if (mq_bytes + mq_treesize < mq_bytes)
372 			goto out_inode;
373 		mq_bytes += mq_treesize;
374 		spin_lock(&mq_lock);
375 		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
376 		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
377 			spin_unlock(&mq_lock);
378 			/* mqueue_evict_inode() releases info->messages */
379 			ret = -EMFILE;
380 			goto out_inode;
381 		}
382 		u->mq_bytes += mq_bytes;
383 		spin_unlock(&mq_lock);
384 
385 		/* all is ok */
386 		info->user = get_uid(u);
387 	} else if (S_ISDIR(mode)) {
388 		inc_nlink(inode);
389 		/* Some things misbehave if size == 0 on a directory */
390 		inode->i_size = 2 * DIRENT_SIZE;
391 		inode->i_op = &mqueue_dir_inode_operations;
392 		inode->i_fop = &simple_dir_operations;
393 	}
394 
395 	return inode;
396 out_inode:
397 	iput(inode);
398 err:
399 	return ERR_PTR(ret);
400 }
401 
402 static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
403 {
404 	struct inode *inode;
405 	struct ipc_namespace *ns = sb->s_fs_info;
406 
407 	sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
408 	sb->s_blocksize = PAGE_SIZE;
409 	sb->s_blocksize_bits = PAGE_SHIFT;
410 	sb->s_magic = MQUEUE_MAGIC;
411 	sb->s_op = &mqueue_super_ops;
412 
413 	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
414 	if (IS_ERR(inode))
415 		return PTR_ERR(inode);
416 
417 	sb->s_root = d_make_root(inode);
418 	if (!sb->s_root)
419 		return -ENOMEM;
420 	return 0;
421 }
422 
423 static int mqueue_get_tree(struct fs_context *fc)
424 {
425 	struct mqueue_fs_context *ctx = fc->fs_private;
426 
427 	return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
428 }
429 
430 static void mqueue_fs_context_free(struct fs_context *fc)
431 {
432 	struct mqueue_fs_context *ctx = fc->fs_private;
433 
434 	put_ipc_ns(ctx->ipc_ns);
435 	kfree(ctx);
436 }
437 
438 static int mqueue_init_fs_context(struct fs_context *fc)
439 {
440 	struct mqueue_fs_context *ctx;
441 
442 	ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
443 	if (!ctx)
444 		return -ENOMEM;
445 
446 	ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
447 	put_user_ns(fc->user_ns);
448 	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
449 	fc->fs_private = ctx;
450 	fc->ops = &mqueue_fs_context_ops;
451 	return 0;
452 }
453 
454 static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
455 {
456 	struct mqueue_fs_context *ctx;
457 	struct fs_context *fc;
458 	struct vfsmount *mnt;
459 
460 	fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
461 	if (IS_ERR(fc))
462 		return ERR_CAST(fc);
463 
464 	ctx = fc->fs_private;
465 	put_ipc_ns(ctx->ipc_ns);
466 	ctx->ipc_ns = get_ipc_ns(ns);
467 	put_user_ns(fc->user_ns);
468 	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
469 
470 	mnt = fc_mount(fc);
471 	put_fs_context(fc);
472 	return mnt;
473 }
474 
475 static void init_once(void *foo)
476 {
477 	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
478 
479 	inode_init_once(&p->vfs_inode);
480 }
481 
482 static struct inode *mqueue_alloc_inode(struct super_block *sb)
483 {
484 	struct mqueue_inode_info *ei;
485 
486 	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
487 	if (!ei)
488 		return NULL;
489 	return &ei->vfs_inode;
490 }
491 
492 static void mqueue_free_inode(struct inode *inode)
493 {
494 	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
495 }
496 
497 static void mqueue_evict_inode(struct inode *inode)
498 {
499 	struct mqueue_inode_info *info;
500 	struct user_struct *user;
501 	struct ipc_namespace *ipc_ns;
502 	struct msg_msg *msg, *nmsg;
503 	LIST_HEAD(tmp_msg);
504 
505 	clear_inode(inode);
506 
507 	if (S_ISDIR(inode->i_mode))
508 		return;
509 
510 	ipc_ns = get_ns_from_inode(inode);
511 	info = MQUEUE_I(inode);
512 	spin_lock(&info->lock);
513 	while ((msg = msg_get(info)) != NULL)
514 		list_add_tail(&msg->m_list, &tmp_msg);
515 	kfree(info->node_cache);
516 	spin_unlock(&info->lock);
517 
518 	list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
519 		list_del(&msg->m_list);
520 		free_msg(msg);
521 	}
522 
523 	user = info->user;
524 	if (user) {
525 		unsigned long mq_bytes, mq_treesize;
526 
527 		/* Total amount of bytes accounted for the mqueue */
528 		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
529 			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
530 			sizeof(struct posix_msg_tree_node);
531 
532 		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
533 					  info->attr.mq_msgsize);
534 
535 		spin_lock(&mq_lock);
536 		user->mq_bytes -= mq_bytes;
537 		/*
538 		 * get_ns_from_inode() ensures that the
539 		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
540 		 * to which we now hold a reference, or it is NULL.
541 		 * We can't put it here under mq_lock, though.
542 		 */
543 		if (ipc_ns)
544 			ipc_ns->mq_queues_count--;
545 		spin_unlock(&mq_lock);
546 		free_uid(user);
547 	}
548 	if (ipc_ns)
549 		put_ipc_ns(ipc_ns);
550 }
551 
552 static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
553 {
554 	struct inode *dir = dentry->d_parent->d_inode;
555 	struct inode *inode;
556 	struct mq_attr *attr = arg;
557 	int error;
558 	struct ipc_namespace *ipc_ns;
559 
560 	spin_lock(&mq_lock);
561 	ipc_ns = __get_ns_from_inode(dir);
562 	if (!ipc_ns) {
563 		error = -EACCES;
564 		goto out_unlock;
565 	}
566 
567 	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
568 	    !capable(CAP_SYS_RESOURCE)) {
569 		error = -ENOSPC;
570 		goto out_unlock;
571 	}
572 	ipc_ns->mq_queues_count++;
573 	spin_unlock(&mq_lock);
574 
575 	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
576 	if (IS_ERR(inode)) {
577 		error = PTR_ERR(inode);
578 		spin_lock(&mq_lock);
579 		ipc_ns->mq_queues_count--;
580 		goto out_unlock;
581 	}
582 
583 	put_ipc_ns(ipc_ns);
584 	dir->i_size += DIRENT_SIZE;
585 	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
586 
587 	d_instantiate(dentry, inode);
588 	dget(dentry);
589 	return 0;
590 out_unlock:
591 	spin_unlock(&mq_lock);
592 	if (ipc_ns)
593 		put_ipc_ns(ipc_ns);
594 	return error;
595 }
596 
597 static int mqueue_create(struct inode *dir, struct dentry *dentry,
598 				umode_t mode, bool excl)
599 {
600 	return mqueue_create_attr(dentry, mode, NULL);
601 }
602 
603 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
604 {
605 	struct inode *inode = d_inode(dentry);
606 
607 	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
608 	dir->i_size -= DIRENT_SIZE;
609 	drop_nlink(inode);
610 	dput(dentry);
611 	return 0;
612 }
613 
614 /*
615 *	This is routine for system read from queue file.
616 *	To avoid mess with doing here some sort of mq_receive we allow
617 *	to read only queue size & notification info (the only values
618 *	that are interesting from user point of view and aren't accessible
619 *	through std routines)
620 */
621 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
622 				size_t count, loff_t *off)
623 {
624 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
625 	char buffer[FILENT_SIZE];
626 	ssize_t ret;
627 
628 	spin_lock(&info->lock);
629 	snprintf(buffer, sizeof(buffer),
630 			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
631 			info->qsize,
632 			info->notify_owner ? info->notify.sigev_notify : 0,
633 			(info->notify_owner &&
634 			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
635 				info->notify.sigev_signo : 0,
636 			pid_vnr(info->notify_owner));
637 	spin_unlock(&info->lock);
638 	buffer[sizeof(buffer)-1] = '\0';
639 
640 	ret = simple_read_from_buffer(u_data, count, off, buffer,
641 				strlen(buffer));
642 	if (ret <= 0)
643 		return ret;
644 
645 	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
646 	return ret;
647 }
648 
649 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
650 {
651 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
652 
653 	spin_lock(&info->lock);
654 	if (task_tgid(current) == info->notify_owner)
655 		remove_notification(info);
656 
657 	spin_unlock(&info->lock);
658 	return 0;
659 }
660 
661 static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
662 {
663 	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
664 	__poll_t retval = 0;
665 
666 	poll_wait(filp, &info->wait_q, poll_tab);
667 
668 	spin_lock(&info->lock);
669 	if (info->attr.mq_curmsgs)
670 		retval = EPOLLIN | EPOLLRDNORM;
671 
672 	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
673 		retval |= EPOLLOUT | EPOLLWRNORM;
674 	spin_unlock(&info->lock);
675 
676 	return retval;
677 }
678 
679 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
680 static void wq_add(struct mqueue_inode_info *info, int sr,
681 			struct ext_wait_queue *ewp)
682 {
683 	struct ext_wait_queue *walk;
684 
685 	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
686 		if (walk->task->prio <= current->prio) {
687 			list_add_tail(&ewp->list, &walk->list);
688 			return;
689 		}
690 	}
691 	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
692 }
693 
694 /*
695  * Puts current task to sleep. Caller must hold queue lock. After return
696  * lock isn't held.
697  * sr: SEND or RECV
698  */
699 static int wq_sleep(struct mqueue_inode_info *info, int sr,
700 		    ktime_t *timeout, struct ext_wait_queue *ewp)
701 	__releases(&info->lock)
702 {
703 	int retval;
704 	signed long time;
705 
706 	wq_add(info, sr, ewp);
707 
708 	for (;;) {
709 		/* memory barrier not required, we hold info->lock */
710 		__set_current_state(TASK_INTERRUPTIBLE);
711 
712 		spin_unlock(&info->lock);
713 		time = schedule_hrtimeout_range_clock(timeout, 0,
714 			HRTIMER_MODE_ABS, CLOCK_REALTIME);
715 
716 		if (READ_ONCE(ewp->state) == STATE_READY) {
717 			/* see MQ_BARRIER for purpose/pairing */
718 			smp_acquire__after_ctrl_dep();
719 			retval = 0;
720 			goto out;
721 		}
722 		spin_lock(&info->lock);
723 
724 		/* we hold info->lock, so no memory barrier required */
725 		if (READ_ONCE(ewp->state) == STATE_READY) {
726 			retval = 0;
727 			goto out_unlock;
728 		}
729 		if (signal_pending(current)) {
730 			retval = -ERESTARTSYS;
731 			break;
732 		}
733 		if (time == 0) {
734 			retval = -ETIMEDOUT;
735 			break;
736 		}
737 	}
738 	list_del(&ewp->list);
739 out_unlock:
740 	spin_unlock(&info->lock);
741 out:
742 	return retval;
743 }
744 
745 /*
746  * Returns waiting task that should be serviced first or NULL if none exists
747  */
748 static struct ext_wait_queue *wq_get_first_waiter(
749 		struct mqueue_inode_info *info, int sr)
750 {
751 	struct list_head *ptr;
752 
753 	ptr = info->e_wait_q[sr].list.prev;
754 	if (ptr == &info->e_wait_q[sr].list)
755 		return NULL;
756 	return list_entry(ptr, struct ext_wait_queue, list);
757 }
758 
759 
760 static inline void set_cookie(struct sk_buff *skb, char code)
761 {
762 	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
763 }
764 
765 /*
766  * The next function is only to split too long sys_mq_timedsend
767  */
768 static void __do_notify(struct mqueue_inode_info *info)
769 {
770 	/* notification
771 	 * invoked when there is registered process and there isn't process
772 	 * waiting synchronously for message AND state of queue changed from
773 	 * empty to not empty. Here we are sure that no one is waiting
774 	 * synchronously. */
775 	if (info->notify_owner &&
776 	    info->attr.mq_curmsgs == 1) {
777 		struct kernel_siginfo sig_i;
778 		switch (info->notify.sigev_notify) {
779 		case SIGEV_NONE:
780 			break;
781 		case SIGEV_SIGNAL:
782 			/* sends signal */
783 
784 			clear_siginfo(&sig_i);
785 			sig_i.si_signo = info->notify.sigev_signo;
786 			sig_i.si_errno = 0;
787 			sig_i.si_code = SI_MESGQ;
788 			sig_i.si_value = info->notify.sigev_value;
789 			/* map current pid/uid into info->owner's namespaces */
790 			rcu_read_lock();
791 			sig_i.si_pid = task_tgid_nr_ns(current,
792 						ns_of_pid(info->notify_owner));
793 			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
794 			rcu_read_unlock();
795 
796 			kill_pid_info(info->notify.sigev_signo,
797 				      &sig_i, info->notify_owner);
798 			break;
799 		case SIGEV_THREAD:
800 			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
801 			netlink_sendskb(info->notify_sock, info->notify_cookie);
802 			break;
803 		}
804 		/* after notification unregisters process */
805 		put_pid(info->notify_owner);
806 		put_user_ns(info->notify_user_ns);
807 		info->notify_owner = NULL;
808 		info->notify_user_ns = NULL;
809 	}
810 	wake_up(&info->wait_q);
811 }
812 
813 static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
814 			   struct timespec64 *ts)
815 {
816 	if (get_timespec64(ts, u_abs_timeout))
817 		return -EFAULT;
818 	if (!timespec64_valid(ts))
819 		return -EINVAL;
820 	return 0;
821 }
822 
823 static void remove_notification(struct mqueue_inode_info *info)
824 {
825 	if (info->notify_owner != NULL &&
826 	    info->notify.sigev_notify == SIGEV_THREAD) {
827 		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
828 		netlink_sendskb(info->notify_sock, info->notify_cookie);
829 	}
830 	put_pid(info->notify_owner);
831 	put_user_ns(info->notify_user_ns);
832 	info->notify_owner = NULL;
833 	info->notify_user_ns = NULL;
834 }
835 
836 static int prepare_open(struct dentry *dentry, int oflag, int ro,
837 			umode_t mode, struct filename *name,
838 			struct mq_attr *attr)
839 {
840 	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
841 						  MAY_READ | MAY_WRITE };
842 	int acc;
843 
844 	if (d_really_is_negative(dentry)) {
845 		if (!(oflag & O_CREAT))
846 			return -ENOENT;
847 		if (ro)
848 			return ro;
849 		audit_inode_parent_hidden(name, dentry->d_parent);
850 		return vfs_mkobj(dentry, mode & ~current_umask(),
851 				  mqueue_create_attr, attr);
852 	}
853 	/* it already existed */
854 	audit_inode(name, dentry, 0);
855 	if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
856 		return -EEXIST;
857 	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
858 		return -EINVAL;
859 	acc = oflag2acc[oflag & O_ACCMODE];
860 	return inode_permission(d_inode(dentry), acc);
861 }
862 
863 static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
864 		      struct mq_attr *attr)
865 {
866 	struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
867 	struct dentry *root = mnt->mnt_root;
868 	struct filename *name;
869 	struct path path;
870 	int fd, error;
871 	int ro;
872 
873 	audit_mq_open(oflag, mode, attr);
874 
875 	if (IS_ERR(name = getname(u_name)))
876 		return PTR_ERR(name);
877 
878 	fd = get_unused_fd_flags(O_CLOEXEC);
879 	if (fd < 0)
880 		goto out_putname;
881 
882 	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
883 	inode_lock(d_inode(root));
884 	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
885 	if (IS_ERR(path.dentry)) {
886 		error = PTR_ERR(path.dentry);
887 		goto out_putfd;
888 	}
889 	path.mnt = mntget(mnt);
890 	error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
891 	if (!error) {
892 		struct file *file = dentry_open(&path, oflag, current_cred());
893 		if (!IS_ERR(file))
894 			fd_install(fd, file);
895 		else
896 			error = PTR_ERR(file);
897 	}
898 	path_put(&path);
899 out_putfd:
900 	if (error) {
901 		put_unused_fd(fd);
902 		fd = error;
903 	}
904 	inode_unlock(d_inode(root));
905 	if (!ro)
906 		mnt_drop_write(mnt);
907 out_putname:
908 	putname(name);
909 	return fd;
910 }
911 
912 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
913 		struct mq_attr __user *, u_attr)
914 {
915 	struct mq_attr attr;
916 	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
917 		return -EFAULT;
918 
919 	return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
920 }
921 
922 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
923 {
924 	int err;
925 	struct filename *name;
926 	struct dentry *dentry;
927 	struct inode *inode = NULL;
928 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
929 	struct vfsmount *mnt = ipc_ns->mq_mnt;
930 
931 	name = getname(u_name);
932 	if (IS_ERR(name))
933 		return PTR_ERR(name);
934 
935 	audit_inode_parent_hidden(name, mnt->mnt_root);
936 	err = mnt_want_write(mnt);
937 	if (err)
938 		goto out_name;
939 	inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
940 	dentry = lookup_one_len(name->name, mnt->mnt_root,
941 				strlen(name->name));
942 	if (IS_ERR(dentry)) {
943 		err = PTR_ERR(dentry);
944 		goto out_unlock;
945 	}
946 
947 	inode = d_inode(dentry);
948 	if (!inode) {
949 		err = -ENOENT;
950 	} else {
951 		ihold(inode);
952 		err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
953 	}
954 	dput(dentry);
955 
956 out_unlock:
957 	inode_unlock(d_inode(mnt->mnt_root));
958 	if (inode)
959 		iput(inode);
960 	mnt_drop_write(mnt);
961 out_name:
962 	putname(name);
963 
964 	return err;
965 }
966 
967 /* Pipelined send and receive functions.
968  *
969  * If a receiver finds no waiting message, then it registers itself in the
970  * list of waiting receivers. A sender checks that list before adding the new
971  * message into the message array. If there is a waiting receiver, then it
972  * bypasses the message array and directly hands the message over to the
973  * receiver. The receiver accepts the message and returns without grabbing the
974  * queue spinlock:
975  *
976  * - Set pointer to message.
977  * - Queue the receiver task for later wakeup (without the info->lock).
978  * - Update its state to STATE_READY. Now the receiver can continue.
979  * - Wake up the process after the lock is dropped. Should the process wake up
980  *   before this wakeup (due to a timeout or a signal) it will either see
981  *   STATE_READY and continue or acquire the lock to check the state again.
982  *
983  * The same algorithm is used for senders.
984  */
985 
986 static inline void __pipelined_op(struct wake_q_head *wake_q,
987 				  struct mqueue_inode_info *info,
988 				  struct ext_wait_queue *this)
989 {
990 	list_del(&this->list);
991 	get_task_struct(this->task);
992 
993 	/* see MQ_BARRIER for purpose/pairing */
994 	smp_store_release(&this->state, STATE_READY);
995 	wake_q_add_safe(wake_q, this->task);
996 }
997 
998 /* pipelined_send() - send a message directly to the task waiting in
999  * sys_mq_timedreceive() (without inserting message into a queue).
1000  */
1001 static inline void pipelined_send(struct wake_q_head *wake_q,
1002 				  struct mqueue_inode_info *info,
1003 				  struct msg_msg *message,
1004 				  struct ext_wait_queue *receiver)
1005 {
1006 	receiver->msg = message;
1007 	__pipelined_op(wake_q, info, receiver);
1008 }
1009 
1010 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
1011  * gets its message and put to the queue (we have one free place for sure). */
1012 static inline void pipelined_receive(struct wake_q_head *wake_q,
1013 				     struct mqueue_inode_info *info)
1014 {
1015 	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
1016 
1017 	if (!sender) {
1018 		/* for poll */
1019 		wake_up_interruptible(&info->wait_q);
1020 		return;
1021 	}
1022 	if (msg_insert(sender->msg, info))
1023 		return;
1024 
1025 	__pipelined_op(wake_q, info, sender);
1026 }
1027 
1028 static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
1029 		size_t msg_len, unsigned int msg_prio,
1030 		struct timespec64 *ts)
1031 {
1032 	struct fd f;
1033 	struct inode *inode;
1034 	struct ext_wait_queue wait;
1035 	struct ext_wait_queue *receiver;
1036 	struct msg_msg *msg_ptr;
1037 	struct mqueue_inode_info *info;
1038 	ktime_t expires, *timeout = NULL;
1039 	struct posix_msg_tree_node *new_leaf = NULL;
1040 	int ret = 0;
1041 	DEFINE_WAKE_Q(wake_q);
1042 
1043 	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
1044 		return -EINVAL;
1045 
1046 	if (ts) {
1047 		expires = timespec64_to_ktime(*ts);
1048 		timeout = &expires;
1049 	}
1050 
1051 	audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
1052 
1053 	f = fdget(mqdes);
1054 	if (unlikely(!f.file)) {
1055 		ret = -EBADF;
1056 		goto out;
1057 	}
1058 
1059 	inode = file_inode(f.file);
1060 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1061 		ret = -EBADF;
1062 		goto out_fput;
1063 	}
1064 	info = MQUEUE_I(inode);
1065 	audit_file(f.file);
1066 
1067 	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1068 		ret = -EBADF;
1069 		goto out_fput;
1070 	}
1071 
1072 	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1073 		ret = -EMSGSIZE;
1074 		goto out_fput;
1075 	}
1076 
1077 	/* First try to allocate memory, before doing anything with
1078 	 * existing queues. */
1079 	msg_ptr = load_msg(u_msg_ptr, msg_len);
1080 	if (IS_ERR(msg_ptr)) {
1081 		ret = PTR_ERR(msg_ptr);
1082 		goto out_fput;
1083 	}
1084 	msg_ptr->m_ts = msg_len;
1085 	msg_ptr->m_type = msg_prio;
1086 
1087 	/*
1088 	 * msg_insert really wants us to have a valid, spare node struct so
1089 	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1090 	 * fall back to that if necessary.
1091 	 */
1092 	if (!info->node_cache)
1093 		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1094 
1095 	spin_lock(&info->lock);
1096 
1097 	if (!info->node_cache && new_leaf) {
1098 		/* Save our speculative allocation into the cache */
1099 		INIT_LIST_HEAD(&new_leaf->msg_list);
1100 		info->node_cache = new_leaf;
1101 		new_leaf = NULL;
1102 	} else {
1103 		kfree(new_leaf);
1104 	}
1105 
1106 	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1107 		if (f.file->f_flags & O_NONBLOCK) {
1108 			ret = -EAGAIN;
1109 		} else {
1110 			wait.task = current;
1111 			wait.msg = (void *) msg_ptr;
1112 
1113 			/* memory barrier not required, we hold info->lock */
1114 			WRITE_ONCE(wait.state, STATE_NONE);
1115 			ret = wq_sleep(info, SEND, timeout, &wait);
1116 			/*
1117 			 * wq_sleep must be called with info->lock held, and
1118 			 * returns with the lock released
1119 			 */
1120 			goto out_free;
1121 		}
1122 	} else {
1123 		receiver = wq_get_first_waiter(info, RECV);
1124 		if (receiver) {
1125 			pipelined_send(&wake_q, info, msg_ptr, receiver);
1126 		} else {
1127 			/* adds message to the queue */
1128 			ret = msg_insert(msg_ptr, info);
1129 			if (ret)
1130 				goto out_unlock;
1131 			__do_notify(info);
1132 		}
1133 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1134 				current_time(inode);
1135 	}
1136 out_unlock:
1137 	spin_unlock(&info->lock);
1138 	wake_up_q(&wake_q);
1139 out_free:
1140 	if (ret)
1141 		free_msg(msg_ptr);
1142 out_fput:
1143 	fdput(f);
1144 out:
1145 	return ret;
1146 }
1147 
1148 static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1149 		size_t msg_len, unsigned int __user *u_msg_prio,
1150 		struct timespec64 *ts)
1151 {
1152 	ssize_t ret;
1153 	struct msg_msg *msg_ptr;
1154 	struct fd f;
1155 	struct inode *inode;
1156 	struct mqueue_inode_info *info;
1157 	struct ext_wait_queue wait;
1158 	ktime_t expires, *timeout = NULL;
1159 	struct posix_msg_tree_node *new_leaf = NULL;
1160 
1161 	if (ts) {
1162 		expires = timespec64_to_ktime(*ts);
1163 		timeout = &expires;
1164 	}
1165 
1166 	audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1167 
1168 	f = fdget(mqdes);
1169 	if (unlikely(!f.file)) {
1170 		ret = -EBADF;
1171 		goto out;
1172 	}
1173 
1174 	inode = file_inode(f.file);
1175 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1176 		ret = -EBADF;
1177 		goto out_fput;
1178 	}
1179 	info = MQUEUE_I(inode);
1180 	audit_file(f.file);
1181 
1182 	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1183 		ret = -EBADF;
1184 		goto out_fput;
1185 	}
1186 
1187 	/* checks if buffer is big enough */
1188 	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1189 		ret = -EMSGSIZE;
1190 		goto out_fput;
1191 	}
1192 
1193 	/*
1194 	 * msg_insert really wants us to have a valid, spare node struct so
1195 	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1196 	 * fall back to that if necessary.
1197 	 */
1198 	if (!info->node_cache)
1199 		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1200 
1201 	spin_lock(&info->lock);
1202 
1203 	if (!info->node_cache && new_leaf) {
1204 		/* Save our speculative allocation into the cache */
1205 		INIT_LIST_HEAD(&new_leaf->msg_list);
1206 		info->node_cache = new_leaf;
1207 	} else {
1208 		kfree(new_leaf);
1209 	}
1210 
1211 	if (info->attr.mq_curmsgs == 0) {
1212 		if (f.file->f_flags & O_NONBLOCK) {
1213 			spin_unlock(&info->lock);
1214 			ret = -EAGAIN;
1215 		} else {
1216 			wait.task = current;
1217 
1218 			/* memory barrier not required, we hold info->lock */
1219 			WRITE_ONCE(wait.state, STATE_NONE);
1220 			ret = wq_sleep(info, RECV, timeout, &wait);
1221 			msg_ptr = wait.msg;
1222 		}
1223 	} else {
1224 		DEFINE_WAKE_Q(wake_q);
1225 
1226 		msg_ptr = msg_get(info);
1227 
1228 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1229 				current_time(inode);
1230 
1231 		/* There is now free space in queue. */
1232 		pipelined_receive(&wake_q, info);
1233 		spin_unlock(&info->lock);
1234 		wake_up_q(&wake_q);
1235 		ret = 0;
1236 	}
1237 	if (ret == 0) {
1238 		ret = msg_ptr->m_ts;
1239 
1240 		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1241 			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1242 			ret = -EFAULT;
1243 		}
1244 		free_msg(msg_ptr);
1245 	}
1246 out_fput:
1247 	fdput(f);
1248 out:
1249 	return ret;
1250 }
1251 
1252 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1253 		size_t, msg_len, unsigned int, msg_prio,
1254 		const struct __kernel_timespec __user *, u_abs_timeout)
1255 {
1256 	struct timespec64 ts, *p = NULL;
1257 	if (u_abs_timeout) {
1258 		int res = prepare_timeout(u_abs_timeout, &ts);
1259 		if (res)
1260 			return res;
1261 		p = &ts;
1262 	}
1263 	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1264 }
1265 
1266 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1267 		size_t, msg_len, unsigned int __user *, u_msg_prio,
1268 		const struct __kernel_timespec __user *, u_abs_timeout)
1269 {
1270 	struct timespec64 ts, *p = NULL;
1271 	if (u_abs_timeout) {
1272 		int res = prepare_timeout(u_abs_timeout, &ts);
1273 		if (res)
1274 			return res;
1275 		p = &ts;
1276 	}
1277 	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1278 }
1279 
1280 /*
1281  * Notes: the case when user wants us to deregister (with NULL as pointer)
1282  * and he isn't currently owner of notification, will be silently discarded.
1283  * It isn't explicitly defined in the POSIX.
1284  */
1285 static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1286 {
1287 	int ret;
1288 	struct fd f;
1289 	struct sock *sock;
1290 	struct inode *inode;
1291 	struct mqueue_inode_info *info;
1292 	struct sk_buff *nc;
1293 
1294 	audit_mq_notify(mqdes, notification);
1295 
1296 	nc = NULL;
1297 	sock = NULL;
1298 	if (notification != NULL) {
1299 		if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1300 			     notification->sigev_notify != SIGEV_SIGNAL &&
1301 			     notification->sigev_notify != SIGEV_THREAD))
1302 			return -EINVAL;
1303 		if (notification->sigev_notify == SIGEV_SIGNAL &&
1304 			!valid_signal(notification->sigev_signo)) {
1305 			return -EINVAL;
1306 		}
1307 		if (notification->sigev_notify == SIGEV_THREAD) {
1308 			long timeo;
1309 
1310 			/* create the notify skb */
1311 			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1312 			if (!nc)
1313 				return -ENOMEM;
1314 
1315 			if (copy_from_user(nc->data,
1316 					notification->sigev_value.sival_ptr,
1317 					NOTIFY_COOKIE_LEN)) {
1318 				ret = -EFAULT;
1319 				goto free_skb;
1320 			}
1321 
1322 			/* TODO: add a header? */
1323 			skb_put(nc, NOTIFY_COOKIE_LEN);
1324 			/* and attach it to the socket */
1325 retry:
1326 			f = fdget(notification->sigev_signo);
1327 			if (!f.file) {
1328 				ret = -EBADF;
1329 				goto out;
1330 			}
1331 			sock = netlink_getsockbyfilp(f.file);
1332 			fdput(f);
1333 			if (IS_ERR(sock)) {
1334 				ret = PTR_ERR(sock);
1335 				goto free_skb;
1336 			}
1337 
1338 			timeo = MAX_SCHEDULE_TIMEOUT;
1339 			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1340 			if (ret == 1) {
1341 				sock = NULL;
1342 				goto retry;
1343 			}
1344 			if (ret)
1345 				return ret;
1346 		}
1347 	}
1348 
1349 	f = fdget(mqdes);
1350 	if (!f.file) {
1351 		ret = -EBADF;
1352 		goto out;
1353 	}
1354 
1355 	inode = file_inode(f.file);
1356 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1357 		ret = -EBADF;
1358 		goto out_fput;
1359 	}
1360 	info = MQUEUE_I(inode);
1361 
1362 	ret = 0;
1363 	spin_lock(&info->lock);
1364 	if (notification == NULL) {
1365 		if (info->notify_owner == task_tgid(current)) {
1366 			remove_notification(info);
1367 			inode->i_atime = inode->i_ctime = current_time(inode);
1368 		}
1369 	} else if (info->notify_owner != NULL) {
1370 		ret = -EBUSY;
1371 	} else {
1372 		switch (notification->sigev_notify) {
1373 		case SIGEV_NONE:
1374 			info->notify.sigev_notify = SIGEV_NONE;
1375 			break;
1376 		case SIGEV_THREAD:
1377 			info->notify_sock = sock;
1378 			info->notify_cookie = nc;
1379 			sock = NULL;
1380 			nc = NULL;
1381 			info->notify.sigev_notify = SIGEV_THREAD;
1382 			break;
1383 		case SIGEV_SIGNAL:
1384 			info->notify.sigev_signo = notification->sigev_signo;
1385 			info->notify.sigev_value = notification->sigev_value;
1386 			info->notify.sigev_notify = SIGEV_SIGNAL;
1387 			break;
1388 		}
1389 
1390 		info->notify_owner = get_pid(task_tgid(current));
1391 		info->notify_user_ns = get_user_ns(current_user_ns());
1392 		inode->i_atime = inode->i_ctime = current_time(inode);
1393 	}
1394 	spin_unlock(&info->lock);
1395 out_fput:
1396 	fdput(f);
1397 out:
1398 	if (sock)
1399 		netlink_detachskb(sock, nc);
1400 	else
1401 free_skb:
1402 		dev_kfree_skb(nc);
1403 
1404 	return ret;
1405 }
1406 
1407 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1408 		const struct sigevent __user *, u_notification)
1409 {
1410 	struct sigevent n, *p = NULL;
1411 	if (u_notification) {
1412 		if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1413 			return -EFAULT;
1414 		p = &n;
1415 	}
1416 	return do_mq_notify(mqdes, p);
1417 }
1418 
1419 static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1420 {
1421 	struct fd f;
1422 	struct inode *inode;
1423 	struct mqueue_inode_info *info;
1424 
1425 	if (new && (new->mq_flags & (~O_NONBLOCK)))
1426 		return -EINVAL;
1427 
1428 	f = fdget(mqdes);
1429 	if (!f.file)
1430 		return -EBADF;
1431 
1432 	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1433 		fdput(f);
1434 		return -EBADF;
1435 	}
1436 
1437 	inode = file_inode(f.file);
1438 	info = MQUEUE_I(inode);
1439 
1440 	spin_lock(&info->lock);
1441 
1442 	if (old) {
1443 		*old = info->attr;
1444 		old->mq_flags = f.file->f_flags & O_NONBLOCK;
1445 	}
1446 	if (new) {
1447 		audit_mq_getsetattr(mqdes, new);
1448 		spin_lock(&f.file->f_lock);
1449 		if (new->mq_flags & O_NONBLOCK)
1450 			f.file->f_flags |= O_NONBLOCK;
1451 		else
1452 			f.file->f_flags &= ~O_NONBLOCK;
1453 		spin_unlock(&f.file->f_lock);
1454 
1455 		inode->i_atime = inode->i_ctime = current_time(inode);
1456 	}
1457 
1458 	spin_unlock(&info->lock);
1459 	fdput(f);
1460 	return 0;
1461 }
1462 
1463 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1464 		const struct mq_attr __user *, u_mqstat,
1465 		struct mq_attr __user *, u_omqstat)
1466 {
1467 	int ret;
1468 	struct mq_attr mqstat, omqstat;
1469 	struct mq_attr *new = NULL, *old = NULL;
1470 
1471 	if (u_mqstat) {
1472 		new = &mqstat;
1473 		if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1474 			return -EFAULT;
1475 	}
1476 	if (u_omqstat)
1477 		old = &omqstat;
1478 
1479 	ret = do_mq_getsetattr(mqdes, new, old);
1480 	if (ret || !old)
1481 		return ret;
1482 
1483 	if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1484 		return -EFAULT;
1485 	return 0;
1486 }
1487 
1488 #ifdef CONFIG_COMPAT
1489 
1490 struct compat_mq_attr {
1491 	compat_long_t mq_flags;      /* message queue flags		     */
1492 	compat_long_t mq_maxmsg;     /* maximum number of messages	     */
1493 	compat_long_t mq_msgsize;    /* maximum message size		     */
1494 	compat_long_t mq_curmsgs;    /* number of messages currently queued  */
1495 	compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1496 };
1497 
1498 static inline int get_compat_mq_attr(struct mq_attr *attr,
1499 			const struct compat_mq_attr __user *uattr)
1500 {
1501 	struct compat_mq_attr v;
1502 
1503 	if (copy_from_user(&v, uattr, sizeof(*uattr)))
1504 		return -EFAULT;
1505 
1506 	memset(attr, 0, sizeof(*attr));
1507 	attr->mq_flags = v.mq_flags;
1508 	attr->mq_maxmsg = v.mq_maxmsg;
1509 	attr->mq_msgsize = v.mq_msgsize;
1510 	attr->mq_curmsgs = v.mq_curmsgs;
1511 	return 0;
1512 }
1513 
1514 static inline int put_compat_mq_attr(const struct mq_attr *attr,
1515 			struct compat_mq_attr __user *uattr)
1516 {
1517 	struct compat_mq_attr v;
1518 
1519 	memset(&v, 0, sizeof(v));
1520 	v.mq_flags = attr->mq_flags;
1521 	v.mq_maxmsg = attr->mq_maxmsg;
1522 	v.mq_msgsize = attr->mq_msgsize;
1523 	v.mq_curmsgs = attr->mq_curmsgs;
1524 	if (copy_to_user(uattr, &v, sizeof(*uattr)))
1525 		return -EFAULT;
1526 	return 0;
1527 }
1528 
1529 COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1530 		       int, oflag, compat_mode_t, mode,
1531 		       struct compat_mq_attr __user *, u_attr)
1532 {
1533 	struct mq_attr attr, *p = NULL;
1534 	if (u_attr && oflag & O_CREAT) {
1535 		p = &attr;
1536 		if (get_compat_mq_attr(&attr, u_attr))
1537 			return -EFAULT;
1538 	}
1539 	return do_mq_open(u_name, oflag, mode, p);
1540 }
1541 
1542 COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1543 		       const struct compat_sigevent __user *, u_notification)
1544 {
1545 	struct sigevent n, *p = NULL;
1546 	if (u_notification) {
1547 		if (get_compat_sigevent(&n, u_notification))
1548 			return -EFAULT;
1549 		if (n.sigev_notify == SIGEV_THREAD)
1550 			n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1551 		p = &n;
1552 	}
1553 	return do_mq_notify(mqdes, p);
1554 }
1555 
1556 COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1557 		       const struct compat_mq_attr __user *, u_mqstat,
1558 		       struct compat_mq_attr __user *, u_omqstat)
1559 {
1560 	int ret;
1561 	struct mq_attr mqstat, omqstat;
1562 	struct mq_attr *new = NULL, *old = NULL;
1563 
1564 	if (u_mqstat) {
1565 		new = &mqstat;
1566 		if (get_compat_mq_attr(new, u_mqstat))
1567 			return -EFAULT;
1568 	}
1569 	if (u_omqstat)
1570 		old = &omqstat;
1571 
1572 	ret = do_mq_getsetattr(mqdes, new, old);
1573 	if (ret || !old)
1574 		return ret;
1575 
1576 	if (put_compat_mq_attr(old, u_omqstat))
1577 		return -EFAULT;
1578 	return 0;
1579 }
1580 #endif
1581 
1582 #ifdef CONFIG_COMPAT_32BIT_TIME
1583 static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1584 				   struct timespec64 *ts)
1585 {
1586 	if (get_old_timespec32(ts, p))
1587 		return -EFAULT;
1588 	if (!timespec64_valid(ts))
1589 		return -EINVAL;
1590 	return 0;
1591 }
1592 
1593 SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1594 		const char __user *, u_msg_ptr,
1595 		unsigned int, msg_len, unsigned int, msg_prio,
1596 		const struct old_timespec32 __user *, u_abs_timeout)
1597 {
1598 	struct timespec64 ts, *p = NULL;
1599 	if (u_abs_timeout) {
1600 		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1601 		if (res)
1602 			return res;
1603 		p = &ts;
1604 	}
1605 	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1606 }
1607 
1608 SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1609 		char __user *, u_msg_ptr,
1610 		unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1611 		const struct old_timespec32 __user *, u_abs_timeout)
1612 {
1613 	struct timespec64 ts, *p = NULL;
1614 	if (u_abs_timeout) {
1615 		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1616 		if (res)
1617 			return res;
1618 		p = &ts;
1619 	}
1620 	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1621 }
1622 #endif
1623 
1624 static const struct inode_operations mqueue_dir_inode_operations = {
1625 	.lookup = simple_lookup,
1626 	.create = mqueue_create,
1627 	.unlink = mqueue_unlink,
1628 };
1629 
1630 static const struct file_operations mqueue_file_operations = {
1631 	.flush = mqueue_flush_file,
1632 	.poll = mqueue_poll_file,
1633 	.read = mqueue_read_file,
1634 	.llseek = default_llseek,
1635 };
1636 
1637 static const struct super_operations mqueue_super_ops = {
1638 	.alloc_inode = mqueue_alloc_inode,
1639 	.free_inode = mqueue_free_inode,
1640 	.evict_inode = mqueue_evict_inode,
1641 	.statfs = simple_statfs,
1642 };
1643 
1644 static const struct fs_context_operations mqueue_fs_context_ops = {
1645 	.free		= mqueue_fs_context_free,
1646 	.get_tree	= mqueue_get_tree,
1647 };
1648 
1649 static struct file_system_type mqueue_fs_type = {
1650 	.name			= "mqueue",
1651 	.init_fs_context	= mqueue_init_fs_context,
1652 	.kill_sb		= kill_litter_super,
1653 	.fs_flags		= FS_USERNS_MOUNT,
1654 };
1655 
1656 int mq_init_ns(struct ipc_namespace *ns)
1657 {
1658 	struct vfsmount *m;
1659 
1660 	ns->mq_queues_count  = 0;
1661 	ns->mq_queues_max    = DFLT_QUEUESMAX;
1662 	ns->mq_msg_max       = DFLT_MSGMAX;
1663 	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1664 	ns->mq_msg_default   = DFLT_MSG;
1665 	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1666 
1667 	m = mq_create_mount(ns);
1668 	if (IS_ERR(m))
1669 		return PTR_ERR(m);
1670 	ns->mq_mnt = m;
1671 	return 0;
1672 }
1673 
1674 void mq_clear_sbinfo(struct ipc_namespace *ns)
1675 {
1676 	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1677 }
1678 
1679 void mq_put_mnt(struct ipc_namespace *ns)
1680 {
1681 	kern_unmount(ns->mq_mnt);
1682 }
1683 
1684 static int __init init_mqueue_fs(void)
1685 {
1686 	int error;
1687 
1688 	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1689 				sizeof(struct mqueue_inode_info), 0,
1690 				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1691 	if (mqueue_inode_cachep == NULL)
1692 		return -ENOMEM;
1693 
1694 	/* ignore failures - they are not fatal */
1695 	mq_sysctl_table = mq_register_sysctl_table();
1696 
1697 	error = register_filesystem(&mqueue_fs_type);
1698 	if (error)
1699 		goto out_sysctl;
1700 
1701 	spin_lock_init(&mq_lock);
1702 
1703 	error = mq_init_ns(&init_ipc_ns);
1704 	if (error)
1705 		goto out_filesystem;
1706 
1707 	return 0;
1708 
1709 out_filesystem:
1710 	unregister_filesystem(&mqueue_fs_type);
1711 out_sysctl:
1712 	if (mq_sysctl_table)
1713 		unregister_sysctl_table(mq_sysctl_table);
1714 	kmem_cache_destroy(mqueue_inode_cachep);
1715 	return error;
1716 }
1717 
1718 device_initcall(init_mqueue_fs);
1719