mqueue.c revision 066cc813
1/*
2 * POSIX message queues filesystem for Linux.
3 *
4 * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5 *                          Michal Wronski          (michal.wronski@gmail.com)
6 *
7 * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
9 *			    Manfred Spraul	    (manfred@colorfullife.com)
10 *
11 * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
12 *
13 * This file is released under the GPL.
14 */
15
16#include <linux/capability.h>
17#include <linux/init.h>
18#include <linux/pagemap.h>
19#include <linux/file.h>
20#include <linux/mount.h>
21#include <linux/namei.h>
22#include <linux/sysctl.h>
23#include <linux/poll.h>
24#include <linux/mqueue.h>
25#include <linux/msg.h>
26#include <linux/skbuff.h>
27#include <linux/vmalloc.h>
28#include <linux/netlink.h>
29#include <linux/syscalls.h>
30#include <linux/audit.h>
31#include <linux/signal.h>
32#include <linux/mutex.h>
33#include <linux/nsproxy.h>
34#include <linux/pid.h>
35#include <linux/ipc_namespace.h>
36#include <linux/user_namespace.h>
37#include <linux/slab.h>
38#include <linux/sched/wake_q.h>
39#include <linux/sched/signal.h>
40#include <linux/sched/user.h>
41
42#include <net/sock.h>
43#include "util.h"
44
45#define MQUEUE_MAGIC	0x19800202
46#define DIRENT_SIZE	20
47#define FILENT_SIZE	80
48
49#define SEND		0
50#define RECV		1
51
52#define STATE_NONE	0
53#define STATE_READY	1
54
55struct posix_msg_tree_node {
56	struct rb_node		rb_node;
57	struct list_head	msg_list;
58	int			priority;
59};
60
61struct ext_wait_queue {		/* queue of sleeping tasks */
62	struct task_struct *task;
63	struct list_head list;
64	struct msg_msg *msg;	/* ptr of loaded message */
65	int state;		/* one of STATE_* values */
66};
67
68struct mqueue_inode_info {
69	spinlock_t lock;
70	struct inode vfs_inode;
71	wait_queue_head_t wait_q;
72
73	struct rb_root msg_tree;
74	struct posix_msg_tree_node *node_cache;
75	struct mq_attr attr;
76
77	struct sigevent notify;
78	struct pid *notify_owner;
79	struct user_namespace *notify_user_ns;
80	struct user_struct *user;	/* user who created, for accounting */
81	struct sock *notify_sock;
82	struct sk_buff *notify_cookie;
83
84	/* for tasks waiting for free space and messages, respectively */
85	struct ext_wait_queue e_wait_q[2];
86
87	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
88};
89
90static const struct inode_operations mqueue_dir_inode_operations;
91static const struct file_operations mqueue_file_operations;
92static const struct super_operations mqueue_super_ops;
93static void remove_notification(struct mqueue_inode_info *info);
94
95static struct kmem_cache *mqueue_inode_cachep;
96
97static struct ctl_table_header *mq_sysctl_table;
98
99static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
100{
101	return container_of(inode, struct mqueue_inode_info, vfs_inode);
102}
103
104/*
105 * This routine should be called with the mq_lock held.
106 */
107static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
108{
109	return get_ipc_ns(inode->i_sb->s_fs_info);
110}
111
112static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
113{
114	struct ipc_namespace *ns;
115
116	spin_lock(&mq_lock);
117	ns = __get_ns_from_inode(inode);
118	spin_unlock(&mq_lock);
119	return ns;
120}
121
122/* Auxiliary functions to manipulate messages' list */
123static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
124{
125	struct rb_node **p, *parent = NULL;
126	struct posix_msg_tree_node *leaf;
127
128	p = &info->msg_tree.rb_node;
129	while (*p) {
130		parent = *p;
131		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
132
133		if (likely(leaf->priority == msg->m_type))
134			goto insert_msg;
135		else if (msg->m_type < leaf->priority)
136			p = &(*p)->rb_left;
137		else
138			p = &(*p)->rb_right;
139	}
140	if (info->node_cache) {
141		leaf = info->node_cache;
142		info->node_cache = NULL;
143	} else {
144		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
145		if (!leaf)
146			return -ENOMEM;
147		INIT_LIST_HEAD(&leaf->msg_list);
148	}
149	leaf->priority = msg->m_type;
150	rb_link_node(&leaf->rb_node, parent, p);
151	rb_insert_color(&leaf->rb_node, &info->msg_tree);
152insert_msg:
153	info->attr.mq_curmsgs++;
154	info->qsize += msg->m_ts;
155	list_add_tail(&msg->m_list, &leaf->msg_list);
156	return 0;
157}
158
159static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
160{
161	struct rb_node **p, *parent = NULL;
162	struct posix_msg_tree_node *leaf;
163	struct msg_msg *msg;
164
165try_again:
166	p = &info->msg_tree.rb_node;
167	while (*p) {
168		parent = *p;
169		/*
170		 * During insert, low priorities go to the left and high to the
171		 * right.  On receive, we want the highest priorities first, so
172		 * walk all the way to the right.
173		 */
174		p = &(*p)->rb_right;
175	}
176	if (!parent) {
177		if (info->attr.mq_curmsgs) {
178			pr_warn_once("Inconsistency in POSIX message queue, "
179				     "no tree element, but supposedly messages "
180				     "should exist!\n");
181			info->attr.mq_curmsgs = 0;
182		}
183		return NULL;
184	}
185	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
186	if (unlikely(list_empty(&leaf->msg_list))) {
187		pr_warn_once("Inconsistency in POSIX message queue, "
188			     "empty leaf node but we haven't implemented "
189			     "lazy leaf delete!\n");
190		rb_erase(&leaf->rb_node, &info->msg_tree);
191		if (info->node_cache) {
192			kfree(leaf);
193		} else {
194			info->node_cache = leaf;
195		}
196		goto try_again;
197	} else {
198		msg = list_first_entry(&leaf->msg_list,
199				       struct msg_msg, m_list);
200		list_del(&msg->m_list);
201		if (list_empty(&leaf->msg_list)) {
202			rb_erase(&leaf->rb_node, &info->msg_tree);
203			if (info->node_cache) {
204				kfree(leaf);
205			} else {
206				info->node_cache = leaf;
207			}
208		}
209	}
210	info->attr.mq_curmsgs--;
211	info->qsize -= msg->m_ts;
212	return msg;
213}
214
215static struct inode *mqueue_get_inode(struct super_block *sb,
216		struct ipc_namespace *ipc_ns, umode_t mode,
217		struct mq_attr *attr)
218{
219	struct user_struct *u = current_user();
220	struct inode *inode;
221	int ret = -ENOMEM;
222
223	inode = new_inode(sb);
224	if (!inode)
225		goto err;
226
227	inode->i_ino = get_next_ino();
228	inode->i_mode = mode;
229	inode->i_uid = current_fsuid();
230	inode->i_gid = current_fsgid();
231	inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
232
233	if (S_ISREG(mode)) {
234		struct mqueue_inode_info *info;
235		unsigned long mq_bytes, mq_treesize;
236
237		inode->i_fop = &mqueue_file_operations;
238		inode->i_size = FILENT_SIZE;
239		/* mqueue specific info */
240		info = MQUEUE_I(inode);
241		spin_lock_init(&info->lock);
242		init_waitqueue_head(&info->wait_q);
243		INIT_LIST_HEAD(&info->e_wait_q[0].list);
244		INIT_LIST_HEAD(&info->e_wait_q[1].list);
245		info->notify_owner = NULL;
246		info->notify_user_ns = NULL;
247		info->qsize = 0;
248		info->user = NULL;	/* set when all is ok */
249		info->msg_tree = RB_ROOT;
250		info->node_cache = NULL;
251		memset(&info->attr, 0, sizeof(info->attr));
252		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
253					   ipc_ns->mq_msg_default);
254		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
255					    ipc_ns->mq_msgsize_default);
256		if (attr) {
257			info->attr.mq_maxmsg = attr->mq_maxmsg;
258			info->attr.mq_msgsize = attr->mq_msgsize;
259		}
260		/*
261		 * We used to allocate a static array of pointers and account
262		 * the size of that array as well as one msg_msg struct per
263		 * possible message into the queue size. That's no longer
264		 * accurate as the queue is now an rbtree and will grow and
265		 * shrink depending on usage patterns.  We can, however, still
266		 * account one msg_msg struct per message, but the nodes are
267		 * allocated depending on priority usage, and most programs
268		 * only use one, or a handful, of priorities.  However, since
269		 * this is pinned memory, we need to assume worst case, so
270		 * that means the min(mq_maxmsg, max_priorities) * struct
271		 * posix_msg_tree_node.
272		 */
273
274		ret = -EINVAL;
275		if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
276			goto out_inode;
277		if (capable(CAP_SYS_RESOURCE)) {
278			if (info->attr.mq_maxmsg > HARD_MSGMAX ||
279			    info->attr.mq_msgsize > HARD_MSGSIZEMAX)
280				goto out_inode;
281		} else {
282			if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
283					info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
284				goto out_inode;
285		}
286		ret = -EOVERFLOW;
287		/* check for overflow */
288		if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
289			goto out_inode;
290		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
291			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
292			sizeof(struct posix_msg_tree_node);
293		mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
294		if (mq_bytes + mq_treesize < mq_bytes)
295			goto out_inode;
296		mq_bytes += mq_treesize;
297		spin_lock(&mq_lock);
298		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
299		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
300			spin_unlock(&mq_lock);
301			/* mqueue_evict_inode() releases info->messages */
302			ret = -EMFILE;
303			goto out_inode;
304		}
305		u->mq_bytes += mq_bytes;
306		spin_unlock(&mq_lock);
307
308		/* all is ok */
309		info->user = get_uid(u);
310	} else if (S_ISDIR(mode)) {
311		inc_nlink(inode);
312		/* Some things misbehave if size == 0 on a directory */
313		inode->i_size = 2 * DIRENT_SIZE;
314		inode->i_op = &mqueue_dir_inode_operations;
315		inode->i_fop = &simple_dir_operations;
316	}
317
318	return inode;
319out_inode:
320	iput(inode);
321err:
322	return ERR_PTR(ret);
323}
324
325static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
326{
327	struct inode *inode;
328	struct ipc_namespace *ns = sb->s_fs_info;
329
330	sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
331	sb->s_blocksize = PAGE_SIZE;
332	sb->s_blocksize_bits = PAGE_SHIFT;
333	sb->s_magic = MQUEUE_MAGIC;
334	sb->s_op = &mqueue_super_ops;
335
336	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
337	if (IS_ERR(inode))
338		return PTR_ERR(inode);
339
340	sb->s_root = d_make_root(inode);
341	if (!sb->s_root)
342		return -ENOMEM;
343	return 0;
344}
345
346static struct dentry *mqueue_mount(struct file_system_type *fs_type,
347			 int flags, const char *dev_name,
348			 void *data)
349{
350	struct ipc_namespace *ns;
351	if (flags & SB_KERNMOUNT) {
352		ns = data;
353		data = NULL;
354	} else {
355		ns = current->nsproxy->ipc_ns;
356	}
357	return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super);
358}
359
360static void init_once(void *foo)
361{
362	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
363
364	inode_init_once(&p->vfs_inode);
365}
366
367static struct inode *mqueue_alloc_inode(struct super_block *sb)
368{
369	struct mqueue_inode_info *ei;
370
371	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
372	if (!ei)
373		return NULL;
374	return &ei->vfs_inode;
375}
376
377static void mqueue_i_callback(struct rcu_head *head)
378{
379	struct inode *inode = container_of(head, struct inode, i_rcu);
380	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
381}
382
383static void mqueue_destroy_inode(struct inode *inode)
384{
385	call_rcu(&inode->i_rcu, mqueue_i_callback);
386}
387
388static void mqueue_evict_inode(struct inode *inode)
389{
390	struct mqueue_inode_info *info;
391	struct user_struct *user;
392	unsigned long mq_bytes, mq_treesize;
393	struct ipc_namespace *ipc_ns;
394	struct msg_msg *msg;
395
396	clear_inode(inode);
397
398	if (S_ISDIR(inode->i_mode))
399		return;
400
401	ipc_ns = get_ns_from_inode(inode);
402	info = MQUEUE_I(inode);
403	spin_lock(&info->lock);
404	while ((msg = msg_get(info)) != NULL)
405		free_msg(msg);
406	kfree(info->node_cache);
407	spin_unlock(&info->lock);
408
409	/* Total amount of bytes accounted for the mqueue */
410	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
411		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
412		sizeof(struct posix_msg_tree_node);
413
414	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
415				  info->attr.mq_msgsize);
416
417	user = info->user;
418	if (user) {
419		spin_lock(&mq_lock);
420		user->mq_bytes -= mq_bytes;
421		/*
422		 * get_ns_from_inode() ensures that the
423		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
424		 * to which we now hold a reference, or it is NULL.
425		 * We can't put it here under mq_lock, though.
426		 */
427		if (ipc_ns)
428			ipc_ns->mq_queues_count--;
429		spin_unlock(&mq_lock);
430		free_uid(user);
431	}
432	if (ipc_ns)
433		put_ipc_ns(ipc_ns);
434}
435
436static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
437{
438	struct inode *dir = dentry->d_parent->d_inode;
439	struct inode *inode;
440	struct mq_attr *attr = arg;
441	int error;
442	struct ipc_namespace *ipc_ns;
443
444	spin_lock(&mq_lock);
445	ipc_ns = __get_ns_from_inode(dir);
446	if (!ipc_ns) {
447		error = -EACCES;
448		goto out_unlock;
449	}
450
451	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
452	    !capable(CAP_SYS_RESOURCE)) {
453		error = -ENOSPC;
454		goto out_unlock;
455	}
456	ipc_ns->mq_queues_count++;
457	spin_unlock(&mq_lock);
458
459	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
460	if (IS_ERR(inode)) {
461		error = PTR_ERR(inode);
462		spin_lock(&mq_lock);
463		ipc_ns->mq_queues_count--;
464		goto out_unlock;
465	}
466
467	put_ipc_ns(ipc_ns);
468	dir->i_size += DIRENT_SIZE;
469	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
470
471	d_instantiate(dentry, inode);
472	dget(dentry);
473	return 0;
474out_unlock:
475	spin_unlock(&mq_lock);
476	if (ipc_ns)
477		put_ipc_ns(ipc_ns);
478	return error;
479}
480
481static int mqueue_create(struct inode *dir, struct dentry *dentry,
482				umode_t mode, bool excl)
483{
484	return mqueue_create_attr(dentry, mode, NULL);
485}
486
487static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
488{
489	struct inode *inode = d_inode(dentry);
490
491	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
492	dir->i_size -= DIRENT_SIZE;
493	drop_nlink(inode);
494	dput(dentry);
495	return 0;
496}
497
498/*
499*	This is routine for system read from queue file.
500*	To avoid mess with doing here some sort of mq_receive we allow
501*	to read only queue size & notification info (the only values
502*	that are interesting from user point of view and aren't accessible
503*	through std routines)
504*/
505static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
506				size_t count, loff_t *off)
507{
508	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
509	char buffer[FILENT_SIZE];
510	ssize_t ret;
511
512	spin_lock(&info->lock);
513	snprintf(buffer, sizeof(buffer),
514			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
515			info->qsize,
516			info->notify_owner ? info->notify.sigev_notify : 0,
517			(info->notify_owner &&
518			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
519				info->notify.sigev_signo : 0,
520			pid_vnr(info->notify_owner));
521	spin_unlock(&info->lock);
522	buffer[sizeof(buffer)-1] = '\0';
523
524	ret = simple_read_from_buffer(u_data, count, off, buffer,
525				strlen(buffer));
526	if (ret <= 0)
527		return ret;
528
529	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
530	return ret;
531}
532
533static int mqueue_flush_file(struct file *filp, fl_owner_t id)
534{
535	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
536
537	spin_lock(&info->lock);
538	if (task_tgid(current) == info->notify_owner)
539		remove_notification(info);
540
541	spin_unlock(&info->lock);
542	return 0;
543}
544
545static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
546{
547	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
548	int retval = 0;
549
550	poll_wait(filp, &info->wait_q, poll_tab);
551
552	spin_lock(&info->lock);
553	if (info->attr.mq_curmsgs)
554		retval = POLLIN | POLLRDNORM;
555
556	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
557		retval |= POLLOUT | POLLWRNORM;
558	spin_unlock(&info->lock);
559
560	return retval;
561}
562
563/* Adds current to info->e_wait_q[sr] before element with smaller prio */
564static void wq_add(struct mqueue_inode_info *info, int sr,
565			struct ext_wait_queue *ewp)
566{
567	struct ext_wait_queue *walk;
568
569	ewp->task = current;
570
571	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
572		if (walk->task->static_prio <= current->static_prio) {
573			list_add_tail(&ewp->list, &walk->list);
574			return;
575		}
576	}
577	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
578}
579
580/*
581 * Puts current task to sleep. Caller must hold queue lock. After return
582 * lock isn't held.
583 * sr: SEND or RECV
584 */
585static int wq_sleep(struct mqueue_inode_info *info, int sr,
586		    ktime_t *timeout, struct ext_wait_queue *ewp)
587	__releases(&info->lock)
588{
589	int retval;
590	signed long time;
591
592	wq_add(info, sr, ewp);
593
594	for (;;) {
595		__set_current_state(TASK_INTERRUPTIBLE);
596
597		spin_unlock(&info->lock);
598		time = schedule_hrtimeout_range_clock(timeout, 0,
599			HRTIMER_MODE_ABS, CLOCK_REALTIME);
600
601		if (ewp->state == STATE_READY) {
602			retval = 0;
603			goto out;
604		}
605		spin_lock(&info->lock);
606		if (ewp->state == STATE_READY) {
607			retval = 0;
608			goto out_unlock;
609		}
610		if (signal_pending(current)) {
611			retval = -ERESTARTSYS;
612			break;
613		}
614		if (time == 0) {
615			retval = -ETIMEDOUT;
616			break;
617		}
618	}
619	list_del(&ewp->list);
620out_unlock:
621	spin_unlock(&info->lock);
622out:
623	return retval;
624}
625
626/*
627 * Returns waiting task that should be serviced first or NULL if none exists
628 */
629static struct ext_wait_queue *wq_get_first_waiter(
630		struct mqueue_inode_info *info, int sr)
631{
632	struct list_head *ptr;
633
634	ptr = info->e_wait_q[sr].list.prev;
635	if (ptr == &info->e_wait_q[sr].list)
636		return NULL;
637	return list_entry(ptr, struct ext_wait_queue, list);
638}
639
640
641static inline void set_cookie(struct sk_buff *skb, char code)
642{
643	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
644}
645
646/*
647 * The next function is only to split too long sys_mq_timedsend
648 */
649static void __do_notify(struct mqueue_inode_info *info)
650{
651	/* notification
652	 * invoked when there is registered process and there isn't process
653	 * waiting synchronously for message AND state of queue changed from
654	 * empty to not empty. Here we are sure that no one is waiting
655	 * synchronously. */
656	if (info->notify_owner &&
657	    info->attr.mq_curmsgs == 1) {
658		struct siginfo sig_i;
659		switch (info->notify.sigev_notify) {
660		case SIGEV_NONE:
661			break;
662		case SIGEV_SIGNAL:
663			/* sends signal */
664
665			sig_i.si_signo = info->notify.sigev_signo;
666			sig_i.si_errno = 0;
667			sig_i.si_code = SI_MESGQ;
668			sig_i.si_value = info->notify.sigev_value;
669			/* map current pid/uid into info->owner's namespaces */
670			rcu_read_lock();
671			sig_i.si_pid = task_tgid_nr_ns(current,
672						ns_of_pid(info->notify_owner));
673			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
674			rcu_read_unlock();
675
676			kill_pid_info(info->notify.sigev_signo,
677				      &sig_i, info->notify_owner);
678			break;
679		case SIGEV_THREAD:
680			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
681			netlink_sendskb(info->notify_sock, info->notify_cookie);
682			break;
683		}
684		/* after notification unregisters process */
685		put_pid(info->notify_owner);
686		put_user_ns(info->notify_user_ns);
687		info->notify_owner = NULL;
688		info->notify_user_ns = NULL;
689	}
690	wake_up(&info->wait_q);
691}
692
693static int prepare_timeout(const struct timespec __user *u_abs_timeout,
694			   struct timespec64 *ts)
695{
696	if (get_timespec64(ts, u_abs_timeout))
697		return -EFAULT;
698	if (!timespec64_valid(ts))
699		return -EINVAL;
700	return 0;
701}
702
703static void remove_notification(struct mqueue_inode_info *info)
704{
705	if (info->notify_owner != NULL &&
706	    info->notify.sigev_notify == SIGEV_THREAD) {
707		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
708		netlink_sendskb(info->notify_sock, info->notify_cookie);
709	}
710	put_pid(info->notify_owner);
711	put_user_ns(info->notify_user_ns);
712	info->notify_owner = NULL;
713	info->notify_user_ns = NULL;
714}
715
716static int prepare_open(struct dentry *dentry, int oflag, int ro,
717			umode_t mode, struct filename *name,
718			struct mq_attr *attr)
719{
720	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
721						  MAY_READ | MAY_WRITE };
722	int acc;
723
724	if (oflag & O_CREAT) {
725		if (d_really_is_positive(dentry)) {	/* entry already exists */
726			audit_inode(name, dentry, 0);
727			if (oflag & O_EXCL)
728				return -EEXIST;
729		} else {
730			if (ro)
731				return ro;
732
733			audit_inode_parent_hidden(name, dentry->d_parent);
734			return vfs_mkobj(dentry, mode & ~current_umask(),
735				  mqueue_create_attr, attr);
736		}
737	} else {
738		if (d_really_is_negative(dentry)) {
739			return -ENOENT;
740		} else {
741			audit_inode(name, dentry, 0);
742		}
743	}
744	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
745		return -EINVAL;
746	acc = oflag2acc[oflag & O_ACCMODE];
747	return inode_permission(d_inode(dentry), acc);
748}
749
750static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
751		      struct mq_attr *attr)
752{
753	struct path path;
754	struct filename *name;
755	int fd, error;
756	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
757	struct vfsmount *mnt = ipc_ns->mq_mnt;
758	struct dentry *root = mnt->mnt_root;
759	int ro;
760
761	audit_mq_open(oflag, mode, attr);
762
763	if (IS_ERR(name = getname(u_name)))
764		return PTR_ERR(name);
765
766	fd = get_unused_fd_flags(O_CLOEXEC);
767	if (fd < 0)
768		goto out_putname;
769
770	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
771	error = 0;
772	inode_lock(d_inode(root));
773	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
774	if (IS_ERR(path.dentry)) {
775		error = PTR_ERR(path.dentry);
776		goto out_putfd;
777	}
778	path.mnt = mntget(mnt);
779
780	error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
781	if (!error) {
782		struct file *file = dentry_open(&path, oflag, current_cred());
783		if (!IS_ERR(file))
784			fd_install(fd, file);
785		else
786			error = PTR_ERR(file);
787	}
788	path_put(&path);
789out_putfd:
790	if (error) {
791		put_unused_fd(fd);
792		fd = error;
793	}
794	inode_unlock(d_inode(root));
795	if (!ro)
796		mnt_drop_write(mnt);
797out_putname:
798	putname(name);
799	return fd;
800}
801
802SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
803		struct mq_attr __user *, u_attr)
804{
805	struct mq_attr attr;
806	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
807		return -EFAULT;
808
809	return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
810}
811
812SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
813{
814	int err;
815	struct filename *name;
816	struct dentry *dentry;
817	struct inode *inode = NULL;
818	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
819	struct vfsmount *mnt = ipc_ns->mq_mnt;
820
821	name = getname(u_name);
822	if (IS_ERR(name))
823		return PTR_ERR(name);
824
825	audit_inode_parent_hidden(name, mnt->mnt_root);
826	err = mnt_want_write(mnt);
827	if (err)
828		goto out_name;
829	inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
830	dentry = lookup_one_len(name->name, mnt->mnt_root,
831				strlen(name->name));
832	if (IS_ERR(dentry)) {
833		err = PTR_ERR(dentry);
834		goto out_unlock;
835	}
836
837	inode = d_inode(dentry);
838	if (!inode) {
839		err = -ENOENT;
840	} else {
841		ihold(inode);
842		err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
843	}
844	dput(dentry);
845
846out_unlock:
847	inode_unlock(d_inode(mnt->mnt_root));
848	if (inode)
849		iput(inode);
850	mnt_drop_write(mnt);
851out_name:
852	putname(name);
853
854	return err;
855}
856
857/* Pipelined send and receive functions.
858 *
859 * If a receiver finds no waiting message, then it registers itself in the
860 * list of waiting receivers. A sender checks that list before adding the new
861 * message into the message array. If there is a waiting receiver, then it
862 * bypasses the message array and directly hands the message over to the
863 * receiver. The receiver accepts the message and returns without grabbing the
864 * queue spinlock:
865 *
866 * - Set pointer to message.
867 * - Queue the receiver task for later wakeup (without the info->lock).
868 * - Update its state to STATE_READY. Now the receiver can continue.
869 * - Wake up the process after the lock is dropped. Should the process wake up
870 *   before this wakeup (due to a timeout or a signal) it will either see
871 *   STATE_READY and continue or acquire the lock to check the state again.
872 *
873 * The same algorithm is used for senders.
874 */
875
876/* pipelined_send() - send a message directly to the task waiting in
877 * sys_mq_timedreceive() (without inserting message into a queue).
878 */
879static inline void pipelined_send(struct wake_q_head *wake_q,
880				  struct mqueue_inode_info *info,
881				  struct msg_msg *message,
882				  struct ext_wait_queue *receiver)
883{
884	receiver->msg = message;
885	list_del(&receiver->list);
886	wake_q_add(wake_q, receiver->task);
887	/*
888	 * Rely on the implicit cmpxchg barrier from wake_q_add such
889	 * that we can ensure that updating receiver->state is the last
890	 * write operation: As once set, the receiver can continue,
891	 * and if we don't have the reference count from the wake_q,
892	 * yet, at that point we can later have a use-after-free
893	 * condition and bogus wakeup.
894	 */
895	receiver->state = STATE_READY;
896}
897
898/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
899 * gets its message and put to the queue (we have one free place for sure). */
900static inline void pipelined_receive(struct wake_q_head *wake_q,
901				     struct mqueue_inode_info *info)
902{
903	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
904
905	if (!sender) {
906		/* for poll */
907		wake_up_interruptible(&info->wait_q);
908		return;
909	}
910	if (msg_insert(sender->msg, info))
911		return;
912
913	list_del(&sender->list);
914	wake_q_add(wake_q, sender->task);
915	sender->state = STATE_READY;
916}
917
918static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
919		size_t msg_len, unsigned int msg_prio,
920		struct timespec64 *ts)
921{
922	struct fd f;
923	struct inode *inode;
924	struct ext_wait_queue wait;
925	struct ext_wait_queue *receiver;
926	struct msg_msg *msg_ptr;
927	struct mqueue_inode_info *info;
928	ktime_t expires, *timeout = NULL;
929	struct posix_msg_tree_node *new_leaf = NULL;
930	int ret = 0;
931	DEFINE_WAKE_Q(wake_q);
932
933	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
934		return -EINVAL;
935
936	if (ts) {
937		expires = timespec64_to_ktime(*ts);
938		timeout = &expires;
939	}
940
941	audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
942
943	f = fdget(mqdes);
944	if (unlikely(!f.file)) {
945		ret = -EBADF;
946		goto out;
947	}
948
949	inode = file_inode(f.file);
950	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
951		ret = -EBADF;
952		goto out_fput;
953	}
954	info = MQUEUE_I(inode);
955	audit_file(f.file);
956
957	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
958		ret = -EBADF;
959		goto out_fput;
960	}
961
962	if (unlikely(msg_len > info->attr.mq_msgsize)) {
963		ret = -EMSGSIZE;
964		goto out_fput;
965	}
966
967	/* First try to allocate memory, before doing anything with
968	 * existing queues. */
969	msg_ptr = load_msg(u_msg_ptr, msg_len);
970	if (IS_ERR(msg_ptr)) {
971		ret = PTR_ERR(msg_ptr);
972		goto out_fput;
973	}
974	msg_ptr->m_ts = msg_len;
975	msg_ptr->m_type = msg_prio;
976
977	/*
978	 * msg_insert really wants us to have a valid, spare node struct so
979	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
980	 * fall back to that if necessary.
981	 */
982	if (!info->node_cache)
983		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
984
985	spin_lock(&info->lock);
986
987	if (!info->node_cache && new_leaf) {
988		/* Save our speculative allocation into the cache */
989		INIT_LIST_HEAD(&new_leaf->msg_list);
990		info->node_cache = new_leaf;
991		new_leaf = NULL;
992	} else {
993		kfree(new_leaf);
994	}
995
996	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
997		if (f.file->f_flags & O_NONBLOCK) {
998			ret = -EAGAIN;
999		} else {
1000			wait.task = current;
1001			wait.msg = (void *) msg_ptr;
1002			wait.state = STATE_NONE;
1003			ret = wq_sleep(info, SEND, timeout, &wait);
1004			/*
1005			 * wq_sleep must be called with info->lock held, and
1006			 * returns with the lock released
1007			 */
1008			goto out_free;
1009		}
1010	} else {
1011		receiver = wq_get_first_waiter(info, RECV);
1012		if (receiver) {
1013			pipelined_send(&wake_q, info, msg_ptr, receiver);
1014		} else {
1015			/* adds message to the queue */
1016			ret = msg_insert(msg_ptr, info);
1017			if (ret)
1018				goto out_unlock;
1019			__do_notify(info);
1020		}
1021		inode->i_atime = inode->i_mtime = inode->i_ctime =
1022				current_time(inode);
1023	}
1024out_unlock:
1025	spin_unlock(&info->lock);
1026	wake_up_q(&wake_q);
1027out_free:
1028	if (ret)
1029		free_msg(msg_ptr);
1030out_fput:
1031	fdput(f);
1032out:
1033	return ret;
1034}
1035
1036static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1037		size_t msg_len, unsigned int __user *u_msg_prio,
1038		struct timespec64 *ts)
1039{
1040	ssize_t ret;
1041	struct msg_msg *msg_ptr;
1042	struct fd f;
1043	struct inode *inode;
1044	struct mqueue_inode_info *info;
1045	struct ext_wait_queue wait;
1046	ktime_t expires, *timeout = NULL;
1047	struct posix_msg_tree_node *new_leaf = NULL;
1048
1049	if (ts) {
1050		expires = timespec64_to_ktime(*ts);
1051		timeout = &expires;
1052	}
1053
1054	audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1055
1056	f = fdget(mqdes);
1057	if (unlikely(!f.file)) {
1058		ret = -EBADF;
1059		goto out;
1060	}
1061
1062	inode = file_inode(f.file);
1063	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1064		ret = -EBADF;
1065		goto out_fput;
1066	}
1067	info = MQUEUE_I(inode);
1068	audit_file(f.file);
1069
1070	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1071		ret = -EBADF;
1072		goto out_fput;
1073	}
1074
1075	/* checks if buffer is big enough */
1076	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1077		ret = -EMSGSIZE;
1078		goto out_fput;
1079	}
1080
1081	/*
1082	 * msg_insert really wants us to have a valid, spare node struct so
1083	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1084	 * fall back to that if necessary.
1085	 */
1086	if (!info->node_cache)
1087		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1088
1089	spin_lock(&info->lock);
1090
1091	if (!info->node_cache && new_leaf) {
1092		/* Save our speculative allocation into the cache */
1093		INIT_LIST_HEAD(&new_leaf->msg_list);
1094		info->node_cache = new_leaf;
1095	} else {
1096		kfree(new_leaf);
1097	}
1098
1099	if (info->attr.mq_curmsgs == 0) {
1100		if (f.file->f_flags & O_NONBLOCK) {
1101			spin_unlock(&info->lock);
1102			ret = -EAGAIN;
1103		} else {
1104			wait.task = current;
1105			wait.state = STATE_NONE;
1106			ret = wq_sleep(info, RECV, timeout, &wait);
1107			msg_ptr = wait.msg;
1108		}
1109	} else {
1110		DEFINE_WAKE_Q(wake_q);
1111
1112		msg_ptr = msg_get(info);
1113
1114		inode->i_atime = inode->i_mtime = inode->i_ctime =
1115				current_time(inode);
1116
1117		/* There is now free space in queue. */
1118		pipelined_receive(&wake_q, info);
1119		spin_unlock(&info->lock);
1120		wake_up_q(&wake_q);
1121		ret = 0;
1122	}
1123	if (ret == 0) {
1124		ret = msg_ptr->m_ts;
1125
1126		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1127			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1128			ret = -EFAULT;
1129		}
1130		free_msg(msg_ptr);
1131	}
1132out_fput:
1133	fdput(f);
1134out:
1135	return ret;
1136}
1137
1138SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1139		size_t, msg_len, unsigned int, msg_prio,
1140		const struct timespec __user *, u_abs_timeout)
1141{
1142	struct timespec64 ts, *p = NULL;
1143	if (u_abs_timeout) {
1144		int res = prepare_timeout(u_abs_timeout, &ts);
1145		if (res)
1146			return res;
1147		p = &ts;
1148	}
1149	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1150}
1151
1152SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1153		size_t, msg_len, unsigned int __user *, u_msg_prio,
1154		const struct timespec __user *, u_abs_timeout)
1155{
1156	struct timespec64 ts, *p = NULL;
1157	if (u_abs_timeout) {
1158		int res = prepare_timeout(u_abs_timeout, &ts);
1159		if (res)
1160			return res;
1161		p = &ts;
1162	}
1163	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1164}
1165
1166/*
1167 * Notes: the case when user wants us to deregister (with NULL as pointer)
1168 * and he isn't currently owner of notification, will be silently discarded.
1169 * It isn't explicitly defined in the POSIX.
1170 */
1171static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1172{
1173	int ret;
1174	struct fd f;
1175	struct sock *sock;
1176	struct inode *inode;
1177	struct mqueue_inode_info *info;
1178	struct sk_buff *nc;
1179
1180	audit_mq_notify(mqdes, notification);
1181
1182	nc = NULL;
1183	sock = NULL;
1184	if (notification != NULL) {
1185		if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1186			     notification->sigev_notify != SIGEV_SIGNAL &&
1187			     notification->sigev_notify != SIGEV_THREAD))
1188			return -EINVAL;
1189		if (notification->sigev_notify == SIGEV_SIGNAL &&
1190			!valid_signal(notification->sigev_signo)) {
1191			return -EINVAL;
1192		}
1193		if (notification->sigev_notify == SIGEV_THREAD) {
1194			long timeo;
1195
1196			/* create the notify skb */
1197			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1198			if (!nc) {
1199				ret = -ENOMEM;
1200				goto out;
1201			}
1202			if (copy_from_user(nc->data,
1203					notification->sigev_value.sival_ptr,
1204					NOTIFY_COOKIE_LEN)) {
1205				ret = -EFAULT;
1206				goto out;
1207			}
1208
1209			/* TODO: add a header? */
1210			skb_put(nc, NOTIFY_COOKIE_LEN);
1211			/* and attach it to the socket */
1212retry:
1213			f = fdget(notification->sigev_signo);
1214			if (!f.file) {
1215				ret = -EBADF;
1216				goto out;
1217			}
1218			sock = netlink_getsockbyfilp(f.file);
1219			fdput(f);
1220			if (IS_ERR(sock)) {
1221				ret = PTR_ERR(sock);
1222				sock = NULL;
1223				goto out;
1224			}
1225
1226			timeo = MAX_SCHEDULE_TIMEOUT;
1227			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1228			if (ret == 1) {
1229				sock = NULL;
1230				goto retry;
1231			}
1232			if (ret) {
1233				sock = NULL;
1234				nc = NULL;
1235				goto out;
1236			}
1237		}
1238	}
1239
1240	f = fdget(mqdes);
1241	if (!f.file) {
1242		ret = -EBADF;
1243		goto out;
1244	}
1245
1246	inode = file_inode(f.file);
1247	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1248		ret = -EBADF;
1249		goto out_fput;
1250	}
1251	info = MQUEUE_I(inode);
1252
1253	ret = 0;
1254	spin_lock(&info->lock);
1255	if (notification == NULL) {
1256		if (info->notify_owner == task_tgid(current)) {
1257			remove_notification(info);
1258			inode->i_atime = inode->i_ctime = current_time(inode);
1259		}
1260	} else if (info->notify_owner != NULL) {
1261		ret = -EBUSY;
1262	} else {
1263		switch (notification->sigev_notify) {
1264		case SIGEV_NONE:
1265			info->notify.sigev_notify = SIGEV_NONE;
1266			break;
1267		case SIGEV_THREAD:
1268			info->notify_sock = sock;
1269			info->notify_cookie = nc;
1270			sock = NULL;
1271			nc = NULL;
1272			info->notify.sigev_notify = SIGEV_THREAD;
1273			break;
1274		case SIGEV_SIGNAL:
1275			info->notify.sigev_signo = notification->sigev_signo;
1276			info->notify.sigev_value = notification->sigev_value;
1277			info->notify.sigev_notify = SIGEV_SIGNAL;
1278			break;
1279		}
1280
1281		info->notify_owner = get_pid(task_tgid(current));
1282		info->notify_user_ns = get_user_ns(current_user_ns());
1283		inode->i_atime = inode->i_ctime = current_time(inode);
1284	}
1285	spin_unlock(&info->lock);
1286out_fput:
1287	fdput(f);
1288out:
1289	if (sock)
1290		netlink_detachskb(sock, nc);
1291	else if (nc)
1292		dev_kfree_skb(nc);
1293
1294	return ret;
1295}
1296
1297SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1298		const struct sigevent __user *, u_notification)
1299{
1300	struct sigevent n, *p = NULL;
1301	if (u_notification) {
1302		if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1303			return -EFAULT;
1304		p = &n;
1305	}
1306	return do_mq_notify(mqdes, p);
1307}
1308
1309static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1310{
1311	struct fd f;
1312	struct inode *inode;
1313	struct mqueue_inode_info *info;
1314
1315	if (new && (new->mq_flags & (~O_NONBLOCK)))
1316		return -EINVAL;
1317
1318	f = fdget(mqdes);
1319	if (!f.file)
1320		return -EBADF;
1321
1322	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1323		fdput(f);
1324		return -EBADF;
1325	}
1326
1327	inode = file_inode(f.file);
1328	info = MQUEUE_I(inode);
1329
1330	spin_lock(&info->lock);
1331
1332	if (old) {
1333		*old = info->attr;
1334		old->mq_flags = f.file->f_flags & O_NONBLOCK;
1335	}
1336	if (new) {
1337		audit_mq_getsetattr(mqdes, new);
1338		spin_lock(&f.file->f_lock);
1339		if (new->mq_flags & O_NONBLOCK)
1340			f.file->f_flags |= O_NONBLOCK;
1341		else
1342			f.file->f_flags &= ~O_NONBLOCK;
1343		spin_unlock(&f.file->f_lock);
1344
1345		inode->i_atime = inode->i_ctime = current_time(inode);
1346	}
1347
1348	spin_unlock(&info->lock);
1349	fdput(f);
1350	return 0;
1351}
1352
1353SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1354		const struct mq_attr __user *, u_mqstat,
1355		struct mq_attr __user *, u_omqstat)
1356{
1357	int ret;
1358	struct mq_attr mqstat, omqstat;
1359	struct mq_attr *new = NULL, *old = NULL;
1360
1361	if (u_mqstat) {
1362		new = &mqstat;
1363		if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1364			return -EFAULT;
1365	}
1366	if (u_omqstat)
1367		old = &omqstat;
1368
1369	ret = do_mq_getsetattr(mqdes, new, old);
1370	if (ret || !old)
1371		return ret;
1372
1373	if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1374		return -EFAULT;
1375	return 0;
1376}
1377
1378#ifdef CONFIG_COMPAT
1379
1380struct compat_mq_attr {
1381	compat_long_t mq_flags;      /* message queue flags		     */
1382	compat_long_t mq_maxmsg;     /* maximum number of messages	     */
1383	compat_long_t mq_msgsize;    /* maximum message size		     */
1384	compat_long_t mq_curmsgs;    /* number of messages currently queued  */
1385	compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1386};
1387
1388static inline int get_compat_mq_attr(struct mq_attr *attr,
1389			const struct compat_mq_attr __user *uattr)
1390{
1391	struct compat_mq_attr v;
1392
1393	if (copy_from_user(&v, uattr, sizeof(*uattr)))
1394		return -EFAULT;
1395
1396	memset(attr, 0, sizeof(*attr));
1397	attr->mq_flags = v.mq_flags;
1398	attr->mq_maxmsg = v.mq_maxmsg;
1399	attr->mq_msgsize = v.mq_msgsize;
1400	attr->mq_curmsgs = v.mq_curmsgs;
1401	return 0;
1402}
1403
1404static inline int put_compat_mq_attr(const struct mq_attr *attr,
1405			struct compat_mq_attr __user *uattr)
1406{
1407	struct compat_mq_attr v;
1408
1409	memset(&v, 0, sizeof(v));
1410	v.mq_flags = attr->mq_flags;
1411	v.mq_maxmsg = attr->mq_maxmsg;
1412	v.mq_msgsize = attr->mq_msgsize;
1413	v.mq_curmsgs = attr->mq_curmsgs;
1414	if (copy_to_user(uattr, &v, sizeof(*uattr)))
1415		return -EFAULT;
1416	return 0;
1417}
1418
1419COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1420		       int, oflag, compat_mode_t, mode,
1421		       struct compat_mq_attr __user *, u_attr)
1422{
1423	struct mq_attr attr, *p = NULL;
1424	if (u_attr && oflag & O_CREAT) {
1425		p = &attr;
1426		if (get_compat_mq_attr(&attr, u_attr))
1427			return -EFAULT;
1428	}
1429	return do_mq_open(u_name, oflag, mode, p);
1430}
1431
1432static int compat_prepare_timeout(const struct compat_timespec __user *p,
1433				   struct timespec64 *ts)
1434{
1435	if (compat_get_timespec64(ts, p))
1436		return -EFAULT;
1437	if (!timespec64_valid(ts))
1438		return -EINVAL;
1439	return 0;
1440}
1441
1442COMPAT_SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes,
1443		       const char __user *, u_msg_ptr,
1444		       compat_size_t, msg_len, unsigned int, msg_prio,
1445		       const struct compat_timespec __user *, u_abs_timeout)
1446{
1447	struct timespec64 ts, *p = NULL;
1448	if (u_abs_timeout) {
1449		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1450		if (res)
1451			return res;
1452		p = &ts;
1453	}
1454	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1455}
1456
1457COMPAT_SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes,
1458		       char __user *, u_msg_ptr,
1459		       compat_size_t, msg_len, unsigned int __user *, u_msg_prio,
1460		       const struct compat_timespec __user *, u_abs_timeout)
1461{
1462	struct timespec64 ts, *p = NULL;
1463	if (u_abs_timeout) {
1464		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1465		if (res)
1466			return res;
1467		p = &ts;
1468	}
1469	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1470}
1471
1472COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1473		       const struct compat_sigevent __user *, u_notification)
1474{
1475	struct sigevent n, *p = NULL;
1476	if (u_notification) {
1477		if (get_compat_sigevent(&n, u_notification))
1478			return -EFAULT;
1479		if (n.sigev_notify == SIGEV_THREAD)
1480			n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1481		p = &n;
1482	}
1483	return do_mq_notify(mqdes, p);
1484}
1485
1486COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1487		       const struct compat_mq_attr __user *, u_mqstat,
1488		       struct compat_mq_attr __user *, u_omqstat)
1489{
1490	int ret;
1491	struct mq_attr mqstat, omqstat;
1492	struct mq_attr *new = NULL, *old = NULL;
1493
1494	if (u_mqstat) {
1495		new = &mqstat;
1496		if (get_compat_mq_attr(new, u_mqstat))
1497			return -EFAULT;
1498	}
1499	if (u_omqstat)
1500		old = &omqstat;
1501
1502	ret = do_mq_getsetattr(mqdes, new, old);
1503	if (ret || !old)
1504		return ret;
1505
1506	if (put_compat_mq_attr(old, u_omqstat))
1507		return -EFAULT;
1508	return 0;
1509}
1510#endif
1511
1512static const struct inode_operations mqueue_dir_inode_operations = {
1513	.lookup = simple_lookup,
1514	.create = mqueue_create,
1515	.unlink = mqueue_unlink,
1516};
1517
1518static const struct file_operations mqueue_file_operations = {
1519	.flush = mqueue_flush_file,
1520	.poll = mqueue_poll_file,
1521	.read = mqueue_read_file,
1522	.llseek = default_llseek,
1523};
1524
1525static const struct super_operations mqueue_super_ops = {
1526	.alloc_inode = mqueue_alloc_inode,
1527	.destroy_inode = mqueue_destroy_inode,
1528	.evict_inode = mqueue_evict_inode,
1529	.statfs = simple_statfs,
1530};
1531
1532static struct file_system_type mqueue_fs_type = {
1533	.name = "mqueue",
1534	.mount = mqueue_mount,
1535	.kill_sb = kill_litter_super,
1536	.fs_flags = FS_USERNS_MOUNT,
1537};
1538
1539int mq_init_ns(struct ipc_namespace *ns)
1540{
1541	ns->mq_queues_count  = 0;
1542	ns->mq_queues_max    = DFLT_QUEUESMAX;
1543	ns->mq_msg_max       = DFLT_MSGMAX;
1544	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1545	ns->mq_msg_default   = DFLT_MSG;
1546	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1547
1548	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1549	if (IS_ERR(ns->mq_mnt)) {
1550		int err = PTR_ERR(ns->mq_mnt);
1551		ns->mq_mnt = NULL;
1552		return err;
1553	}
1554	return 0;
1555}
1556
1557void mq_clear_sbinfo(struct ipc_namespace *ns)
1558{
1559	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1560}
1561
1562void mq_put_mnt(struct ipc_namespace *ns)
1563{
1564	kern_unmount(ns->mq_mnt);
1565}
1566
1567static int __init init_mqueue_fs(void)
1568{
1569	int error;
1570
1571	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1572				sizeof(struct mqueue_inode_info), 0,
1573				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1574	if (mqueue_inode_cachep == NULL)
1575		return -ENOMEM;
1576
1577	/* ignore failures - they are not fatal */
1578	mq_sysctl_table = mq_register_sysctl_table();
1579
1580	error = register_filesystem(&mqueue_fs_type);
1581	if (error)
1582		goto out_sysctl;
1583
1584	spin_lock_init(&mq_lock);
1585
1586	error = mq_init_ns(&init_ipc_ns);
1587	if (error)
1588		goto out_filesystem;
1589
1590	return 0;
1591
1592out_filesystem:
1593	unregister_filesystem(&mqueue_fs_type);
1594out_sysctl:
1595	if (mq_sysctl_table)
1596		unregister_sysctl_table(mq_sysctl_table);
1597	kmem_cache_destroy(mqueue_inode_cachep);
1598	return error;
1599}
1600
1601device_initcall(init_mqueue_fs);
1602