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	bool newns; /* Set if newly created ipc namespace */
49	};
50
51	#define MQUEUE_MAGIC 0x19800202
52	#define DIRENT_SIZE 20
53	#define FILENT_SIZE 80
54
55	#define SEND 0
56	#define RECV 1
57
58	#define STATE_NONE 0
59	#define STATE_READY 1
60
61	struct posix_msg_tree_node {
62	struct rb_node rb_node;
63	struct list_head msg_list;
64	int priority;
65	};
66
67	/*
68	* Locking:
69	*
70	* Accesses to a message queue are synchronized by acquiring info->lock.
71	*
72	* There are two notable exceptions:
73	* - The actual wakeup of a sleeping task is performed using the wake_q
74	* framework. info->lock is already released when wake_up_q is called.
75	* - The exit codepaths after sleeping check ext_wait_queue->state without
76	* any locks. If it is STATE_READY, then the syscall is completed without
77	* acquiring info->lock.
78	*
79	* MQ_BARRIER:
80	* To achieve proper release/acquire memory barrier pairing, the state is set to
81	* STATE_READY with smp_store_release(), and it is read with READ_ONCE followed
82	* by smp_acquire__after_ctrl_dep(). In addition, wake_q_add_safe() is used.
83	*
84	* This prevents the following races:
85	*
86	* 1) With the simple wake_q_add(), the task could be gone already before
87	* the increase of the reference happens
88	* Thread A
89	* Thread B
90	* WRITE_ONCE(wait.state, STATE_NONE);
91	* schedule_hrtimeout()
92	* wake_q_add(A)
93	* if (cmpxchg()) // success
94	* ->state = STATE_READY (reordered)
95	* <timeout returns>
96	* if (wait.state == STATE_READY) return;
97	* sysret to user space
98	* sys_exit()
99	* get_task_struct() // UaF
100	*
101	* Solution: Use wake_q_add_safe() and perform the get_task_struct() before
102	* the smp_store_release() that does ->state = STATE_READY.
103	*
104	* 2) Without proper _release/_acquire barriers, the woken up task
105	* could read stale data
106	*
107	* Thread A
108	* Thread B
109	* do_mq_timedreceive
110	* WRITE_ONCE(wait.state, STATE_NONE);
111	* schedule_hrtimeout()
112	* state = STATE_READY;
113	* <timeout returns>
114	* if (wait.state == STATE_READY) return;
115	* msg_ptr = wait.msg; // Access to stale data!
116	* receiver->msg = message; (reordered)
117	*
118	* Solution: use _release and _acquire barriers.
119	*
120	* 3) There is intentionally no barrier when setting current->state
121	* to TASK_INTERRUPTIBLE: spin_unlock(&info->lock) provides the
122	* release memory barrier, and the wakeup is triggered when holding
123	* info->lock, i.e. spin_lock(&info->lock) provided a pairing
124	* acquire memory barrier.
125	*/
126
127	struct ext_wait_queue { /* queue of sleeping tasks */
128	struct task_struct *task;
129	struct list_head list;
130	struct msg_msg *msg; /* ptr of loaded message */
131	int state; /* one of STATE_* values */
132	};
133
134	struct mqueue_inode_info {
135	spinlock_t lock;
136	struct inode vfs_inode;
137	wait_queue_head_t wait_q;
138
139	struct rb_root msg_tree;
140	struct rb_node *msg_tree_rightmost;
141	struct posix_msg_tree_node *node_cache;
142	struct mq_attr attr;
143
144	struct sigevent notify;
145	struct pid *notify_owner;
146	u32 notify_self_exec_id;
147	struct user_namespace *notify_user_ns;
148	struct ucounts *ucounts; /* user who created, for accounting */
149	struct sock *notify_sock;
150	struct sk_buff *notify_cookie;
151
152	/* for tasks waiting for free space and messages, respectively */
153	struct ext_wait_queue e_wait_q[2];
154
155	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
156	};
157
158	static struct file_system_type mqueue_fs_type;
159	static const struct inode_operations mqueue_dir_inode_operations;
160	static const struct file_operations mqueue_file_operations;
161	static const struct super_operations mqueue_super_ops;
162	static const struct fs_context_operations mqueue_fs_context_ops;
163	static void remove_notification(struct mqueue_inode_info *info);
164
165	static struct kmem_cache *mqueue_inode_cachep;
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(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(lock: &mq_lock);
185	ns = __get_ns_from_inode(inode);
186	spin_unlock(lock: &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(list: &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(node: &leaf->rb_node, parent, rb_link: 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(new: &msg->m_list, head: &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(objp: leaf);
244	else
245	info->node_cache = leaf;
246	}
247
248	static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
249	{
250	struct rb_node *parent = NULL;
251	struct posix_msg_tree_node *leaf;
252	struct msg_msg *msg;
253
254	try_again:
255	/*
256	* During insert, low priorities go to the left and high to the
257	* right. On receive, we want the highest priorities first, so
258	* walk all the way to the right.
259	*/
260	parent = info->msg_tree_rightmost;
261	if (!parent) {
262	if (info->attr.mq_curmsgs) {
263	pr_warn_once("Inconsistency in POSIX message queue, "
264	"no tree element, but supposedly messages "
265	"should exist!\n");
266	info->attr.mq_curmsgs = 0;
267	}
268	return NULL;
269	}
270	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
271	if (unlikely(list_empty(&leaf->msg_list))) {
272	pr_warn_once("Inconsistency in POSIX message queue, "
273	"empty leaf node but we haven't implemented "
274	"lazy leaf delete!\n");
275	msg_tree_erase(leaf, info);
276	goto try_again;
277	} else {
278	msg = list_first_entry(&leaf->msg_list,
279	struct msg_msg, m_list);
280	list_del(entry: &msg->m_list);
281	if (list_empty(head: &leaf->msg_list)) {
282	msg_tree_erase(leaf, info);
283	}
284	}
285	info->attr.mq_curmsgs--;
286	info->qsize -= msg->m_ts;
287	return msg;
288	}
289
290	static struct inode *mqueue_get_inode(struct super_block *sb,
291	struct ipc_namespace *ipc_ns, umode_t mode,
292	struct mq_attr *attr)
293	{
294	struct inode *inode;
295	int ret = -ENOMEM;
296
297	inode = new_inode(sb);
298	if (!inode)
299	goto err;
300
301	inode->i_ino = get_next_ino();
302	inode->i_mode = mode;
303	inode->i_uid = current_fsuid();
304	inode->i_gid = current_fsgid();
305	simple_inode_init_ts(inode);
306
307	if (S_ISREG(mode)) {
308	struct mqueue_inode_info *info;
309	unsigned long mq_bytes, mq_treesize;
310
311	inode->i_fop = &mqueue_file_operations;
312	inode->i_size = FILENT_SIZE;
313	/* mqueue specific info */
314	info = MQUEUE_I(inode);
315	spin_lock_init(&info->lock);
316	init_waitqueue_head(&info->wait_q);
317	INIT_LIST_HEAD(list: &info->e_wait_q[0].list);
318	INIT_LIST_HEAD(list: &info->e_wait_q[1].list);
319	info->notify_owner = NULL;
320	info->notify_user_ns = NULL;
321	info->qsize = 0;
322	info->ucounts = NULL; /* set when all is ok */
323	info->msg_tree = RB_ROOT;
324	info->msg_tree_rightmost = NULL;
325	info->node_cache = NULL;
326	memset(&info->attr, 0, sizeof(info->attr));
327	info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
328	ipc_ns->mq_msg_default);
329	info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
330	ipc_ns->mq_msgsize_default);
331	if (attr) {
332	info->attr.mq_maxmsg = attr->mq_maxmsg;
333	info->attr.mq_msgsize = attr->mq_msgsize;
334	}
335	/*
336	* We used to allocate a static array of pointers and account
337	* the size of that array as well as one msg_msg struct per
338	* possible message into the queue size. That's no longer
339	* accurate as the queue is now an rbtree and will grow and
340	* shrink depending on usage patterns. We can, however, still
341	* account one msg_msg struct per message, but the nodes are
342	* allocated depending on priority usage, and most programs
343	* only use one, or a handful, of priorities. However, since
344	* this is pinned memory, we need to assume worst case, so
345	* that means the min(mq_maxmsg, max_priorities) * struct
346	* posix_msg_tree_node.
347	*/
348
349	ret = -EINVAL;
350	if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
351	goto out_inode;
352	if (capable(CAP_SYS_RESOURCE)) {
353	if (info->attr.mq_maxmsg > HARD_MSGMAX ||
354	info->attr.mq_msgsize > HARD_MSGSIZEMAX)
355	goto out_inode;
356	} else {
357	if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
358	info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
359	goto out_inode;
360	}
361	ret = -EOVERFLOW;
362	/* check for overflow */
363	if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
364	goto out_inode;
365	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
366	min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
367	sizeof(struct posix_msg_tree_node);
368	mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
369	if (mq_bytes + mq_treesize < mq_bytes)
370	goto out_inode;
371	mq_bytes += mq_treesize;
372	info->ucounts = get_ucounts(current_ucounts());
373	if (info->ucounts) {
374	long msgqueue;
375
376	spin_lock(lock: &mq_lock);
377	msgqueue = inc_rlimit_ucounts(ucounts: info->ucounts, type: UCOUNT_RLIMIT_MSGQUEUE, v: mq_bytes);
378	if (msgqueue == LONG_MAX || msgqueue > rlimit(RLIMIT_MSGQUEUE)) {
379	dec_rlimit_ucounts(ucounts: info->ucounts, type: UCOUNT_RLIMIT_MSGQUEUE, v: mq_bytes);
380	spin_unlock(lock: &mq_lock);
381	put_ucounts(ucounts: info->ucounts);
382	info->ucounts = NULL;
383	/* mqueue_evict_inode() releases info->messages */
384	ret = -EMFILE;
385	goto out_inode;
386	}
387	spin_unlock(lock: &mq_lock);
388	}
389	} else if (S_ISDIR(mode)) {
390	inc_nlink(inode);
391	/* Some things misbehave if size == 0 on a directory */
392	inode->i_size = 2 * DIRENT_SIZE;
393	inode->i_op = &mqueue_dir_inode_operations;
394	inode->i_fop = &simple_dir_operations;
395	}
396
397	return inode;
398	out_inode:
399	iput(inode);
400	err:
401	return ERR_PTR(error: ret);
402	}
403
404	static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
405	{
406	struct inode *inode;
407	struct ipc_namespace *ns = sb->s_fs_info;
408
409	sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
410	sb->s_blocksize = PAGE_SIZE;
411	sb->s_blocksize_bits = PAGE_SHIFT;
412	sb->s_magic = MQUEUE_MAGIC;
413	sb->s_op = &mqueue_super_ops;
414
415	inode = mqueue_get_inode(sb, ipc_ns: ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
416	if (IS_ERR(ptr: inode))
417	return PTR_ERR(ptr: inode);
418
419	sb->s_root = d_make_root(inode);
420	if (!sb->s_root)
421	return -ENOMEM;
422	return 0;
423	}
424
425	static int mqueue_get_tree(struct fs_context *fc)
426	{
427	struct mqueue_fs_context *ctx = fc->fs_private;
428
429	/*
430	* With a newly created ipc namespace, we don't need to do a search
431	* for an ipc namespace match, but we still need to set s_fs_info.
432	*/
433	if (ctx->newns) {
434	fc->s_fs_info = ctx->ipc_ns;
435	return get_tree_nodev(fc, fill_super: mqueue_fill_super);
436	}
437	return get_tree_keyed(fc, fill_super: mqueue_fill_super, key: ctx->ipc_ns);
438	}
439
440	static void mqueue_fs_context_free(struct fs_context *fc)
441	{
442	struct mqueue_fs_context *ctx = fc->fs_private;
443
444	put_ipc_ns(ns: ctx->ipc_ns);
445	kfree(objp: ctx);
446	}
447
448	static int mqueue_init_fs_context(struct fs_context *fc)
449	{
450	struct mqueue_fs_context *ctx;
451
452	ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
453	if (!ctx)
454	return -ENOMEM;
455
456	ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
457	put_user_ns(ns: fc->user_ns);
458	fc->user_ns = get_user_ns(ns: ctx->ipc_ns->user_ns);
459	fc->fs_private = ctx;
460	fc->ops = &mqueue_fs_context_ops;
461	return 0;
462	}
463
464	/*
465	* mq_init_ns() is currently the only caller of mq_create_mount().
466	* So the ns parameter is always a newly created ipc namespace.
467	*/
468	static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
469	{
470	struct mqueue_fs_context *ctx;
471	struct fs_context *fc;
472	struct vfsmount *mnt;
473
474	fc = fs_context_for_mount(fs_type: &mqueue_fs_type, SB_KERNMOUNT);
475	if (IS_ERR(ptr: fc))
476	return ERR_CAST(ptr: fc);
477
478	ctx = fc->fs_private;
479	ctx->newns = true;
480	put_ipc_ns(ns: ctx->ipc_ns);
481	ctx->ipc_ns = get_ipc_ns(ns);
482	put_user_ns(ns: fc->user_ns);
483	fc->user_ns = get_user_ns(ns: ctx->ipc_ns->user_ns);
484
485	mnt = fc_mount(fc);
486	put_fs_context(fc);
487	return mnt;
488	}
489
490	static void init_once(void *foo)
491	{
492	struct mqueue_inode_info *p = foo;
493
494	inode_init_once(&p->vfs_inode);
495	}
496
497	static struct inode *mqueue_alloc_inode(struct super_block *sb)
498	{
499	struct mqueue_inode_info *ei;
500
501	ei = alloc_inode_sb(sb, mqueue_inode_cachep, GFP_KERNEL);
502	if (!ei)
503	return NULL;
504	return &ei->vfs_inode;
505	}
506
507	static void mqueue_free_inode(struct inode *inode)
508	{
509	kmem_cache_free(s: mqueue_inode_cachep, objp: MQUEUE_I(inode));
510	}
511
512	static void mqueue_evict_inode(struct inode *inode)
513	{
514	struct mqueue_inode_info *info;
515	struct ipc_namespace *ipc_ns;
516	struct msg_msg *msg, *nmsg;
517	LIST_HEAD(tmp_msg);
518
519	clear_inode(inode);
520
521	if (S_ISDIR(inode->i_mode))
522	return;
523
524	ipc_ns = get_ns_from_inode(inode);
525	info = MQUEUE_I(inode);
526	spin_lock(lock: &info->lock);
527	while ((msg = msg_get(info)) != NULL)
528	list_add_tail(new: &msg->m_list, head: &tmp_msg);
529	kfree(objp: info->node_cache);
530	spin_unlock(lock: &info->lock);
531
532	list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
533	list_del(entry: &msg->m_list);
534	free_msg(msg);
535	}
536
537	if (info->ucounts) {
538	unsigned long mq_bytes, mq_treesize;
539
540	/* Total amount of bytes accounted for the mqueue */
541	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
542	min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
543	sizeof(struct posix_msg_tree_node);
544
545	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
546	info->attr.mq_msgsize);
547
548	spin_lock(lock: &mq_lock);
549	dec_rlimit_ucounts(ucounts: info->ucounts, type: UCOUNT_RLIMIT_MSGQUEUE, v: mq_bytes);
550	/*
551	* get_ns_from_inode() ensures that the
552	* (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
553	* to which we now hold a reference, or it is NULL.
554	* We can't put it here under mq_lock, though.
555	*/
556	if (ipc_ns)
557	ipc_ns->mq_queues_count--;
558	spin_unlock(lock: &mq_lock);
559	put_ucounts(ucounts: info->ucounts);
560	info->ucounts = NULL;
561	}
562	if (ipc_ns)
563	put_ipc_ns(ns: ipc_ns);
564	}
565
566	static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
567	{
568	struct inode *dir = dentry->d_parent->d_inode;
569	struct inode *inode;
570	struct mq_attr *attr = arg;
571	int error;
572	struct ipc_namespace *ipc_ns;
573
574	spin_lock(lock: &mq_lock);
575	ipc_ns = __get_ns_from_inode(inode: dir);
576	if (!ipc_ns) {
577	error = -EACCES;
578	goto out_unlock;
579	}
580
581	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
582	!capable(CAP_SYS_RESOURCE)) {
583	error = -ENOSPC;
584	goto out_unlock;
585	}
586	ipc_ns->mq_queues_count++;
587	spin_unlock(lock: &mq_lock);
588
589	inode = mqueue_get_inode(sb: dir->i_sb, ipc_ns, mode, attr);
590	if (IS_ERR(ptr: inode)) {
591	error = PTR_ERR(ptr: inode);
592	spin_lock(lock: &mq_lock);
593	ipc_ns->mq_queues_count--;
594	goto out_unlock;
595	}
596
597	put_ipc_ns(ns: ipc_ns);
598	dir->i_size += DIRENT_SIZE;
599	simple_inode_init_ts(inode: dir);
600
601	d_instantiate(dentry, inode);
602	dget(dentry);
603	return 0;
604	out_unlock:
605	spin_unlock(lock: &mq_lock);
606	if (ipc_ns)
607	put_ipc_ns(ns: ipc_ns);
608	return error;
609	}
610
611	static int mqueue_create(struct mnt_idmap *idmap, struct inode *dir,
612	struct dentry *dentry, umode_t mode, bool excl)
613	{
614	return mqueue_create_attr(dentry, mode, NULL);
615	}
616
617	static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
618	{
619	struct inode *inode = d_inode(dentry);
620
621	simple_inode_init_ts(inode: dir);
622	dir->i_size -= DIRENT_SIZE;
623	drop_nlink(inode);
624	dput(dentry);
625	return 0;
626	}
627
628	/*
629	* This is routine for system read from queue file.
630	* To avoid mess with doing here some sort of mq_receive we allow
631	* to read only queue size & notification info (the only values
632	* that are interesting from user point of view and aren't accessible
633	* through std routines)
634	*/
635	static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
636	size_t count, loff_t *off)
637	{
638	struct inode *inode = file_inode(f: filp);
639	struct mqueue_inode_info *info = MQUEUE_I(inode);
640	char buffer[FILENT_SIZE];
641	ssize_t ret;
642
643	spin_lock(lock: &info->lock);
644	snprintf(buf: buffer, size: sizeof(buffer),
645	fmt: "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
646	info->qsize,
647	info->notify_owner ? info->notify.sigev_notify : 0,
648	(info->notify_owner &&
649	info->notify.sigev_notify == SIGEV_SIGNAL) ?
650	info->notify.sigev_signo : 0,
651	pid_vnr(pid: info->notify_owner));
652	spin_unlock(lock: &info->lock);
653	buffer[sizeof(buffer)-1] = '\0';
654
655	ret = simple_read_from_buffer(to: u_data, count, ppos: off, from: buffer,
656	strlen(buffer));
657	if (ret <= 0)
658	return ret;
659
660	inode_set_atime_to_ts(inode, ts: inode_set_ctime_current(inode));
661	return ret;
662	}
663
664	static int mqueue_flush_file(struct file *filp, fl_owner_t id)
665	{
666	struct mqueue_inode_info *info = MQUEUE_I(inode: file_inode(f: filp));
667
668	spin_lock(lock: &info->lock);
669	if (task_tgid(current) == info->notify_owner)
670	remove_notification(info);
671
672	spin_unlock(lock: &info->lock);
673	return 0;
674	}
675
676	static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
677	{
678	struct mqueue_inode_info *info = MQUEUE_I(inode: file_inode(f: filp));
679	__poll_t retval = 0;
680
681	poll_wait(filp, wait_address: &info->wait_q, p: poll_tab);
682
683	spin_lock(lock: &info->lock);
684	if (info->attr.mq_curmsgs)
685	retval = EPOLLIN | EPOLLRDNORM;
686
687	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
688	retval |= EPOLLOUT | EPOLLWRNORM;
689	spin_unlock(lock: &info->lock);
690
691	return retval;
692	}
693
694	/* Adds current to info->e_wait_q[sr] before element with smaller prio */
695	static void wq_add(struct mqueue_inode_info *info, int sr,
696	struct ext_wait_queue *ewp)
697	{
698	struct ext_wait_queue *walk;
699
700	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
701	if (walk->task->prio <= current->prio) {
702	list_add_tail(new: &ewp->list, head: &walk->list);
703	return;
704	}
705	}
706	list_add_tail(new: &ewp->list, head: &info->e_wait_q[sr].list);
707	}
708
709	/*
710	* Puts current task to sleep. Caller must hold queue lock. After return
711	* lock isn't held.
712	* sr: SEND or RECV
713	*/
714	static int wq_sleep(struct mqueue_inode_info *info, int sr,
715	ktime_t *timeout, struct ext_wait_queue *ewp)
716	__releases(&info->lock)
717	{
718	int retval;
719	signed long time;
720
721	wq_add(info, sr, ewp);
722
723	for (;;) {
724	/* memory barrier not required, we hold info->lock */
725	__set_current_state(TASK_INTERRUPTIBLE);
726
727	spin_unlock(lock: &info->lock);
728	time = schedule_hrtimeout_range_clock(expires: timeout, delta: 0,
729	mode: HRTIMER_MODE_ABS, CLOCK_REALTIME);
730
731	if (READ_ONCE(ewp->state) == STATE_READY) {
732	/* see MQ_BARRIER for purpose/pairing */
733	smp_acquire__after_ctrl_dep();
734	retval = 0;
735	goto out;
736	}
737	spin_lock(lock: &info->lock);
738
739	/* we hold info->lock, so no memory barrier required */
740	if (READ_ONCE(ewp->state) == STATE_READY) {
741	retval = 0;
742	goto out_unlock;
743	}
744	if (signal_pending(current)) {
745	retval = -ERESTARTSYS;
746	break;
747	}
748	if (time == 0) {
749	retval = -ETIMEDOUT;
750	break;
751	}
752	}
753	list_del(entry: &ewp->list);
754	out_unlock:
755	spin_unlock(lock: &info->lock);
756	out:
757	return retval;
758	}
759
760	/*
761	* Returns waiting task that should be serviced first or NULL if none exists
762	*/
763	static struct ext_wait_queue *wq_get_first_waiter(
764	struct mqueue_inode_info *info, int sr)
765	{
766	struct list_head *ptr;
767
768	ptr = info->e_wait_q[sr].list.prev;
769	if (ptr == &info->e_wait_q[sr].list)
770	return NULL;
771	return list_entry(ptr, struct ext_wait_queue, list);
772	}
773
774
775	static inline void set_cookie(struct sk_buff *skb, char code)
776	{
777	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
778	}
779
780	/*
781	* The next function is only to split too long sys_mq_timedsend
782	*/
783	static void __do_notify(struct mqueue_inode_info *info)
784	{
785	/* notification
786	* invoked when there is registered process and there isn't process
787	* waiting synchronously for message AND state of queue changed from
788	* empty to not empty. Here we are sure that no one is waiting
789	* synchronously. */
790	if (info->notify_owner &&
791	info->attr.mq_curmsgs == 1) {
792	switch (info->notify.sigev_notify) {
793	case SIGEV_NONE:
794	break;
795	case SIGEV_SIGNAL: {
796	struct kernel_siginfo sig_i;
797	struct task_struct *task;
798
799	/* do_mq_notify() accepts sigev_signo == 0, why?? */
800	if (!info->notify.sigev_signo)
801	break;
802
803	clear_siginfo(info: &sig_i);
804	sig_i.si_signo = info->notify.sigev_signo;
805	sig_i.si_errno = 0;
806	sig_i.si_code = SI_MESGQ;
807	sig_i.si_value = info->notify.sigev_value;
808	rcu_read_lock();
809	/* map current pid/uid into info->owner's namespaces */
810	sig_i.si_pid = task_tgid_nr_ns(current,
811	ns: ns_of_pid(pid: info->notify_owner));
812	sig_i.si_uid = from_kuid_munged(to: info->notify_user_ns,
813	current_uid());
814	/*
815	* We can't use kill_pid_info(), this signal should
816	* bypass check_kill_permission(). It is from kernel
817	* but si_fromuser() can't know this.
818	* We do check the self_exec_id, to avoid sending
819	* signals to programs that don't expect them.
820	*/
821	task = pid_task(pid: info->notify_owner, PIDTYPE_TGID);
822	if (task && task->self_exec_id ==
823	info->notify_self_exec_id) {
824	do_send_sig_info(sig: info->notify.sigev_signo,
825	info: &sig_i, p: task, type: PIDTYPE_TGID);
826	}
827	rcu_read_unlock();
828	break;
829	}
830	case SIGEV_THREAD:
831	set_cookie(skb: info->notify_cookie, NOTIFY_WOKENUP);
832	netlink_sendskb(sk: info->notify_sock, skb: info->notify_cookie);
833	break;
834	}
835	/* after notification unregisters process */
836	put_pid(pid: info->notify_owner);
837	put_user_ns(ns: info->notify_user_ns);
838	info->notify_owner = NULL;
839	info->notify_user_ns = NULL;
840	}
841	wake_up(&info->wait_q);
842	}
843
844	static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
845	struct timespec64 *ts)
846	{
847	if (get_timespec64(ts, uts: u_abs_timeout))
848	return -EFAULT;
849	if (!timespec64_valid(ts))
850	return -EINVAL;
851	return 0;
852	}
853
854	static void remove_notification(struct mqueue_inode_info *info)
855	{
856	if (info->notify_owner != NULL &&
857	info->notify.sigev_notify == SIGEV_THREAD) {
858	set_cookie(skb: info->notify_cookie, NOTIFY_REMOVED);
859	netlink_sendskb(sk: info->notify_sock, skb: info->notify_cookie);
860	}
861	put_pid(pid: info->notify_owner);
862	put_user_ns(ns: info->notify_user_ns);
863	info->notify_owner = NULL;
864	info->notify_user_ns = NULL;
865	}
866
867	static int prepare_open(struct dentry *dentry, int oflag, int ro,
868	umode_t mode, struct filename *name,
869	struct mq_attr *attr)
870	{
871	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
872	MAY_READ | MAY_WRITE };
873	int acc;
874
875	if (d_really_is_negative(dentry)) {
876	if (!(oflag & O_CREAT))
877	return -ENOENT;
878	if (ro)
879	return ro;
880	audit_inode_parent_hidden(name, dentry: dentry->d_parent);
881	return vfs_mkobj(dentry, mode & ~current_umask(),
882	f: mqueue_create_attr, attr);
883	}
884	/* it already existed */
885	audit_inode(name, dentry, aflags: 0);
886	if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
887	return -EEXIST;
888	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
889	return -EINVAL;
890	acc = oflag2acc[oflag & O_ACCMODE];
891	return inode_permission(&nop_mnt_idmap, d_inode(dentry), acc);
892	}
893
894	static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
895	struct mq_attr *attr)
896	{
897	struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
898	struct dentry *root = mnt->mnt_root;
899	struct filename *name;
900	struct path path;
901	int fd, error;
902	int ro;
903
904	audit_mq_open(oflag, mode, attr);
905
906	name = getname(name: u_name);
907	if (IS_ERR(ptr: name))
908	return PTR_ERR(ptr: name);
909
910	fd = get_unused_fd_flags(O_CLOEXEC);
911	if (fd < 0)
912	goto out_putname;
913
914	ro = mnt_want_write(mnt); /* we'll drop it in any case */
915	inode_lock(inode: d_inode(dentry: root));
916	path.dentry = lookup_noperm(&QSTR(name->name), root);
917	if (IS_ERR(ptr: path.dentry)) {
918	error = PTR_ERR(ptr: path.dentry);
919	goto out_putfd;
920	}
921	path.mnt = mntget(mnt);
922	error = prepare_open(dentry: path.dentry, oflag, ro, mode, name, attr);
923	if (!error) {
924	struct file *file = dentry_open(path: &path, flags: oflag, current_cred());
925	if (!IS_ERR(ptr: file))
926	fd_install(fd, file);
927	else
928	error = PTR_ERR(ptr: file);
929	}
930	path_put(&path);
931	out_putfd:
932	if (error) {
933	put_unused_fd(fd);
934	fd = error;
935	}
936	inode_unlock(inode: d_inode(dentry: root));
937	if (!ro)
938	mnt_drop_write(mnt);
939	out_putname:
940	putname(name);
941	return fd;
942	}
943
944	SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
945	struct mq_attr __user *, u_attr)
946	{
947	struct mq_attr attr;
948	if (u_attr && copy_from_user(to: &attr, from: u_attr, n: sizeof(struct mq_attr)))
949	return -EFAULT;
950
951	return do_mq_open(u_name, oflag, mode, attr: u_attr ? &attr : NULL);
952	}
953
954	SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
955	{
956	int err;
957	struct filename *name;
958	struct dentry *dentry;
959	struct inode *inode = NULL;
960	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
961	struct vfsmount *mnt = ipc_ns->mq_mnt;
962
963	name = getname(name: u_name);
964	if (IS_ERR(ptr: name))
965	return PTR_ERR(ptr: name);
966
967	audit_inode_parent_hidden(name, dentry: mnt->mnt_root);
968	err = mnt_want_write(mnt);
969	if (err)
970	goto out_name;
971	inode_lock_nested(inode: d_inode(dentry: mnt->mnt_root), subclass: I_MUTEX_PARENT);
972	dentry = lookup_noperm(&QSTR(name->name), mnt->mnt_root);
973	if (IS_ERR(ptr: dentry)) {
974	err = PTR_ERR(ptr: dentry);
975	goto out_unlock;
976	}
977
978	inode = d_inode(dentry);
979	if (!inode) {
980	err = -ENOENT;
981	} else {
982	ihold(inode);
983	err = vfs_unlink(&nop_mnt_idmap, d_inode(dentry: dentry->d_parent),
984	dentry, NULL);
985	}
986	dput(dentry);
987
988	out_unlock:
989	inode_unlock(inode: d_inode(dentry: mnt->mnt_root));
990	iput(inode);
991	mnt_drop_write(mnt);
992	out_name:
993	putname(name);
994
995	return err;
996	}
997
998	/* Pipelined send and receive functions.
999	*
1000	* If a receiver finds no waiting message, then it registers itself in the
1001	* list of waiting receivers. A sender checks that list before adding the new
1002	* message into the message array. If there is a waiting receiver, then it
1003	* bypasses the message array and directly hands the message over to the
1004	* receiver. The receiver accepts the message and returns without grabbing the
1005	* queue spinlock:
1006	*
1007	* - Set pointer to message.
1008	* - Queue the receiver task for later wakeup (without the info->lock).
1009	* - Update its state to STATE_READY. Now the receiver can continue.
1010	* - Wake up the process after the lock is dropped. Should the process wake up
1011	* before this wakeup (due to a timeout or a signal) it will either see
1012	* STATE_READY and continue or acquire the lock to check the state again.
1013	*
1014	* The same algorithm is used for senders.
1015	*/
1016
1017	static inline void __pipelined_op(struct wake_q_head *wake_q,
1018	struct mqueue_inode_info *info,
1019	struct ext_wait_queue *this)
1020	{
1021	struct task_struct *task;
1022
1023	list_del(entry: &this->list);
1024	task = get_task_struct(t: this->task);
1025
1026	/* see MQ_BARRIER for purpose/pairing */
1027	smp_store_release(&this->state, STATE_READY);
1028	wake_q_add_safe(head: wake_q, task);
1029	}
1030
1031	/* pipelined_send() - send a message directly to the task waiting in
1032	* sys_mq_timedreceive() (without inserting message into a queue).
1033	*/
1034	static inline void pipelined_send(struct wake_q_head *wake_q,
1035	struct mqueue_inode_info *info,
1036	struct msg_msg *message,
1037	struct ext_wait_queue *receiver)
1038	{
1039	receiver->msg = message;
1040	__pipelined_op(wake_q, info, this: receiver);
1041	}
1042
1043	/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
1044	* gets its message and put to the queue (we have one free place for sure). */
1045	static inline void pipelined_receive(struct wake_q_head *wake_q,
1046	struct mqueue_inode_info *info)
1047	{
1048	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
1049
1050	if (!sender) {
1051	/* for poll */
1052	wake_up_interruptible(&info->wait_q);
1053	return;
1054	}
1055	if (msg_insert(msg: sender->msg, info))
1056	return;
1057
1058	__pipelined_op(wake_q, info, this: sender);
1059	}
1060
1061	static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
1062	size_t msg_len, unsigned int msg_prio,
1063	struct timespec64 *ts)
1064	{
1065	struct inode *inode;
1066	struct ext_wait_queue wait;
1067	struct ext_wait_queue *receiver;
1068	struct msg_msg *msg_ptr;
1069	struct mqueue_inode_info *info;
1070	ktime_t expires, *timeout = NULL;
1071	struct posix_msg_tree_node *new_leaf = NULL;
1072	int ret = 0;
1073	DEFINE_WAKE_Q(wake_q);
1074
1075	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
1076	return -EINVAL;
1077
1078	if (ts) {
1079	expires = timespec64_to_ktime(ts: *ts);
1080	timeout = &expires;
1081	}
1082
1083	audit_mq_sendrecv(mqdes, msg_len, msg_prio, abs_timeout: ts);
1084
1085	CLASS(fd, f)(fd: mqdes);
1086	if (fd_empty(f))
1087	return -EBADF;
1088
1089	inode = file_inode(fd_file(f));
1090	if (unlikely(fd_file(f)->f_op != &mqueue_file_operations))
1091	return -EBADF;
1092	info = MQUEUE_I(inode);
1093	audit_file(fd_file(f));
1094
1095	if (unlikely(!(fd_file(f)->f_mode & FMODE_WRITE)))
1096	return -EBADF;
1097
1098	if (unlikely(msg_len > info->attr.mq_msgsize))
1099	return -EMSGSIZE;
1100
1101	/* First try to allocate memory, before doing anything with
1102	* existing queues. */
1103	msg_ptr = load_msg(src: u_msg_ptr, len: msg_len);
1104	if (IS_ERR(ptr: msg_ptr))
1105	return PTR_ERR(ptr: msg_ptr);
1106	msg_ptr->m_ts = msg_len;
1107	msg_ptr->m_type = msg_prio;
1108
1109	/*
1110	* msg_insert really wants us to have a valid, spare node struct so
1111	* it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1112	* fall back to that if necessary.
1113	*/
1114	if (!info->node_cache)
1115	new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1116
1117	spin_lock(lock: &info->lock);
1118
1119	if (!info->node_cache && new_leaf) {
1120	/* Save our speculative allocation into the cache */
1121	INIT_LIST_HEAD(list: &new_leaf->msg_list);
1122	info->node_cache = new_leaf;
1123	new_leaf = NULL;
1124	} else {
1125	kfree(objp: new_leaf);
1126	}
1127
1128	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1129	if (fd_file(f)->f_flags & O_NONBLOCK) {
1130	ret = -EAGAIN;
1131	} else {
1132	wait.task = current;
1133	wait.msg = (void *) msg_ptr;
1134
1135	/* memory barrier not required, we hold info->lock */
1136	WRITE_ONCE(wait.state, STATE_NONE);
1137	ret = wq_sleep(info, SEND, timeout, ewp: &wait);
1138	/*
1139	* wq_sleep must be called with info->lock held, and
1140	* returns with the lock released
1141	*/
1142	goto out_free;
1143	}
1144	} else {
1145	receiver = wq_get_first_waiter(info, RECV);
1146	if (receiver) {
1147	pipelined_send(wake_q: &wake_q, info, message: msg_ptr, receiver);
1148	} else {
1149	/* adds message to the queue */
1150	ret = msg_insert(msg: msg_ptr, info);
1151	if (ret)
1152	goto out_unlock;
1153	__do_notify(info);
1154	}
1155	simple_inode_init_ts(inode);
1156	}
1157	out_unlock:
1158	spin_unlock(lock: &info->lock);
1159	wake_up_q(head: &wake_q);
1160	out_free:
1161	if (ret)
1162	free_msg(msg: msg_ptr);
1163	return ret;
1164	}
1165
1166	static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1167	size_t msg_len, unsigned int __user *u_msg_prio,
1168	struct timespec64 *ts)
1169	{
1170	ssize_t ret;
1171	struct msg_msg *msg_ptr;
1172	struct inode *inode;
1173	struct mqueue_inode_info *info;
1174	struct ext_wait_queue wait;
1175	ktime_t expires, *timeout = NULL;
1176	struct posix_msg_tree_node *new_leaf = NULL;
1177
1178	if (ts) {
1179	expires = timespec64_to_ktime(ts: *ts);
1180	timeout = &expires;
1181	}
1182
1183	audit_mq_sendrecv(mqdes, msg_len, msg_prio: 0, abs_timeout: ts);
1184
1185	CLASS(fd, f)(fd: mqdes);
1186	if (fd_empty(f))
1187	return -EBADF;
1188
1189	inode = file_inode(fd_file(f));
1190	if (unlikely(fd_file(f)->f_op != &mqueue_file_operations))
1191	return -EBADF;
1192	info = MQUEUE_I(inode);
1193	audit_file(fd_file(f));
1194
1195	if (unlikely(!(fd_file(f)->f_mode & FMODE_READ)))
1196	return -EBADF;
1197
1198	/* checks if buffer is big enough */
1199	if (unlikely(msg_len < info->attr.mq_msgsize))
1200	return -EMSGSIZE;
1201
1202	/*
1203	* msg_insert really wants us to have a valid, spare node struct so
1204	* it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1205	* fall back to that if necessary.
1206	*/
1207	if (!info->node_cache)
1208	new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1209
1210	spin_lock(lock: &info->lock);
1211
1212	if (!info->node_cache && new_leaf) {
1213	/* Save our speculative allocation into the cache */
1214	INIT_LIST_HEAD(list: &new_leaf->msg_list);
1215	info->node_cache = new_leaf;
1216	} else {
1217	kfree(objp: new_leaf);
1218	}
1219
1220	if (info->attr.mq_curmsgs == 0) {
1221	if (fd_file(f)->f_flags & O_NONBLOCK) {
1222	spin_unlock(lock: &info->lock);
1223	ret = -EAGAIN;
1224	} else {
1225	wait.task = current;
1226
1227	/* memory barrier not required, we hold info->lock */
1228	WRITE_ONCE(wait.state, STATE_NONE);
1229	ret = wq_sleep(info, RECV, timeout, ewp: &wait);
1230	msg_ptr = wait.msg;
1231	}
1232	} else {
1233	DEFINE_WAKE_Q(wake_q);
1234
1235	msg_ptr = msg_get(info);
1236
1237	simple_inode_init_ts(inode);
1238
1239	/* There is now free space in queue. */
1240	pipelined_receive(wake_q: &wake_q, info);
1241	spin_unlock(lock: &info->lock);
1242	wake_up_q(head: &wake_q);
1243	ret = 0;
1244	}
1245	if (ret == 0) {
1246	ret = msg_ptr->m_ts;
1247
1248	if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1249	store_msg(dest: u_msg_ptr, msg: msg_ptr, len: msg_ptr->m_ts)) {
1250	ret = -EFAULT;
1251	}
1252	free_msg(msg: msg_ptr);
1253	}
1254	return ret;
1255	}
1256
1257	SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1258	size_t, msg_len, unsigned int, msg_prio,
1259	const struct __kernel_timespec __user *, u_abs_timeout)
1260	{
1261	struct timespec64 ts, *p = NULL;
1262	if (u_abs_timeout) {
1263	int res = prepare_timeout(u_abs_timeout, ts: &ts);
1264	if (res)
1265	return res;
1266	p = &ts;
1267	}
1268	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, ts: p);
1269	}
1270
1271	SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1272	size_t, msg_len, unsigned int __user *, u_msg_prio,
1273	const struct __kernel_timespec __user *, u_abs_timeout)
1274	{
1275	struct timespec64 ts, *p = NULL;
1276	if (u_abs_timeout) {
1277	int res = prepare_timeout(u_abs_timeout, ts: &ts);
1278	if (res)
1279	return res;
1280	p = &ts;
1281	}
1282	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, ts: p);
1283	}
1284
1285	/*
1286	* Notes: the case when user wants us to deregister (with NULL as pointer)
1287	* and he isn't currently owner of notification, will be silently discarded.
1288	* It isn't explicitly defined in the POSIX.
1289	*/
1290	static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1291	{
1292	int ret;
1293	struct sock *sock;
1294	struct inode *inode;
1295	struct mqueue_inode_info *info;
1296	struct sk_buff *nc;
1297
1298	audit_mq_notify(mqdes, notification);
1299
1300	nc = NULL;
1301	sock = NULL;
1302	if (notification != NULL) {
1303	if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1304	notification->sigev_notify != SIGEV_SIGNAL &&
1305	notification->sigev_notify != SIGEV_THREAD))
1306	return -EINVAL;
1307	if (notification->sigev_notify == SIGEV_SIGNAL &&
1308	!valid_signal(sig: notification->sigev_signo)) {
1309	return -EINVAL;
1310	}
1311	if (notification->sigev_notify == SIGEV_THREAD) {
1312	long timeo;
1313
1314	/* create the notify skb */
1315	nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1316	if (!nc)
1317	return -ENOMEM;
1318
1319	if (copy_from_user(to: nc->data,
1320	from: notification->sigev_value.sival_ptr,
1321	NOTIFY_COOKIE_LEN)) {
1322	kfree_skb(skb: nc);
1323	return -EFAULT;
1324	}
1325
1326	/* TODO: add a header? */
1327	skb_put(skb: nc, NOTIFY_COOKIE_LEN);
1328	/* and attach it to the socket */
1329	retry:
1330	sock = netlink_getsockbyfd(fd: notification->sigev_signo);
1331	if (IS_ERR(ptr: sock)) {
1332	kfree_skb(skb: nc);
1333	return PTR_ERR(ptr: sock);
1334	}
1335
1336	timeo = MAX_SCHEDULE_TIMEOUT;
1337	ret = netlink_attachskb(sk: sock, skb: nc, timeo: &timeo, NULL);
1338	if (ret == 1)
1339	goto retry;
1340	if (ret)
1341	return ret;
1342	}
1343	}
1344
1345	CLASS(fd, f)(fd: mqdes);
1346	if (fd_empty(f)) {
1347	ret = -EBADF;
1348	goto out;
1349	}
1350
1351	inode = file_inode(fd_file(f));
1352	if (unlikely(fd_file(f)->f_op != &mqueue_file_operations)) {
1353	ret = -EBADF;
1354	goto out;
1355	}
1356	info = MQUEUE_I(inode);
1357
1358	ret = 0;
1359	spin_lock(lock: &info->lock);
1360	if (notification == NULL) {
1361	if (info->notify_owner == task_tgid(current)) {
1362	remove_notification(info);
1363	inode_set_atime_to_ts(inode,
1364	ts: inode_set_ctime_current(inode));
1365	}
1366	} else if (info->notify_owner != NULL) {
1367	ret = -EBUSY;
1368	} else {
1369	switch (notification->sigev_notify) {
1370	case SIGEV_NONE:
1371	info->notify.sigev_notify = SIGEV_NONE;
1372	break;
1373	case SIGEV_THREAD:
1374	info->notify_sock = sock;
1375	info->notify_cookie = nc;
1376	sock = NULL;
1377	nc = NULL;
1378	info->notify.sigev_notify = SIGEV_THREAD;
1379	break;
1380	case SIGEV_SIGNAL:
1381	info->notify.sigev_signo = notification->sigev_signo;
1382	info->notify.sigev_value = notification->sigev_value;
1383	info->notify.sigev_notify = SIGEV_SIGNAL;
1384	info->notify_self_exec_id = current->self_exec_id;
1385	break;
1386	}
1387
1388	info->notify_owner = get_pid(pid: task_tgid(current));
1389	info->notify_user_ns = get_user_ns(current_user_ns());
1390	inode_set_atime_to_ts(inode, ts: inode_set_ctime_current(inode));
1391	}
1392	spin_unlock(lock: &info->lock);
1393	out:
1394	if (sock)
1395	netlink_detachskb(sk: sock, skb: nc);
1396	return ret;
1397	}
1398
1399	SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1400	const struct sigevent __user *, u_notification)
1401	{
1402	struct sigevent n, *p = NULL;
1403	if (u_notification) {
1404	if (copy_from_user(to: &n, from: u_notification, n: sizeof(struct sigevent)))
1405	return -EFAULT;
1406	p = &n;
1407	}
1408	return do_mq_notify(mqdes, notification: p);
1409	}
1410
1411	static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1412	{
1413	struct inode *inode;
1414	struct mqueue_inode_info *info;
1415
1416	if (new && (new->mq_flags & (~O_NONBLOCK)))
1417	return -EINVAL;
1418
1419	CLASS(fd, f)(fd: mqdes);
1420	if (fd_empty(f))
1421	return -EBADF;
1422
1423	if (unlikely(fd_file(f)->f_op != &mqueue_file_operations))
1424	return -EBADF;
1425
1426	inode = file_inode(fd_file(f));
1427	info = MQUEUE_I(inode);
1428
1429	spin_lock(lock: &info->lock);
1430
1431	if (old) {
1432	*old = info->attr;
1433	old->mq_flags = fd_file(f)->f_flags & O_NONBLOCK;
1434	}
1435	if (new) {
1436	audit_mq_getsetattr(mqdes, mqstat: new);
1437	spin_lock(lock: &fd_file(f)->f_lock);
1438	if (new->mq_flags & O_NONBLOCK)
1439	fd_file(f)->f_flags |= O_NONBLOCK;
1440	else
1441	fd_file(f)->f_flags &= ~O_NONBLOCK;
1442	spin_unlock(lock: &fd_file(f)->f_lock);
1443
1444	inode_set_atime_to_ts(inode, ts: inode_set_ctime_current(inode));
1445	}
1446
1447	spin_unlock(lock: &info->lock);
1448	return 0;
1449	}
1450
1451	SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1452	const struct mq_attr __user *, u_mqstat,
1453	struct mq_attr __user *, u_omqstat)
1454	{
1455	int ret;
1456	struct mq_attr mqstat, omqstat;
1457	struct mq_attr *new = NULL, *old = NULL;
1458
1459	if (u_mqstat) {
1460	new = &mqstat;
1461	if (copy_from_user(to: new, from: u_mqstat, n: sizeof(struct mq_attr)))
1462	return -EFAULT;
1463	}
1464	if (u_omqstat)
1465	old = &omqstat;
1466
1467	ret = do_mq_getsetattr(mqdes, new, old);
1468	if (ret || !old)
1469	return ret;
1470
1471	if (copy_to_user(to: u_omqstat, from: old, n: sizeof(struct mq_attr)))
1472	return -EFAULT;
1473	return 0;
1474	}
1475
1476	#ifdef CONFIG_COMPAT
1477
1478	struct compat_mq_attr {
1479	compat_long_t mq_flags; /* message queue flags */
1480	compat_long_t mq_maxmsg; /* maximum number of messages */
1481	compat_long_t mq_msgsize; /* maximum message size */
1482	compat_long_t mq_curmsgs; /* number of messages currently queued */
1483	compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1484	};
1485
1486	static inline int get_compat_mq_attr(struct mq_attr *attr,
1487	const struct compat_mq_attr __user *uattr)
1488	{
1489	struct compat_mq_attr v;
1490
1491	if (copy_from_user(to: &v, from: uattr, n: sizeof(*uattr)))
1492	return -EFAULT;
1493
1494	memset(attr, 0, sizeof(*attr));
1495	attr->mq_flags = v.mq_flags;
1496	attr->mq_maxmsg = v.mq_maxmsg;
1497	attr->mq_msgsize = v.mq_msgsize;
1498	attr->mq_curmsgs = v.mq_curmsgs;
1499	return 0;
1500	}
1501
1502	static inline int put_compat_mq_attr(const struct mq_attr *attr,
1503	struct compat_mq_attr __user *uattr)
1504	{
1505	struct compat_mq_attr v;
1506
1507	memset(&v, 0, sizeof(v));
1508	v.mq_flags = attr->mq_flags;
1509	v.mq_maxmsg = attr->mq_maxmsg;
1510	v.mq_msgsize = attr->mq_msgsize;
1511	v.mq_curmsgs = attr->mq_curmsgs;
1512	if (copy_to_user(to: uattr, from: &v, n: sizeof(*uattr)))
1513	return -EFAULT;
1514	return 0;
1515	}
1516
1517	COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1518	int, oflag, compat_mode_t, mode,
1519	struct compat_mq_attr __user *, u_attr)
1520	{
1521	struct mq_attr attr, *p = NULL;
1522	if (u_attr && oflag & O_CREAT) {
1523	p = &attr;
1524	if (get_compat_mq_attr(attr: &attr, uattr: u_attr))
1525	return -EFAULT;
1526	}
1527	return do_mq_open(u_name, oflag, mode, attr: p);
1528	}
1529
1530	COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1531	const struct compat_sigevent __user *, u_notification)
1532	{
1533	struct sigevent n, *p = NULL;
1534	if (u_notification) {
1535	if (get_compat_sigevent(event: &n, u_event: u_notification))
1536	return -EFAULT;
1537	if (n.sigev_notify == SIGEV_THREAD)
1538	n.sigev_value.sival_ptr = compat_ptr(uptr: n.sigev_value.sival_int);
1539	p = &n;
1540	}
1541	return do_mq_notify(mqdes, notification: p);
1542	}
1543
1544	COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1545	const struct compat_mq_attr __user *, u_mqstat,
1546	struct compat_mq_attr __user *, u_omqstat)
1547	{
1548	int ret;
1549	struct mq_attr mqstat, omqstat;
1550	struct mq_attr *new = NULL, *old = NULL;
1551
1552	if (u_mqstat) {
1553	new = &mqstat;
1554	if (get_compat_mq_attr(attr: new, uattr: u_mqstat))
1555	return -EFAULT;
1556	}
1557	if (u_omqstat)
1558	old = &omqstat;
1559
1560	ret = do_mq_getsetattr(mqdes, new, old);
1561	if (ret || !old)
1562	return ret;
1563
1564	if (put_compat_mq_attr(attr: old, uattr: u_omqstat))
1565	return -EFAULT;
1566	return 0;
1567	}
1568	#endif
1569
1570	#ifdef CONFIG_COMPAT_32BIT_TIME
1571	static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1572	struct timespec64 *ts)
1573	{
1574	if (get_old_timespec32(ts, p))
1575	return -EFAULT;
1576	if (!timespec64_valid(ts))
1577	return -EINVAL;
1578	return 0;
1579	}
1580
1581	SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1582	const char __user *, u_msg_ptr,
1583	unsigned int, msg_len, unsigned int, msg_prio,
1584	const struct old_timespec32 __user *, u_abs_timeout)
1585	{
1586	struct timespec64 ts, *p = NULL;
1587	if (u_abs_timeout) {
1588	int res = compat_prepare_timeout(p: u_abs_timeout, ts: &ts);
1589	if (res)
1590	return res;
1591	p = &ts;
1592	}
1593	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, ts: p);
1594	}
1595
1596	SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1597	char __user *, u_msg_ptr,
1598	unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1599	const struct old_timespec32 __user *, u_abs_timeout)
1600	{
1601	struct timespec64 ts, *p = NULL;
1602	if (u_abs_timeout) {
1603	int res = compat_prepare_timeout(p: u_abs_timeout, ts: &ts);
1604	if (res)
1605	return res;
1606	p = &ts;
1607	}
1608	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, ts: p);
1609	}
1610	#endif
1611
1612	static const struct inode_operations mqueue_dir_inode_operations = {
1613	.lookup = simple_lookup,
1614	.create = mqueue_create,
1615	.unlink = mqueue_unlink,
1616	};
1617
1618	static const struct file_operations mqueue_file_operations = {
1619	.flush = mqueue_flush_file,
1620	.poll = mqueue_poll_file,
1621	.read = mqueue_read_file,
1622	.llseek = default_llseek,
1623	};
1624
1625	static const struct super_operations mqueue_super_ops = {
1626	.alloc_inode = mqueue_alloc_inode,
1627	.free_inode = mqueue_free_inode,
1628	.evict_inode = mqueue_evict_inode,
1629	.statfs = simple_statfs,
1630	};
1631
1632	static const struct fs_context_operations mqueue_fs_context_ops = {
1633	.free = mqueue_fs_context_free,
1634	.get_tree = mqueue_get_tree,
1635	};
1636
1637	static struct file_system_type mqueue_fs_type = {
1638	.name = "mqueue",
1639	.init_fs_context = mqueue_init_fs_context,
1640	.kill_sb = kill_litter_super,
1641	.fs_flags = FS_USERNS_MOUNT,
1642	};
1643
1644	int mq_init_ns(struct ipc_namespace *ns)
1645	{
1646	struct vfsmount *m;
1647
1648	ns->mq_queues_count = 0;
1649	ns->mq_queues_max = DFLT_QUEUESMAX;
1650	ns->mq_msg_max = DFLT_MSGMAX;
1651	ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1652	ns->mq_msg_default = DFLT_MSG;
1653	ns->mq_msgsize_default = DFLT_MSGSIZE;
1654
1655	m = mq_create_mount(ns);
1656	if (IS_ERR(ptr: m))
1657	return PTR_ERR(ptr: m);
1658	ns->mq_mnt = m;
1659	return 0;
1660	}
1661
1662	void mq_clear_sbinfo(struct ipc_namespace *ns)
1663	{
1664	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1665	}
1666
1667	static int __init init_mqueue_fs(void)
1668	{
1669	int error;
1670
1671	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1672	sizeof(struct mqueue_inode_info), 0,
1673	SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1674	if (mqueue_inode_cachep == NULL)
1675	return -ENOMEM;
1676
1677	if (!setup_mq_sysctls(&init_ipc_ns)) {
1678	pr_warn("sysctl registration failed\n");
1679	error = -ENOMEM;
1680	goto out_kmem;
1681	}
1682
1683	error = register_filesystem(&mqueue_fs_type);
1684	if (error)
1685	goto out_sysctl;
1686
1687	spin_lock_init(&mq_lock);
1688
1689	error = mq_init_ns(ns: &init_ipc_ns);
1690	if (error)
1691	goto out_filesystem;
1692
1693	return 0;
1694
1695	out_filesystem:
1696	unregister_filesystem(&mqueue_fs_type);
1697	out_sysctl:
1698	retire_mq_sysctls(ns: &init_ipc_ns);
1699	out_kmem:
1700	kmem_cache_destroy(s: mqueue_inode_cachep);
1701	return error;
1702	}
1703
1704	device_initcall(init_mqueue_fs);
1705