xref: /linux/ipc/mqueue.c (revision b43ab901d671e3e3cad425ea5e9a3c74e266dcdd)
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/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_struct *user;	/* user who created, for accounting */
70 	struct sock *notify_sock;
71 	struct sk_buff *notify_cookie;
72 
73 	/* for tasks waiting for free space and messages, respectively */
74 	struct ext_wait_queue e_wait_q[2];
75 
76 	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
77 };
78 
79 static const struct inode_operations mqueue_dir_inode_operations;
80 static const struct file_operations mqueue_file_operations;
81 static const struct super_operations mqueue_super_ops;
82 static void remove_notification(struct mqueue_inode_info *info);
83 
84 static struct kmem_cache *mqueue_inode_cachep;
85 
86 static struct ctl_table_header * mq_sysctl_table;
87 
88 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
89 {
90 	return container_of(inode, struct mqueue_inode_info, vfs_inode);
91 }
92 
93 /*
94  * This routine should be called with the mq_lock held.
95  */
96 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
97 {
98 	return get_ipc_ns(inode->i_sb->s_fs_info);
99 }
100 
101 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
102 {
103 	struct ipc_namespace *ns;
104 
105 	spin_lock(&mq_lock);
106 	ns = __get_ns_from_inode(inode);
107 	spin_unlock(&mq_lock);
108 	return ns;
109 }
110 
111 static struct inode *mqueue_get_inode(struct super_block *sb,
112 		struct ipc_namespace *ipc_ns, umode_t mode,
113 		struct mq_attr *attr)
114 {
115 	struct user_struct *u = current_user();
116 	struct inode *inode;
117 	int ret = -ENOMEM;
118 
119 	inode = new_inode(sb);
120 	if (!inode)
121 		goto err;
122 
123 	inode->i_ino = get_next_ino();
124 	inode->i_mode = mode;
125 	inode->i_uid = current_fsuid();
126 	inode->i_gid = current_fsgid();
127 	inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
128 
129 	if (S_ISREG(mode)) {
130 		struct mqueue_inode_info *info;
131 		struct task_struct *p = current;
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->qsize = 0;
144 		info->user = NULL;	/* set when all is ok */
145 		memset(&info->attr, 0, sizeof(info->attr));
146 		info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
147 		info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
148 		if (attr) {
149 			info->attr.mq_maxmsg = attr->mq_maxmsg;
150 			info->attr.mq_msgsize = attr->mq_msgsize;
151 		}
152 		mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
153 		info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
154 		if (!info->messages)
155 			goto out_inode;
156 
157 		mq_bytes = (mq_msg_tblsz +
158 			(info->attr.mq_maxmsg * info->attr.mq_msgsize));
159 
160 		spin_lock(&mq_lock);
161 		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
162 		    u->mq_bytes + mq_bytes > task_rlimit(p, RLIMIT_MSGQUEUE)) {
163 			spin_unlock(&mq_lock);
164 			/* mqueue_evict_inode() releases info->messages */
165 			ret = -EMFILE;
166 			goto out_inode;
167 		}
168 		u->mq_bytes += mq_bytes;
169 		spin_unlock(&mq_lock);
170 
171 		/* all is ok */
172 		info->user = get_uid(u);
173 	} else if (S_ISDIR(mode)) {
174 		inc_nlink(inode);
175 		/* Some things misbehave if size == 0 on a directory */
176 		inode->i_size = 2 * DIRENT_SIZE;
177 		inode->i_op = &mqueue_dir_inode_operations;
178 		inode->i_fop = &simple_dir_operations;
179 	}
180 
181 	return inode;
182 out_inode:
183 	iput(inode);
184 err:
185 	return ERR_PTR(ret);
186 }
187 
188 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
189 {
190 	struct inode *inode;
191 	struct ipc_namespace *ns = data;
192 	int error;
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,
200 				NULL);
201 	if (IS_ERR(inode)) {
202 		error = PTR_ERR(inode);
203 		goto out;
204 	}
205 
206 	sb->s_root = d_alloc_root(inode);
207 	if (!sb->s_root) {
208 		iput(inode);
209 		error = -ENOMEM;
210 		goto out;
211 	}
212 	error = 0;
213 
214 out:
215 	return error;
216 }
217 
218 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
219 			 int flags, const char *dev_name,
220 			 void *data)
221 {
222 	if (!(flags & MS_KERNMOUNT))
223 		data = current->nsproxy->ipc_ns;
224 	return mount_ns(fs_type, flags, data, mqueue_fill_super);
225 }
226 
227 static void init_once(void *foo)
228 {
229 	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
230 
231 	inode_init_once(&p->vfs_inode);
232 }
233 
234 static struct inode *mqueue_alloc_inode(struct super_block *sb)
235 {
236 	struct mqueue_inode_info *ei;
237 
238 	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
239 	if (!ei)
240 		return NULL;
241 	return &ei->vfs_inode;
242 }
243 
244 static void mqueue_i_callback(struct rcu_head *head)
245 {
246 	struct inode *inode = container_of(head, struct inode, i_rcu);
247 	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
248 }
249 
250 static void mqueue_destroy_inode(struct inode *inode)
251 {
252 	call_rcu(&inode->i_rcu, mqueue_i_callback);
253 }
254 
255 static void mqueue_evict_inode(struct inode *inode)
256 {
257 	struct mqueue_inode_info *info;
258 	struct user_struct *user;
259 	unsigned long mq_bytes;
260 	int i;
261 	struct ipc_namespace *ipc_ns;
262 
263 	end_writeback(inode);
264 
265 	if (S_ISDIR(inode->i_mode))
266 		return;
267 
268 	ipc_ns = get_ns_from_inode(inode);
269 	info = MQUEUE_I(inode);
270 	spin_lock(&info->lock);
271 	for (i = 0; i < info->attr.mq_curmsgs; i++)
272 		free_msg(info->messages[i]);
273 	kfree(info->messages);
274 	spin_unlock(&info->lock);
275 
276 	/* Total amount of bytes accounted for the mqueue */
277 	mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
278 	    + info->attr.mq_msgsize);
279 	user = info->user;
280 	if (user) {
281 		spin_lock(&mq_lock);
282 		user->mq_bytes -= mq_bytes;
283 		/*
284 		 * get_ns_from_inode() ensures that the
285 		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
286 		 * to which we now hold a reference, or it is NULL.
287 		 * We can't put it here under mq_lock, though.
288 		 */
289 		if (ipc_ns)
290 			ipc_ns->mq_queues_count--;
291 		spin_unlock(&mq_lock);
292 		free_uid(user);
293 	}
294 	if (ipc_ns)
295 		put_ipc_ns(ipc_ns);
296 }
297 
298 static int mqueue_create(struct inode *dir, struct dentry *dentry,
299 				umode_t mode, struct nameidata *nd)
300 {
301 	struct inode *inode;
302 	struct mq_attr *attr = dentry->d_fsdata;
303 	int error;
304 	struct ipc_namespace *ipc_ns;
305 
306 	spin_lock(&mq_lock);
307 	ipc_ns = __get_ns_from_inode(dir);
308 	if (!ipc_ns) {
309 		error = -EACCES;
310 		goto out_unlock;
311 	}
312 	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
313 			!capable(CAP_SYS_RESOURCE)) {
314 		error = -ENOSPC;
315 		goto out_unlock;
316 	}
317 	ipc_ns->mq_queues_count++;
318 	spin_unlock(&mq_lock);
319 
320 	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
321 	if (IS_ERR(inode)) {
322 		error = PTR_ERR(inode);
323 		spin_lock(&mq_lock);
324 		ipc_ns->mq_queues_count--;
325 		goto out_unlock;
326 	}
327 
328 	put_ipc_ns(ipc_ns);
329 	dir->i_size += DIRENT_SIZE;
330 	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
331 
332 	d_instantiate(dentry, inode);
333 	dget(dentry);
334 	return 0;
335 out_unlock:
336 	spin_unlock(&mq_lock);
337 	if (ipc_ns)
338 		put_ipc_ns(ipc_ns);
339 	return error;
340 }
341 
342 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
343 {
344   	struct inode *inode = dentry->d_inode;
345 
346 	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
347 	dir->i_size -= DIRENT_SIZE;
348   	drop_nlink(inode);
349   	dput(dentry);
350   	return 0;
351 }
352 
353 /*
354 *	This is routine for system read from queue file.
355 *	To avoid mess with doing here some sort of mq_receive we allow
356 *	to read only queue size & notification info (the only values
357 *	that are interesting from user point of view and aren't accessible
358 *	through std routines)
359 */
360 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
361 				size_t count, loff_t *off)
362 {
363 	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
364 	char buffer[FILENT_SIZE];
365 	ssize_t ret;
366 
367 	spin_lock(&info->lock);
368 	snprintf(buffer, sizeof(buffer),
369 			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
370 			info->qsize,
371 			info->notify_owner ? info->notify.sigev_notify : 0,
372 			(info->notify_owner &&
373 			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
374 				info->notify.sigev_signo : 0,
375 			pid_vnr(info->notify_owner));
376 	spin_unlock(&info->lock);
377 	buffer[sizeof(buffer)-1] = '\0';
378 
379 	ret = simple_read_from_buffer(u_data, count, off, buffer,
380 				strlen(buffer));
381 	if (ret <= 0)
382 		return ret;
383 
384 	filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
385 	return ret;
386 }
387 
388 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
389 {
390 	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
391 
392 	spin_lock(&info->lock);
393 	if (task_tgid(current) == info->notify_owner)
394 		remove_notification(info);
395 
396 	spin_unlock(&info->lock);
397 	return 0;
398 }
399 
400 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
401 {
402 	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
403 	int retval = 0;
404 
405 	poll_wait(filp, &info->wait_q, poll_tab);
406 
407 	spin_lock(&info->lock);
408 	if (info->attr.mq_curmsgs)
409 		retval = POLLIN | POLLRDNORM;
410 
411 	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
412 		retval |= POLLOUT | POLLWRNORM;
413 	spin_unlock(&info->lock);
414 
415 	return retval;
416 }
417 
418 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
419 static void wq_add(struct mqueue_inode_info *info, int sr,
420 			struct ext_wait_queue *ewp)
421 {
422 	struct ext_wait_queue *walk;
423 
424 	ewp->task = current;
425 
426 	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
427 		if (walk->task->static_prio <= current->static_prio) {
428 			list_add_tail(&ewp->list, &walk->list);
429 			return;
430 		}
431 	}
432 	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
433 }
434 
435 /*
436  * Puts current task to sleep. Caller must hold queue lock. After return
437  * lock isn't held.
438  * sr: SEND or RECV
439  */
440 static int wq_sleep(struct mqueue_inode_info *info, int sr,
441 		    ktime_t *timeout, struct ext_wait_queue *ewp)
442 {
443 	int retval;
444 	signed long time;
445 
446 	wq_add(info, sr, ewp);
447 
448 	for (;;) {
449 		set_current_state(TASK_INTERRUPTIBLE);
450 
451 		spin_unlock(&info->lock);
452 		time = schedule_hrtimeout_range_clock(timeout, 0,
453 			HRTIMER_MODE_ABS, CLOCK_REALTIME);
454 
455 		while (ewp->state == STATE_PENDING)
456 			cpu_relax();
457 
458 		if (ewp->state == STATE_READY) {
459 			retval = 0;
460 			goto out;
461 		}
462 		spin_lock(&info->lock);
463 		if (ewp->state == STATE_READY) {
464 			retval = 0;
465 			goto out_unlock;
466 		}
467 		if (signal_pending(current)) {
468 			retval = -ERESTARTSYS;
469 			break;
470 		}
471 		if (time == 0) {
472 			retval = -ETIMEDOUT;
473 			break;
474 		}
475 	}
476 	list_del(&ewp->list);
477 out_unlock:
478 	spin_unlock(&info->lock);
479 out:
480 	return retval;
481 }
482 
483 /*
484  * Returns waiting task that should be serviced first or NULL if none exists
485  */
486 static struct ext_wait_queue *wq_get_first_waiter(
487 		struct mqueue_inode_info *info, int sr)
488 {
489 	struct list_head *ptr;
490 
491 	ptr = info->e_wait_q[sr].list.prev;
492 	if (ptr == &info->e_wait_q[sr].list)
493 		return NULL;
494 	return list_entry(ptr, struct ext_wait_queue, list);
495 }
496 
497 /* Auxiliary functions to manipulate messages' list */
498 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
499 {
500 	int k;
501 
502 	k = info->attr.mq_curmsgs - 1;
503 	while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
504 		info->messages[k + 1] = info->messages[k];
505 		k--;
506 	}
507 	info->attr.mq_curmsgs++;
508 	info->qsize += ptr->m_ts;
509 	info->messages[k + 1] = ptr;
510 }
511 
512 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
513 {
514 	info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
515 	return info->messages[info->attr.mq_curmsgs];
516 }
517 
518 static inline void set_cookie(struct sk_buff *skb, char code)
519 {
520 	((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
521 }
522 
523 /*
524  * The next function is only to split too long sys_mq_timedsend
525  */
526 static void __do_notify(struct mqueue_inode_info *info)
527 {
528 	/* notification
529 	 * invoked when there is registered process and there isn't process
530 	 * waiting synchronously for message AND state of queue changed from
531 	 * empty to not empty. Here we are sure that no one is waiting
532 	 * synchronously. */
533 	if (info->notify_owner &&
534 	    info->attr.mq_curmsgs == 1) {
535 		struct siginfo sig_i;
536 		switch (info->notify.sigev_notify) {
537 		case SIGEV_NONE:
538 			break;
539 		case SIGEV_SIGNAL:
540 			/* sends signal */
541 
542 			sig_i.si_signo = info->notify.sigev_signo;
543 			sig_i.si_errno = 0;
544 			sig_i.si_code = SI_MESGQ;
545 			sig_i.si_value = info->notify.sigev_value;
546 			/* map current pid/uid into info->owner's namespaces */
547 			rcu_read_lock();
548 			sig_i.si_pid = task_tgid_nr_ns(current,
549 						ns_of_pid(info->notify_owner));
550 			sig_i.si_uid = user_ns_map_uid(info->user->user_ns,
551 						current_cred(), current_uid());
552 			rcu_read_unlock();
553 
554 			kill_pid_info(info->notify.sigev_signo,
555 				      &sig_i, info->notify_owner);
556 			break;
557 		case SIGEV_THREAD:
558 			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
559 			netlink_sendskb(info->notify_sock, info->notify_cookie);
560 			break;
561 		}
562 		/* after notification unregisters process */
563 		put_pid(info->notify_owner);
564 		info->notify_owner = NULL;
565 	}
566 	wake_up(&info->wait_q);
567 }
568 
569 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
570 			   ktime_t *expires, struct timespec *ts)
571 {
572 	if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
573 		return -EFAULT;
574 	if (!timespec_valid(ts))
575 		return -EINVAL;
576 
577 	*expires = timespec_to_ktime(*ts);
578 	return 0;
579 }
580 
581 static void remove_notification(struct mqueue_inode_info *info)
582 {
583 	if (info->notify_owner != NULL &&
584 	    info->notify.sigev_notify == SIGEV_THREAD) {
585 		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
586 		netlink_sendskb(info->notify_sock, info->notify_cookie);
587 	}
588 	put_pid(info->notify_owner);
589 	info->notify_owner = NULL;
590 }
591 
592 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
593 {
594 	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
595 		return 0;
596 	if (capable(CAP_SYS_RESOURCE)) {
597 		if (attr->mq_maxmsg > HARD_MSGMAX)
598 			return 0;
599 	} else {
600 		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
601 				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
602 			return 0;
603 	}
604 	/* check for overflow */
605 	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
606 		return 0;
607 	if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
608 	    + sizeof (struct msg_msg *))) <
609 	    (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
610 		return 0;
611 	return 1;
612 }
613 
614 /*
615  * Invoked when creating a new queue via sys_mq_open
616  */
617 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
618 			struct dentry *dentry, int oflag, umode_t mode,
619 			struct mq_attr *attr)
620 {
621 	const struct cred *cred = current_cred();
622 	struct file *result;
623 	int ret;
624 
625 	if (attr) {
626 		if (!mq_attr_ok(ipc_ns, attr)) {
627 			ret = -EINVAL;
628 			goto out;
629 		}
630 		/* store for use during create */
631 		dentry->d_fsdata = attr;
632 	}
633 
634 	mode &= ~current_umask();
635 	ret = mnt_want_write(ipc_ns->mq_mnt);
636 	if (ret)
637 		goto out;
638 	ret = vfs_create(dir->d_inode, dentry, mode, NULL);
639 	dentry->d_fsdata = NULL;
640 	if (ret)
641 		goto out_drop_write;
642 
643 	result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
644 	/*
645 	 * dentry_open() took a persistent mnt_want_write(),
646 	 * so we can now drop this one.
647 	 */
648 	mnt_drop_write(ipc_ns->mq_mnt);
649 	return result;
650 
651 out_drop_write:
652 	mnt_drop_write(ipc_ns->mq_mnt);
653 out:
654 	dput(dentry);
655 	mntput(ipc_ns->mq_mnt);
656 	return ERR_PTR(ret);
657 }
658 
659 /* Opens existing queue */
660 static struct file *do_open(struct ipc_namespace *ipc_ns,
661 				struct dentry *dentry, int oflag)
662 {
663 	int ret;
664 	const struct cred *cred = current_cred();
665 
666 	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
667 						  MAY_READ | MAY_WRITE };
668 
669 	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
670 		ret = -EINVAL;
671 		goto err;
672 	}
673 
674 	if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
675 		ret = -EACCES;
676 		goto err;
677 	}
678 
679 	return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
680 
681 err:
682 	dput(dentry);
683 	mntput(ipc_ns->mq_mnt);
684 	return ERR_PTR(ret);
685 }
686 
687 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
688 		struct mq_attr __user *, u_attr)
689 {
690 	struct dentry *dentry;
691 	struct file *filp;
692 	char *name;
693 	struct mq_attr attr;
694 	int fd, error;
695 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
696 
697 	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
698 		return -EFAULT;
699 
700 	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
701 
702 	if (IS_ERR(name = getname(u_name)))
703 		return PTR_ERR(name);
704 
705 	fd = get_unused_fd_flags(O_CLOEXEC);
706 	if (fd < 0)
707 		goto out_putname;
708 
709 	mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
710 	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
711 	if (IS_ERR(dentry)) {
712 		error = PTR_ERR(dentry);
713 		goto out_putfd;
714 	}
715 	mntget(ipc_ns->mq_mnt);
716 
717 	if (oflag & O_CREAT) {
718 		if (dentry->d_inode) {	/* entry already exists */
719 			audit_inode(name, dentry);
720 			if (oflag & O_EXCL) {
721 				error = -EEXIST;
722 				goto out;
723 			}
724 			filp = do_open(ipc_ns, dentry, oflag);
725 		} else {
726 			filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
727 						dentry, oflag, mode,
728 						u_attr ? &attr : NULL);
729 		}
730 	} else {
731 		if (!dentry->d_inode) {
732 			error = -ENOENT;
733 			goto out;
734 		}
735 		audit_inode(name, dentry);
736 		filp = do_open(ipc_ns, dentry, oflag);
737 	}
738 
739 	if (IS_ERR(filp)) {
740 		error = PTR_ERR(filp);
741 		goto out_putfd;
742 	}
743 
744 	fd_install(fd, filp);
745 	goto out_upsem;
746 
747 out:
748 	dput(dentry);
749 	mntput(ipc_ns->mq_mnt);
750 out_putfd:
751 	put_unused_fd(fd);
752 	fd = error;
753 out_upsem:
754 	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
755 out_putname:
756 	putname(name);
757 	return fd;
758 }
759 
760 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
761 {
762 	int err;
763 	char *name;
764 	struct dentry *dentry;
765 	struct inode *inode = NULL;
766 	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
767 
768 	name = getname(u_name);
769 	if (IS_ERR(name))
770 		return PTR_ERR(name);
771 
772 	mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
773 			I_MUTEX_PARENT);
774 	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
775 	if (IS_ERR(dentry)) {
776 		err = PTR_ERR(dentry);
777 		goto out_unlock;
778 	}
779 
780 	if (!dentry->d_inode) {
781 		err = -ENOENT;
782 		goto out_err;
783 	}
784 
785 	inode = dentry->d_inode;
786 	if (inode)
787 		ihold(inode);
788 	err = mnt_want_write(ipc_ns->mq_mnt);
789 	if (err)
790 		goto out_err;
791 	err = vfs_unlink(dentry->d_parent->d_inode, dentry);
792 	mnt_drop_write(ipc_ns->mq_mnt);
793 out_err:
794 	dput(dentry);
795 
796 out_unlock:
797 	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
798 	putname(name);
799 	if (inode)
800 		iput(inode);
801 
802 	return err;
803 }
804 
805 /* Pipelined send and receive functions.
806  *
807  * If a receiver finds no waiting message, then it registers itself in the
808  * list of waiting receivers. A sender checks that list before adding the new
809  * message into the message array. If there is a waiting receiver, then it
810  * bypasses the message array and directly hands the message over to the
811  * receiver.
812  * The receiver accepts the message and returns without grabbing the queue
813  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
814  * are necessary. The same algorithm is used for sysv semaphores, see
815  * ipc/sem.c for more details.
816  *
817  * The same algorithm is used for senders.
818  */
819 
820 /* pipelined_send() - send a message directly to the task waiting in
821  * sys_mq_timedreceive() (without inserting message into a queue).
822  */
823 static inline void pipelined_send(struct mqueue_inode_info *info,
824 				  struct msg_msg *message,
825 				  struct ext_wait_queue *receiver)
826 {
827 	receiver->msg = message;
828 	list_del(&receiver->list);
829 	receiver->state = STATE_PENDING;
830 	wake_up_process(receiver->task);
831 	smp_wmb();
832 	receiver->state = STATE_READY;
833 }
834 
835 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
836  * gets its message and put to the queue (we have one free place for sure). */
837 static inline void pipelined_receive(struct mqueue_inode_info *info)
838 {
839 	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
840 
841 	if (!sender) {
842 		/* for poll */
843 		wake_up_interruptible(&info->wait_q);
844 		return;
845 	}
846 	msg_insert(sender->msg, info);
847 	list_del(&sender->list);
848 	sender->state = STATE_PENDING;
849 	wake_up_process(sender->task);
850 	smp_wmb();
851 	sender->state = STATE_READY;
852 }
853 
854 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
855 		size_t, msg_len, unsigned int, msg_prio,
856 		const struct timespec __user *, u_abs_timeout)
857 {
858 	struct file *filp;
859 	struct inode *inode;
860 	struct ext_wait_queue wait;
861 	struct ext_wait_queue *receiver;
862 	struct msg_msg *msg_ptr;
863 	struct mqueue_inode_info *info;
864 	ktime_t expires, *timeout = NULL;
865 	struct timespec ts;
866 	int ret;
867 
868 	if (u_abs_timeout) {
869 		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
870 		if (res)
871 			return res;
872 		timeout = &expires;
873 	}
874 
875 	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
876 		return -EINVAL;
877 
878 	audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
879 
880 	filp = fget(mqdes);
881 	if (unlikely(!filp)) {
882 		ret = -EBADF;
883 		goto out;
884 	}
885 
886 	inode = filp->f_path.dentry->d_inode;
887 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
888 		ret = -EBADF;
889 		goto out_fput;
890 	}
891 	info = MQUEUE_I(inode);
892 	audit_inode(NULL, filp->f_path.dentry);
893 
894 	if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
895 		ret = -EBADF;
896 		goto out_fput;
897 	}
898 
899 	if (unlikely(msg_len > info->attr.mq_msgsize)) {
900 		ret = -EMSGSIZE;
901 		goto out_fput;
902 	}
903 
904 	/* First try to allocate memory, before doing anything with
905 	 * existing queues. */
906 	msg_ptr = load_msg(u_msg_ptr, msg_len);
907 	if (IS_ERR(msg_ptr)) {
908 		ret = PTR_ERR(msg_ptr);
909 		goto out_fput;
910 	}
911 	msg_ptr->m_ts = msg_len;
912 	msg_ptr->m_type = msg_prio;
913 
914 	spin_lock(&info->lock);
915 
916 	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
917 		if (filp->f_flags & O_NONBLOCK) {
918 			spin_unlock(&info->lock);
919 			ret = -EAGAIN;
920 		} else {
921 			wait.task = current;
922 			wait.msg = (void *) msg_ptr;
923 			wait.state = STATE_NONE;
924 			ret = wq_sleep(info, SEND, timeout, &wait);
925 		}
926 		if (ret < 0)
927 			free_msg(msg_ptr);
928 	} else {
929 		receiver = wq_get_first_waiter(info, RECV);
930 		if (receiver) {
931 			pipelined_send(info, msg_ptr, receiver);
932 		} else {
933 			/* adds message to the queue */
934 			msg_insert(msg_ptr, info);
935 			__do_notify(info);
936 		}
937 		inode->i_atime = inode->i_mtime = inode->i_ctime =
938 				CURRENT_TIME;
939 		spin_unlock(&info->lock);
940 		ret = 0;
941 	}
942 out_fput:
943 	fput(filp);
944 out:
945 	return ret;
946 }
947 
948 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
949 		size_t, msg_len, unsigned int __user *, u_msg_prio,
950 		const struct timespec __user *, u_abs_timeout)
951 {
952 	ssize_t ret;
953 	struct msg_msg *msg_ptr;
954 	struct file *filp;
955 	struct inode *inode;
956 	struct mqueue_inode_info *info;
957 	struct ext_wait_queue wait;
958 	ktime_t expires, *timeout = NULL;
959 	struct timespec ts;
960 
961 	if (u_abs_timeout) {
962 		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
963 		if (res)
964 			return res;
965 		timeout = &expires;
966 	}
967 
968 	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
969 
970 	filp = fget(mqdes);
971 	if (unlikely(!filp)) {
972 		ret = -EBADF;
973 		goto out;
974 	}
975 
976 	inode = filp->f_path.dentry->d_inode;
977 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
978 		ret = -EBADF;
979 		goto out_fput;
980 	}
981 	info = MQUEUE_I(inode);
982 	audit_inode(NULL, filp->f_path.dentry);
983 
984 	if (unlikely(!(filp->f_mode & FMODE_READ))) {
985 		ret = -EBADF;
986 		goto out_fput;
987 	}
988 
989 	/* checks if buffer is big enough */
990 	if (unlikely(msg_len < info->attr.mq_msgsize)) {
991 		ret = -EMSGSIZE;
992 		goto out_fput;
993 	}
994 
995 	spin_lock(&info->lock);
996 	if (info->attr.mq_curmsgs == 0) {
997 		if (filp->f_flags & O_NONBLOCK) {
998 			spin_unlock(&info->lock);
999 			ret = -EAGAIN;
1000 		} else {
1001 			wait.task = current;
1002 			wait.state = STATE_NONE;
1003 			ret = wq_sleep(info, RECV, timeout, &wait);
1004 			msg_ptr = wait.msg;
1005 		}
1006 	} else {
1007 		msg_ptr = msg_get(info);
1008 
1009 		inode->i_atime = inode->i_mtime = inode->i_ctime =
1010 				CURRENT_TIME;
1011 
1012 		/* There is now free space in queue. */
1013 		pipelined_receive(info);
1014 		spin_unlock(&info->lock);
1015 		ret = 0;
1016 	}
1017 	if (ret == 0) {
1018 		ret = msg_ptr->m_ts;
1019 
1020 		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1021 			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1022 			ret = -EFAULT;
1023 		}
1024 		free_msg(msg_ptr);
1025 	}
1026 out_fput:
1027 	fput(filp);
1028 out:
1029 	return ret;
1030 }
1031 
1032 /*
1033  * Notes: the case when user wants us to deregister (with NULL as pointer)
1034  * and he isn't currently owner of notification, will be silently discarded.
1035  * It isn't explicitly defined in the POSIX.
1036  */
1037 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1038 		const struct sigevent __user *, u_notification)
1039 {
1040 	int ret;
1041 	struct file *filp;
1042 	struct sock *sock;
1043 	struct inode *inode;
1044 	struct sigevent notification;
1045 	struct mqueue_inode_info *info;
1046 	struct sk_buff *nc;
1047 
1048 	if (u_notification) {
1049 		if (copy_from_user(&notification, u_notification,
1050 					sizeof(struct sigevent)))
1051 			return -EFAULT;
1052 	}
1053 
1054 	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1055 
1056 	nc = NULL;
1057 	sock = NULL;
1058 	if (u_notification != NULL) {
1059 		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1060 			     notification.sigev_notify != SIGEV_SIGNAL &&
1061 			     notification.sigev_notify != SIGEV_THREAD))
1062 			return -EINVAL;
1063 		if (notification.sigev_notify == SIGEV_SIGNAL &&
1064 			!valid_signal(notification.sigev_signo)) {
1065 			return -EINVAL;
1066 		}
1067 		if (notification.sigev_notify == SIGEV_THREAD) {
1068 			long timeo;
1069 
1070 			/* create the notify skb */
1071 			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1072 			if (!nc) {
1073 				ret = -ENOMEM;
1074 				goto out;
1075 			}
1076 			if (copy_from_user(nc->data,
1077 					notification.sigev_value.sival_ptr,
1078 					NOTIFY_COOKIE_LEN)) {
1079 				ret = -EFAULT;
1080 				goto out;
1081 			}
1082 
1083 			/* TODO: add a header? */
1084 			skb_put(nc, NOTIFY_COOKIE_LEN);
1085 			/* and attach it to the socket */
1086 retry:
1087 			filp = fget(notification.sigev_signo);
1088 			if (!filp) {
1089 				ret = -EBADF;
1090 				goto out;
1091 			}
1092 			sock = netlink_getsockbyfilp(filp);
1093 			fput(filp);
1094 			if (IS_ERR(sock)) {
1095 				ret = PTR_ERR(sock);
1096 				sock = NULL;
1097 				goto out;
1098 			}
1099 
1100 			timeo = MAX_SCHEDULE_TIMEOUT;
1101 			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1102 			if (ret == 1)
1103 				goto retry;
1104 			if (ret) {
1105 				sock = NULL;
1106 				nc = NULL;
1107 				goto out;
1108 			}
1109 		}
1110 	}
1111 
1112 	filp = fget(mqdes);
1113 	if (!filp) {
1114 		ret = -EBADF;
1115 		goto out;
1116 	}
1117 
1118 	inode = filp->f_path.dentry->d_inode;
1119 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1120 		ret = -EBADF;
1121 		goto out_fput;
1122 	}
1123 	info = MQUEUE_I(inode);
1124 
1125 	ret = 0;
1126 	spin_lock(&info->lock);
1127 	if (u_notification == NULL) {
1128 		if (info->notify_owner == task_tgid(current)) {
1129 			remove_notification(info);
1130 			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1131 		}
1132 	} else if (info->notify_owner != NULL) {
1133 		ret = -EBUSY;
1134 	} else {
1135 		switch (notification.sigev_notify) {
1136 		case SIGEV_NONE:
1137 			info->notify.sigev_notify = SIGEV_NONE;
1138 			break;
1139 		case SIGEV_THREAD:
1140 			info->notify_sock = sock;
1141 			info->notify_cookie = nc;
1142 			sock = NULL;
1143 			nc = NULL;
1144 			info->notify.sigev_notify = SIGEV_THREAD;
1145 			break;
1146 		case SIGEV_SIGNAL:
1147 			info->notify.sigev_signo = notification.sigev_signo;
1148 			info->notify.sigev_value = notification.sigev_value;
1149 			info->notify.sigev_notify = SIGEV_SIGNAL;
1150 			break;
1151 		}
1152 
1153 		info->notify_owner = get_pid(task_tgid(current));
1154 		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1155 	}
1156 	spin_unlock(&info->lock);
1157 out_fput:
1158 	fput(filp);
1159 out:
1160 	if (sock) {
1161 		netlink_detachskb(sock, nc);
1162 	} else if (nc) {
1163 		dev_kfree_skb(nc);
1164 	}
1165 	return ret;
1166 }
1167 
1168 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1169 		const struct mq_attr __user *, u_mqstat,
1170 		struct mq_attr __user *, u_omqstat)
1171 {
1172 	int ret;
1173 	struct mq_attr mqstat, omqstat;
1174 	struct file *filp;
1175 	struct inode *inode;
1176 	struct mqueue_inode_info *info;
1177 
1178 	if (u_mqstat != NULL) {
1179 		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1180 			return -EFAULT;
1181 		if (mqstat.mq_flags & (~O_NONBLOCK))
1182 			return -EINVAL;
1183 	}
1184 
1185 	filp = fget(mqdes);
1186 	if (!filp) {
1187 		ret = -EBADF;
1188 		goto out;
1189 	}
1190 
1191 	inode = filp->f_path.dentry->d_inode;
1192 	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1193 		ret = -EBADF;
1194 		goto out_fput;
1195 	}
1196 	info = MQUEUE_I(inode);
1197 
1198 	spin_lock(&info->lock);
1199 
1200 	omqstat = info->attr;
1201 	omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1202 	if (u_mqstat) {
1203 		audit_mq_getsetattr(mqdes, &mqstat);
1204 		spin_lock(&filp->f_lock);
1205 		if (mqstat.mq_flags & O_NONBLOCK)
1206 			filp->f_flags |= O_NONBLOCK;
1207 		else
1208 			filp->f_flags &= ~O_NONBLOCK;
1209 		spin_unlock(&filp->f_lock);
1210 
1211 		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1212 	}
1213 
1214 	spin_unlock(&info->lock);
1215 
1216 	ret = 0;
1217 	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1218 						sizeof(struct mq_attr)))
1219 		ret = -EFAULT;
1220 
1221 out_fput:
1222 	fput(filp);
1223 out:
1224 	return ret;
1225 }
1226 
1227 static const struct inode_operations mqueue_dir_inode_operations = {
1228 	.lookup = simple_lookup,
1229 	.create = mqueue_create,
1230 	.unlink = mqueue_unlink,
1231 };
1232 
1233 static const struct file_operations mqueue_file_operations = {
1234 	.flush = mqueue_flush_file,
1235 	.poll = mqueue_poll_file,
1236 	.read = mqueue_read_file,
1237 	.llseek = default_llseek,
1238 };
1239 
1240 static const struct super_operations mqueue_super_ops = {
1241 	.alloc_inode = mqueue_alloc_inode,
1242 	.destroy_inode = mqueue_destroy_inode,
1243 	.evict_inode = mqueue_evict_inode,
1244 	.statfs = simple_statfs,
1245 };
1246 
1247 static struct file_system_type mqueue_fs_type = {
1248 	.name = "mqueue",
1249 	.mount = mqueue_mount,
1250 	.kill_sb = kill_litter_super,
1251 };
1252 
1253 int mq_init_ns(struct ipc_namespace *ns)
1254 {
1255 	ns->mq_queues_count  = 0;
1256 	ns->mq_queues_max    = DFLT_QUEUESMAX;
1257 	ns->mq_msg_max       = DFLT_MSGMAX;
1258 	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1259 
1260 	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1261 	if (IS_ERR(ns->mq_mnt)) {
1262 		int err = PTR_ERR(ns->mq_mnt);
1263 		ns->mq_mnt = NULL;
1264 		return err;
1265 	}
1266 	return 0;
1267 }
1268 
1269 void mq_clear_sbinfo(struct ipc_namespace *ns)
1270 {
1271 	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1272 }
1273 
1274 void mq_put_mnt(struct ipc_namespace *ns)
1275 {
1276 	kern_unmount(ns->mq_mnt);
1277 }
1278 
1279 static int __init init_mqueue_fs(void)
1280 {
1281 	int error;
1282 
1283 	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1284 				sizeof(struct mqueue_inode_info), 0,
1285 				SLAB_HWCACHE_ALIGN, init_once);
1286 	if (mqueue_inode_cachep == NULL)
1287 		return -ENOMEM;
1288 
1289 	/* ignore failures - they are not fatal */
1290 	mq_sysctl_table = mq_register_sysctl_table();
1291 
1292 	error = register_filesystem(&mqueue_fs_type);
1293 	if (error)
1294 		goto out_sysctl;
1295 
1296 	spin_lock_init(&mq_lock);
1297 
1298 	error = mq_init_ns(&init_ipc_ns);
1299 	if (error)
1300 		goto out_filesystem;
1301 
1302 	return 0;
1303 
1304 out_filesystem:
1305 	unregister_filesystem(&mqueue_fs_type);
1306 out_sysctl:
1307 	if (mq_sysctl_table)
1308 		unregister_sysctl_table(mq_sysctl_table);
1309 	kmem_cache_destroy(mqueue_inode_cachep);
1310 	return error;
1311 }
1312 
1313 __initcall(init_mqueue_fs);
1314