mqueue.c revision eecec19d
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		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
274			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
275			sizeof(struct posix_msg_tree_node);
276
277		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
278					  info->attr.mq_msgsize);
279
280		spin_lock(&mq_lock);
281		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
282		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
283			spin_unlock(&mq_lock);
284			/* mqueue_evict_inode() releases info->messages */
285			ret = -EMFILE;
286			goto out_inode;
287		}
288		u->mq_bytes += mq_bytes;
289		spin_unlock(&mq_lock);
290
291		/* all is ok */
292		info->user = get_uid(u);
293	} else if (S_ISDIR(mode)) {
294		inc_nlink(inode);
295		/* Some things misbehave if size == 0 on a directory */
296		inode->i_size = 2 * DIRENT_SIZE;
297		inode->i_op = &mqueue_dir_inode_operations;
298		inode->i_fop = &simple_dir_operations;
299	}
300
301	return inode;
302out_inode:
303	iput(inode);
304err:
305	return ERR_PTR(ret);
306}
307
308static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
309{
310	struct inode *inode;
311	struct ipc_namespace *ns = sb->s_fs_info;
312
313	sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
314	sb->s_blocksize = PAGE_SIZE;
315	sb->s_blocksize_bits = PAGE_SHIFT;
316	sb->s_magic = MQUEUE_MAGIC;
317	sb->s_op = &mqueue_super_ops;
318
319	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
320	if (IS_ERR(inode))
321		return PTR_ERR(inode);
322
323	sb->s_root = d_make_root(inode);
324	if (!sb->s_root)
325		return -ENOMEM;
326	return 0;
327}
328
329static struct dentry *mqueue_mount(struct file_system_type *fs_type,
330			 int flags, const char *dev_name,
331			 void *data)
332{
333	struct ipc_namespace *ns;
334	if (flags & SB_KERNMOUNT) {
335		ns = data;
336		data = NULL;
337	} else {
338		ns = current->nsproxy->ipc_ns;
339	}
340	return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super);
341}
342
343static void init_once(void *foo)
344{
345	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
346
347	inode_init_once(&p->vfs_inode);
348}
349
350static struct inode *mqueue_alloc_inode(struct super_block *sb)
351{
352	struct mqueue_inode_info *ei;
353
354	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
355	if (!ei)
356		return NULL;
357	return &ei->vfs_inode;
358}
359
360static void mqueue_i_callback(struct rcu_head *head)
361{
362	struct inode *inode = container_of(head, struct inode, i_rcu);
363	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
364}
365
366static void mqueue_destroy_inode(struct inode *inode)
367{
368	call_rcu(&inode->i_rcu, mqueue_i_callback);
369}
370
371static void mqueue_evict_inode(struct inode *inode)
372{
373	struct mqueue_inode_info *info;
374	struct user_struct *user;
375	unsigned long mq_bytes, mq_treesize;
376	struct ipc_namespace *ipc_ns;
377	struct msg_msg *msg;
378
379	clear_inode(inode);
380
381	if (S_ISDIR(inode->i_mode))
382		return;
383
384	ipc_ns = get_ns_from_inode(inode);
385	info = MQUEUE_I(inode);
386	spin_lock(&info->lock);
387	while ((msg = msg_get(info)) != NULL)
388		free_msg(msg);
389	kfree(info->node_cache);
390	spin_unlock(&info->lock);
391
392	/* Total amount of bytes accounted for the mqueue */
393	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
394		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
395		sizeof(struct posix_msg_tree_node);
396
397	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
398				  info->attr.mq_msgsize);
399
400	user = info->user;
401	if (user) {
402		spin_lock(&mq_lock);
403		user->mq_bytes -= mq_bytes;
404		/*
405		 * get_ns_from_inode() ensures that the
406		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
407		 * to which we now hold a reference, or it is NULL.
408		 * We can't put it here under mq_lock, though.
409		 */
410		if (ipc_ns)
411			ipc_ns->mq_queues_count--;
412		spin_unlock(&mq_lock);
413		free_uid(user);
414	}
415	if (ipc_ns)
416		put_ipc_ns(ipc_ns);
417}
418
419static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
420{
421	struct inode *dir = dentry->d_parent->d_inode;
422	struct inode *inode;
423	struct mq_attr *attr = arg;
424	int error;
425	struct ipc_namespace *ipc_ns;
426
427	spin_lock(&mq_lock);
428	ipc_ns = __get_ns_from_inode(dir);
429	if (!ipc_ns) {
430		error = -EACCES;
431		goto out_unlock;
432	}
433
434	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
435	    !capable(CAP_SYS_RESOURCE)) {
436		error = -ENOSPC;
437		goto out_unlock;
438	}
439	ipc_ns->mq_queues_count++;
440	spin_unlock(&mq_lock);
441
442	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
443	if (IS_ERR(inode)) {
444		error = PTR_ERR(inode);
445		spin_lock(&mq_lock);
446		ipc_ns->mq_queues_count--;
447		goto out_unlock;
448	}
449
450	put_ipc_ns(ipc_ns);
451	dir->i_size += DIRENT_SIZE;
452	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
453
454	d_instantiate(dentry, inode);
455	dget(dentry);
456	return 0;
457out_unlock:
458	spin_unlock(&mq_lock);
459	if (ipc_ns)
460		put_ipc_ns(ipc_ns);
461	return error;
462}
463
464static int mqueue_create(struct inode *dir, struct dentry *dentry,
465				umode_t mode, bool excl)
466{
467	return mqueue_create_attr(dentry, mode, NULL);
468}
469
470static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
471{
472	struct inode *inode = d_inode(dentry);
473
474	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
475	dir->i_size -= DIRENT_SIZE;
476	drop_nlink(inode);
477	dput(dentry);
478	return 0;
479}
480
481/*
482*	This is routine for system read from queue file.
483*	To avoid mess with doing here some sort of mq_receive we allow
484*	to read only queue size & notification info (the only values
485*	that are interesting from user point of view and aren't accessible
486*	through std routines)
487*/
488static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
489				size_t count, loff_t *off)
490{
491	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
492	char buffer[FILENT_SIZE];
493	ssize_t ret;
494
495	spin_lock(&info->lock);
496	snprintf(buffer, sizeof(buffer),
497			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
498			info->qsize,
499			info->notify_owner ? info->notify.sigev_notify : 0,
500			(info->notify_owner &&
501			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
502				info->notify.sigev_signo : 0,
503			pid_vnr(info->notify_owner));
504	spin_unlock(&info->lock);
505	buffer[sizeof(buffer)-1] = '\0';
506
507	ret = simple_read_from_buffer(u_data, count, off, buffer,
508				strlen(buffer));
509	if (ret <= 0)
510		return ret;
511
512	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
513	return ret;
514}
515
516static int mqueue_flush_file(struct file *filp, fl_owner_t id)
517{
518	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
519
520	spin_lock(&info->lock);
521	if (task_tgid(current) == info->notify_owner)
522		remove_notification(info);
523
524	spin_unlock(&info->lock);
525	return 0;
526}
527
528static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
529{
530	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
531	int retval = 0;
532
533	poll_wait(filp, &info->wait_q, poll_tab);
534
535	spin_lock(&info->lock);
536	if (info->attr.mq_curmsgs)
537		retval = POLLIN | POLLRDNORM;
538
539	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
540		retval |= POLLOUT | POLLWRNORM;
541	spin_unlock(&info->lock);
542
543	return retval;
544}
545
546/* Adds current to info->e_wait_q[sr] before element with smaller prio */
547static void wq_add(struct mqueue_inode_info *info, int sr,
548			struct ext_wait_queue *ewp)
549{
550	struct ext_wait_queue *walk;
551
552	ewp->task = current;
553
554	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
555		if (walk->task->static_prio <= current->static_prio) {
556			list_add_tail(&ewp->list, &walk->list);
557			return;
558		}
559	}
560	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
561}
562
563/*
564 * Puts current task to sleep. Caller must hold queue lock. After return
565 * lock isn't held.
566 * sr: SEND or RECV
567 */
568static int wq_sleep(struct mqueue_inode_info *info, int sr,
569		    ktime_t *timeout, struct ext_wait_queue *ewp)
570	__releases(&info->lock)
571{
572	int retval;
573	signed long time;
574
575	wq_add(info, sr, ewp);
576
577	for (;;) {
578		__set_current_state(TASK_INTERRUPTIBLE);
579
580		spin_unlock(&info->lock);
581		time = schedule_hrtimeout_range_clock(timeout, 0,
582			HRTIMER_MODE_ABS, CLOCK_REALTIME);
583
584		if (ewp->state == STATE_READY) {
585			retval = 0;
586			goto out;
587		}
588		spin_lock(&info->lock);
589		if (ewp->state == STATE_READY) {
590			retval = 0;
591			goto out_unlock;
592		}
593		if (signal_pending(current)) {
594			retval = -ERESTARTSYS;
595			break;
596		}
597		if (time == 0) {
598			retval = -ETIMEDOUT;
599			break;
600		}
601	}
602	list_del(&ewp->list);
603out_unlock:
604	spin_unlock(&info->lock);
605out:
606	return retval;
607}
608
609/*
610 * Returns waiting task that should be serviced first or NULL if none exists
611 */
612static struct ext_wait_queue *wq_get_first_waiter(
613		struct mqueue_inode_info *info, int sr)
614{
615	struct list_head *ptr;
616
617	ptr = info->e_wait_q[sr].list.prev;
618	if (ptr == &info->e_wait_q[sr].list)
619		return NULL;
620	return list_entry(ptr, struct ext_wait_queue, list);
621}
622
623
624static inline void set_cookie(struct sk_buff *skb, char code)
625{
626	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
627}
628
629/*
630 * The next function is only to split too long sys_mq_timedsend
631 */
632static void __do_notify(struct mqueue_inode_info *info)
633{
634	/* notification
635	 * invoked when there is registered process and there isn't process
636	 * waiting synchronously for message AND state of queue changed from
637	 * empty to not empty. Here we are sure that no one is waiting
638	 * synchronously. */
639	if (info->notify_owner &&
640	    info->attr.mq_curmsgs == 1) {
641		struct siginfo sig_i;
642		switch (info->notify.sigev_notify) {
643		case SIGEV_NONE:
644			break;
645		case SIGEV_SIGNAL:
646			/* sends signal */
647
648			sig_i.si_signo = info->notify.sigev_signo;
649			sig_i.si_errno = 0;
650			sig_i.si_code = SI_MESGQ;
651			sig_i.si_value = info->notify.sigev_value;
652			/* map current pid/uid into info->owner's namespaces */
653			rcu_read_lock();
654			sig_i.si_pid = task_tgid_nr_ns(current,
655						ns_of_pid(info->notify_owner));
656			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
657			rcu_read_unlock();
658
659			kill_pid_info(info->notify.sigev_signo,
660				      &sig_i, info->notify_owner);
661			break;
662		case SIGEV_THREAD:
663			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
664			netlink_sendskb(info->notify_sock, info->notify_cookie);
665			break;
666		}
667		/* after notification unregisters process */
668		put_pid(info->notify_owner);
669		put_user_ns(info->notify_user_ns);
670		info->notify_owner = NULL;
671		info->notify_user_ns = NULL;
672	}
673	wake_up(&info->wait_q);
674}
675
676static int prepare_timeout(const struct timespec __user *u_abs_timeout,
677			   struct timespec64 *ts)
678{
679	if (get_timespec64(ts, u_abs_timeout))
680		return -EFAULT;
681	if (!timespec64_valid(ts))
682		return -EINVAL;
683	return 0;
684}
685
686static void remove_notification(struct mqueue_inode_info *info)
687{
688	if (info->notify_owner != NULL &&
689	    info->notify.sigev_notify == SIGEV_THREAD) {
690		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
691		netlink_sendskb(info->notify_sock, info->notify_cookie);
692	}
693	put_pid(info->notify_owner);
694	put_user_ns(info->notify_user_ns);
695	info->notify_owner = NULL;
696	info->notify_user_ns = NULL;
697}
698
699static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
700{
701	int mq_treesize;
702	unsigned long total_size;
703
704	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
705		return -EINVAL;
706	if (capable(CAP_SYS_RESOURCE)) {
707		if (attr->mq_maxmsg > HARD_MSGMAX ||
708		    attr->mq_msgsize > HARD_MSGSIZEMAX)
709			return -EINVAL;
710	} else {
711		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
712				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
713			return -EINVAL;
714	}
715	/* check for overflow */
716	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
717		return -EOVERFLOW;
718	mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
719		min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
720		sizeof(struct posix_msg_tree_node);
721	total_size = attr->mq_maxmsg * attr->mq_msgsize;
722	if (total_size + mq_treesize < total_size)
723		return -EOVERFLOW;
724	return 0;
725}
726
727/*
728 * Invoked when creating a new queue via sys_mq_open
729 */
730static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
731			struct path *path, int oflag, umode_t mode,
732			struct mq_attr *attr)
733{
734	const struct cred *cred = current_cred();
735	int ret;
736
737	if (attr) {
738		ret = mq_attr_ok(ipc_ns, attr);
739		if (ret)
740			return ERR_PTR(ret);
741	} else {
742		struct mq_attr def_attr;
743
744		def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
745					 ipc_ns->mq_msg_default);
746		def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
747					  ipc_ns->mq_msgsize_default);
748		ret = mq_attr_ok(ipc_ns, &def_attr);
749		if (ret)
750			return ERR_PTR(ret);
751	}
752
753	ret = vfs_mkobj(path->dentry, mode & ~current_umask(),
754			mqueue_create_attr, attr);
755	if (ret)
756		return ERR_PTR(ret);
757	return dentry_open(path, oflag, cred);
758}
759
760/* Opens existing queue */
761static struct file *do_open(struct path *path, int oflag)
762{
763	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
764						  MAY_READ | MAY_WRITE };
765	int acc;
766	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
767		return ERR_PTR(-EINVAL);
768	acc = oflag2acc[oflag & O_ACCMODE];
769	if (inode_permission(d_inode(path->dentry), acc))
770		return ERR_PTR(-EACCES);
771	return dentry_open(path, oflag, current_cred());
772}
773
774static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
775		      struct mq_attr *attr)
776{
777	struct path path;
778	struct file *filp;
779	struct filename *name;
780	int fd, error;
781	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
782	struct vfsmount *mnt = ipc_ns->mq_mnt;
783	struct dentry *root = mnt->mnt_root;
784	int ro;
785
786	audit_mq_open(oflag, mode, attr);
787
788	if (IS_ERR(name = getname(u_name)))
789		return PTR_ERR(name);
790
791	fd = get_unused_fd_flags(O_CLOEXEC);
792	if (fd < 0)
793		goto out_putname;
794
795	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
796	error = 0;
797	inode_lock(d_inode(root));
798	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
799	if (IS_ERR(path.dentry)) {
800		error = PTR_ERR(path.dentry);
801		goto out_putfd;
802	}
803	path.mnt = mntget(mnt);
804
805	if (oflag & O_CREAT) {
806		if (d_really_is_positive(path.dentry)) {	/* entry already exists */
807			audit_inode(name, path.dentry, 0);
808			if (oflag & O_EXCL) {
809				error = -EEXIST;
810				goto out;
811			}
812			filp = do_open(&path, oflag);
813		} else {
814			if (ro) {
815				error = ro;
816				goto out;
817			}
818			audit_inode_parent_hidden(name, root);
819			filp = do_create(ipc_ns, d_inode(root), &path,
820					 oflag, mode, attr);
821		}
822	} else {
823		if (d_really_is_negative(path.dentry)) {
824			error = -ENOENT;
825			goto out;
826		}
827		audit_inode(name, path.dentry, 0);
828		filp = do_open(&path, oflag);
829	}
830
831	if (!IS_ERR(filp))
832		fd_install(fd, filp);
833	else
834		error = PTR_ERR(filp);
835out:
836	path_put(&path);
837out_putfd:
838	if (error) {
839		put_unused_fd(fd);
840		fd = error;
841	}
842	inode_unlock(d_inode(root));
843	if (!ro)
844		mnt_drop_write(mnt);
845out_putname:
846	putname(name);
847	return fd;
848}
849
850SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
851		struct mq_attr __user *, u_attr)
852{
853	struct mq_attr attr;
854	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
855		return -EFAULT;
856
857	return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
858}
859
860SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
861{
862	int err;
863	struct filename *name;
864	struct dentry *dentry;
865	struct inode *inode = NULL;
866	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
867	struct vfsmount *mnt = ipc_ns->mq_mnt;
868
869	name = getname(u_name);
870	if (IS_ERR(name))
871		return PTR_ERR(name);
872
873	audit_inode_parent_hidden(name, mnt->mnt_root);
874	err = mnt_want_write(mnt);
875	if (err)
876		goto out_name;
877	inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
878	dentry = lookup_one_len(name->name, mnt->mnt_root,
879				strlen(name->name));
880	if (IS_ERR(dentry)) {
881		err = PTR_ERR(dentry);
882		goto out_unlock;
883	}
884
885	inode = d_inode(dentry);
886	if (!inode) {
887		err = -ENOENT;
888	} else {
889		ihold(inode);
890		err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
891	}
892	dput(dentry);
893
894out_unlock:
895	inode_unlock(d_inode(mnt->mnt_root));
896	if (inode)
897		iput(inode);
898	mnt_drop_write(mnt);
899out_name:
900	putname(name);
901
902	return err;
903}
904
905/* Pipelined send and receive functions.
906 *
907 * If a receiver finds no waiting message, then it registers itself in the
908 * list of waiting receivers. A sender checks that list before adding the new
909 * message into the message array. If there is a waiting receiver, then it
910 * bypasses the message array and directly hands the message over to the
911 * receiver. The receiver accepts the message and returns without grabbing the
912 * queue spinlock:
913 *
914 * - Set pointer to message.
915 * - Queue the receiver task for later wakeup (without the info->lock).
916 * - Update its state to STATE_READY. Now the receiver can continue.
917 * - Wake up the process after the lock is dropped. Should the process wake up
918 *   before this wakeup (due to a timeout or a signal) it will either see
919 *   STATE_READY and continue or acquire the lock to check the state again.
920 *
921 * The same algorithm is used for senders.
922 */
923
924/* pipelined_send() - send a message directly to the task waiting in
925 * sys_mq_timedreceive() (without inserting message into a queue).
926 */
927static inline void pipelined_send(struct wake_q_head *wake_q,
928				  struct mqueue_inode_info *info,
929				  struct msg_msg *message,
930				  struct ext_wait_queue *receiver)
931{
932	receiver->msg = message;
933	list_del(&receiver->list);
934	wake_q_add(wake_q, receiver->task);
935	/*
936	 * Rely on the implicit cmpxchg barrier from wake_q_add such
937	 * that we can ensure that updating receiver->state is the last
938	 * write operation: As once set, the receiver can continue,
939	 * and if we don't have the reference count from the wake_q,
940	 * yet, at that point we can later have a use-after-free
941	 * condition and bogus wakeup.
942	 */
943	receiver->state = STATE_READY;
944}
945
946/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
947 * gets its message and put to the queue (we have one free place for sure). */
948static inline void pipelined_receive(struct wake_q_head *wake_q,
949				     struct mqueue_inode_info *info)
950{
951	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
952
953	if (!sender) {
954		/* for poll */
955		wake_up_interruptible(&info->wait_q);
956		return;
957	}
958	if (msg_insert(sender->msg, info))
959		return;
960
961	list_del(&sender->list);
962	wake_q_add(wake_q, sender->task);
963	sender->state = STATE_READY;
964}
965
966static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
967		size_t msg_len, unsigned int msg_prio,
968		struct timespec64 *ts)
969{
970	struct fd f;
971	struct inode *inode;
972	struct ext_wait_queue wait;
973	struct ext_wait_queue *receiver;
974	struct msg_msg *msg_ptr;
975	struct mqueue_inode_info *info;
976	ktime_t expires, *timeout = NULL;
977	struct posix_msg_tree_node *new_leaf = NULL;
978	int ret = 0;
979	DEFINE_WAKE_Q(wake_q);
980
981	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
982		return -EINVAL;
983
984	if (ts) {
985		expires = timespec64_to_ktime(*ts);
986		timeout = &expires;
987	}
988
989	audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
990
991	f = fdget(mqdes);
992	if (unlikely(!f.file)) {
993		ret = -EBADF;
994		goto out;
995	}
996
997	inode = file_inode(f.file);
998	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
999		ret = -EBADF;
1000		goto out_fput;
1001	}
1002	info = MQUEUE_I(inode);
1003	audit_file(f.file);
1004
1005	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1006		ret = -EBADF;
1007		goto out_fput;
1008	}
1009
1010	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1011		ret = -EMSGSIZE;
1012		goto out_fput;
1013	}
1014
1015	/* First try to allocate memory, before doing anything with
1016	 * existing queues. */
1017	msg_ptr = load_msg(u_msg_ptr, msg_len);
1018	if (IS_ERR(msg_ptr)) {
1019		ret = PTR_ERR(msg_ptr);
1020		goto out_fput;
1021	}
1022	msg_ptr->m_ts = msg_len;
1023	msg_ptr->m_type = msg_prio;
1024
1025	/*
1026	 * msg_insert really wants us to have a valid, spare node struct so
1027	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1028	 * fall back to that if necessary.
1029	 */
1030	if (!info->node_cache)
1031		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1032
1033	spin_lock(&info->lock);
1034
1035	if (!info->node_cache && new_leaf) {
1036		/* Save our speculative allocation into the cache */
1037		INIT_LIST_HEAD(&new_leaf->msg_list);
1038		info->node_cache = new_leaf;
1039		new_leaf = NULL;
1040	} else {
1041		kfree(new_leaf);
1042	}
1043
1044	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1045		if (f.file->f_flags & O_NONBLOCK) {
1046			ret = -EAGAIN;
1047		} else {
1048			wait.task = current;
1049			wait.msg = (void *) msg_ptr;
1050			wait.state = STATE_NONE;
1051			ret = wq_sleep(info, SEND, timeout, &wait);
1052			/*
1053			 * wq_sleep must be called with info->lock held, and
1054			 * returns with the lock released
1055			 */
1056			goto out_free;
1057		}
1058	} else {
1059		receiver = wq_get_first_waiter(info, RECV);
1060		if (receiver) {
1061			pipelined_send(&wake_q, info, msg_ptr, receiver);
1062		} else {
1063			/* adds message to the queue */
1064			ret = msg_insert(msg_ptr, info);
1065			if (ret)
1066				goto out_unlock;
1067			__do_notify(info);
1068		}
1069		inode->i_atime = inode->i_mtime = inode->i_ctime =
1070				current_time(inode);
1071	}
1072out_unlock:
1073	spin_unlock(&info->lock);
1074	wake_up_q(&wake_q);
1075out_free:
1076	if (ret)
1077		free_msg(msg_ptr);
1078out_fput:
1079	fdput(f);
1080out:
1081	return ret;
1082}
1083
1084static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1085		size_t msg_len, unsigned int __user *u_msg_prio,
1086		struct timespec64 *ts)
1087{
1088	ssize_t ret;
1089	struct msg_msg *msg_ptr;
1090	struct fd f;
1091	struct inode *inode;
1092	struct mqueue_inode_info *info;
1093	struct ext_wait_queue wait;
1094	ktime_t expires, *timeout = NULL;
1095	struct posix_msg_tree_node *new_leaf = NULL;
1096
1097	if (ts) {
1098		expires = timespec64_to_ktime(*ts);
1099		timeout = &expires;
1100	}
1101
1102	audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1103
1104	f = fdget(mqdes);
1105	if (unlikely(!f.file)) {
1106		ret = -EBADF;
1107		goto out;
1108	}
1109
1110	inode = file_inode(f.file);
1111	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1112		ret = -EBADF;
1113		goto out_fput;
1114	}
1115	info = MQUEUE_I(inode);
1116	audit_file(f.file);
1117
1118	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1119		ret = -EBADF;
1120		goto out_fput;
1121	}
1122
1123	/* checks if buffer is big enough */
1124	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1125		ret = -EMSGSIZE;
1126		goto out_fput;
1127	}
1128
1129	/*
1130	 * msg_insert really wants us to have a valid, spare node struct so
1131	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1132	 * fall back to that if necessary.
1133	 */
1134	if (!info->node_cache)
1135		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1136
1137	spin_lock(&info->lock);
1138
1139	if (!info->node_cache && new_leaf) {
1140		/* Save our speculative allocation into the cache */
1141		INIT_LIST_HEAD(&new_leaf->msg_list);
1142		info->node_cache = new_leaf;
1143	} else {
1144		kfree(new_leaf);
1145	}
1146
1147	if (info->attr.mq_curmsgs == 0) {
1148		if (f.file->f_flags & O_NONBLOCK) {
1149			spin_unlock(&info->lock);
1150			ret = -EAGAIN;
1151		} else {
1152			wait.task = current;
1153			wait.state = STATE_NONE;
1154			ret = wq_sleep(info, RECV, timeout, &wait);
1155			msg_ptr = wait.msg;
1156		}
1157	} else {
1158		DEFINE_WAKE_Q(wake_q);
1159
1160		msg_ptr = msg_get(info);
1161
1162		inode->i_atime = inode->i_mtime = inode->i_ctime =
1163				current_time(inode);
1164
1165		/* There is now free space in queue. */
1166		pipelined_receive(&wake_q, info);
1167		spin_unlock(&info->lock);
1168		wake_up_q(&wake_q);
1169		ret = 0;
1170	}
1171	if (ret == 0) {
1172		ret = msg_ptr->m_ts;
1173
1174		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1175			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1176			ret = -EFAULT;
1177		}
1178		free_msg(msg_ptr);
1179	}
1180out_fput:
1181	fdput(f);
1182out:
1183	return ret;
1184}
1185
1186SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1187		size_t, msg_len, unsigned int, msg_prio,
1188		const struct timespec __user *, u_abs_timeout)
1189{
1190	struct timespec64 ts, *p = NULL;
1191	if (u_abs_timeout) {
1192		int res = prepare_timeout(u_abs_timeout, &ts);
1193		if (res)
1194			return res;
1195		p = &ts;
1196	}
1197	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1198}
1199
1200SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1201		size_t, msg_len, unsigned int __user *, u_msg_prio,
1202		const struct timespec __user *, u_abs_timeout)
1203{
1204	struct timespec64 ts, *p = NULL;
1205	if (u_abs_timeout) {
1206		int res = prepare_timeout(u_abs_timeout, &ts);
1207		if (res)
1208			return res;
1209		p = &ts;
1210	}
1211	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1212}
1213
1214/*
1215 * Notes: the case when user wants us to deregister (with NULL as pointer)
1216 * and he isn't currently owner of notification, will be silently discarded.
1217 * It isn't explicitly defined in the POSIX.
1218 */
1219static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1220{
1221	int ret;
1222	struct fd f;
1223	struct sock *sock;
1224	struct inode *inode;
1225	struct mqueue_inode_info *info;
1226	struct sk_buff *nc;
1227
1228	audit_mq_notify(mqdes, notification);
1229
1230	nc = NULL;
1231	sock = NULL;
1232	if (notification != NULL) {
1233		if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1234			     notification->sigev_notify != SIGEV_SIGNAL &&
1235			     notification->sigev_notify != SIGEV_THREAD))
1236			return -EINVAL;
1237		if (notification->sigev_notify == SIGEV_SIGNAL &&
1238			!valid_signal(notification->sigev_signo)) {
1239			return -EINVAL;
1240		}
1241		if (notification->sigev_notify == SIGEV_THREAD) {
1242			long timeo;
1243
1244			/* create the notify skb */
1245			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1246			if (!nc) {
1247				ret = -ENOMEM;
1248				goto out;
1249			}
1250			if (copy_from_user(nc->data,
1251					notification->sigev_value.sival_ptr,
1252					NOTIFY_COOKIE_LEN)) {
1253				ret = -EFAULT;
1254				goto out;
1255			}
1256
1257			/* TODO: add a header? */
1258			skb_put(nc, NOTIFY_COOKIE_LEN);
1259			/* and attach it to the socket */
1260retry:
1261			f = fdget(notification->sigev_signo);
1262			if (!f.file) {
1263				ret = -EBADF;
1264				goto out;
1265			}
1266			sock = netlink_getsockbyfilp(f.file);
1267			fdput(f);
1268			if (IS_ERR(sock)) {
1269				ret = PTR_ERR(sock);
1270				sock = NULL;
1271				goto out;
1272			}
1273
1274			timeo = MAX_SCHEDULE_TIMEOUT;
1275			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1276			if (ret == 1) {
1277				sock = NULL;
1278				goto retry;
1279			}
1280			if (ret) {
1281				sock = NULL;
1282				nc = NULL;
1283				goto out;
1284			}
1285		}
1286	}
1287
1288	f = fdget(mqdes);
1289	if (!f.file) {
1290		ret = -EBADF;
1291		goto out;
1292	}
1293
1294	inode = file_inode(f.file);
1295	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1296		ret = -EBADF;
1297		goto out_fput;
1298	}
1299	info = MQUEUE_I(inode);
1300
1301	ret = 0;
1302	spin_lock(&info->lock);
1303	if (notification == NULL) {
1304		if (info->notify_owner == task_tgid(current)) {
1305			remove_notification(info);
1306			inode->i_atime = inode->i_ctime = current_time(inode);
1307		}
1308	} else if (info->notify_owner != NULL) {
1309		ret = -EBUSY;
1310	} else {
1311		switch (notification->sigev_notify) {
1312		case SIGEV_NONE:
1313			info->notify.sigev_notify = SIGEV_NONE;
1314			break;
1315		case SIGEV_THREAD:
1316			info->notify_sock = sock;
1317			info->notify_cookie = nc;
1318			sock = NULL;
1319			nc = NULL;
1320			info->notify.sigev_notify = SIGEV_THREAD;
1321			break;
1322		case SIGEV_SIGNAL:
1323			info->notify.sigev_signo = notification->sigev_signo;
1324			info->notify.sigev_value = notification->sigev_value;
1325			info->notify.sigev_notify = SIGEV_SIGNAL;
1326			break;
1327		}
1328
1329		info->notify_owner = get_pid(task_tgid(current));
1330		info->notify_user_ns = get_user_ns(current_user_ns());
1331		inode->i_atime = inode->i_ctime = current_time(inode);
1332	}
1333	spin_unlock(&info->lock);
1334out_fput:
1335	fdput(f);
1336out:
1337	if (sock)
1338		netlink_detachskb(sock, nc);
1339	else if (nc)
1340		dev_kfree_skb(nc);
1341
1342	return ret;
1343}
1344
1345SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1346		const struct sigevent __user *, u_notification)
1347{
1348	struct sigevent n, *p = NULL;
1349	if (u_notification) {
1350		if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1351			return -EFAULT;
1352		p = &n;
1353	}
1354	return do_mq_notify(mqdes, p);
1355}
1356
1357static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1358{
1359	struct fd f;
1360	struct inode *inode;
1361	struct mqueue_inode_info *info;
1362
1363	if (new && (new->mq_flags & (~O_NONBLOCK)))
1364		return -EINVAL;
1365
1366	f = fdget(mqdes);
1367	if (!f.file)
1368		return -EBADF;
1369
1370	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1371		fdput(f);
1372		return -EBADF;
1373	}
1374
1375	inode = file_inode(f.file);
1376	info = MQUEUE_I(inode);
1377
1378	spin_lock(&info->lock);
1379
1380	if (old) {
1381		*old = info->attr;
1382		old->mq_flags = f.file->f_flags & O_NONBLOCK;
1383	}
1384	if (new) {
1385		audit_mq_getsetattr(mqdes, new);
1386		spin_lock(&f.file->f_lock);
1387		if (new->mq_flags & O_NONBLOCK)
1388			f.file->f_flags |= O_NONBLOCK;
1389		else
1390			f.file->f_flags &= ~O_NONBLOCK;
1391		spin_unlock(&f.file->f_lock);
1392
1393		inode->i_atime = inode->i_ctime = current_time(inode);
1394	}
1395
1396	spin_unlock(&info->lock);
1397	fdput(f);
1398	return 0;
1399}
1400
1401SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1402		const struct mq_attr __user *, u_mqstat,
1403		struct mq_attr __user *, u_omqstat)
1404{
1405	int ret;
1406	struct mq_attr mqstat, omqstat;
1407	struct mq_attr *new = NULL, *old = NULL;
1408
1409	if (u_mqstat) {
1410		new = &mqstat;
1411		if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1412			return -EFAULT;
1413	}
1414	if (u_omqstat)
1415		old = &omqstat;
1416
1417	ret = do_mq_getsetattr(mqdes, new, old);
1418	if (ret || !old)
1419		return ret;
1420
1421	if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1422		return -EFAULT;
1423	return 0;
1424}
1425
1426#ifdef CONFIG_COMPAT
1427
1428struct compat_mq_attr {
1429	compat_long_t mq_flags;      /* message queue flags		     */
1430	compat_long_t mq_maxmsg;     /* maximum number of messages	     */
1431	compat_long_t mq_msgsize;    /* maximum message size		     */
1432	compat_long_t mq_curmsgs;    /* number of messages currently queued  */
1433	compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1434};
1435
1436static inline int get_compat_mq_attr(struct mq_attr *attr,
1437			const struct compat_mq_attr __user *uattr)
1438{
1439	struct compat_mq_attr v;
1440
1441	if (copy_from_user(&v, uattr, sizeof(*uattr)))
1442		return -EFAULT;
1443
1444	memset(attr, 0, sizeof(*attr));
1445	attr->mq_flags = v.mq_flags;
1446	attr->mq_maxmsg = v.mq_maxmsg;
1447	attr->mq_msgsize = v.mq_msgsize;
1448	attr->mq_curmsgs = v.mq_curmsgs;
1449	return 0;
1450}
1451
1452static inline int put_compat_mq_attr(const struct mq_attr *attr,
1453			struct compat_mq_attr __user *uattr)
1454{
1455	struct compat_mq_attr v;
1456
1457	memset(&v, 0, sizeof(v));
1458	v.mq_flags = attr->mq_flags;
1459	v.mq_maxmsg = attr->mq_maxmsg;
1460	v.mq_msgsize = attr->mq_msgsize;
1461	v.mq_curmsgs = attr->mq_curmsgs;
1462	if (copy_to_user(uattr, &v, sizeof(*uattr)))
1463		return -EFAULT;
1464	return 0;
1465}
1466
1467COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1468		       int, oflag, compat_mode_t, mode,
1469		       struct compat_mq_attr __user *, u_attr)
1470{
1471	struct mq_attr attr, *p = NULL;
1472	if (u_attr && oflag & O_CREAT) {
1473		p = &attr;
1474		if (get_compat_mq_attr(&attr, u_attr))
1475			return -EFAULT;
1476	}
1477	return do_mq_open(u_name, oflag, mode, p);
1478}
1479
1480static int compat_prepare_timeout(const struct compat_timespec __user *p,
1481				   struct timespec64 *ts)
1482{
1483	if (compat_get_timespec64(ts, p))
1484		return -EFAULT;
1485	if (!timespec64_valid(ts))
1486		return -EINVAL;
1487	return 0;
1488}
1489
1490COMPAT_SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes,
1491		       const char __user *, u_msg_ptr,
1492		       compat_size_t, msg_len, unsigned int, msg_prio,
1493		       const struct compat_timespec __user *, u_abs_timeout)
1494{
1495	struct timespec64 ts, *p = NULL;
1496	if (u_abs_timeout) {
1497		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1498		if (res)
1499			return res;
1500		p = &ts;
1501	}
1502	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1503}
1504
1505COMPAT_SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes,
1506		       char __user *, u_msg_ptr,
1507		       compat_size_t, msg_len, unsigned int __user *, u_msg_prio,
1508		       const struct compat_timespec __user *, u_abs_timeout)
1509{
1510	struct timespec64 ts, *p = NULL;
1511	if (u_abs_timeout) {
1512		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1513		if (res)
1514			return res;
1515		p = &ts;
1516	}
1517	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1518}
1519
1520COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1521		       const struct compat_sigevent __user *, u_notification)
1522{
1523	struct sigevent n, *p = NULL;
1524	if (u_notification) {
1525		if (get_compat_sigevent(&n, u_notification))
1526			return -EFAULT;
1527		if (n.sigev_notify == SIGEV_THREAD)
1528			n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1529		p = &n;
1530	}
1531	return do_mq_notify(mqdes, p);
1532}
1533
1534COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1535		       const struct compat_mq_attr __user *, u_mqstat,
1536		       struct compat_mq_attr __user *, u_omqstat)
1537{
1538	int ret;
1539	struct mq_attr mqstat, omqstat;
1540	struct mq_attr *new = NULL, *old = NULL;
1541
1542	if (u_mqstat) {
1543		new = &mqstat;
1544		if (get_compat_mq_attr(new, u_mqstat))
1545			return -EFAULT;
1546	}
1547	if (u_omqstat)
1548		old = &omqstat;
1549
1550	ret = do_mq_getsetattr(mqdes, new, old);
1551	if (ret || !old)
1552		return ret;
1553
1554	if (put_compat_mq_attr(old, u_omqstat))
1555		return -EFAULT;
1556	return 0;
1557}
1558#endif
1559
1560static const struct inode_operations mqueue_dir_inode_operations = {
1561	.lookup = simple_lookup,
1562	.create = mqueue_create,
1563	.unlink = mqueue_unlink,
1564};
1565
1566static const struct file_operations mqueue_file_operations = {
1567	.flush = mqueue_flush_file,
1568	.poll = mqueue_poll_file,
1569	.read = mqueue_read_file,
1570	.llseek = default_llseek,
1571};
1572
1573static const struct super_operations mqueue_super_ops = {
1574	.alloc_inode = mqueue_alloc_inode,
1575	.destroy_inode = mqueue_destroy_inode,
1576	.evict_inode = mqueue_evict_inode,
1577	.statfs = simple_statfs,
1578};
1579
1580static struct file_system_type mqueue_fs_type = {
1581	.name = "mqueue",
1582	.mount = mqueue_mount,
1583	.kill_sb = kill_litter_super,
1584	.fs_flags = FS_USERNS_MOUNT,
1585};
1586
1587int mq_init_ns(struct ipc_namespace *ns)
1588{
1589	ns->mq_queues_count  = 0;
1590	ns->mq_queues_max    = DFLT_QUEUESMAX;
1591	ns->mq_msg_max       = DFLT_MSGMAX;
1592	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1593	ns->mq_msg_default   = DFLT_MSG;
1594	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1595
1596	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1597	if (IS_ERR(ns->mq_mnt)) {
1598		int err = PTR_ERR(ns->mq_mnt);
1599		ns->mq_mnt = NULL;
1600		return err;
1601	}
1602	return 0;
1603}
1604
1605void mq_clear_sbinfo(struct ipc_namespace *ns)
1606{
1607	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1608}
1609
1610void mq_put_mnt(struct ipc_namespace *ns)
1611{
1612	kern_unmount(ns->mq_mnt);
1613}
1614
1615static int __init init_mqueue_fs(void)
1616{
1617	int error;
1618
1619	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1620				sizeof(struct mqueue_inode_info), 0,
1621				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1622	if (mqueue_inode_cachep == NULL)
1623		return -ENOMEM;
1624
1625	/* ignore failures - they are not fatal */
1626	mq_sysctl_table = mq_register_sysctl_table();
1627
1628	error = register_filesystem(&mqueue_fs_type);
1629	if (error)
1630		goto out_sysctl;
1631
1632	spin_lock_init(&mq_lock);
1633
1634	error = mq_init_ns(&init_ipc_ns);
1635	if (error)
1636		goto out_filesystem;
1637
1638	return 0;
1639
1640out_filesystem:
1641	unregister_filesystem(&mqueue_fs_type);
1642out_sysctl:
1643	if (mq_sysctl_table)
1644		unregister_sysctl_table(mq_sysctl_table);
1645	kmem_cache_destroy(mqueue_inode_cachep);
1646	return error;
1647}
1648
1649device_initcall(init_mqueue_fs);
1650