mqueue.c revision 935c6912
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
46struct mqueue_fs_context {
47	struct ipc_namespace	*ipc_ns;
48};
49
50#define MQUEUE_MAGIC	0x19800202
51#define DIRENT_SIZE	20
52#define FILENT_SIZE	80
53
54#define SEND		0
55#define RECV		1
56
57#define STATE_NONE	0
58#define STATE_READY	1
59
60struct posix_msg_tree_node {
61	struct rb_node		rb_node;
62	struct list_head	msg_list;
63	int			priority;
64};
65
66struct ext_wait_queue {		/* queue of sleeping tasks */
67	struct task_struct *task;
68	struct list_head list;
69	struct msg_msg *msg;	/* ptr of loaded message */
70	int state;		/* one of STATE_* values */
71};
72
73struct mqueue_inode_info {
74	spinlock_t lock;
75	struct inode vfs_inode;
76	wait_queue_head_t wait_q;
77
78	struct rb_root msg_tree;
79	struct posix_msg_tree_node *node_cache;
80	struct mq_attr attr;
81
82	struct sigevent notify;
83	struct pid *notify_owner;
84	struct user_namespace *notify_user_ns;
85	struct user_struct *user;	/* user who created, for accounting */
86	struct sock *notify_sock;
87	struct sk_buff *notify_cookie;
88
89	/* for tasks waiting for free space and messages, respectively */
90	struct ext_wait_queue e_wait_q[2];
91
92	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
93};
94
95static struct file_system_type mqueue_fs_type;
96static const struct inode_operations mqueue_dir_inode_operations;
97static const struct file_operations mqueue_file_operations;
98static const struct super_operations mqueue_super_ops;
99static const struct fs_context_operations mqueue_fs_context_ops;
100static void remove_notification(struct mqueue_inode_info *info);
101
102static struct kmem_cache *mqueue_inode_cachep;
103
104static struct ctl_table_header *mq_sysctl_table;
105
106static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
107{
108	return container_of(inode, struct mqueue_inode_info, vfs_inode);
109}
110
111/*
112 * This routine should be called with the mq_lock held.
113 */
114static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
115{
116	return get_ipc_ns(inode->i_sb->s_fs_info);
117}
118
119static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
120{
121	struct ipc_namespace *ns;
122
123	spin_lock(&mq_lock);
124	ns = __get_ns_from_inode(inode);
125	spin_unlock(&mq_lock);
126	return ns;
127}
128
129/* Auxiliary functions to manipulate messages' list */
130static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
131{
132	struct rb_node **p, *parent = NULL;
133	struct posix_msg_tree_node *leaf;
134
135	p = &info->msg_tree.rb_node;
136	while (*p) {
137		parent = *p;
138		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
139
140		if (likely(leaf->priority == msg->m_type))
141			goto insert_msg;
142		else if (msg->m_type < leaf->priority)
143			p = &(*p)->rb_left;
144		else
145			p = &(*p)->rb_right;
146	}
147	if (info->node_cache) {
148		leaf = info->node_cache;
149		info->node_cache = NULL;
150	} else {
151		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
152		if (!leaf)
153			return -ENOMEM;
154		INIT_LIST_HEAD(&leaf->msg_list);
155	}
156	leaf->priority = msg->m_type;
157	rb_link_node(&leaf->rb_node, parent, p);
158	rb_insert_color(&leaf->rb_node, &info->msg_tree);
159insert_msg:
160	info->attr.mq_curmsgs++;
161	info->qsize += msg->m_ts;
162	list_add_tail(&msg->m_list, &leaf->msg_list);
163	return 0;
164}
165
166static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
167{
168	struct rb_node **p, *parent = NULL;
169	struct posix_msg_tree_node *leaf;
170	struct msg_msg *msg;
171
172try_again:
173	p = &info->msg_tree.rb_node;
174	while (*p) {
175		parent = *p;
176		/*
177		 * During insert, low priorities go to the left and high to the
178		 * right.  On receive, we want the highest priorities first, so
179		 * walk all the way to the right.
180		 */
181		p = &(*p)->rb_right;
182	}
183	if (!parent) {
184		if (info->attr.mq_curmsgs) {
185			pr_warn_once("Inconsistency in POSIX message queue, "
186				     "no tree element, but supposedly messages "
187				     "should exist!\n");
188			info->attr.mq_curmsgs = 0;
189		}
190		return NULL;
191	}
192	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
193	if (unlikely(list_empty(&leaf->msg_list))) {
194		pr_warn_once("Inconsistency in POSIX message queue, "
195			     "empty leaf node but we haven't implemented "
196			     "lazy leaf delete!\n");
197		rb_erase(&leaf->rb_node, &info->msg_tree);
198		if (info->node_cache) {
199			kfree(leaf);
200		} else {
201			info->node_cache = leaf;
202		}
203		goto try_again;
204	} else {
205		msg = list_first_entry(&leaf->msg_list,
206				       struct msg_msg, m_list);
207		list_del(&msg->m_list);
208		if (list_empty(&leaf->msg_list)) {
209			rb_erase(&leaf->rb_node, &info->msg_tree);
210			if (info->node_cache) {
211				kfree(leaf);
212			} else {
213				info->node_cache = leaf;
214			}
215		}
216	}
217	info->attr.mq_curmsgs--;
218	info->qsize -= msg->m_ts;
219	return msg;
220}
221
222static struct inode *mqueue_get_inode(struct super_block *sb,
223		struct ipc_namespace *ipc_ns, umode_t mode,
224		struct mq_attr *attr)
225{
226	struct user_struct *u = current_user();
227	struct inode *inode;
228	int ret = -ENOMEM;
229
230	inode = new_inode(sb);
231	if (!inode)
232		goto err;
233
234	inode->i_ino = get_next_ino();
235	inode->i_mode = mode;
236	inode->i_uid = current_fsuid();
237	inode->i_gid = current_fsgid();
238	inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
239
240	if (S_ISREG(mode)) {
241		struct mqueue_inode_info *info;
242		unsigned long mq_bytes, mq_treesize;
243
244		inode->i_fop = &mqueue_file_operations;
245		inode->i_size = FILENT_SIZE;
246		/* mqueue specific info */
247		info = MQUEUE_I(inode);
248		spin_lock_init(&info->lock);
249		init_waitqueue_head(&info->wait_q);
250		INIT_LIST_HEAD(&info->e_wait_q[0].list);
251		INIT_LIST_HEAD(&info->e_wait_q[1].list);
252		info->notify_owner = NULL;
253		info->notify_user_ns = NULL;
254		info->qsize = 0;
255		info->user = NULL;	/* set when all is ok */
256		info->msg_tree = RB_ROOT;
257		info->node_cache = NULL;
258		memset(&info->attr, 0, sizeof(info->attr));
259		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
260					   ipc_ns->mq_msg_default);
261		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
262					    ipc_ns->mq_msgsize_default);
263		if (attr) {
264			info->attr.mq_maxmsg = attr->mq_maxmsg;
265			info->attr.mq_msgsize = attr->mq_msgsize;
266		}
267		/*
268		 * We used to allocate a static array of pointers and account
269		 * the size of that array as well as one msg_msg struct per
270		 * possible message into the queue size. That's no longer
271		 * accurate as the queue is now an rbtree and will grow and
272		 * shrink depending on usage patterns.  We can, however, still
273		 * account one msg_msg struct per message, but the nodes are
274		 * allocated depending on priority usage, and most programs
275		 * only use one, or a handful, of priorities.  However, since
276		 * this is pinned memory, we need to assume worst case, so
277		 * that means the min(mq_maxmsg, max_priorities) * struct
278		 * posix_msg_tree_node.
279		 */
280
281		ret = -EINVAL;
282		if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
283			goto out_inode;
284		if (capable(CAP_SYS_RESOURCE)) {
285			if (info->attr.mq_maxmsg > HARD_MSGMAX ||
286			    info->attr.mq_msgsize > HARD_MSGSIZEMAX)
287				goto out_inode;
288		} else {
289			if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
290					info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
291				goto out_inode;
292		}
293		ret = -EOVERFLOW;
294		/* check for overflow */
295		if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
296			goto out_inode;
297		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
298			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
299			sizeof(struct posix_msg_tree_node);
300		mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
301		if (mq_bytes + mq_treesize < mq_bytes)
302			goto out_inode;
303		mq_bytes += mq_treesize;
304		spin_lock(&mq_lock);
305		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
306		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
307			spin_unlock(&mq_lock);
308			/* mqueue_evict_inode() releases info->messages */
309			ret = -EMFILE;
310			goto out_inode;
311		}
312		u->mq_bytes += mq_bytes;
313		spin_unlock(&mq_lock);
314
315		/* all is ok */
316		info->user = get_uid(u);
317	} else if (S_ISDIR(mode)) {
318		inc_nlink(inode);
319		/* Some things misbehave if size == 0 on a directory */
320		inode->i_size = 2 * DIRENT_SIZE;
321		inode->i_op = &mqueue_dir_inode_operations;
322		inode->i_fop = &simple_dir_operations;
323	}
324
325	return inode;
326out_inode:
327	iput(inode);
328err:
329	return ERR_PTR(ret);
330}
331
332static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
333{
334	struct inode *inode;
335	struct ipc_namespace *ns = sb->s_fs_info;
336
337	sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
338	sb->s_blocksize = PAGE_SIZE;
339	sb->s_blocksize_bits = PAGE_SHIFT;
340	sb->s_magic = MQUEUE_MAGIC;
341	sb->s_op = &mqueue_super_ops;
342
343	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
344	if (IS_ERR(inode))
345		return PTR_ERR(inode);
346
347	sb->s_root = d_make_root(inode);
348	if (!sb->s_root)
349		return -ENOMEM;
350	return 0;
351}
352
353static int mqueue_get_tree(struct fs_context *fc)
354{
355	struct mqueue_fs_context *ctx = fc->fs_private;
356
357	put_user_ns(fc->user_ns);
358	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
359	fc->s_fs_info = ctx->ipc_ns;
360	return vfs_get_super(fc, vfs_get_keyed_super, mqueue_fill_super);
361}
362
363static void mqueue_fs_context_free(struct fs_context *fc)
364{
365	struct mqueue_fs_context *ctx = fc->fs_private;
366
367	if (ctx->ipc_ns)
368		put_ipc_ns(ctx->ipc_ns);
369	kfree(ctx);
370}
371
372static int mqueue_init_fs_context(struct fs_context *fc)
373{
374	struct mqueue_fs_context *ctx;
375
376	ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
377	if (!ctx)
378		return -ENOMEM;
379
380	ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
381	fc->fs_private = ctx;
382	fc->ops = &mqueue_fs_context_ops;
383	return 0;
384}
385
386static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
387{
388	struct mqueue_fs_context *ctx;
389	struct fs_context *fc;
390	struct vfsmount *mnt;
391
392	fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
393	if (IS_ERR(fc))
394		return ERR_CAST(fc);
395
396	ctx = fc->fs_private;
397	put_ipc_ns(ctx->ipc_ns);
398	ctx->ipc_ns = get_ipc_ns(ns);
399
400	mnt = fc_mount(fc);
401	put_fs_context(fc);
402	return mnt;
403}
404
405static void init_once(void *foo)
406{
407	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
408
409	inode_init_once(&p->vfs_inode);
410}
411
412static struct inode *mqueue_alloc_inode(struct super_block *sb)
413{
414	struct mqueue_inode_info *ei;
415
416	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
417	if (!ei)
418		return NULL;
419	return &ei->vfs_inode;
420}
421
422static void mqueue_i_callback(struct rcu_head *head)
423{
424	struct inode *inode = container_of(head, struct inode, i_rcu);
425	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
426}
427
428static void mqueue_destroy_inode(struct inode *inode)
429{
430	call_rcu(&inode->i_rcu, mqueue_i_callback);
431}
432
433static void mqueue_evict_inode(struct inode *inode)
434{
435	struct mqueue_inode_info *info;
436	struct user_struct *user;
437	unsigned long mq_bytes, mq_treesize;
438	struct ipc_namespace *ipc_ns;
439	struct msg_msg *msg;
440
441	clear_inode(inode);
442
443	if (S_ISDIR(inode->i_mode))
444		return;
445
446	ipc_ns = get_ns_from_inode(inode);
447	info = MQUEUE_I(inode);
448	spin_lock(&info->lock);
449	while ((msg = msg_get(info)) != NULL)
450		free_msg(msg);
451	kfree(info->node_cache);
452	spin_unlock(&info->lock);
453
454	/* Total amount of bytes accounted for the mqueue */
455	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
456		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
457		sizeof(struct posix_msg_tree_node);
458
459	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
460				  info->attr.mq_msgsize);
461
462	user = info->user;
463	if (user) {
464		spin_lock(&mq_lock);
465		user->mq_bytes -= mq_bytes;
466		/*
467		 * get_ns_from_inode() ensures that the
468		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
469		 * to which we now hold a reference, or it is NULL.
470		 * We can't put it here under mq_lock, though.
471		 */
472		if (ipc_ns)
473			ipc_ns->mq_queues_count--;
474		spin_unlock(&mq_lock);
475		free_uid(user);
476	}
477	if (ipc_ns)
478		put_ipc_ns(ipc_ns);
479}
480
481static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
482{
483	struct inode *dir = dentry->d_parent->d_inode;
484	struct inode *inode;
485	struct mq_attr *attr = arg;
486	int error;
487	struct ipc_namespace *ipc_ns;
488
489	spin_lock(&mq_lock);
490	ipc_ns = __get_ns_from_inode(dir);
491	if (!ipc_ns) {
492		error = -EACCES;
493		goto out_unlock;
494	}
495
496	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
497	    !capable(CAP_SYS_RESOURCE)) {
498		error = -ENOSPC;
499		goto out_unlock;
500	}
501	ipc_ns->mq_queues_count++;
502	spin_unlock(&mq_lock);
503
504	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
505	if (IS_ERR(inode)) {
506		error = PTR_ERR(inode);
507		spin_lock(&mq_lock);
508		ipc_ns->mq_queues_count--;
509		goto out_unlock;
510	}
511
512	put_ipc_ns(ipc_ns);
513	dir->i_size += DIRENT_SIZE;
514	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
515
516	d_instantiate(dentry, inode);
517	dget(dentry);
518	return 0;
519out_unlock:
520	spin_unlock(&mq_lock);
521	if (ipc_ns)
522		put_ipc_ns(ipc_ns);
523	return error;
524}
525
526static int mqueue_create(struct inode *dir, struct dentry *dentry,
527				umode_t mode, bool excl)
528{
529	return mqueue_create_attr(dentry, mode, NULL);
530}
531
532static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
533{
534	struct inode *inode = d_inode(dentry);
535
536	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
537	dir->i_size -= DIRENT_SIZE;
538	drop_nlink(inode);
539	dput(dentry);
540	return 0;
541}
542
543/*
544*	This is routine for system read from queue file.
545*	To avoid mess with doing here some sort of mq_receive we allow
546*	to read only queue size & notification info (the only values
547*	that are interesting from user point of view and aren't accessible
548*	through std routines)
549*/
550static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
551				size_t count, loff_t *off)
552{
553	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
554	char buffer[FILENT_SIZE];
555	ssize_t ret;
556
557	spin_lock(&info->lock);
558	snprintf(buffer, sizeof(buffer),
559			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
560			info->qsize,
561			info->notify_owner ? info->notify.sigev_notify : 0,
562			(info->notify_owner &&
563			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
564				info->notify.sigev_signo : 0,
565			pid_vnr(info->notify_owner));
566	spin_unlock(&info->lock);
567	buffer[sizeof(buffer)-1] = '\0';
568
569	ret = simple_read_from_buffer(u_data, count, off, buffer,
570				strlen(buffer));
571	if (ret <= 0)
572		return ret;
573
574	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
575	return ret;
576}
577
578static int mqueue_flush_file(struct file *filp, fl_owner_t id)
579{
580	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
581
582	spin_lock(&info->lock);
583	if (task_tgid(current) == info->notify_owner)
584		remove_notification(info);
585
586	spin_unlock(&info->lock);
587	return 0;
588}
589
590static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
591{
592	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
593	__poll_t retval = 0;
594
595	poll_wait(filp, &info->wait_q, poll_tab);
596
597	spin_lock(&info->lock);
598	if (info->attr.mq_curmsgs)
599		retval = EPOLLIN | EPOLLRDNORM;
600
601	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
602		retval |= EPOLLOUT | EPOLLWRNORM;
603	spin_unlock(&info->lock);
604
605	return retval;
606}
607
608/* Adds current to info->e_wait_q[sr] before element with smaller prio */
609static void wq_add(struct mqueue_inode_info *info, int sr,
610			struct ext_wait_queue *ewp)
611{
612	struct ext_wait_queue *walk;
613
614	ewp->task = current;
615
616	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
617		if (walk->task->prio <= current->prio) {
618			list_add_tail(&ewp->list, &walk->list);
619			return;
620		}
621	}
622	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
623}
624
625/*
626 * Puts current task to sleep. Caller must hold queue lock. After return
627 * lock isn't held.
628 * sr: SEND or RECV
629 */
630static int wq_sleep(struct mqueue_inode_info *info, int sr,
631		    ktime_t *timeout, struct ext_wait_queue *ewp)
632	__releases(&info->lock)
633{
634	int retval;
635	signed long time;
636
637	wq_add(info, sr, ewp);
638
639	for (;;) {
640		__set_current_state(TASK_INTERRUPTIBLE);
641
642		spin_unlock(&info->lock);
643		time = schedule_hrtimeout_range_clock(timeout, 0,
644			HRTIMER_MODE_ABS, CLOCK_REALTIME);
645
646		if (ewp->state == STATE_READY) {
647			retval = 0;
648			goto out;
649		}
650		spin_lock(&info->lock);
651		if (ewp->state == STATE_READY) {
652			retval = 0;
653			goto out_unlock;
654		}
655		if (signal_pending(current)) {
656			retval = -ERESTARTSYS;
657			break;
658		}
659		if (time == 0) {
660			retval = -ETIMEDOUT;
661			break;
662		}
663	}
664	list_del(&ewp->list);
665out_unlock:
666	spin_unlock(&info->lock);
667out:
668	return retval;
669}
670
671/*
672 * Returns waiting task that should be serviced first or NULL if none exists
673 */
674static struct ext_wait_queue *wq_get_first_waiter(
675		struct mqueue_inode_info *info, int sr)
676{
677	struct list_head *ptr;
678
679	ptr = info->e_wait_q[sr].list.prev;
680	if (ptr == &info->e_wait_q[sr].list)
681		return NULL;
682	return list_entry(ptr, struct ext_wait_queue, list);
683}
684
685
686static inline void set_cookie(struct sk_buff *skb, char code)
687{
688	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
689}
690
691/*
692 * The next function is only to split too long sys_mq_timedsend
693 */
694static void __do_notify(struct mqueue_inode_info *info)
695{
696	/* notification
697	 * invoked when there is registered process and there isn't process
698	 * waiting synchronously for message AND state of queue changed from
699	 * empty to not empty. Here we are sure that no one is waiting
700	 * synchronously. */
701	if (info->notify_owner &&
702	    info->attr.mq_curmsgs == 1) {
703		struct kernel_siginfo sig_i;
704		switch (info->notify.sigev_notify) {
705		case SIGEV_NONE:
706			break;
707		case SIGEV_SIGNAL:
708			/* sends signal */
709
710			clear_siginfo(&sig_i);
711			sig_i.si_signo = info->notify.sigev_signo;
712			sig_i.si_errno = 0;
713			sig_i.si_code = SI_MESGQ;
714			sig_i.si_value = info->notify.sigev_value;
715			/* map current pid/uid into info->owner's namespaces */
716			rcu_read_lock();
717			sig_i.si_pid = task_tgid_nr_ns(current,
718						ns_of_pid(info->notify_owner));
719			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
720			rcu_read_unlock();
721
722			kill_pid_info(info->notify.sigev_signo,
723				      &sig_i, info->notify_owner);
724			break;
725		case SIGEV_THREAD:
726			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
727			netlink_sendskb(info->notify_sock, info->notify_cookie);
728			break;
729		}
730		/* after notification unregisters process */
731		put_pid(info->notify_owner);
732		put_user_ns(info->notify_user_ns);
733		info->notify_owner = NULL;
734		info->notify_user_ns = NULL;
735	}
736	wake_up(&info->wait_q);
737}
738
739static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
740			   struct timespec64 *ts)
741{
742	if (get_timespec64(ts, u_abs_timeout))
743		return -EFAULT;
744	if (!timespec64_valid(ts))
745		return -EINVAL;
746	return 0;
747}
748
749static void remove_notification(struct mqueue_inode_info *info)
750{
751	if (info->notify_owner != NULL &&
752	    info->notify.sigev_notify == SIGEV_THREAD) {
753		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
754		netlink_sendskb(info->notify_sock, info->notify_cookie);
755	}
756	put_pid(info->notify_owner);
757	put_user_ns(info->notify_user_ns);
758	info->notify_owner = NULL;
759	info->notify_user_ns = NULL;
760}
761
762static int prepare_open(struct dentry *dentry, int oflag, int ro,
763			umode_t mode, struct filename *name,
764			struct mq_attr *attr)
765{
766	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
767						  MAY_READ | MAY_WRITE };
768	int acc;
769
770	if (d_really_is_negative(dentry)) {
771		if (!(oflag & O_CREAT))
772			return -ENOENT;
773		if (ro)
774			return ro;
775		audit_inode_parent_hidden(name, dentry->d_parent);
776		return vfs_mkobj(dentry, mode & ~current_umask(),
777				  mqueue_create_attr, attr);
778	}
779	/* it already existed */
780	audit_inode(name, dentry, 0);
781	if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
782		return -EEXIST;
783	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
784		return -EINVAL;
785	acc = oflag2acc[oflag & O_ACCMODE];
786	return inode_permission(d_inode(dentry), acc);
787}
788
789static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
790		      struct mq_attr *attr)
791{
792	struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
793	struct dentry *root = mnt->mnt_root;
794	struct filename *name;
795	struct path path;
796	int fd, error;
797	int ro;
798
799	audit_mq_open(oflag, mode, attr);
800
801	if (IS_ERR(name = getname(u_name)))
802		return PTR_ERR(name);
803
804	fd = get_unused_fd_flags(O_CLOEXEC);
805	if (fd < 0)
806		goto out_putname;
807
808	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
809	inode_lock(d_inode(root));
810	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
811	if (IS_ERR(path.dentry)) {
812		error = PTR_ERR(path.dentry);
813		goto out_putfd;
814	}
815	path.mnt = mntget(mnt);
816	error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
817	if (!error) {
818		struct file *file = dentry_open(&path, oflag, current_cred());
819		if (!IS_ERR(file))
820			fd_install(fd, file);
821		else
822			error = PTR_ERR(file);
823	}
824	path_put(&path);
825out_putfd:
826	if (error) {
827		put_unused_fd(fd);
828		fd = error;
829	}
830	inode_unlock(d_inode(root));
831	if (!ro)
832		mnt_drop_write(mnt);
833out_putname:
834	putname(name);
835	return fd;
836}
837
838SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
839		struct mq_attr __user *, u_attr)
840{
841	struct mq_attr attr;
842	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
843		return -EFAULT;
844
845	return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
846}
847
848SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
849{
850	int err;
851	struct filename *name;
852	struct dentry *dentry;
853	struct inode *inode = NULL;
854	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
855	struct vfsmount *mnt = ipc_ns->mq_mnt;
856
857	name = getname(u_name);
858	if (IS_ERR(name))
859		return PTR_ERR(name);
860
861	audit_inode_parent_hidden(name, mnt->mnt_root);
862	err = mnt_want_write(mnt);
863	if (err)
864		goto out_name;
865	inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
866	dentry = lookup_one_len(name->name, mnt->mnt_root,
867				strlen(name->name));
868	if (IS_ERR(dentry)) {
869		err = PTR_ERR(dentry);
870		goto out_unlock;
871	}
872
873	inode = d_inode(dentry);
874	if (!inode) {
875		err = -ENOENT;
876	} else {
877		ihold(inode);
878		err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
879	}
880	dput(dentry);
881
882out_unlock:
883	inode_unlock(d_inode(mnt->mnt_root));
884	if (inode)
885		iput(inode);
886	mnt_drop_write(mnt);
887out_name:
888	putname(name);
889
890	return err;
891}
892
893/* Pipelined send and receive functions.
894 *
895 * If a receiver finds no waiting message, then it registers itself in the
896 * list of waiting receivers. A sender checks that list before adding the new
897 * message into the message array. If there is a waiting receiver, then it
898 * bypasses the message array and directly hands the message over to the
899 * receiver. The receiver accepts the message and returns without grabbing the
900 * queue spinlock:
901 *
902 * - Set pointer to message.
903 * - Queue the receiver task for later wakeup (without the info->lock).
904 * - Update its state to STATE_READY. Now the receiver can continue.
905 * - Wake up the process after the lock is dropped. Should the process wake up
906 *   before this wakeup (due to a timeout or a signal) it will either see
907 *   STATE_READY and continue or acquire the lock to check the state again.
908 *
909 * The same algorithm is used for senders.
910 */
911
912/* pipelined_send() - send a message directly to the task waiting in
913 * sys_mq_timedreceive() (without inserting message into a queue).
914 */
915static inline void pipelined_send(struct wake_q_head *wake_q,
916				  struct mqueue_inode_info *info,
917				  struct msg_msg *message,
918				  struct ext_wait_queue *receiver)
919{
920	receiver->msg = message;
921	list_del(&receiver->list);
922	wake_q_add(wake_q, receiver->task);
923	/*
924	 * Rely on the implicit cmpxchg barrier from wake_q_add such
925	 * that we can ensure that updating receiver->state is the last
926	 * write operation: As once set, the receiver can continue,
927	 * and if we don't have the reference count from the wake_q,
928	 * yet, at that point we can later have a use-after-free
929	 * condition and bogus wakeup.
930	 */
931	receiver->state = STATE_READY;
932}
933
934/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
935 * gets its message and put to the queue (we have one free place for sure). */
936static inline void pipelined_receive(struct wake_q_head *wake_q,
937				     struct mqueue_inode_info *info)
938{
939	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
940
941	if (!sender) {
942		/* for poll */
943		wake_up_interruptible(&info->wait_q);
944		return;
945	}
946	if (msg_insert(sender->msg, info))
947		return;
948
949	list_del(&sender->list);
950	wake_q_add(wake_q, sender->task);
951	sender->state = STATE_READY;
952}
953
954static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
955		size_t msg_len, unsigned int msg_prio,
956		struct timespec64 *ts)
957{
958	struct fd f;
959	struct inode *inode;
960	struct ext_wait_queue wait;
961	struct ext_wait_queue *receiver;
962	struct msg_msg *msg_ptr;
963	struct mqueue_inode_info *info;
964	ktime_t expires, *timeout = NULL;
965	struct posix_msg_tree_node *new_leaf = NULL;
966	int ret = 0;
967	DEFINE_WAKE_Q(wake_q);
968
969	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
970		return -EINVAL;
971
972	if (ts) {
973		expires = timespec64_to_ktime(*ts);
974		timeout = &expires;
975	}
976
977	audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
978
979	f = fdget(mqdes);
980	if (unlikely(!f.file)) {
981		ret = -EBADF;
982		goto out;
983	}
984
985	inode = file_inode(f.file);
986	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
987		ret = -EBADF;
988		goto out_fput;
989	}
990	info = MQUEUE_I(inode);
991	audit_file(f.file);
992
993	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
994		ret = -EBADF;
995		goto out_fput;
996	}
997
998	if (unlikely(msg_len > info->attr.mq_msgsize)) {
999		ret = -EMSGSIZE;
1000		goto out_fput;
1001	}
1002
1003	/* First try to allocate memory, before doing anything with
1004	 * existing queues. */
1005	msg_ptr = load_msg(u_msg_ptr, msg_len);
1006	if (IS_ERR(msg_ptr)) {
1007		ret = PTR_ERR(msg_ptr);
1008		goto out_fput;
1009	}
1010	msg_ptr->m_ts = msg_len;
1011	msg_ptr->m_type = msg_prio;
1012
1013	/*
1014	 * msg_insert really wants us to have a valid, spare node struct so
1015	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1016	 * fall back to that if necessary.
1017	 */
1018	if (!info->node_cache)
1019		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1020
1021	spin_lock(&info->lock);
1022
1023	if (!info->node_cache && new_leaf) {
1024		/* Save our speculative allocation into the cache */
1025		INIT_LIST_HEAD(&new_leaf->msg_list);
1026		info->node_cache = new_leaf;
1027		new_leaf = NULL;
1028	} else {
1029		kfree(new_leaf);
1030	}
1031
1032	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1033		if (f.file->f_flags & O_NONBLOCK) {
1034			ret = -EAGAIN;
1035		} else {
1036			wait.task = current;
1037			wait.msg = (void *) msg_ptr;
1038			wait.state = STATE_NONE;
1039			ret = wq_sleep(info, SEND, timeout, &wait);
1040			/*
1041			 * wq_sleep must be called with info->lock held, and
1042			 * returns with the lock released
1043			 */
1044			goto out_free;
1045		}
1046	} else {
1047		receiver = wq_get_first_waiter(info, RECV);
1048		if (receiver) {
1049			pipelined_send(&wake_q, info, msg_ptr, receiver);
1050		} else {
1051			/* adds message to the queue */
1052			ret = msg_insert(msg_ptr, info);
1053			if (ret)
1054				goto out_unlock;
1055			__do_notify(info);
1056		}
1057		inode->i_atime = inode->i_mtime = inode->i_ctime =
1058				current_time(inode);
1059	}
1060out_unlock:
1061	spin_unlock(&info->lock);
1062	wake_up_q(&wake_q);
1063out_free:
1064	if (ret)
1065		free_msg(msg_ptr);
1066out_fput:
1067	fdput(f);
1068out:
1069	return ret;
1070}
1071
1072static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1073		size_t msg_len, unsigned int __user *u_msg_prio,
1074		struct timespec64 *ts)
1075{
1076	ssize_t ret;
1077	struct msg_msg *msg_ptr;
1078	struct fd f;
1079	struct inode *inode;
1080	struct mqueue_inode_info *info;
1081	struct ext_wait_queue wait;
1082	ktime_t expires, *timeout = NULL;
1083	struct posix_msg_tree_node *new_leaf = NULL;
1084
1085	if (ts) {
1086		expires = timespec64_to_ktime(*ts);
1087		timeout = &expires;
1088	}
1089
1090	audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1091
1092	f = fdget(mqdes);
1093	if (unlikely(!f.file)) {
1094		ret = -EBADF;
1095		goto out;
1096	}
1097
1098	inode = file_inode(f.file);
1099	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1100		ret = -EBADF;
1101		goto out_fput;
1102	}
1103	info = MQUEUE_I(inode);
1104	audit_file(f.file);
1105
1106	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1107		ret = -EBADF;
1108		goto out_fput;
1109	}
1110
1111	/* checks if buffer is big enough */
1112	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1113		ret = -EMSGSIZE;
1114		goto out_fput;
1115	}
1116
1117	/*
1118	 * msg_insert really wants us to have a valid, spare node struct so
1119	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1120	 * fall back to that if necessary.
1121	 */
1122	if (!info->node_cache)
1123		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1124
1125	spin_lock(&info->lock);
1126
1127	if (!info->node_cache && new_leaf) {
1128		/* Save our speculative allocation into the cache */
1129		INIT_LIST_HEAD(&new_leaf->msg_list);
1130		info->node_cache = new_leaf;
1131	} else {
1132		kfree(new_leaf);
1133	}
1134
1135	if (info->attr.mq_curmsgs == 0) {
1136		if (f.file->f_flags & O_NONBLOCK) {
1137			spin_unlock(&info->lock);
1138			ret = -EAGAIN;
1139		} else {
1140			wait.task = current;
1141			wait.state = STATE_NONE;
1142			ret = wq_sleep(info, RECV, timeout, &wait);
1143			msg_ptr = wait.msg;
1144		}
1145	} else {
1146		DEFINE_WAKE_Q(wake_q);
1147
1148		msg_ptr = msg_get(info);
1149
1150		inode->i_atime = inode->i_mtime = inode->i_ctime =
1151				current_time(inode);
1152
1153		/* There is now free space in queue. */
1154		pipelined_receive(&wake_q, info);
1155		spin_unlock(&info->lock);
1156		wake_up_q(&wake_q);
1157		ret = 0;
1158	}
1159	if (ret == 0) {
1160		ret = msg_ptr->m_ts;
1161
1162		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1163			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1164			ret = -EFAULT;
1165		}
1166		free_msg(msg_ptr);
1167	}
1168out_fput:
1169	fdput(f);
1170out:
1171	return ret;
1172}
1173
1174SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1175		size_t, msg_len, unsigned int, msg_prio,
1176		const struct __kernel_timespec __user *, u_abs_timeout)
1177{
1178	struct timespec64 ts, *p = NULL;
1179	if (u_abs_timeout) {
1180		int res = prepare_timeout(u_abs_timeout, &ts);
1181		if (res)
1182			return res;
1183		p = &ts;
1184	}
1185	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1186}
1187
1188SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1189		size_t, msg_len, unsigned int __user *, u_msg_prio,
1190		const struct __kernel_timespec __user *, u_abs_timeout)
1191{
1192	struct timespec64 ts, *p = NULL;
1193	if (u_abs_timeout) {
1194		int res = prepare_timeout(u_abs_timeout, &ts);
1195		if (res)
1196			return res;
1197		p = &ts;
1198	}
1199	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1200}
1201
1202/*
1203 * Notes: the case when user wants us to deregister (with NULL as pointer)
1204 * and he isn't currently owner of notification, will be silently discarded.
1205 * It isn't explicitly defined in the POSIX.
1206 */
1207static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1208{
1209	int ret;
1210	struct fd f;
1211	struct sock *sock;
1212	struct inode *inode;
1213	struct mqueue_inode_info *info;
1214	struct sk_buff *nc;
1215
1216	audit_mq_notify(mqdes, notification);
1217
1218	nc = NULL;
1219	sock = NULL;
1220	if (notification != NULL) {
1221		if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1222			     notification->sigev_notify != SIGEV_SIGNAL &&
1223			     notification->sigev_notify != SIGEV_THREAD))
1224			return -EINVAL;
1225		if (notification->sigev_notify == SIGEV_SIGNAL &&
1226			!valid_signal(notification->sigev_signo)) {
1227			return -EINVAL;
1228		}
1229		if (notification->sigev_notify == SIGEV_THREAD) {
1230			long timeo;
1231
1232			/* create the notify skb */
1233			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1234			if (!nc) {
1235				ret = -ENOMEM;
1236				goto out;
1237			}
1238			if (copy_from_user(nc->data,
1239					notification->sigev_value.sival_ptr,
1240					NOTIFY_COOKIE_LEN)) {
1241				ret = -EFAULT;
1242				goto out;
1243			}
1244
1245			/* TODO: add a header? */
1246			skb_put(nc, NOTIFY_COOKIE_LEN);
1247			/* and attach it to the socket */
1248retry:
1249			f = fdget(notification->sigev_signo);
1250			if (!f.file) {
1251				ret = -EBADF;
1252				goto out;
1253			}
1254			sock = netlink_getsockbyfilp(f.file);
1255			fdput(f);
1256			if (IS_ERR(sock)) {
1257				ret = PTR_ERR(sock);
1258				sock = NULL;
1259				goto out;
1260			}
1261
1262			timeo = MAX_SCHEDULE_TIMEOUT;
1263			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1264			if (ret == 1) {
1265				sock = NULL;
1266				goto retry;
1267			}
1268			if (ret) {
1269				sock = NULL;
1270				nc = NULL;
1271				goto out;
1272			}
1273		}
1274	}
1275
1276	f = fdget(mqdes);
1277	if (!f.file) {
1278		ret = -EBADF;
1279		goto out;
1280	}
1281
1282	inode = file_inode(f.file);
1283	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1284		ret = -EBADF;
1285		goto out_fput;
1286	}
1287	info = MQUEUE_I(inode);
1288
1289	ret = 0;
1290	spin_lock(&info->lock);
1291	if (notification == NULL) {
1292		if (info->notify_owner == task_tgid(current)) {
1293			remove_notification(info);
1294			inode->i_atime = inode->i_ctime = current_time(inode);
1295		}
1296	} else if (info->notify_owner != NULL) {
1297		ret = -EBUSY;
1298	} else {
1299		switch (notification->sigev_notify) {
1300		case SIGEV_NONE:
1301			info->notify.sigev_notify = SIGEV_NONE;
1302			break;
1303		case SIGEV_THREAD:
1304			info->notify_sock = sock;
1305			info->notify_cookie = nc;
1306			sock = NULL;
1307			nc = NULL;
1308			info->notify.sigev_notify = SIGEV_THREAD;
1309			break;
1310		case SIGEV_SIGNAL:
1311			info->notify.sigev_signo = notification->sigev_signo;
1312			info->notify.sigev_value = notification->sigev_value;
1313			info->notify.sigev_notify = SIGEV_SIGNAL;
1314			break;
1315		}
1316
1317		info->notify_owner = get_pid(task_tgid(current));
1318		info->notify_user_ns = get_user_ns(current_user_ns());
1319		inode->i_atime = inode->i_ctime = current_time(inode);
1320	}
1321	spin_unlock(&info->lock);
1322out_fput:
1323	fdput(f);
1324out:
1325	if (sock)
1326		netlink_detachskb(sock, nc);
1327	else if (nc)
1328		dev_kfree_skb(nc);
1329
1330	return ret;
1331}
1332
1333SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1334		const struct sigevent __user *, u_notification)
1335{
1336	struct sigevent n, *p = NULL;
1337	if (u_notification) {
1338		if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1339			return -EFAULT;
1340		p = &n;
1341	}
1342	return do_mq_notify(mqdes, p);
1343}
1344
1345static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1346{
1347	struct fd f;
1348	struct inode *inode;
1349	struct mqueue_inode_info *info;
1350
1351	if (new && (new->mq_flags & (~O_NONBLOCK)))
1352		return -EINVAL;
1353
1354	f = fdget(mqdes);
1355	if (!f.file)
1356		return -EBADF;
1357
1358	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1359		fdput(f);
1360		return -EBADF;
1361	}
1362
1363	inode = file_inode(f.file);
1364	info = MQUEUE_I(inode);
1365
1366	spin_lock(&info->lock);
1367
1368	if (old) {
1369		*old = info->attr;
1370		old->mq_flags = f.file->f_flags & O_NONBLOCK;
1371	}
1372	if (new) {
1373		audit_mq_getsetattr(mqdes, new);
1374		spin_lock(&f.file->f_lock);
1375		if (new->mq_flags & O_NONBLOCK)
1376			f.file->f_flags |= O_NONBLOCK;
1377		else
1378			f.file->f_flags &= ~O_NONBLOCK;
1379		spin_unlock(&f.file->f_lock);
1380
1381		inode->i_atime = inode->i_ctime = current_time(inode);
1382	}
1383
1384	spin_unlock(&info->lock);
1385	fdput(f);
1386	return 0;
1387}
1388
1389SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1390		const struct mq_attr __user *, u_mqstat,
1391		struct mq_attr __user *, u_omqstat)
1392{
1393	int ret;
1394	struct mq_attr mqstat, omqstat;
1395	struct mq_attr *new = NULL, *old = NULL;
1396
1397	if (u_mqstat) {
1398		new = &mqstat;
1399		if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1400			return -EFAULT;
1401	}
1402	if (u_omqstat)
1403		old = &omqstat;
1404
1405	ret = do_mq_getsetattr(mqdes, new, old);
1406	if (ret || !old)
1407		return ret;
1408
1409	if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1410		return -EFAULT;
1411	return 0;
1412}
1413
1414#ifdef CONFIG_COMPAT
1415
1416struct compat_mq_attr {
1417	compat_long_t mq_flags;      /* message queue flags		     */
1418	compat_long_t mq_maxmsg;     /* maximum number of messages	     */
1419	compat_long_t mq_msgsize;    /* maximum message size		     */
1420	compat_long_t mq_curmsgs;    /* number of messages currently queued  */
1421	compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1422};
1423
1424static inline int get_compat_mq_attr(struct mq_attr *attr,
1425			const struct compat_mq_attr __user *uattr)
1426{
1427	struct compat_mq_attr v;
1428
1429	if (copy_from_user(&v, uattr, sizeof(*uattr)))
1430		return -EFAULT;
1431
1432	memset(attr, 0, sizeof(*attr));
1433	attr->mq_flags = v.mq_flags;
1434	attr->mq_maxmsg = v.mq_maxmsg;
1435	attr->mq_msgsize = v.mq_msgsize;
1436	attr->mq_curmsgs = v.mq_curmsgs;
1437	return 0;
1438}
1439
1440static inline int put_compat_mq_attr(const struct mq_attr *attr,
1441			struct compat_mq_attr __user *uattr)
1442{
1443	struct compat_mq_attr v;
1444
1445	memset(&v, 0, sizeof(v));
1446	v.mq_flags = attr->mq_flags;
1447	v.mq_maxmsg = attr->mq_maxmsg;
1448	v.mq_msgsize = attr->mq_msgsize;
1449	v.mq_curmsgs = attr->mq_curmsgs;
1450	if (copy_to_user(uattr, &v, sizeof(*uattr)))
1451		return -EFAULT;
1452	return 0;
1453}
1454
1455COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1456		       int, oflag, compat_mode_t, mode,
1457		       struct compat_mq_attr __user *, u_attr)
1458{
1459	struct mq_attr attr, *p = NULL;
1460	if (u_attr && oflag & O_CREAT) {
1461		p = &attr;
1462		if (get_compat_mq_attr(&attr, u_attr))
1463			return -EFAULT;
1464	}
1465	return do_mq_open(u_name, oflag, mode, p);
1466}
1467
1468COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1469		       const struct compat_sigevent __user *, u_notification)
1470{
1471	struct sigevent n, *p = NULL;
1472	if (u_notification) {
1473		if (get_compat_sigevent(&n, u_notification))
1474			return -EFAULT;
1475		if (n.sigev_notify == SIGEV_THREAD)
1476			n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1477		p = &n;
1478	}
1479	return do_mq_notify(mqdes, p);
1480}
1481
1482COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1483		       const struct compat_mq_attr __user *, u_mqstat,
1484		       struct compat_mq_attr __user *, u_omqstat)
1485{
1486	int ret;
1487	struct mq_attr mqstat, omqstat;
1488	struct mq_attr *new = NULL, *old = NULL;
1489
1490	if (u_mqstat) {
1491		new = &mqstat;
1492		if (get_compat_mq_attr(new, u_mqstat))
1493			return -EFAULT;
1494	}
1495	if (u_omqstat)
1496		old = &omqstat;
1497
1498	ret = do_mq_getsetattr(mqdes, new, old);
1499	if (ret || !old)
1500		return ret;
1501
1502	if (put_compat_mq_attr(old, u_omqstat))
1503		return -EFAULT;
1504	return 0;
1505}
1506#endif
1507
1508#ifdef CONFIG_COMPAT_32BIT_TIME
1509static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1510				   struct timespec64 *ts)
1511{
1512	if (get_old_timespec32(ts, p))
1513		return -EFAULT;
1514	if (!timespec64_valid(ts))
1515		return -EINVAL;
1516	return 0;
1517}
1518
1519COMPAT_SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes,
1520		       const char __user *, u_msg_ptr,
1521		       compat_size_t, msg_len, unsigned int, msg_prio,
1522		       const struct old_timespec32 __user *, u_abs_timeout)
1523{
1524	struct timespec64 ts, *p = NULL;
1525	if (u_abs_timeout) {
1526		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1527		if (res)
1528			return res;
1529		p = &ts;
1530	}
1531	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1532}
1533
1534COMPAT_SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes,
1535		       char __user *, u_msg_ptr,
1536		       compat_size_t, msg_len, unsigned int __user *, u_msg_prio,
1537		       const struct old_timespec32 __user *, u_abs_timeout)
1538{
1539	struct timespec64 ts, *p = NULL;
1540	if (u_abs_timeout) {
1541		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1542		if (res)
1543			return res;
1544		p = &ts;
1545	}
1546	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1547}
1548#endif
1549
1550static const struct inode_operations mqueue_dir_inode_operations = {
1551	.lookup = simple_lookup,
1552	.create = mqueue_create,
1553	.unlink = mqueue_unlink,
1554};
1555
1556static const struct file_operations mqueue_file_operations = {
1557	.flush = mqueue_flush_file,
1558	.poll = mqueue_poll_file,
1559	.read = mqueue_read_file,
1560	.llseek = default_llseek,
1561};
1562
1563static const struct super_operations mqueue_super_ops = {
1564	.alloc_inode = mqueue_alloc_inode,
1565	.destroy_inode = mqueue_destroy_inode,
1566	.evict_inode = mqueue_evict_inode,
1567	.statfs = simple_statfs,
1568};
1569
1570static const struct fs_context_operations mqueue_fs_context_ops = {
1571	.free		= mqueue_fs_context_free,
1572	.get_tree	= mqueue_get_tree,
1573};
1574
1575static struct file_system_type mqueue_fs_type = {
1576	.name			= "mqueue",
1577	.init_fs_context	= mqueue_init_fs_context,
1578	.kill_sb		= kill_litter_super,
1579	.fs_flags		= FS_USERNS_MOUNT,
1580};
1581
1582int mq_init_ns(struct ipc_namespace *ns)
1583{
1584	struct vfsmount *m;
1585
1586	ns->mq_queues_count  = 0;
1587	ns->mq_queues_max    = DFLT_QUEUESMAX;
1588	ns->mq_msg_max       = DFLT_MSGMAX;
1589	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1590	ns->mq_msg_default   = DFLT_MSG;
1591	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1592
1593	m = mq_create_mount(ns);
1594	if (IS_ERR(m))
1595		return PTR_ERR(m);
1596	ns->mq_mnt = m;
1597	return 0;
1598}
1599
1600void mq_clear_sbinfo(struct ipc_namespace *ns)
1601{
1602	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1603}
1604
1605void mq_put_mnt(struct ipc_namespace *ns)
1606{
1607	kern_unmount(ns->mq_mnt);
1608}
1609
1610static int __init init_mqueue_fs(void)
1611{
1612	int error;
1613
1614	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1615				sizeof(struct mqueue_inode_info), 0,
1616				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1617	if (mqueue_inode_cachep == NULL)
1618		return -ENOMEM;
1619
1620	/* ignore failures - they are not fatal */
1621	mq_sysctl_table = mq_register_sysctl_table();
1622
1623	error = register_filesystem(&mqueue_fs_type);
1624	if (error)
1625		goto out_sysctl;
1626
1627	spin_lock_init(&mq_lock);
1628
1629	error = mq_init_ns(&init_ipc_ns);
1630	if (error)
1631		goto out_filesystem;
1632
1633	return 0;
1634
1635out_filesystem:
1636	unregister_filesystem(&mqueue_fs_type);
1637out_sysctl:
1638	if (mq_sysctl_table)
1639		unregister_sysctl_table(mq_sysctl_table);
1640	kmem_cache_destroy(mqueue_inode_cachep);
1641	return error;
1642}
1643
1644device_initcall(init_mqueue_fs);
1645