xref: /linux-6.15/ipc/mqueue.c (revision 74ca4313)
1 /*
2  * POSIX message queues filesystem for Linux.
3  *
4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    ([email protected])
5  *                          Michal Wronski          ([email protected])
6  *
7  * Spinlocks:               Mohamed Abbas           ([email protected])
8  * Lockless receive & send, fd based notify:
9  * 			    Manfred Spraul	    ([email protected])
10  *
11  * Audit:                   George Wilson           ([email protected])
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/netlink.h>
28 #include <linux/syscalls.h>
29 #include <linux/audit.h>
30 #include <linux/signal.h>
31 #include <linux/mutex.h>
32 #include <linux/nsproxy.h>
33 #include <linux/pid.h>
34 #include <linux/ipc_namespace.h>
35 #include <linux/user_namespace.h>
36 #include <linux/slab.h>
37 
38 #include <net/sock.h>
39 #include "util.h"
40 
41 #define MQUEUE_MAGIC	0x19800202
42 #define DIRENT_SIZE	20
43 #define FILENT_SIZE	80
44 
45 #define SEND		0
46 #define RECV		1
47 
48 #define STATE_NONE	0
49 #define STATE_PENDING	1
50 #define STATE_READY	2
51 
52 struct ext_wait_queue {		/* queue of sleeping tasks */
53 	struct task_struct *task;
54 	struct list_head list;
55 	struct msg_msg *msg;	/* ptr of loaded message */
56 	int state;		/* one of STATE_* values */
57 };
58 
59 struct mqueue_inode_info {
60 	spinlock_t lock;
61 	struct inode vfs_inode;
62 	wait_queue_head_t wait_q;
63 
64 	struct msg_msg **messages;
65 	struct mq_attr attr;
66 
67 	struct sigevent notify;
68 	struct pid* notify_owner;
69 	struct user_namespace *notify_user_ns;
70 	struct user_struct *user;	/* user who created, for accounting */
71 	struct sock *notify_sock;
72 	struct sk_buff *notify_cookie;
73 
74 	/* for tasks waiting for free space and messages, respectively */
75 	struct ext_wait_queue e_wait_q[2];
76 
77 	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
78 };
79 
80 static const struct inode_operations mqueue_dir_inode_operations;
81 static const struct file_operations mqueue_file_operations;
82 static const struct super_operations mqueue_super_ops;
83 static void remove_notification(struct mqueue_inode_info *info);
84 
85 static struct kmem_cache *mqueue_inode_cachep;
86 
87 static struct ctl_table_header * mq_sysctl_table;
88 
89 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
90 {
91 	return container_of(inode, struct mqueue_inode_info, vfs_inode);
92 }
93 
94 /*
95  * This routine should be called with the mq_lock held.
96  */
97 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
98 {
99 	return get_ipc_ns(inode->i_sb->s_fs_info);
100 }
101 
102 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
103 {
104 	struct ipc_namespace *ns;
105 
106 	spin_lock(&mq_lock);
107 	ns = __get_ns_from_inode(inode);
108 	spin_unlock(&mq_lock);
109 	return ns;
110 }
111 
112 static struct inode *mqueue_get_inode(struct super_block *sb,
113 		struct ipc_namespace *ipc_ns, umode_t mode,
114 		struct mq_attr *attr)
115 {
116 	struct user_struct *u = current_user();
117 	struct inode *inode;
118 	int ret = -ENOMEM;
119 
120 	inode = new_inode(sb);
121 	if (!inode)
122 		goto err;
123 
124 	inode->i_ino = get_next_ino();
125 	inode->i_mode = mode;
126 	inode->i_uid = current_fsuid();
127 	inode->i_gid = current_fsgid();
128 	inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
129 
130 	if (S_ISREG(mode)) {
131 		struct mqueue_inode_info *info;
132 		unsigned long mq_bytes, mq_msg_tblsz;
133 
134 		inode->i_fop = &mqueue_file_operations;
135 		inode->i_size = FILENT_SIZE;
136 		/* mqueue specific info */
137 		info = MQUEUE_I(inode);
138 		spin_lock_init(&info->lock);
139 		init_waitqueue_head(&info->wait_q);
140 		INIT_LIST_HEAD(&info->e_wait_q[0].list);
141 		INIT_LIST_HEAD(&info->e_wait_q[1].list);
142 		info->notify_owner = NULL;
143 		info->notify_user_ns = NULL;
144 		info->qsize = 0;
145 		info->user = NULL;	/* set when all is ok */
146 		memset(&info->attr, 0, sizeof(info->attr));
147 		info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
148 		info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
149 		if (attr) {
150 			info->attr.mq_maxmsg = attr->mq_maxmsg;
151 			info->attr.mq_msgsize = attr->mq_msgsize;
152 		}
153 		mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
154 		info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
155 		if (!info->messages)
156 			goto out_inode;
157 
158 		mq_bytes = (mq_msg_tblsz +
159 			(info->attr.mq_maxmsg * info->attr.mq_msgsize));
160 
161 		spin_lock(&mq_lock);
162 		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
163 		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
164 			spin_unlock(&mq_lock);
165 			/* mqueue_evict_inode() releases info->messages */
166 			ret = -EMFILE;
167 			goto out_inode;
168 		}
169 		u->mq_bytes += mq_bytes;
170 		spin_unlock(&mq_lock);
171 
172 		/* all is ok */
173 		info->user = get_uid(u);
174 	} else if (S_ISDIR(mode)) {
175 		inc_nlink(inode);
176 		/* Some things misbehave if size == 0 on a directory */
177 		inode->i_size = 2 * DIRENT_SIZE;
178 		inode->i_op = &mqueue_dir_inode_operations;
179 		inode->i_fop = &simple_dir_operations;
180 	}
181 
182 	return inode;
183 out_inode:
184 	iput(inode);
185 err:
186 	return ERR_PTR(ret);
187 }
188 
189 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
190 {
191 	struct inode *inode;
192 	struct ipc_namespace *ns = data;
193 
194 	sb->s_blocksize = PAGE_CACHE_SIZE;
195 	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
196 	sb->s_magic = MQUEUE_MAGIC;
197 	sb->s_op = &mqueue_super_ops;
198 
199 	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
200 	if (IS_ERR(inode))
201 		return PTR_ERR(inode);
202 
203 	sb->s_root = d_make_root(inode);
204 	if (!sb->s_root)
205 		return -ENOMEM;
206 	return 0;
207 }
208 
209 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
210 			 int flags, const char *dev_name,
211 			 void *data)
212 {
213 	if (!(flags & MS_KERNMOUNT))
214 		data = current->nsproxy->ipc_ns;
215 	return mount_ns(fs_type, flags, data, mqueue_fill_super);
216 }
217 
218 static void init_once(void *foo)
219 {
220 	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
221 
222 	inode_init_once(&p->vfs_inode);
223 }
224 
225 static struct inode *mqueue_alloc_inode(struct super_block *sb)
226 {
227 	struct mqueue_inode_info *ei;
228 
229 	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
230 	if (!ei)
231 		return NULL;
232 	return &ei->vfs_inode;
233 }
234 
235 static void mqueue_i_callback(struct rcu_head *head)
236 {
237 	struct inode *inode = container_of(head, struct inode, i_rcu);
238 	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
239 }
240 
241 static void mqueue_destroy_inode(struct inode *inode)
242 {
243 	call_rcu(&inode->i_rcu, mqueue_i_callback);
244 }
245 
246 static void mqueue_evict_inode(struct inode *inode)
247 {
248 	struct mqueue_inode_info *info;
249 	struct user_struct *user;
250 	unsigned long mq_bytes;
251 	int i;
252 	struct ipc_namespace *ipc_ns;
253 
254 	end_writeback(inode);
255 
256 	if (S_ISDIR(inode->i_mode))
257 		return;
258 
259 	ipc_ns = get_ns_from_inode(inode);
260 	info = MQUEUE_I(inode);
261 	spin_lock(&info->lock);
262 	for (i = 0; i < info->attr.mq_curmsgs; i++)
263 		free_msg(info->messages[i]);
264 	kfree(info->messages);
265 	spin_unlock(&info->lock);
266 
267 	/* Total amount of bytes accounted for the mqueue */
268 	mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
269 	    + info->attr.mq_msgsize);
270 	user = info->user;
271 	if (user) {
272 		spin_lock(&mq_lock);
273 		user->mq_bytes -= mq_bytes;
274 		/*
275 		 * get_ns_from_inode() ensures that the
276 		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
277 		 * to which we now hold a reference, or it is NULL.
278 		 * We can't put it here under mq_lock, though.
279 		 */
280 		if (ipc_ns)
281 			ipc_ns->mq_queues_count--;
282 		spin_unlock(&mq_lock);
283 		free_uid(user);
284 	}
285 	if (ipc_ns)
286 		put_ipc_ns(ipc_ns);
287 }
288 
289 static int mqueue_create(struct inode *dir, struct dentry *dentry,
290 				umode_t mode, struct nameidata *nd)
291 {
292 	struct inode *inode;
293 	struct mq_attr *attr = dentry->d_fsdata;
294 	int error;
295 	struct ipc_namespace *ipc_ns;
296 
297 	spin_lock(&mq_lock);
298 	ipc_ns = __get_ns_from_inode(dir);
299 	if (!ipc_ns) {
300 		error = -EACCES;
301 		goto out_unlock;
302 	}
303 	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
304 			!capable(CAP_SYS_RESOURCE)) {
305 		error = -ENOSPC;
306 		goto out_unlock;
307 	}
308 	ipc_ns->mq_queues_count++;
309 	spin_unlock(&mq_lock);
310 
311 	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
312 	if (IS_ERR(inode)) {
313 		error = PTR_ERR(inode);
314 		spin_lock(&mq_lock);
315 		ipc_ns->mq_queues_count--;
316 		goto out_unlock;
317 	}
318 
319 	put_ipc_ns(ipc_ns);
320 	dir->i_size += DIRENT_SIZE;
321 	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
322 
323 	d_instantiate(dentry, inode);
324 	dget(dentry);
325 	return 0;
326 out_unlock:
327 	spin_unlock(&mq_lock);
328 	if (ipc_ns)
329 		put_ipc_ns(ipc_ns);
330 	return error;
331 }
332 
333 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
334 {
335   	struct inode *inode = dentry->d_inode;
336 
337 	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
338 	dir->i_size -= DIRENT_SIZE;
339   	drop_nlink(inode);
340   	dput(dentry);
341   	return 0;
342 }
343 
344 /*
345 *	This is routine for system read from queue file.
346 *	To avoid mess with doing here some sort of mq_receive we allow
347 *	to read only queue size & notification info (the only values
348 *	that are interesting from user point of view and aren't accessible
349 *	through std routines)
350 */
351 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
352 				size_t count, loff_t *off)
353 {
354 	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
355 	char buffer[FILENT_SIZE];
356 	ssize_t ret;
357 
358 	spin_lock(&info->lock);
359 	snprintf(buffer, sizeof(buffer),
360 			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
361 			info->qsize,
362 			info->notify_owner ? info->notify.sigev_notify : 0,
363 			(info->notify_owner &&
364 			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
365 				info->notify.sigev_signo : 0,
366 			pid_vnr(info->notify_owner));
367 	spin_unlock(&info->lock);
368 	buffer[sizeof(buffer)-1] = '\0';
369 
370 	ret = simple_read_from_buffer(u_data, count, off, buffer,
371 				strlen(buffer));
372 	if (ret <= 0)
373 		return ret;
374 
375 	filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
376 	return ret;
377 }
378 
379 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
380 {
381 	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
382 
383 	spin_lock(&info->lock);
384 	if (task_tgid(current) == info->notify_owner)
385 		remove_notification(info);
386 
387 	spin_unlock(&info->lock);
388 	return 0;
389 }
390 
391 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
392 {
393 	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
394 	int retval = 0;
395 
396 	poll_wait(filp, &info->wait_q, poll_tab);
397 
398 	spin_lock(&info->lock);
399 	if (info->attr.mq_curmsgs)
400 		retval = POLLIN | POLLRDNORM;
401 
402 	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
403 		retval |= POLLOUT | POLLWRNORM;
404 	spin_unlock(&info->lock);
405 
406 	return retval;
407 }
408 
409 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
410 static void wq_add(struct mqueue_inode_info *info, int sr,
411 			struct ext_wait_queue *ewp)
412 {
413 	struct ext_wait_queue *walk;
414 
415 	ewp->task = current;
416 
417 	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
418 		if (walk->task->static_prio <= current->static_prio) {
419 			list_add_tail(&ewp->list, &walk->list);
420 			return;
421 		}
422 	}
423 	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
424 }
425 
426 /*
427  * Puts current task to sleep. Caller must hold queue lock. After return
428  * lock isn't held.
429  * sr: SEND or RECV
430  */
431 static int wq_sleep(struct mqueue_inode_info *info, int sr,
432 		    ktime_t *timeout, struct ext_wait_queue *ewp)
433 {
434 	int retval;
435 	signed long time;
436 
437 	wq_add(info, sr, ewp);
438 
439 	for (;;) {
440 		set_current_state(TASK_INTERRUPTIBLE);
441 
442 		spin_unlock(&info->lock);
443 		time = schedule_hrtimeout_range_clock(timeout, 0,
444 			HRTIMER_MODE_ABS, CLOCK_REALTIME);
445 
446 		while (ewp->state == STATE_PENDING)
447 			cpu_relax();
448 
449 		if (ewp->state == STATE_READY) {
450 			retval = 0;
451 			goto out;
452 		}
453 		spin_lock(&info->lock);
454 		if (ewp->state == STATE_READY) {
455 			retval = 0;
456 			goto out_unlock;
457 		}
458 		if (signal_pending(current)) {
459 			retval = -ERESTARTSYS;
460 			break;
461 		}
462 		if (time == 0) {
463 			retval = -ETIMEDOUT;
464 			break;
465 		}
466 	}
467 	list_del(&ewp->list);
468 out_unlock:
469 	spin_unlock(&info->lock);
470 out:
471 	return retval;
472 }
473 
474 /*
475  * Returns waiting task that should be serviced first or NULL if none exists
476  */
477 static struct ext_wait_queue *wq_get_first_waiter(
478 		struct mqueue_inode_info *info, int sr)
479 {
480 	struct list_head *ptr;
481 
482 	ptr = info->e_wait_q[sr].list.prev;
483 	if (ptr == &info->e_wait_q[sr].list)
484 		return NULL;
485 	return list_entry(ptr, struct ext_wait_queue, list);
486 }
487 
488 /* Auxiliary functions to manipulate messages' list */
489 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
490 {
491 	int k;
492 
493 	k = info->attr.mq_curmsgs - 1;
494 	while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
495 		info->messages[k + 1] = info->messages[k];
496 		k--;
497 	}
498 	info->attr.mq_curmsgs++;
499 	info->qsize += ptr->m_ts;
500 	info->messages[k + 1] = ptr;
501 }
502 
503 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
504 {
505 	info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
506 	return info->messages[info->attr.mq_curmsgs];
507 }
508 
509 static inline void set_cookie(struct sk_buff *skb, char code)
510 {
511 	((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
512 }
513 
514 /*
515  * The next function is only to split too long sys_mq_timedsend
516  */
517 static void __do_notify(struct mqueue_inode_info *info)
518 {
519 	/* notification
520 	 * invoked when there is registered process and there isn't process
521 	 * waiting synchronously for message AND state of queue changed from
522 	 * empty to not empty. Here we are sure that no one is waiting
523 	 * synchronously. */
524 	if (info->notify_owner &&
525 	    info->attr.mq_curmsgs == 1) {
526 		struct siginfo sig_i;
527 		switch (info->notify.sigev_notify) {
528 		case SIGEV_NONE:
529 			break;
530 		case SIGEV_SIGNAL:
531 			/* sends signal */
532 
533 			sig_i.si_signo = info->notify.sigev_signo;
534 			sig_i.si_errno = 0;
535 			sig_i.si_code = SI_MESGQ;
536 			sig_i.si_value = info->notify.sigev_value;
537 			/* map current pid/uid into info->owner's namespaces */
538 			rcu_read_lock();
539 			sig_i.si_pid = task_tgid_nr_ns(current,
540 						ns_of_pid(info->notify_owner));
541 			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
542 			rcu_read_unlock();
543 
544 			kill_pid_info(info->notify.sigev_signo,
545 				      &sig_i, info->notify_owner);
546 			break;
547 		case SIGEV_THREAD:
548 			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
549 			netlink_sendskb(info->notify_sock, info->notify_cookie);
550 			break;
551 		}
552 		/* after notification unregisters process */
553 		put_pid(info->notify_owner);
554 		put_user_ns(info->notify_user_ns);
555 		info->notify_owner = NULL;
556 		info->notify_user_ns = NULL;
557 	}
558 	wake_up(&info->wait_q);
559 }
560 
561 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
562 			   ktime_t *expires, struct timespec *ts)
563 {
564 	if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
565 		return -EFAULT;
566 	if (!timespec_valid(ts))
567 		return -EINVAL;
568 
569 	*expires = timespec_to_ktime(*ts);
570 	return 0;
571 }
572 
573 static void remove_notification(struct mqueue_inode_info *info)
574 {
575 	if (info->notify_owner != NULL &&
576 	    info->notify.sigev_notify == SIGEV_THREAD) {
577 		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
578 		netlink_sendskb(info->notify_sock, info->notify_cookie);
579 	}
580 	put_pid(info->notify_owner);
581 	put_user_ns(info->notify_user_ns);
582 	info->notify_owner = NULL;
583 	info->notify_user_ns = NULL;
584 }
585 
586 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
587 {
588 	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
589 		return 0;
590 	if (capable(CAP_SYS_RESOURCE)) {
591 		if (attr->mq_maxmsg > HARD_MSGMAX)
592 			return 0;
593 	} else {
594 		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
595 				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
596 			return 0;
597 	}
598 	/* check for overflow */
599 	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
600 		return 0;
601 	if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
602 	    + sizeof (struct msg_msg *))) <
603 	    (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
604 		return 0;
605 	return 1;
606 }
607 
608 /*
609  * Invoked when creating a new queue via sys_mq_open
610  */
611 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
612 			struct dentry *dentry, int oflag, umode_t mode,
613 			struct mq_attr *attr)
614 {
615 	const struct cred *cred = current_cred();
616 	struct file *result;
617 	int ret;
618 
619 	if (attr) {
620 		if (!mq_attr_ok(ipc_ns, attr)) {
621 			ret = -EINVAL;
622 			goto out;
623 		}
624 		/* store for use during create */
625 		dentry->d_fsdata = attr;
626 	}
627 
628 	mode &= ~current_umask();
629 	ret = mnt_want_write(ipc_ns->mq_mnt);
630 	if (ret)
631 		goto out;
632 	ret = vfs_create(dir->d_inode, dentry, mode, NULL);
633 	dentry->d_fsdata = NULL;
634 	if (ret)
635 		goto out_drop_write;
636 
637 	result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
638 	/*
639 	 * dentry_open() took a persistent mnt_want_write(),
640 	 * so we can now drop this one.
641 	 */
642 	mnt_drop_write(ipc_ns->mq_mnt);
643 	return result;
644 
645 out_drop_write:
646 	mnt_drop_write(ipc_ns->mq_mnt);
647 out:
648 	dput(dentry);
649 	mntput(ipc_ns->mq_mnt);
650 	return ERR_PTR(ret);
651 }
652 
653 /* Opens existing queue */
654 static struct file *do_open(struct ipc_namespace *ipc_ns,
655 				struct dentry *dentry, int oflag)
656 {
657 	int ret;
658 	const struct cred *cred = current_cred();
659 
660 	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
661 						  MAY_READ | MAY_WRITE };
662 
663 	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
664 		ret = -EINVAL;
665 		goto err;
666 	}
667 
668 	if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
669 		ret = -EACCES;
670 		goto err;
671 	}
672 
673 	return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
674 
675 err:
676 	dput(dentry);
677 	mntput(ipc_ns->mq_mnt);
678 	return ERR_PTR(ret);
679 }
680 
681 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
682 		struct mq_attr __user *, u_attr)
683 {
684 	struct dentry *dentry;
685 	struct file *filp;
686 	char *name;
687 	struct mq_attr attr;
688 	int fd, error;
689 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
690 
691 	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
692 		return -EFAULT;
693 
694 	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
695 
696 	if (IS_ERR(name = getname(u_name)))
697 		return PTR_ERR(name);
698 
699 	fd = get_unused_fd_flags(O_CLOEXEC);
700 	if (fd < 0)
701 		goto out_putname;
702 
703 	mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
704 	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
705 	if (IS_ERR(dentry)) {
706 		error = PTR_ERR(dentry);
707 		goto out_putfd;
708 	}
709 	mntget(ipc_ns->mq_mnt);
710 
711 	if (oflag & O_CREAT) {
712 		if (dentry->d_inode) {	/* entry already exists */
713 			audit_inode(name, dentry);
714 			if (oflag & O_EXCL) {
715 				error = -EEXIST;
716 				goto out;
717 			}
718 			filp = do_open(ipc_ns, dentry, oflag);
719 		} else {
720 			filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
721 						dentry, oflag, mode,
722 						u_attr ? &attr : NULL);
723 		}
724 	} else {
725 		if (!dentry->d_inode) {
726 			error = -ENOENT;
727 			goto out;
728 		}
729 		audit_inode(name, dentry);
730 		filp = do_open(ipc_ns, dentry, oflag);
731 	}
732 
733 	if (IS_ERR(filp)) {
734 		error = PTR_ERR(filp);
735 		goto out_putfd;
736 	}
737 
738 	fd_install(fd, filp);
739 	goto out_upsem;
740 
741 out:
742 	dput(dentry);
743 	mntput(ipc_ns->mq_mnt);
744 out_putfd:
745 	put_unused_fd(fd);
746 	fd = error;
747 out_upsem:
748 	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
749 out_putname:
750 	putname(name);
751 	return fd;
752 }
753 
754 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
755 {
756 	int err;
757 	char *name;
758 	struct dentry *dentry;
759 	struct inode *inode = NULL;
760 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
761 
762 	name = getname(u_name);
763 	if (IS_ERR(name))
764 		return PTR_ERR(name);
765 
766 	mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
767 			I_MUTEX_PARENT);
768 	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
769 	if (IS_ERR(dentry)) {
770 		err = PTR_ERR(dentry);
771 		goto out_unlock;
772 	}
773 
774 	if (!dentry->d_inode) {
775 		err = -ENOENT;
776 		goto out_err;
777 	}
778 
779 	inode = dentry->d_inode;
780 	if (inode)
781 		ihold(inode);
782 	err = mnt_want_write(ipc_ns->mq_mnt);
783 	if (err)
784 		goto out_err;
785 	err = vfs_unlink(dentry->d_parent->d_inode, dentry);
786 	mnt_drop_write(ipc_ns->mq_mnt);
787 out_err:
788 	dput(dentry);
789 
790 out_unlock:
791 	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
792 	putname(name);
793 	if (inode)
794 		iput(inode);
795 
796 	return err;
797 }
798 
799 /* Pipelined send and receive functions.
800  *
801  * If a receiver finds no waiting message, then it registers itself in the
802  * list of waiting receivers. A sender checks that list before adding the new
803  * message into the message array. If there is a waiting receiver, then it
804  * bypasses the message array and directly hands the message over to the
805  * receiver.
806  * The receiver accepts the message and returns without grabbing the queue
807  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
808  * are necessary. The same algorithm is used for sysv semaphores, see
809  * ipc/sem.c for more details.
810  *
811  * The same algorithm is used for senders.
812  */
813 
814 /* pipelined_send() - send a message directly to the task waiting in
815  * sys_mq_timedreceive() (without inserting message into a queue).
816  */
817 static inline void pipelined_send(struct mqueue_inode_info *info,
818 				  struct msg_msg *message,
819 				  struct ext_wait_queue *receiver)
820 {
821 	receiver->msg = message;
822 	list_del(&receiver->list);
823 	receiver->state = STATE_PENDING;
824 	wake_up_process(receiver->task);
825 	smp_wmb();
826 	receiver->state = STATE_READY;
827 }
828 
829 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
830  * gets its message and put to the queue (we have one free place for sure). */
831 static inline void pipelined_receive(struct mqueue_inode_info *info)
832 {
833 	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
834 
835 	if (!sender) {
836 		/* for poll */
837 		wake_up_interruptible(&info->wait_q);
838 		return;
839 	}
840 	msg_insert(sender->msg, info);
841 	list_del(&sender->list);
842 	sender->state = STATE_PENDING;
843 	wake_up_process(sender->task);
844 	smp_wmb();
845 	sender->state = STATE_READY;
846 }
847 
848 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
849 		size_t, msg_len, unsigned int, msg_prio,
850 		const struct timespec __user *, u_abs_timeout)
851 {
852 	struct file *filp;
853 	struct inode *inode;
854 	struct ext_wait_queue wait;
855 	struct ext_wait_queue *receiver;
856 	struct msg_msg *msg_ptr;
857 	struct mqueue_inode_info *info;
858 	ktime_t expires, *timeout = NULL;
859 	struct timespec ts;
860 	int ret;
861 
862 	if (u_abs_timeout) {
863 		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
864 		if (res)
865 			return res;
866 		timeout = &expires;
867 	}
868 
869 	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
870 		return -EINVAL;
871 
872 	audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
873 
874 	filp = fget(mqdes);
875 	if (unlikely(!filp)) {
876 		ret = -EBADF;
877 		goto out;
878 	}
879 
880 	inode = filp->f_path.dentry->d_inode;
881 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
882 		ret = -EBADF;
883 		goto out_fput;
884 	}
885 	info = MQUEUE_I(inode);
886 	audit_inode(NULL, filp->f_path.dentry);
887 
888 	if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
889 		ret = -EBADF;
890 		goto out_fput;
891 	}
892 
893 	if (unlikely(msg_len > info->attr.mq_msgsize)) {
894 		ret = -EMSGSIZE;
895 		goto out_fput;
896 	}
897 
898 	/* First try to allocate memory, before doing anything with
899 	 * existing queues. */
900 	msg_ptr = load_msg(u_msg_ptr, msg_len);
901 	if (IS_ERR(msg_ptr)) {
902 		ret = PTR_ERR(msg_ptr);
903 		goto out_fput;
904 	}
905 	msg_ptr->m_ts = msg_len;
906 	msg_ptr->m_type = msg_prio;
907 
908 	spin_lock(&info->lock);
909 
910 	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
911 		if (filp->f_flags & O_NONBLOCK) {
912 			spin_unlock(&info->lock);
913 			ret = -EAGAIN;
914 		} else {
915 			wait.task = current;
916 			wait.msg = (void *) msg_ptr;
917 			wait.state = STATE_NONE;
918 			ret = wq_sleep(info, SEND, timeout, &wait);
919 		}
920 		if (ret < 0)
921 			free_msg(msg_ptr);
922 	} else {
923 		receiver = wq_get_first_waiter(info, RECV);
924 		if (receiver) {
925 			pipelined_send(info, msg_ptr, receiver);
926 		} else {
927 			/* adds message to the queue */
928 			msg_insert(msg_ptr, info);
929 			__do_notify(info);
930 		}
931 		inode->i_atime = inode->i_mtime = inode->i_ctime =
932 				CURRENT_TIME;
933 		spin_unlock(&info->lock);
934 		ret = 0;
935 	}
936 out_fput:
937 	fput(filp);
938 out:
939 	return ret;
940 }
941 
942 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
943 		size_t, msg_len, unsigned int __user *, u_msg_prio,
944 		const struct timespec __user *, u_abs_timeout)
945 {
946 	ssize_t ret;
947 	struct msg_msg *msg_ptr;
948 	struct file *filp;
949 	struct inode *inode;
950 	struct mqueue_inode_info *info;
951 	struct ext_wait_queue wait;
952 	ktime_t expires, *timeout = NULL;
953 	struct timespec ts;
954 
955 	if (u_abs_timeout) {
956 		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
957 		if (res)
958 			return res;
959 		timeout = &expires;
960 	}
961 
962 	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
963 
964 	filp = fget(mqdes);
965 	if (unlikely(!filp)) {
966 		ret = -EBADF;
967 		goto out;
968 	}
969 
970 	inode = filp->f_path.dentry->d_inode;
971 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
972 		ret = -EBADF;
973 		goto out_fput;
974 	}
975 	info = MQUEUE_I(inode);
976 	audit_inode(NULL, filp->f_path.dentry);
977 
978 	if (unlikely(!(filp->f_mode & FMODE_READ))) {
979 		ret = -EBADF;
980 		goto out_fput;
981 	}
982 
983 	/* checks if buffer is big enough */
984 	if (unlikely(msg_len < info->attr.mq_msgsize)) {
985 		ret = -EMSGSIZE;
986 		goto out_fput;
987 	}
988 
989 	spin_lock(&info->lock);
990 	if (info->attr.mq_curmsgs == 0) {
991 		if (filp->f_flags & O_NONBLOCK) {
992 			spin_unlock(&info->lock);
993 			ret = -EAGAIN;
994 		} else {
995 			wait.task = current;
996 			wait.state = STATE_NONE;
997 			ret = wq_sleep(info, RECV, timeout, &wait);
998 			msg_ptr = wait.msg;
999 		}
1000 	} else {
1001 		msg_ptr = msg_get(info);
1002 
1003 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1004 				CURRENT_TIME;
1005 
1006 		/* There is now free space in queue. */
1007 		pipelined_receive(info);
1008 		spin_unlock(&info->lock);
1009 		ret = 0;
1010 	}
1011 	if (ret == 0) {
1012 		ret = msg_ptr->m_ts;
1013 
1014 		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1015 			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1016 			ret = -EFAULT;
1017 		}
1018 		free_msg(msg_ptr);
1019 	}
1020 out_fput:
1021 	fput(filp);
1022 out:
1023 	return ret;
1024 }
1025 
1026 /*
1027  * Notes: the case when user wants us to deregister (with NULL as pointer)
1028  * and he isn't currently owner of notification, will be silently discarded.
1029  * It isn't explicitly defined in the POSIX.
1030  */
1031 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1032 		const struct sigevent __user *, u_notification)
1033 {
1034 	int ret;
1035 	struct file *filp;
1036 	struct sock *sock;
1037 	struct inode *inode;
1038 	struct sigevent notification;
1039 	struct mqueue_inode_info *info;
1040 	struct sk_buff *nc;
1041 
1042 	if (u_notification) {
1043 		if (copy_from_user(&notification, u_notification,
1044 					sizeof(struct sigevent)))
1045 			return -EFAULT;
1046 	}
1047 
1048 	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1049 
1050 	nc = NULL;
1051 	sock = NULL;
1052 	if (u_notification != NULL) {
1053 		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1054 			     notification.sigev_notify != SIGEV_SIGNAL &&
1055 			     notification.sigev_notify != SIGEV_THREAD))
1056 			return -EINVAL;
1057 		if (notification.sigev_notify == SIGEV_SIGNAL &&
1058 			!valid_signal(notification.sigev_signo)) {
1059 			return -EINVAL;
1060 		}
1061 		if (notification.sigev_notify == SIGEV_THREAD) {
1062 			long timeo;
1063 
1064 			/* create the notify skb */
1065 			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1066 			if (!nc) {
1067 				ret = -ENOMEM;
1068 				goto out;
1069 			}
1070 			if (copy_from_user(nc->data,
1071 					notification.sigev_value.sival_ptr,
1072 					NOTIFY_COOKIE_LEN)) {
1073 				ret = -EFAULT;
1074 				goto out;
1075 			}
1076 
1077 			/* TODO: add a header? */
1078 			skb_put(nc, NOTIFY_COOKIE_LEN);
1079 			/* and attach it to the socket */
1080 retry:
1081 			filp = fget(notification.sigev_signo);
1082 			if (!filp) {
1083 				ret = -EBADF;
1084 				goto out;
1085 			}
1086 			sock = netlink_getsockbyfilp(filp);
1087 			fput(filp);
1088 			if (IS_ERR(sock)) {
1089 				ret = PTR_ERR(sock);
1090 				sock = NULL;
1091 				goto out;
1092 			}
1093 
1094 			timeo = MAX_SCHEDULE_TIMEOUT;
1095 			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1096 			if (ret == 1)
1097 				goto retry;
1098 			if (ret) {
1099 				sock = NULL;
1100 				nc = NULL;
1101 				goto out;
1102 			}
1103 		}
1104 	}
1105 
1106 	filp = fget(mqdes);
1107 	if (!filp) {
1108 		ret = -EBADF;
1109 		goto out;
1110 	}
1111 
1112 	inode = filp->f_path.dentry->d_inode;
1113 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1114 		ret = -EBADF;
1115 		goto out_fput;
1116 	}
1117 	info = MQUEUE_I(inode);
1118 
1119 	ret = 0;
1120 	spin_lock(&info->lock);
1121 	if (u_notification == NULL) {
1122 		if (info->notify_owner == task_tgid(current)) {
1123 			remove_notification(info);
1124 			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1125 		}
1126 	} else if (info->notify_owner != NULL) {
1127 		ret = -EBUSY;
1128 	} else {
1129 		switch (notification.sigev_notify) {
1130 		case SIGEV_NONE:
1131 			info->notify.sigev_notify = SIGEV_NONE;
1132 			break;
1133 		case SIGEV_THREAD:
1134 			info->notify_sock = sock;
1135 			info->notify_cookie = nc;
1136 			sock = NULL;
1137 			nc = NULL;
1138 			info->notify.sigev_notify = SIGEV_THREAD;
1139 			break;
1140 		case SIGEV_SIGNAL:
1141 			info->notify.sigev_signo = notification.sigev_signo;
1142 			info->notify.sigev_value = notification.sigev_value;
1143 			info->notify.sigev_notify = SIGEV_SIGNAL;
1144 			break;
1145 		}
1146 
1147 		info->notify_owner = get_pid(task_tgid(current));
1148 		info->notify_user_ns = get_user_ns(current_user_ns());
1149 		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1150 	}
1151 	spin_unlock(&info->lock);
1152 out_fput:
1153 	fput(filp);
1154 out:
1155 	if (sock) {
1156 		netlink_detachskb(sock, nc);
1157 	} else if (nc) {
1158 		dev_kfree_skb(nc);
1159 	}
1160 	return ret;
1161 }
1162 
1163 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1164 		const struct mq_attr __user *, u_mqstat,
1165 		struct mq_attr __user *, u_omqstat)
1166 {
1167 	int ret;
1168 	struct mq_attr mqstat, omqstat;
1169 	struct file *filp;
1170 	struct inode *inode;
1171 	struct mqueue_inode_info *info;
1172 
1173 	if (u_mqstat != NULL) {
1174 		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1175 			return -EFAULT;
1176 		if (mqstat.mq_flags & (~O_NONBLOCK))
1177 			return -EINVAL;
1178 	}
1179 
1180 	filp = fget(mqdes);
1181 	if (!filp) {
1182 		ret = -EBADF;
1183 		goto out;
1184 	}
1185 
1186 	inode = filp->f_path.dentry->d_inode;
1187 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1188 		ret = -EBADF;
1189 		goto out_fput;
1190 	}
1191 	info = MQUEUE_I(inode);
1192 
1193 	spin_lock(&info->lock);
1194 
1195 	omqstat = info->attr;
1196 	omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1197 	if (u_mqstat) {
1198 		audit_mq_getsetattr(mqdes, &mqstat);
1199 		spin_lock(&filp->f_lock);
1200 		if (mqstat.mq_flags & O_NONBLOCK)
1201 			filp->f_flags |= O_NONBLOCK;
1202 		else
1203 			filp->f_flags &= ~O_NONBLOCK;
1204 		spin_unlock(&filp->f_lock);
1205 
1206 		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1207 	}
1208 
1209 	spin_unlock(&info->lock);
1210 
1211 	ret = 0;
1212 	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1213 						sizeof(struct mq_attr)))
1214 		ret = -EFAULT;
1215 
1216 out_fput:
1217 	fput(filp);
1218 out:
1219 	return ret;
1220 }
1221 
1222 static const struct inode_operations mqueue_dir_inode_operations = {
1223 	.lookup = simple_lookup,
1224 	.create = mqueue_create,
1225 	.unlink = mqueue_unlink,
1226 };
1227 
1228 static const struct file_operations mqueue_file_operations = {
1229 	.flush = mqueue_flush_file,
1230 	.poll = mqueue_poll_file,
1231 	.read = mqueue_read_file,
1232 	.llseek = default_llseek,
1233 };
1234 
1235 static const struct super_operations mqueue_super_ops = {
1236 	.alloc_inode = mqueue_alloc_inode,
1237 	.destroy_inode = mqueue_destroy_inode,
1238 	.evict_inode = mqueue_evict_inode,
1239 	.statfs = simple_statfs,
1240 };
1241 
1242 static struct file_system_type mqueue_fs_type = {
1243 	.name = "mqueue",
1244 	.mount = mqueue_mount,
1245 	.kill_sb = kill_litter_super,
1246 };
1247 
1248 int mq_init_ns(struct ipc_namespace *ns)
1249 {
1250 	ns->mq_queues_count  = 0;
1251 	ns->mq_queues_max    = DFLT_QUEUESMAX;
1252 	ns->mq_msg_max       = DFLT_MSGMAX;
1253 	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1254 
1255 	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1256 	if (IS_ERR(ns->mq_mnt)) {
1257 		int err = PTR_ERR(ns->mq_mnt);
1258 		ns->mq_mnt = NULL;
1259 		return err;
1260 	}
1261 	return 0;
1262 }
1263 
1264 void mq_clear_sbinfo(struct ipc_namespace *ns)
1265 {
1266 	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1267 }
1268 
1269 void mq_put_mnt(struct ipc_namespace *ns)
1270 {
1271 	kern_unmount(ns->mq_mnt);
1272 }
1273 
1274 static int __init init_mqueue_fs(void)
1275 {
1276 	int error;
1277 
1278 	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1279 				sizeof(struct mqueue_inode_info), 0,
1280 				SLAB_HWCACHE_ALIGN, init_once);
1281 	if (mqueue_inode_cachep == NULL)
1282 		return -ENOMEM;
1283 
1284 	/* ignore failures - they are not fatal */
1285 	mq_sysctl_table = mq_register_sysctl_table();
1286 
1287 	error = register_filesystem(&mqueue_fs_type);
1288 	if (error)
1289 		goto out_sysctl;
1290 
1291 	spin_lock_init(&mq_lock);
1292 
1293 	error = mq_init_ns(&init_ipc_ns);
1294 	if (error)
1295 		goto out_filesystem;
1296 
1297 	return 0;
1298 
1299 out_filesystem:
1300 	unregister_filesystem(&mqueue_fs_type);
1301 out_sysctl:
1302 	if (mq_sysctl_table)
1303 		unregister_sysctl_table(mq_sysctl_table);
1304 	kmem_cache_destroy(mqueue_inode_cachep);
1305 	return error;
1306 }
1307 
1308 __initcall(init_mqueue_fs);
1309