mqueue.c revision 68e34f4e
1/*
2 * POSIX message queues filesystem for Linux.
3 *
4 * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
5 *                          Michal Wronski          (michal.wronski@gmail.com)
6 *
7 * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
9 *			    Manfred Spraul	    (manfred@colorfullife.com)
10 *
11 * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
12 *
13 * This file is released under the GPL.
14 */
15
16#include <linux/capability.h>
17#include <linux/init.h>
18#include <linux/pagemap.h>
19#include <linux/file.h>
20#include <linux/mount.h>
21#include <linux/namei.h>
22#include <linux/sysctl.h>
23#include <linux/poll.h>
24#include <linux/mqueue.h>
25#include <linux/msg.h>
26#include <linux/skbuff.h>
27#include <linux/vmalloc.h>
28#include <linux/netlink.h>
29#include <linux/syscalls.h>
30#include <linux/audit.h>
31#include <linux/signal.h>
32#include <linux/mutex.h>
33#include <linux/nsproxy.h>
34#include <linux/pid.h>
35#include <linux/ipc_namespace.h>
36#include <linux/user_namespace.h>
37#include <linux/slab.h>
38#include <linux/sched/wake_q.h>
39#include <linux/sched/signal.h>
40#include <linux/sched/user.h>
41
42#include <net/sock.h>
43#include "util.h"
44
45#define MQUEUE_MAGIC	0x19800202
46#define DIRENT_SIZE	20
47#define FILENT_SIZE	80
48
49#define SEND		0
50#define RECV		1
51
52#define STATE_NONE	0
53#define STATE_READY	1
54
55struct posix_msg_tree_node {
56	struct rb_node		rb_node;
57	struct list_head	msg_list;
58	int			priority;
59};
60
61struct ext_wait_queue {		/* queue of sleeping tasks */
62	struct task_struct *task;
63	struct list_head list;
64	struct msg_msg *msg;	/* ptr of loaded message */
65	int state;		/* one of STATE_* values */
66};
67
68struct mqueue_inode_info {
69	spinlock_t lock;
70	struct inode vfs_inode;
71	wait_queue_head_t wait_q;
72
73	struct rb_root msg_tree;
74	struct posix_msg_tree_node *node_cache;
75	struct mq_attr attr;
76
77	struct sigevent notify;
78	struct pid *notify_owner;
79	struct user_namespace *notify_user_ns;
80	struct user_struct *user;	/* user who created, for accounting */
81	struct sock *notify_sock;
82	struct sk_buff *notify_cookie;
83
84	/* for tasks waiting for free space and messages, respectively */
85	struct ext_wait_queue e_wait_q[2];
86
87	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
88};
89
90static const struct inode_operations mqueue_dir_inode_operations;
91static const struct file_operations mqueue_file_operations;
92static const struct super_operations mqueue_super_ops;
93static void remove_notification(struct mqueue_inode_info *info);
94
95static struct kmem_cache *mqueue_inode_cachep;
96
97static struct ctl_table_header *mq_sysctl_table;
98
99static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
100{
101	return container_of(inode, struct mqueue_inode_info, vfs_inode);
102}
103
104/*
105 * This routine should be called with the mq_lock held.
106 */
107static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
108{
109	return get_ipc_ns(inode->i_sb->s_fs_info);
110}
111
112static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
113{
114	struct ipc_namespace *ns;
115
116	spin_lock(&mq_lock);
117	ns = __get_ns_from_inode(inode);
118	spin_unlock(&mq_lock);
119	return ns;
120}
121
122/* Auxiliary functions to manipulate messages' list */
123static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
124{
125	struct rb_node **p, *parent = NULL;
126	struct posix_msg_tree_node *leaf;
127
128	p = &info->msg_tree.rb_node;
129	while (*p) {
130		parent = *p;
131		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
132
133		if (likely(leaf->priority == msg->m_type))
134			goto insert_msg;
135		else if (msg->m_type < leaf->priority)
136			p = &(*p)->rb_left;
137		else
138			p = &(*p)->rb_right;
139	}
140	if (info->node_cache) {
141		leaf = info->node_cache;
142		info->node_cache = NULL;
143	} else {
144		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
145		if (!leaf)
146			return -ENOMEM;
147		INIT_LIST_HEAD(&leaf->msg_list);
148	}
149	leaf->priority = msg->m_type;
150	rb_link_node(&leaf->rb_node, parent, p);
151	rb_insert_color(&leaf->rb_node, &info->msg_tree);
152insert_msg:
153	info->attr.mq_curmsgs++;
154	info->qsize += msg->m_ts;
155	list_add_tail(&msg->m_list, &leaf->msg_list);
156	return 0;
157}
158
159static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
160{
161	struct rb_node **p, *parent = NULL;
162	struct posix_msg_tree_node *leaf;
163	struct msg_msg *msg;
164
165try_again:
166	p = &info->msg_tree.rb_node;
167	while (*p) {
168		parent = *p;
169		/*
170		 * During insert, low priorities go to the left and high to the
171		 * right.  On receive, we want the highest priorities first, so
172		 * walk all the way to the right.
173		 */
174		p = &(*p)->rb_right;
175	}
176	if (!parent) {
177		if (info->attr.mq_curmsgs) {
178			pr_warn_once("Inconsistency in POSIX message queue, "
179				     "no tree element, but supposedly messages "
180				     "should exist!\n");
181			info->attr.mq_curmsgs = 0;
182		}
183		return NULL;
184	}
185	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
186	if (unlikely(list_empty(&leaf->msg_list))) {
187		pr_warn_once("Inconsistency in POSIX message queue, "
188			     "empty leaf node but we haven't implemented "
189			     "lazy leaf delete!\n");
190		rb_erase(&leaf->rb_node, &info->msg_tree);
191		if (info->node_cache) {
192			kfree(leaf);
193		} else {
194			info->node_cache = leaf;
195		}
196		goto try_again;
197	} else {
198		msg = list_first_entry(&leaf->msg_list,
199				       struct msg_msg, m_list);
200		list_del(&msg->m_list);
201		if (list_empty(&leaf->msg_list)) {
202			rb_erase(&leaf->rb_node, &info->msg_tree);
203			if (info->node_cache) {
204				kfree(leaf);
205			} else {
206				info->node_cache = leaf;
207			}
208		}
209	}
210	info->attr.mq_curmsgs--;
211	info->qsize -= msg->m_ts;
212	return msg;
213}
214
215static struct inode *mqueue_get_inode(struct super_block *sb,
216		struct ipc_namespace *ipc_ns, umode_t mode,
217		struct mq_attr *attr)
218{
219	struct user_struct *u = current_user();
220	struct inode *inode;
221	int ret = -ENOMEM;
222
223	inode = new_inode(sb);
224	if (!inode)
225		goto err;
226
227	inode->i_ino = get_next_ino();
228	inode->i_mode = mode;
229	inode->i_uid = current_fsuid();
230	inode->i_gid = current_fsgid();
231	inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
232
233	if (S_ISREG(mode)) {
234		struct mqueue_inode_info *info;
235		unsigned long mq_bytes, mq_treesize;
236
237		inode->i_fop = &mqueue_file_operations;
238		inode->i_size = FILENT_SIZE;
239		/* mqueue specific info */
240		info = MQUEUE_I(inode);
241		spin_lock_init(&info->lock);
242		init_waitqueue_head(&info->wait_q);
243		INIT_LIST_HEAD(&info->e_wait_q[0].list);
244		INIT_LIST_HEAD(&info->e_wait_q[1].list);
245		info->notify_owner = NULL;
246		info->notify_user_ns = NULL;
247		info->qsize = 0;
248		info->user = NULL;	/* set when all is ok */
249		info->msg_tree = RB_ROOT;
250		info->node_cache = NULL;
251		memset(&info->attr, 0, sizeof(info->attr));
252		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
253					   ipc_ns->mq_msg_default);
254		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
255					    ipc_ns->mq_msgsize_default);
256		if (attr) {
257			info->attr.mq_maxmsg = attr->mq_maxmsg;
258			info->attr.mq_msgsize = attr->mq_msgsize;
259		}
260		/*
261		 * We used to allocate a static array of pointers and account
262		 * the size of that array as well as one msg_msg struct per
263		 * possible message into the queue size. That's no longer
264		 * accurate as the queue is now an rbtree and will grow and
265		 * shrink depending on usage patterns.  We can, however, still
266		 * account one msg_msg struct per message, but the nodes are
267		 * allocated depending on priority usage, and most programs
268		 * only use one, or a handful, of priorities.  However, since
269		 * this is pinned memory, we need to assume worst case, so
270		 * that means the min(mq_maxmsg, max_priorities) * struct
271		 * posix_msg_tree_node.
272		 */
273
274		ret = -EINVAL;
275		if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
276			goto out_inode;
277		if (capable(CAP_SYS_RESOURCE)) {
278			if (info->attr.mq_maxmsg > HARD_MSGMAX ||
279			    info->attr.mq_msgsize > HARD_MSGSIZEMAX)
280				goto out_inode;
281		} else {
282			if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
283					info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
284				goto out_inode;
285		}
286		ret = -EOVERFLOW;
287		/* check for overflow */
288		if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
289			goto out_inode;
290		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
291			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
292			sizeof(struct posix_msg_tree_node);
293		mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
294		if (mq_bytes + mq_treesize < mq_bytes)
295			goto out_inode;
296		mq_bytes += mq_treesize;
297		spin_lock(&mq_lock);
298		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
299		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
300			spin_unlock(&mq_lock);
301			/* mqueue_evict_inode() releases info->messages */
302			ret = -EMFILE;
303			goto out_inode;
304		}
305		u->mq_bytes += mq_bytes;
306		spin_unlock(&mq_lock);
307
308		/* all is ok */
309		info->user = get_uid(u);
310	} else if (S_ISDIR(mode)) {
311		inc_nlink(inode);
312		/* Some things misbehave if size == 0 on a directory */
313		inode->i_size = 2 * DIRENT_SIZE;
314		inode->i_op = &mqueue_dir_inode_operations;
315		inode->i_fop = &simple_dir_operations;
316	}
317
318	return inode;
319out_inode:
320	iput(inode);
321err:
322	return ERR_PTR(ret);
323}
324
325static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
326{
327	struct inode *inode;
328	struct ipc_namespace *ns = data;
329
330	sb->s_fs_info = ns;
331	sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
332	sb->s_blocksize = PAGE_SIZE;
333	sb->s_blocksize_bits = PAGE_SHIFT;
334	sb->s_magic = MQUEUE_MAGIC;
335	sb->s_op = &mqueue_super_ops;
336
337	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
338	if (IS_ERR(inode))
339		return PTR_ERR(inode);
340
341	sb->s_root = d_make_root(inode);
342	if (!sb->s_root)
343		return -ENOMEM;
344	return 0;
345}
346
347static struct file_system_type mqueue_fs_type;
348/*
349 * Return value is pinned only by reference in ->mq_mnt; it will
350 * live until ipcns dies.  Caller does not need to drop it.
351 */
352static struct vfsmount *mq_internal_mount(void)
353{
354	struct ipc_namespace *ns = current->nsproxy->ipc_ns;
355	struct vfsmount *m = ns->mq_mnt;
356	if (m)
357		return m;
358	m = kern_mount_data(&mqueue_fs_type, ns);
359	spin_lock(&mq_lock);
360	if (unlikely(ns->mq_mnt)) {
361		spin_unlock(&mq_lock);
362		if (!IS_ERR(m))
363			kern_unmount(m);
364		return ns->mq_mnt;
365	}
366	if (!IS_ERR(m))
367		ns->mq_mnt = m;
368	spin_unlock(&mq_lock);
369	return m;
370}
371
372static struct dentry *mqueue_mount(struct file_system_type *fs_type,
373			 int flags, const char *dev_name,
374			 void *data)
375{
376	struct vfsmount *m;
377	if (flags & SB_KERNMOUNT)
378		return mount_nodev(fs_type, flags, data, mqueue_fill_super);
379	m = mq_internal_mount();
380	if (IS_ERR(m))
381		return ERR_CAST(m);
382	atomic_inc(&m->mnt_sb->s_active);
383	down_write(&m->mnt_sb->s_umount);
384	return dget(m->mnt_root);
385}
386
387static void init_once(void *foo)
388{
389	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
390
391	inode_init_once(&p->vfs_inode);
392}
393
394static struct inode *mqueue_alloc_inode(struct super_block *sb)
395{
396	struct mqueue_inode_info *ei;
397
398	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
399	if (!ei)
400		return NULL;
401	return &ei->vfs_inode;
402}
403
404static void mqueue_i_callback(struct rcu_head *head)
405{
406	struct inode *inode = container_of(head, struct inode, i_rcu);
407	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
408}
409
410static void mqueue_destroy_inode(struct inode *inode)
411{
412	call_rcu(&inode->i_rcu, mqueue_i_callback);
413}
414
415static void mqueue_evict_inode(struct inode *inode)
416{
417	struct mqueue_inode_info *info;
418	struct user_struct *user;
419	unsigned long mq_bytes, mq_treesize;
420	struct ipc_namespace *ipc_ns;
421	struct msg_msg *msg;
422
423	clear_inode(inode);
424
425	if (S_ISDIR(inode->i_mode))
426		return;
427
428	ipc_ns = get_ns_from_inode(inode);
429	info = MQUEUE_I(inode);
430	spin_lock(&info->lock);
431	while ((msg = msg_get(info)) != NULL)
432		free_msg(msg);
433	kfree(info->node_cache);
434	spin_unlock(&info->lock);
435
436	/* Total amount of bytes accounted for the mqueue */
437	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
438		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
439		sizeof(struct posix_msg_tree_node);
440
441	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
442				  info->attr.mq_msgsize);
443
444	user = info->user;
445	if (user) {
446		spin_lock(&mq_lock);
447		user->mq_bytes -= mq_bytes;
448		/*
449		 * get_ns_from_inode() ensures that the
450		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
451		 * to which we now hold a reference, or it is NULL.
452		 * We can't put it here under mq_lock, though.
453		 */
454		if (ipc_ns)
455			ipc_ns->mq_queues_count--;
456		spin_unlock(&mq_lock);
457		free_uid(user);
458	}
459	if (ipc_ns)
460		put_ipc_ns(ipc_ns);
461}
462
463static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
464{
465	struct inode *dir = dentry->d_parent->d_inode;
466	struct inode *inode;
467	struct mq_attr *attr = arg;
468	int error;
469	struct ipc_namespace *ipc_ns;
470
471	spin_lock(&mq_lock);
472	ipc_ns = __get_ns_from_inode(dir);
473	if (!ipc_ns) {
474		error = -EACCES;
475		goto out_unlock;
476	}
477
478	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
479	    !capable(CAP_SYS_RESOURCE)) {
480		error = -ENOSPC;
481		goto out_unlock;
482	}
483	ipc_ns->mq_queues_count++;
484	spin_unlock(&mq_lock);
485
486	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
487	if (IS_ERR(inode)) {
488		error = PTR_ERR(inode);
489		spin_lock(&mq_lock);
490		ipc_ns->mq_queues_count--;
491		goto out_unlock;
492	}
493
494	put_ipc_ns(ipc_ns);
495	dir->i_size += DIRENT_SIZE;
496	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
497
498	d_instantiate(dentry, inode);
499	dget(dentry);
500	return 0;
501out_unlock:
502	spin_unlock(&mq_lock);
503	if (ipc_ns)
504		put_ipc_ns(ipc_ns);
505	return error;
506}
507
508static int mqueue_create(struct inode *dir, struct dentry *dentry,
509				umode_t mode, bool excl)
510{
511	return mqueue_create_attr(dentry, mode, NULL);
512}
513
514static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
515{
516	struct inode *inode = d_inode(dentry);
517
518	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
519	dir->i_size -= DIRENT_SIZE;
520	drop_nlink(inode);
521	dput(dentry);
522	return 0;
523}
524
525/*
526*	This is routine for system read from queue file.
527*	To avoid mess with doing here some sort of mq_receive we allow
528*	to read only queue size & notification info (the only values
529*	that are interesting from user point of view and aren't accessible
530*	through std routines)
531*/
532static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
533				size_t count, loff_t *off)
534{
535	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
536	char buffer[FILENT_SIZE];
537	ssize_t ret;
538
539	spin_lock(&info->lock);
540	snprintf(buffer, sizeof(buffer),
541			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
542			info->qsize,
543			info->notify_owner ? info->notify.sigev_notify : 0,
544			(info->notify_owner &&
545			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
546				info->notify.sigev_signo : 0,
547			pid_vnr(info->notify_owner));
548	spin_unlock(&info->lock);
549	buffer[sizeof(buffer)-1] = '\0';
550
551	ret = simple_read_from_buffer(u_data, count, off, buffer,
552				strlen(buffer));
553	if (ret <= 0)
554		return ret;
555
556	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
557	return ret;
558}
559
560static int mqueue_flush_file(struct file *filp, fl_owner_t id)
561{
562	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
563
564	spin_lock(&info->lock);
565	if (task_tgid(current) == info->notify_owner)
566		remove_notification(info);
567
568	spin_unlock(&info->lock);
569	return 0;
570}
571
572static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
573{
574	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
575	__poll_t retval = 0;
576
577	poll_wait(filp, &info->wait_q, poll_tab);
578
579	spin_lock(&info->lock);
580	if (info->attr.mq_curmsgs)
581		retval = POLLIN | POLLRDNORM;
582
583	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
584		retval |= POLLOUT | POLLWRNORM;
585	spin_unlock(&info->lock);
586
587	return retval;
588}
589
590/* Adds current to info->e_wait_q[sr] before element with smaller prio */
591static void wq_add(struct mqueue_inode_info *info, int sr,
592			struct ext_wait_queue *ewp)
593{
594	struct ext_wait_queue *walk;
595
596	ewp->task = current;
597
598	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
599		if (walk->task->prio <= current->prio) {
600			list_add_tail(&ewp->list, &walk->list);
601			return;
602		}
603	}
604	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
605}
606
607/*
608 * Puts current task to sleep. Caller must hold queue lock. After return
609 * lock isn't held.
610 * sr: SEND or RECV
611 */
612static int wq_sleep(struct mqueue_inode_info *info, int sr,
613		    ktime_t *timeout, struct ext_wait_queue *ewp)
614	__releases(&info->lock)
615{
616	int retval;
617	signed long time;
618
619	wq_add(info, sr, ewp);
620
621	for (;;) {
622		__set_current_state(TASK_INTERRUPTIBLE);
623
624		spin_unlock(&info->lock);
625		time = schedule_hrtimeout_range_clock(timeout, 0,
626			HRTIMER_MODE_ABS, CLOCK_REALTIME);
627
628		if (ewp->state == STATE_READY) {
629			retval = 0;
630			goto out;
631		}
632		spin_lock(&info->lock);
633		if (ewp->state == STATE_READY) {
634			retval = 0;
635			goto out_unlock;
636		}
637		if (signal_pending(current)) {
638			retval = -ERESTARTSYS;
639			break;
640		}
641		if (time == 0) {
642			retval = -ETIMEDOUT;
643			break;
644		}
645	}
646	list_del(&ewp->list);
647out_unlock:
648	spin_unlock(&info->lock);
649out:
650	return retval;
651}
652
653/*
654 * Returns waiting task that should be serviced first or NULL if none exists
655 */
656static struct ext_wait_queue *wq_get_first_waiter(
657		struct mqueue_inode_info *info, int sr)
658{
659	struct list_head *ptr;
660
661	ptr = info->e_wait_q[sr].list.prev;
662	if (ptr == &info->e_wait_q[sr].list)
663		return NULL;
664	return list_entry(ptr, struct ext_wait_queue, list);
665}
666
667
668static inline void set_cookie(struct sk_buff *skb, char code)
669{
670	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
671}
672
673/*
674 * The next function is only to split too long sys_mq_timedsend
675 */
676static void __do_notify(struct mqueue_inode_info *info)
677{
678	/* notification
679	 * invoked when there is registered process and there isn't process
680	 * waiting synchronously for message AND state of queue changed from
681	 * empty to not empty. Here we are sure that no one is waiting
682	 * synchronously. */
683	if (info->notify_owner &&
684	    info->attr.mq_curmsgs == 1) {
685		struct siginfo sig_i;
686		switch (info->notify.sigev_notify) {
687		case SIGEV_NONE:
688			break;
689		case SIGEV_SIGNAL:
690			/* sends signal */
691
692			clear_siginfo(&sig_i);
693			sig_i.si_signo = info->notify.sigev_signo;
694			sig_i.si_errno = 0;
695			sig_i.si_code = SI_MESGQ;
696			sig_i.si_value = info->notify.sigev_value;
697			/* map current pid/uid into info->owner's namespaces */
698			rcu_read_lock();
699			sig_i.si_pid = task_tgid_nr_ns(current,
700						ns_of_pid(info->notify_owner));
701			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
702			rcu_read_unlock();
703
704			kill_pid_info(info->notify.sigev_signo,
705				      &sig_i, info->notify_owner);
706			break;
707		case SIGEV_THREAD:
708			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
709			netlink_sendskb(info->notify_sock, info->notify_cookie);
710			break;
711		}
712		/* after notification unregisters process */
713		put_pid(info->notify_owner);
714		put_user_ns(info->notify_user_ns);
715		info->notify_owner = NULL;
716		info->notify_user_ns = NULL;
717	}
718	wake_up(&info->wait_q);
719}
720
721static int prepare_timeout(const struct timespec __user *u_abs_timeout,
722			   struct timespec64 *ts)
723{
724	if (get_timespec64(ts, u_abs_timeout))
725		return -EFAULT;
726	if (!timespec64_valid(ts))
727		return -EINVAL;
728	return 0;
729}
730
731static void remove_notification(struct mqueue_inode_info *info)
732{
733	if (info->notify_owner != NULL &&
734	    info->notify.sigev_notify == SIGEV_THREAD) {
735		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
736		netlink_sendskb(info->notify_sock, info->notify_cookie);
737	}
738	put_pid(info->notify_owner);
739	put_user_ns(info->notify_user_ns);
740	info->notify_owner = NULL;
741	info->notify_user_ns = NULL;
742}
743
744static int prepare_open(struct dentry *dentry, int oflag, int ro,
745			umode_t mode, struct filename *name,
746			struct mq_attr *attr)
747{
748	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
749						  MAY_READ | MAY_WRITE };
750	int acc;
751
752	if (d_really_is_negative(dentry)) {
753		if (!(oflag & O_CREAT))
754			return -ENOENT;
755		if (ro)
756			return ro;
757		audit_inode_parent_hidden(name, dentry->d_parent);
758		return vfs_mkobj(dentry, mode & ~current_umask(),
759				  mqueue_create_attr, attr);
760	}
761	/* it already existed */
762	audit_inode(name, dentry, 0);
763	if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
764		return -EEXIST;
765	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
766		return -EINVAL;
767	acc = oflag2acc[oflag & O_ACCMODE];
768	return inode_permission(d_inode(dentry), acc);
769}
770
771static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
772		      struct mq_attr *attr)
773{
774	struct vfsmount *mnt = mq_internal_mount();
775	struct dentry *root;
776	struct filename *name;
777	struct path path;
778	int fd, error;
779	int ro;
780
781	if (IS_ERR(mnt))
782		return PTR_ERR(mnt);
783
784	audit_mq_open(oflag, mode, attr);
785
786	if (IS_ERR(name = getname(u_name)))
787		return PTR_ERR(name);
788
789	fd = get_unused_fd_flags(O_CLOEXEC);
790	if (fd < 0)
791		goto out_putname;
792
793	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
794	root = mnt->mnt_root;
795	inode_lock(d_inode(root));
796	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
797	if (IS_ERR(path.dentry)) {
798		error = PTR_ERR(path.dentry);
799		goto out_putfd;
800	}
801	path.mnt = mntget(mnt);
802	error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
803	if (!error) {
804		struct file *file = dentry_open(&path, oflag, current_cred());
805		if (!IS_ERR(file))
806			fd_install(fd, file);
807		else
808			error = PTR_ERR(file);
809	}
810	path_put(&path);
811out_putfd:
812	if (error) {
813		put_unused_fd(fd);
814		fd = error;
815	}
816	inode_unlock(d_inode(root));
817	if (!ro)
818		mnt_drop_write(mnt);
819out_putname:
820	putname(name);
821	return fd;
822}
823
824SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
825		struct mq_attr __user *, u_attr)
826{
827	struct mq_attr attr;
828	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
829		return -EFAULT;
830
831	return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
832}
833
834SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
835{
836	int err;
837	struct filename *name;
838	struct dentry *dentry;
839	struct inode *inode = NULL;
840	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
841	struct vfsmount *mnt = ipc_ns->mq_mnt;
842
843	if (!mnt)
844		return -ENOENT;
845
846	name = getname(u_name);
847	if (IS_ERR(name))
848		return PTR_ERR(name);
849
850	audit_inode_parent_hidden(name, mnt->mnt_root);
851	err = mnt_want_write(mnt);
852	if (err)
853		goto out_name;
854	inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
855	dentry = lookup_one_len(name->name, mnt->mnt_root,
856				strlen(name->name));
857	if (IS_ERR(dentry)) {
858		err = PTR_ERR(dentry);
859		goto out_unlock;
860	}
861
862	inode = d_inode(dentry);
863	if (!inode) {
864		err = -ENOENT;
865	} else {
866		ihold(inode);
867		err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
868	}
869	dput(dentry);
870
871out_unlock:
872	inode_unlock(d_inode(mnt->mnt_root));
873	if (inode)
874		iput(inode);
875	mnt_drop_write(mnt);
876out_name:
877	putname(name);
878
879	return err;
880}
881
882/* Pipelined send and receive functions.
883 *
884 * If a receiver finds no waiting message, then it registers itself in the
885 * list of waiting receivers. A sender checks that list before adding the new
886 * message into the message array. If there is a waiting receiver, then it
887 * bypasses the message array and directly hands the message over to the
888 * receiver. The receiver accepts the message and returns without grabbing the
889 * queue spinlock:
890 *
891 * - Set pointer to message.
892 * - Queue the receiver task for later wakeup (without the info->lock).
893 * - Update its state to STATE_READY. Now the receiver can continue.
894 * - Wake up the process after the lock is dropped. Should the process wake up
895 *   before this wakeup (due to a timeout or a signal) it will either see
896 *   STATE_READY and continue or acquire the lock to check the state again.
897 *
898 * The same algorithm is used for senders.
899 */
900
901/* pipelined_send() - send a message directly to the task waiting in
902 * sys_mq_timedreceive() (without inserting message into a queue).
903 */
904static inline void pipelined_send(struct wake_q_head *wake_q,
905				  struct mqueue_inode_info *info,
906				  struct msg_msg *message,
907				  struct ext_wait_queue *receiver)
908{
909	receiver->msg = message;
910	list_del(&receiver->list);
911	wake_q_add(wake_q, receiver->task);
912	/*
913	 * Rely on the implicit cmpxchg barrier from wake_q_add such
914	 * that we can ensure that updating receiver->state is the last
915	 * write operation: As once set, the receiver can continue,
916	 * and if we don't have the reference count from the wake_q,
917	 * yet, at that point we can later have a use-after-free
918	 * condition and bogus wakeup.
919	 */
920	receiver->state = STATE_READY;
921}
922
923/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
924 * gets its message and put to the queue (we have one free place for sure). */
925static inline void pipelined_receive(struct wake_q_head *wake_q,
926				     struct mqueue_inode_info *info)
927{
928	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
929
930	if (!sender) {
931		/* for poll */
932		wake_up_interruptible(&info->wait_q);
933		return;
934	}
935	if (msg_insert(sender->msg, info))
936		return;
937
938	list_del(&sender->list);
939	wake_q_add(wake_q, sender->task);
940	sender->state = STATE_READY;
941}
942
943static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
944		size_t msg_len, unsigned int msg_prio,
945		struct timespec64 *ts)
946{
947	struct fd f;
948	struct inode *inode;
949	struct ext_wait_queue wait;
950	struct ext_wait_queue *receiver;
951	struct msg_msg *msg_ptr;
952	struct mqueue_inode_info *info;
953	ktime_t expires, *timeout = NULL;
954	struct posix_msg_tree_node *new_leaf = NULL;
955	int ret = 0;
956	DEFINE_WAKE_Q(wake_q);
957
958	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
959		return -EINVAL;
960
961	if (ts) {
962		expires = timespec64_to_ktime(*ts);
963		timeout = &expires;
964	}
965
966	audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
967
968	f = fdget(mqdes);
969	if (unlikely(!f.file)) {
970		ret = -EBADF;
971		goto out;
972	}
973
974	inode = file_inode(f.file);
975	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
976		ret = -EBADF;
977		goto out_fput;
978	}
979	info = MQUEUE_I(inode);
980	audit_file(f.file);
981
982	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
983		ret = -EBADF;
984		goto out_fput;
985	}
986
987	if (unlikely(msg_len > info->attr.mq_msgsize)) {
988		ret = -EMSGSIZE;
989		goto out_fput;
990	}
991
992	/* First try to allocate memory, before doing anything with
993	 * existing queues. */
994	msg_ptr = load_msg(u_msg_ptr, msg_len);
995	if (IS_ERR(msg_ptr)) {
996		ret = PTR_ERR(msg_ptr);
997		goto out_fput;
998	}
999	msg_ptr->m_ts = msg_len;
1000	msg_ptr->m_type = msg_prio;
1001
1002	/*
1003	 * msg_insert really wants us to have a valid, spare node struct so
1004	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1005	 * fall back to that if necessary.
1006	 */
1007	if (!info->node_cache)
1008		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1009
1010	spin_lock(&info->lock);
1011
1012	if (!info->node_cache && new_leaf) {
1013		/* Save our speculative allocation into the cache */
1014		INIT_LIST_HEAD(&new_leaf->msg_list);
1015		info->node_cache = new_leaf;
1016		new_leaf = NULL;
1017	} else {
1018		kfree(new_leaf);
1019	}
1020
1021	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1022		if (f.file->f_flags & O_NONBLOCK) {
1023			ret = -EAGAIN;
1024		} else {
1025			wait.task = current;
1026			wait.msg = (void *) msg_ptr;
1027			wait.state = STATE_NONE;
1028			ret = wq_sleep(info, SEND, timeout, &wait);
1029			/*
1030			 * wq_sleep must be called with info->lock held, and
1031			 * returns with the lock released
1032			 */
1033			goto out_free;
1034		}
1035	} else {
1036		receiver = wq_get_first_waiter(info, RECV);
1037		if (receiver) {
1038			pipelined_send(&wake_q, info, msg_ptr, receiver);
1039		} else {
1040			/* adds message to the queue */
1041			ret = msg_insert(msg_ptr, info);
1042			if (ret)
1043				goto out_unlock;
1044			__do_notify(info);
1045		}
1046		inode->i_atime = inode->i_mtime = inode->i_ctime =
1047				current_time(inode);
1048	}
1049out_unlock:
1050	spin_unlock(&info->lock);
1051	wake_up_q(&wake_q);
1052out_free:
1053	if (ret)
1054		free_msg(msg_ptr);
1055out_fput:
1056	fdput(f);
1057out:
1058	return ret;
1059}
1060
1061static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1062		size_t msg_len, unsigned int __user *u_msg_prio,
1063		struct timespec64 *ts)
1064{
1065	ssize_t ret;
1066	struct msg_msg *msg_ptr;
1067	struct fd f;
1068	struct inode *inode;
1069	struct mqueue_inode_info *info;
1070	struct ext_wait_queue wait;
1071	ktime_t expires, *timeout = NULL;
1072	struct posix_msg_tree_node *new_leaf = NULL;
1073
1074	if (ts) {
1075		expires = timespec64_to_ktime(*ts);
1076		timeout = &expires;
1077	}
1078
1079	audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1080
1081	f = fdget(mqdes);
1082	if (unlikely(!f.file)) {
1083		ret = -EBADF;
1084		goto out;
1085	}
1086
1087	inode = file_inode(f.file);
1088	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1089		ret = -EBADF;
1090		goto out_fput;
1091	}
1092	info = MQUEUE_I(inode);
1093	audit_file(f.file);
1094
1095	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1096		ret = -EBADF;
1097		goto out_fput;
1098	}
1099
1100	/* checks if buffer is big enough */
1101	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1102		ret = -EMSGSIZE;
1103		goto out_fput;
1104	}
1105
1106	/*
1107	 * msg_insert really wants us to have a valid, spare node struct so
1108	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1109	 * fall back to that if necessary.
1110	 */
1111	if (!info->node_cache)
1112		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1113
1114	spin_lock(&info->lock);
1115
1116	if (!info->node_cache && new_leaf) {
1117		/* Save our speculative allocation into the cache */
1118		INIT_LIST_HEAD(&new_leaf->msg_list);
1119		info->node_cache = new_leaf;
1120	} else {
1121		kfree(new_leaf);
1122	}
1123
1124	if (info->attr.mq_curmsgs == 0) {
1125		if (f.file->f_flags & O_NONBLOCK) {
1126			spin_unlock(&info->lock);
1127			ret = -EAGAIN;
1128		} else {
1129			wait.task = current;
1130			wait.state = STATE_NONE;
1131			ret = wq_sleep(info, RECV, timeout, &wait);
1132			msg_ptr = wait.msg;
1133		}
1134	} else {
1135		DEFINE_WAKE_Q(wake_q);
1136
1137		msg_ptr = msg_get(info);
1138
1139		inode->i_atime = inode->i_mtime = inode->i_ctime =
1140				current_time(inode);
1141
1142		/* There is now free space in queue. */
1143		pipelined_receive(&wake_q, info);
1144		spin_unlock(&info->lock);
1145		wake_up_q(&wake_q);
1146		ret = 0;
1147	}
1148	if (ret == 0) {
1149		ret = msg_ptr->m_ts;
1150
1151		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1152			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1153			ret = -EFAULT;
1154		}
1155		free_msg(msg_ptr);
1156	}
1157out_fput:
1158	fdput(f);
1159out:
1160	return ret;
1161}
1162
1163SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1164		size_t, msg_len, unsigned int, msg_prio,
1165		const struct timespec __user *, u_abs_timeout)
1166{
1167	struct timespec64 ts, *p = NULL;
1168	if (u_abs_timeout) {
1169		int res = prepare_timeout(u_abs_timeout, &ts);
1170		if (res)
1171			return res;
1172		p = &ts;
1173	}
1174	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1175}
1176
1177SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1178		size_t, msg_len, unsigned int __user *, u_msg_prio,
1179		const struct timespec __user *, u_abs_timeout)
1180{
1181	struct timespec64 ts, *p = NULL;
1182	if (u_abs_timeout) {
1183		int res = prepare_timeout(u_abs_timeout, &ts);
1184		if (res)
1185			return res;
1186		p = &ts;
1187	}
1188	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1189}
1190
1191/*
1192 * Notes: the case when user wants us to deregister (with NULL as pointer)
1193 * and he isn't currently owner of notification, will be silently discarded.
1194 * It isn't explicitly defined in the POSIX.
1195 */
1196static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1197{
1198	int ret;
1199	struct fd f;
1200	struct sock *sock;
1201	struct inode *inode;
1202	struct mqueue_inode_info *info;
1203	struct sk_buff *nc;
1204
1205	audit_mq_notify(mqdes, notification);
1206
1207	nc = NULL;
1208	sock = NULL;
1209	if (notification != NULL) {
1210		if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1211			     notification->sigev_notify != SIGEV_SIGNAL &&
1212			     notification->sigev_notify != SIGEV_THREAD))
1213			return -EINVAL;
1214		if (notification->sigev_notify == SIGEV_SIGNAL &&
1215			!valid_signal(notification->sigev_signo)) {
1216			return -EINVAL;
1217		}
1218		if (notification->sigev_notify == SIGEV_THREAD) {
1219			long timeo;
1220
1221			/* create the notify skb */
1222			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1223			if (!nc) {
1224				ret = -ENOMEM;
1225				goto out;
1226			}
1227			if (copy_from_user(nc->data,
1228					notification->sigev_value.sival_ptr,
1229					NOTIFY_COOKIE_LEN)) {
1230				ret = -EFAULT;
1231				goto out;
1232			}
1233
1234			/* TODO: add a header? */
1235			skb_put(nc, NOTIFY_COOKIE_LEN);
1236			/* and attach it to the socket */
1237retry:
1238			f = fdget(notification->sigev_signo);
1239			if (!f.file) {
1240				ret = -EBADF;
1241				goto out;
1242			}
1243			sock = netlink_getsockbyfilp(f.file);
1244			fdput(f);
1245			if (IS_ERR(sock)) {
1246				ret = PTR_ERR(sock);
1247				sock = NULL;
1248				goto out;
1249			}
1250
1251			timeo = MAX_SCHEDULE_TIMEOUT;
1252			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1253			if (ret == 1) {
1254				sock = NULL;
1255				goto retry;
1256			}
1257			if (ret) {
1258				sock = NULL;
1259				nc = NULL;
1260				goto out;
1261			}
1262		}
1263	}
1264
1265	f = fdget(mqdes);
1266	if (!f.file) {
1267		ret = -EBADF;
1268		goto out;
1269	}
1270
1271	inode = file_inode(f.file);
1272	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1273		ret = -EBADF;
1274		goto out_fput;
1275	}
1276	info = MQUEUE_I(inode);
1277
1278	ret = 0;
1279	spin_lock(&info->lock);
1280	if (notification == NULL) {
1281		if (info->notify_owner == task_tgid(current)) {
1282			remove_notification(info);
1283			inode->i_atime = inode->i_ctime = current_time(inode);
1284		}
1285	} else if (info->notify_owner != NULL) {
1286		ret = -EBUSY;
1287	} else {
1288		switch (notification->sigev_notify) {
1289		case SIGEV_NONE:
1290			info->notify.sigev_notify = SIGEV_NONE;
1291			break;
1292		case SIGEV_THREAD:
1293			info->notify_sock = sock;
1294			info->notify_cookie = nc;
1295			sock = NULL;
1296			nc = NULL;
1297			info->notify.sigev_notify = SIGEV_THREAD;
1298			break;
1299		case SIGEV_SIGNAL:
1300			info->notify.sigev_signo = notification->sigev_signo;
1301			info->notify.sigev_value = notification->sigev_value;
1302			info->notify.sigev_notify = SIGEV_SIGNAL;
1303			break;
1304		}
1305
1306		info->notify_owner = get_pid(task_tgid(current));
1307		info->notify_user_ns = get_user_ns(current_user_ns());
1308		inode->i_atime = inode->i_ctime = current_time(inode);
1309	}
1310	spin_unlock(&info->lock);
1311out_fput:
1312	fdput(f);
1313out:
1314	if (sock)
1315		netlink_detachskb(sock, nc);
1316	else if (nc)
1317		dev_kfree_skb(nc);
1318
1319	return ret;
1320}
1321
1322SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1323		const struct sigevent __user *, u_notification)
1324{
1325	struct sigevent n, *p = NULL;
1326	if (u_notification) {
1327		if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1328			return -EFAULT;
1329		p = &n;
1330	}
1331	return do_mq_notify(mqdes, p);
1332}
1333
1334static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1335{
1336	struct fd f;
1337	struct inode *inode;
1338	struct mqueue_inode_info *info;
1339
1340	if (new && (new->mq_flags & (~O_NONBLOCK)))
1341		return -EINVAL;
1342
1343	f = fdget(mqdes);
1344	if (!f.file)
1345		return -EBADF;
1346
1347	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1348		fdput(f);
1349		return -EBADF;
1350	}
1351
1352	inode = file_inode(f.file);
1353	info = MQUEUE_I(inode);
1354
1355	spin_lock(&info->lock);
1356
1357	if (old) {
1358		*old = info->attr;
1359		old->mq_flags = f.file->f_flags & O_NONBLOCK;
1360	}
1361	if (new) {
1362		audit_mq_getsetattr(mqdes, new);
1363		spin_lock(&f.file->f_lock);
1364		if (new->mq_flags & O_NONBLOCK)
1365			f.file->f_flags |= O_NONBLOCK;
1366		else
1367			f.file->f_flags &= ~O_NONBLOCK;
1368		spin_unlock(&f.file->f_lock);
1369
1370		inode->i_atime = inode->i_ctime = current_time(inode);
1371	}
1372
1373	spin_unlock(&info->lock);
1374	fdput(f);
1375	return 0;
1376}
1377
1378SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1379		const struct mq_attr __user *, u_mqstat,
1380		struct mq_attr __user *, u_omqstat)
1381{
1382	int ret;
1383	struct mq_attr mqstat, omqstat;
1384	struct mq_attr *new = NULL, *old = NULL;
1385
1386	if (u_mqstat) {
1387		new = &mqstat;
1388		if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1389			return -EFAULT;
1390	}
1391	if (u_omqstat)
1392		old = &omqstat;
1393
1394	ret = do_mq_getsetattr(mqdes, new, old);
1395	if (ret || !old)
1396		return ret;
1397
1398	if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1399		return -EFAULT;
1400	return 0;
1401}
1402
1403#ifdef CONFIG_COMPAT
1404
1405struct compat_mq_attr {
1406	compat_long_t mq_flags;      /* message queue flags		     */
1407	compat_long_t mq_maxmsg;     /* maximum number of messages	     */
1408	compat_long_t mq_msgsize;    /* maximum message size		     */
1409	compat_long_t mq_curmsgs;    /* number of messages currently queued  */
1410	compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1411};
1412
1413static inline int get_compat_mq_attr(struct mq_attr *attr,
1414			const struct compat_mq_attr __user *uattr)
1415{
1416	struct compat_mq_attr v;
1417
1418	if (copy_from_user(&v, uattr, sizeof(*uattr)))
1419		return -EFAULT;
1420
1421	memset(attr, 0, sizeof(*attr));
1422	attr->mq_flags = v.mq_flags;
1423	attr->mq_maxmsg = v.mq_maxmsg;
1424	attr->mq_msgsize = v.mq_msgsize;
1425	attr->mq_curmsgs = v.mq_curmsgs;
1426	return 0;
1427}
1428
1429static inline int put_compat_mq_attr(const struct mq_attr *attr,
1430			struct compat_mq_attr __user *uattr)
1431{
1432	struct compat_mq_attr v;
1433
1434	memset(&v, 0, sizeof(v));
1435	v.mq_flags = attr->mq_flags;
1436	v.mq_maxmsg = attr->mq_maxmsg;
1437	v.mq_msgsize = attr->mq_msgsize;
1438	v.mq_curmsgs = attr->mq_curmsgs;
1439	if (copy_to_user(uattr, &v, sizeof(*uattr)))
1440		return -EFAULT;
1441	return 0;
1442}
1443
1444COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1445		       int, oflag, compat_mode_t, mode,
1446		       struct compat_mq_attr __user *, u_attr)
1447{
1448	struct mq_attr attr, *p = NULL;
1449	if (u_attr && oflag & O_CREAT) {
1450		p = &attr;
1451		if (get_compat_mq_attr(&attr, u_attr))
1452			return -EFAULT;
1453	}
1454	return do_mq_open(u_name, oflag, mode, p);
1455}
1456
1457static int compat_prepare_timeout(const struct compat_timespec __user *p,
1458				   struct timespec64 *ts)
1459{
1460	if (compat_get_timespec64(ts, p))
1461		return -EFAULT;
1462	if (!timespec64_valid(ts))
1463		return -EINVAL;
1464	return 0;
1465}
1466
1467COMPAT_SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes,
1468		       const char __user *, u_msg_ptr,
1469		       compat_size_t, msg_len, unsigned int, msg_prio,
1470		       const struct compat_timespec __user *, u_abs_timeout)
1471{
1472	struct timespec64 ts, *p = NULL;
1473	if (u_abs_timeout) {
1474		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1475		if (res)
1476			return res;
1477		p = &ts;
1478	}
1479	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1480}
1481
1482COMPAT_SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes,
1483		       char __user *, u_msg_ptr,
1484		       compat_size_t, msg_len, unsigned int __user *, u_msg_prio,
1485		       const struct compat_timespec __user *, u_abs_timeout)
1486{
1487	struct timespec64 ts, *p = NULL;
1488	if (u_abs_timeout) {
1489		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1490		if (res)
1491			return res;
1492		p = &ts;
1493	}
1494	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1495}
1496
1497COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1498		       const struct compat_sigevent __user *, u_notification)
1499{
1500	struct sigevent n, *p = NULL;
1501	if (u_notification) {
1502		if (get_compat_sigevent(&n, u_notification))
1503			return -EFAULT;
1504		if (n.sigev_notify == SIGEV_THREAD)
1505			n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1506		p = &n;
1507	}
1508	return do_mq_notify(mqdes, p);
1509}
1510
1511COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1512		       const struct compat_mq_attr __user *, u_mqstat,
1513		       struct compat_mq_attr __user *, u_omqstat)
1514{
1515	int ret;
1516	struct mq_attr mqstat, omqstat;
1517	struct mq_attr *new = NULL, *old = NULL;
1518
1519	if (u_mqstat) {
1520		new = &mqstat;
1521		if (get_compat_mq_attr(new, u_mqstat))
1522			return -EFAULT;
1523	}
1524	if (u_omqstat)
1525		old = &omqstat;
1526
1527	ret = do_mq_getsetattr(mqdes, new, old);
1528	if (ret || !old)
1529		return ret;
1530
1531	if (put_compat_mq_attr(old, u_omqstat))
1532		return -EFAULT;
1533	return 0;
1534}
1535#endif
1536
1537static const struct inode_operations mqueue_dir_inode_operations = {
1538	.lookup = simple_lookup,
1539	.create = mqueue_create,
1540	.unlink = mqueue_unlink,
1541};
1542
1543static const struct file_operations mqueue_file_operations = {
1544	.flush = mqueue_flush_file,
1545	.poll = mqueue_poll_file,
1546	.read = mqueue_read_file,
1547	.llseek = default_llseek,
1548};
1549
1550static const struct super_operations mqueue_super_ops = {
1551	.alloc_inode = mqueue_alloc_inode,
1552	.destroy_inode = mqueue_destroy_inode,
1553	.evict_inode = mqueue_evict_inode,
1554	.statfs = simple_statfs,
1555};
1556
1557static struct file_system_type mqueue_fs_type = {
1558	.name = "mqueue",
1559	.mount = mqueue_mount,
1560	.kill_sb = kill_litter_super,
1561	.fs_flags = FS_USERNS_MOUNT,
1562};
1563
1564int mq_init_ns(struct ipc_namespace *ns)
1565{
1566	ns->mq_queues_count  = 0;
1567	ns->mq_queues_max    = DFLT_QUEUESMAX;
1568	ns->mq_msg_max       = DFLT_MSGMAX;
1569	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1570	ns->mq_msg_default   = DFLT_MSG;
1571	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1572	ns->mq_mnt = NULL;
1573
1574	return 0;
1575}
1576
1577void mq_clear_sbinfo(struct ipc_namespace *ns)
1578{
1579	if (ns->mq_mnt)
1580		ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1581}
1582
1583void mq_put_mnt(struct ipc_namespace *ns)
1584{
1585	if (ns->mq_mnt)
1586		kern_unmount(ns->mq_mnt);
1587}
1588
1589static int __init init_mqueue_fs(void)
1590{
1591	struct vfsmount *m;
1592	int error;
1593
1594	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1595				sizeof(struct mqueue_inode_info), 0,
1596				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1597	if (mqueue_inode_cachep == NULL)
1598		return -ENOMEM;
1599
1600	/* ignore failures - they are not fatal */
1601	mq_sysctl_table = mq_register_sysctl_table();
1602
1603	error = register_filesystem(&mqueue_fs_type);
1604	if (error)
1605		goto out_sysctl;
1606
1607	spin_lock_init(&mq_lock);
1608
1609	error = mq_init_ns(&init_ipc_ns);
1610	if (error)
1611		goto out_filesystem;
1612
1613	m = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1614	if (IS_ERR(m))
1615		goto out_filesystem;
1616	init_ipc_ns.mq_mnt = m;
1617	return 0;
1618
1619out_filesystem:
1620	unregister_filesystem(&mqueue_fs_type);
1621out_sysctl:
1622	if (mq_sysctl_table)
1623		unregister_sysctl_table(mq_sysctl_table);
1624	kmem_cache_destroy(mqueue_inode_cachep);
1625	return error;
1626}
1627
1628device_initcall(init_mqueue_fs);
1629