mqueue.c revision 2c12ea49
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
39#include <net/sock.h>
40#include "util.h"
41
42#define MQUEUE_MAGIC	0x19800202
43#define DIRENT_SIZE	20
44#define FILENT_SIZE	80
45
46#define SEND		0
47#define RECV		1
48
49#define STATE_NONE	0
50#define STATE_PENDING	1
51#define STATE_READY	2
52
53struct posix_msg_tree_node {
54	struct rb_node		rb_node;
55	struct list_head	msg_list;
56	int			priority;
57};
58
59struct ext_wait_queue {		/* queue of sleeping tasks */
60	struct task_struct *task;
61	struct list_head list;
62	struct msg_msg *msg;	/* ptr of loaded message */
63	int state;		/* one of STATE_* values */
64};
65
66struct mqueue_inode_info {
67	spinlock_t lock;
68	struct inode vfs_inode;
69	wait_queue_head_t wait_q;
70
71	struct rb_root msg_tree;
72	struct mq_attr attr;
73
74	struct sigevent notify;
75	struct pid* notify_owner;
76	struct user_namespace *notify_user_ns;
77	struct user_struct *user;	/* user who created, for accounting */
78	struct sock *notify_sock;
79	struct sk_buff *notify_cookie;
80
81	/* for tasks waiting for free space and messages, respectively */
82	struct ext_wait_queue e_wait_q[2];
83
84	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
85};
86
87static const struct inode_operations mqueue_dir_inode_operations;
88static const struct file_operations mqueue_file_operations;
89static const struct super_operations mqueue_super_ops;
90static void remove_notification(struct mqueue_inode_info *info);
91
92static struct kmem_cache *mqueue_inode_cachep;
93
94static struct ctl_table_header * mq_sysctl_table;
95
96static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
97{
98	return container_of(inode, struct mqueue_inode_info, vfs_inode);
99}
100
101/*
102 * This routine should be called with the mq_lock held.
103 */
104static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
105{
106	return get_ipc_ns(inode->i_sb->s_fs_info);
107}
108
109static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
110{
111	struct ipc_namespace *ns;
112
113	spin_lock(&mq_lock);
114	ns = __get_ns_from_inode(inode);
115	spin_unlock(&mq_lock);
116	return ns;
117}
118
119/* Auxiliary functions to manipulate messages' list */
120static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
121{
122	struct rb_node **p, *parent = NULL;
123	struct posix_msg_tree_node *leaf;
124
125	p = &info->msg_tree.rb_node;
126	while (*p) {
127		parent = *p;
128		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
129
130		if (likely(leaf->priority == msg->m_type))
131			goto insert_msg;
132		else if (msg->m_type < leaf->priority)
133			p = &(*p)->rb_left;
134		else
135			p = &(*p)->rb_right;
136	}
137	leaf = kzalloc(sizeof(*leaf), GFP_ATOMIC);
138	if (!leaf)
139		return -ENOMEM;
140	rb_init_node(&leaf->rb_node);
141	INIT_LIST_HEAD(&leaf->msg_list);
142	leaf->priority = msg->m_type;
143	rb_link_node(&leaf->rb_node, parent, p);
144	rb_insert_color(&leaf->rb_node, &info->msg_tree);
145	info->qsize += sizeof(struct posix_msg_tree_node);
146insert_msg:
147	info->attr.mq_curmsgs++;
148	info->qsize += msg->m_ts;
149	list_add_tail(&msg->m_list, &leaf->msg_list);
150	return 0;
151}
152
153static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
154{
155	struct rb_node **p, *parent = NULL;
156	struct posix_msg_tree_node *leaf;
157	struct msg_msg *msg;
158
159try_again:
160	p = &info->msg_tree.rb_node;
161	while (*p) {
162		parent = *p;
163		/*
164		 * During insert, low priorities go to the left and high to the
165		 * right.  On receive, we want the highest priorities first, so
166		 * walk all the way to the right.
167		 */
168		p = &(*p)->rb_right;
169	}
170	if (!parent) {
171		if (info->attr.mq_curmsgs) {
172			pr_warn_once("Inconsistency in POSIX message queue, "
173				     "no tree element, but supposedly messages "
174				     "should exist!\n");
175			info->attr.mq_curmsgs = 0;
176		}
177		return NULL;
178	}
179	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
180	if (list_empty(&leaf->msg_list)) {
181		pr_warn_once("Inconsistency in POSIX message queue, "
182			     "empty leaf node but we haven't implemented "
183			     "lazy leaf delete!\n");
184		rb_erase(&leaf->rb_node, &info->msg_tree);
185		info->qsize -= sizeof(struct posix_msg_tree_node);
186		kfree(leaf);
187		goto try_again;
188	} else {
189		msg = list_first_entry(&leaf->msg_list,
190				       struct msg_msg, m_list);
191		list_del(&msg->m_list);
192		if (list_empty(&leaf->msg_list)) {
193			rb_erase(&leaf->rb_node, &info->msg_tree);
194			info->qsize -= sizeof(struct posix_msg_tree_node);
195			kfree(leaf);
196		}
197	}
198	info->attr.mq_curmsgs--;
199	info->qsize -= msg->m_ts;
200	return msg;
201}
202
203static struct inode *mqueue_get_inode(struct super_block *sb,
204		struct ipc_namespace *ipc_ns, umode_t mode,
205		struct mq_attr *attr)
206{
207	struct user_struct *u = current_user();
208	struct inode *inode;
209	int ret = -ENOMEM;
210
211	inode = new_inode(sb);
212	if (!inode)
213		goto err;
214
215	inode->i_ino = get_next_ino();
216	inode->i_mode = mode;
217	inode->i_uid = current_fsuid();
218	inode->i_gid = current_fsgid();
219	inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
220
221	if (S_ISREG(mode)) {
222		struct mqueue_inode_info *info;
223		unsigned long mq_bytes, mq_treesize;
224
225		inode->i_fop = &mqueue_file_operations;
226		inode->i_size = FILENT_SIZE;
227		/* mqueue specific info */
228		info = MQUEUE_I(inode);
229		spin_lock_init(&info->lock);
230		init_waitqueue_head(&info->wait_q);
231		INIT_LIST_HEAD(&info->e_wait_q[0].list);
232		INIT_LIST_HEAD(&info->e_wait_q[1].list);
233		info->notify_owner = NULL;
234		info->notify_user_ns = NULL;
235		info->qsize = 0;
236		info->user = NULL;	/* set when all is ok */
237		info->msg_tree = RB_ROOT;
238		memset(&info->attr, 0, sizeof(info->attr));
239		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
240					   ipc_ns->mq_msg_default);
241		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
242					    ipc_ns->mq_msgsize_default);
243		if (attr) {
244			info->attr.mq_maxmsg = attr->mq_maxmsg;
245			info->attr.mq_msgsize = attr->mq_msgsize;
246		}
247		/*
248		 * We used to allocate a static array of pointers and account
249		 * the size of that array as well as one msg_msg struct per
250		 * possible message into the queue size. That's no longer
251		 * accurate as the queue is now an rbtree and will grow and
252		 * shrink depending on usage patterns.  We can, however, still
253		 * account one msg_msg struct per message, but the nodes are
254		 * allocated depending on priority usage, and most programs
255		 * only use one, or a handful, of priorities.  However, since
256		 * this is pinned memory, we need to assume worst case, so
257		 * that means the min(mq_maxmsg, max_priorities) * struct
258		 * posix_msg_tree_node.
259		 */
260		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
261			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
262			sizeof(struct posix_msg_tree_node);
263
264		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
265					  info->attr.mq_msgsize);
266
267		spin_lock(&mq_lock);
268		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
269		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
270			spin_unlock(&mq_lock);
271			/* mqueue_evict_inode() releases info->messages */
272			ret = -EMFILE;
273			goto out_inode;
274		}
275		u->mq_bytes += mq_bytes;
276		spin_unlock(&mq_lock);
277
278		/* all is ok */
279		info->user = get_uid(u);
280	} else if (S_ISDIR(mode)) {
281		inc_nlink(inode);
282		/* Some things misbehave if size == 0 on a directory */
283		inode->i_size = 2 * DIRENT_SIZE;
284		inode->i_op = &mqueue_dir_inode_operations;
285		inode->i_fop = &simple_dir_operations;
286	}
287
288	return inode;
289out_inode:
290	iput(inode);
291err:
292	return ERR_PTR(ret);
293}
294
295static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
296{
297	struct inode *inode;
298	struct ipc_namespace *ns = data;
299
300	sb->s_blocksize = PAGE_CACHE_SIZE;
301	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
302	sb->s_magic = MQUEUE_MAGIC;
303	sb->s_op = &mqueue_super_ops;
304
305	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
306	if (IS_ERR(inode))
307		return PTR_ERR(inode);
308
309	sb->s_root = d_make_root(inode);
310	if (!sb->s_root)
311		return -ENOMEM;
312	return 0;
313}
314
315static struct dentry *mqueue_mount(struct file_system_type *fs_type,
316			 int flags, const char *dev_name,
317			 void *data)
318{
319	if (!(flags & MS_KERNMOUNT))
320		data = current->nsproxy->ipc_ns;
321	return mount_ns(fs_type, flags, data, mqueue_fill_super);
322}
323
324static void init_once(void *foo)
325{
326	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
327
328	inode_init_once(&p->vfs_inode);
329}
330
331static struct inode *mqueue_alloc_inode(struct super_block *sb)
332{
333	struct mqueue_inode_info *ei;
334
335	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
336	if (!ei)
337		return NULL;
338	return &ei->vfs_inode;
339}
340
341static void mqueue_i_callback(struct rcu_head *head)
342{
343	struct inode *inode = container_of(head, struct inode, i_rcu);
344	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
345}
346
347static void mqueue_destroy_inode(struct inode *inode)
348{
349	call_rcu(&inode->i_rcu, mqueue_i_callback);
350}
351
352static void mqueue_evict_inode(struct inode *inode)
353{
354	struct mqueue_inode_info *info;
355	struct user_struct *user;
356	unsigned long mq_bytes, mq_treesize;
357	struct ipc_namespace *ipc_ns;
358	struct msg_msg *msg;
359
360	clear_inode(inode);
361
362	if (S_ISDIR(inode->i_mode))
363		return;
364
365	ipc_ns = get_ns_from_inode(inode);
366	info = MQUEUE_I(inode);
367	spin_lock(&info->lock);
368	while ((msg = msg_get(info)) != NULL)
369		free_msg(msg);
370	spin_unlock(&info->lock);
371
372	/* Total amount of bytes accounted for the mqueue */
373	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
374		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
375		sizeof(struct posix_msg_tree_node);
376
377	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
378				  info->attr.mq_msgsize);
379
380	user = info->user;
381	if (user) {
382		spin_lock(&mq_lock);
383		user->mq_bytes -= mq_bytes;
384		/*
385		 * get_ns_from_inode() ensures that the
386		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
387		 * to which we now hold a reference, or it is NULL.
388		 * We can't put it here under mq_lock, though.
389		 */
390		if (ipc_ns)
391			ipc_ns->mq_queues_count--;
392		spin_unlock(&mq_lock);
393		free_uid(user);
394	}
395	if (ipc_ns)
396		put_ipc_ns(ipc_ns);
397}
398
399static int mqueue_create(struct inode *dir, struct dentry *dentry,
400				umode_t mode, struct nameidata *nd)
401{
402	struct inode *inode;
403	struct mq_attr *attr = dentry->d_fsdata;
404	int error;
405	struct ipc_namespace *ipc_ns;
406
407	spin_lock(&mq_lock);
408	ipc_ns = __get_ns_from_inode(dir);
409	if (!ipc_ns) {
410		error = -EACCES;
411		goto out_unlock;
412	}
413	if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX ||
414	    (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
415	     !capable(CAP_SYS_RESOURCE))) {
416		error = -ENOSPC;
417		goto out_unlock;
418	}
419	ipc_ns->mq_queues_count++;
420	spin_unlock(&mq_lock);
421
422	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
423	if (IS_ERR(inode)) {
424		error = PTR_ERR(inode);
425		spin_lock(&mq_lock);
426		ipc_ns->mq_queues_count--;
427		goto out_unlock;
428	}
429
430	put_ipc_ns(ipc_ns);
431	dir->i_size += DIRENT_SIZE;
432	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
433
434	d_instantiate(dentry, inode);
435	dget(dentry);
436	return 0;
437out_unlock:
438	spin_unlock(&mq_lock);
439	if (ipc_ns)
440		put_ipc_ns(ipc_ns);
441	return error;
442}
443
444static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
445{
446  	struct inode *inode = dentry->d_inode;
447
448	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
449	dir->i_size -= DIRENT_SIZE;
450  	drop_nlink(inode);
451  	dput(dentry);
452  	return 0;
453}
454
455/*
456*	This is routine for system read from queue file.
457*	To avoid mess with doing here some sort of mq_receive we allow
458*	to read only queue size & notification info (the only values
459*	that are interesting from user point of view and aren't accessible
460*	through std routines)
461*/
462static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
463				size_t count, loff_t *off)
464{
465	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
466	char buffer[FILENT_SIZE];
467	ssize_t ret;
468
469	spin_lock(&info->lock);
470	snprintf(buffer, sizeof(buffer),
471			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
472			info->qsize,
473			info->notify_owner ? info->notify.sigev_notify : 0,
474			(info->notify_owner &&
475			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
476				info->notify.sigev_signo : 0,
477			pid_vnr(info->notify_owner));
478	spin_unlock(&info->lock);
479	buffer[sizeof(buffer)-1] = '\0';
480
481	ret = simple_read_from_buffer(u_data, count, off, buffer,
482				strlen(buffer));
483	if (ret <= 0)
484		return ret;
485
486	filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
487	return ret;
488}
489
490static int mqueue_flush_file(struct file *filp, fl_owner_t id)
491{
492	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
493
494	spin_lock(&info->lock);
495	if (task_tgid(current) == info->notify_owner)
496		remove_notification(info);
497
498	spin_unlock(&info->lock);
499	return 0;
500}
501
502static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
503{
504	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
505	int retval = 0;
506
507	poll_wait(filp, &info->wait_q, poll_tab);
508
509	spin_lock(&info->lock);
510	if (info->attr.mq_curmsgs)
511		retval = POLLIN | POLLRDNORM;
512
513	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
514		retval |= POLLOUT | POLLWRNORM;
515	spin_unlock(&info->lock);
516
517	return retval;
518}
519
520/* Adds current to info->e_wait_q[sr] before element with smaller prio */
521static void wq_add(struct mqueue_inode_info *info, int sr,
522			struct ext_wait_queue *ewp)
523{
524	struct ext_wait_queue *walk;
525
526	ewp->task = current;
527
528	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
529		if (walk->task->static_prio <= current->static_prio) {
530			list_add_tail(&ewp->list, &walk->list);
531			return;
532		}
533	}
534	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
535}
536
537/*
538 * Puts current task to sleep. Caller must hold queue lock. After return
539 * lock isn't held.
540 * sr: SEND or RECV
541 */
542static int wq_sleep(struct mqueue_inode_info *info, int sr,
543		    ktime_t *timeout, struct ext_wait_queue *ewp)
544{
545	int retval;
546	signed long time;
547
548	wq_add(info, sr, ewp);
549
550	for (;;) {
551		set_current_state(TASK_INTERRUPTIBLE);
552
553		spin_unlock(&info->lock);
554		time = schedule_hrtimeout_range_clock(timeout, 0,
555			HRTIMER_MODE_ABS, CLOCK_REALTIME);
556
557		while (ewp->state == STATE_PENDING)
558			cpu_relax();
559
560		if (ewp->state == STATE_READY) {
561			retval = 0;
562			goto out;
563		}
564		spin_lock(&info->lock);
565		if (ewp->state == STATE_READY) {
566			retval = 0;
567			goto out_unlock;
568		}
569		if (signal_pending(current)) {
570			retval = -ERESTARTSYS;
571			break;
572		}
573		if (time == 0) {
574			retval = -ETIMEDOUT;
575			break;
576		}
577	}
578	list_del(&ewp->list);
579out_unlock:
580	spin_unlock(&info->lock);
581out:
582	return retval;
583}
584
585/*
586 * Returns waiting task that should be serviced first or NULL if none exists
587 */
588static struct ext_wait_queue *wq_get_first_waiter(
589		struct mqueue_inode_info *info, int sr)
590{
591	struct list_head *ptr;
592
593	ptr = info->e_wait_q[sr].list.prev;
594	if (ptr == &info->e_wait_q[sr].list)
595		return NULL;
596	return list_entry(ptr, struct ext_wait_queue, list);
597}
598
599
600static inline void set_cookie(struct sk_buff *skb, char code)
601{
602	((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
603}
604
605/*
606 * The next function is only to split too long sys_mq_timedsend
607 */
608static void __do_notify(struct mqueue_inode_info *info)
609{
610	/* notification
611	 * invoked when there is registered process and there isn't process
612	 * waiting synchronously for message AND state of queue changed from
613	 * empty to not empty. Here we are sure that no one is waiting
614	 * synchronously. */
615	if (info->notify_owner &&
616	    info->attr.mq_curmsgs == 1) {
617		struct siginfo sig_i;
618		switch (info->notify.sigev_notify) {
619		case SIGEV_NONE:
620			break;
621		case SIGEV_SIGNAL:
622			/* sends signal */
623
624			sig_i.si_signo = info->notify.sigev_signo;
625			sig_i.si_errno = 0;
626			sig_i.si_code = SI_MESGQ;
627			sig_i.si_value = info->notify.sigev_value;
628			/* map current pid/uid into info->owner's namespaces */
629			rcu_read_lock();
630			sig_i.si_pid = task_tgid_nr_ns(current,
631						ns_of_pid(info->notify_owner));
632			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
633			rcu_read_unlock();
634
635			kill_pid_info(info->notify.sigev_signo,
636				      &sig_i, info->notify_owner);
637			break;
638		case SIGEV_THREAD:
639			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
640			netlink_sendskb(info->notify_sock, info->notify_cookie);
641			break;
642		}
643		/* after notification unregisters process */
644		put_pid(info->notify_owner);
645		put_user_ns(info->notify_user_ns);
646		info->notify_owner = NULL;
647		info->notify_user_ns = NULL;
648	}
649	wake_up(&info->wait_q);
650}
651
652static int prepare_timeout(const struct timespec __user *u_abs_timeout,
653			   ktime_t *expires, struct timespec *ts)
654{
655	if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
656		return -EFAULT;
657	if (!timespec_valid(ts))
658		return -EINVAL;
659
660	*expires = timespec_to_ktime(*ts);
661	return 0;
662}
663
664static void remove_notification(struct mqueue_inode_info *info)
665{
666	if (info->notify_owner != NULL &&
667	    info->notify.sigev_notify == SIGEV_THREAD) {
668		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
669		netlink_sendskb(info->notify_sock, info->notify_cookie);
670	}
671	put_pid(info->notify_owner);
672	put_user_ns(info->notify_user_ns);
673	info->notify_owner = NULL;
674	info->notify_user_ns = NULL;
675}
676
677static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
678{
679	int mq_treesize;
680	unsigned long total_size;
681
682	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
683		return 0;
684	if (capable(CAP_SYS_RESOURCE)) {
685		if (attr->mq_maxmsg > HARD_MSGMAX ||
686		    attr->mq_msgsize > HARD_MSGSIZEMAX)
687			return 0;
688	} else {
689		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
690				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
691			return 0;
692	}
693	/* check for overflow */
694	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
695		return 0;
696	mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
697		min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
698		sizeof(struct posix_msg_tree_node);
699	total_size = attr->mq_maxmsg * attr->mq_msgsize;
700	if (total_size + mq_treesize < total_size)
701		return 0;
702	return 1;
703}
704
705/*
706 * Invoked when creating a new queue via sys_mq_open
707 */
708static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
709			struct dentry *dentry, int oflag, umode_t mode,
710			struct mq_attr *attr)
711{
712	const struct cred *cred = current_cred();
713	struct file *result;
714	int ret;
715
716	if (attr) {
717		if (!mq_attr_ok(ipc_ns, attr)) {
718			ret = -EINVAL;
719			goto out;
720		}
721		/* store for use during create */
722		dentry->d_fsdata = attr;
723	}
724
725	mode &= ~current_umask();
726	ret = mnt_want_write(ipc_ns->mq_mnt);
727	if (ret)
728		goto out;
729	ret = vfs_create(dir->d_inode, dentry, mode, NULL);
730	dentry->d_fsdata = NULL;
731	if (ret)
732		goto out_drop_write;
733
734	result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
735	/*
736	 * dentry_open() took a persistent mnt_want_write(),
737	 * so we can now drop this one.
738	 */
739	mnt_drop_write(ipc_ns->mq_mnt);
740	return result;
741
742out_drop_write:
743	mnt_drop_write(ipc_ns->mq_mnt);
744out:
745	dput(dentry);
746	mntput(ipc_ns->mq_mnt);
747	return ERR_PTR(ret);
748}
749
750/* Opens existing queue */
751static struct file *do_open(struct ipc_namespace *ipc_ns,
752				struct dentry *dentry, int oflag)
753{
754	int ret;
755	const struct cred *cred = current_cred();
756
757	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
758						  MAY_READ | MAY_WRITE };
759
760	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
761		ret = -EINVAL;
762		goto err;
763	}
764
765	if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
766		ret = -EACCES;
767		goto err;
768	}
769
770	return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
771
772err:
773	dput(dentry);
774	mntput(ipc_ns->mq_mnt);
775	return ERR_PTR(ret);
776}
777
778SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
779		struct mq_attr __user *, u_attr)
780{
781	struct dentry *dentry;
782	struct file *filp;
783	char *name;
784	struct mq_attr attr;
785	int fd, error;
786	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
787
788	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
789		return -EFAULT;
790
791	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
792
793	if (IS_ERR(name = getname(u_name)))
794		return PTR_ERR(name);
795
796	fd = get_unused_fd_flags(O_CLOEXEC);
797	if (fd < 0)
798		goto out_putname;
799
800	mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
801	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
802	if (IS_ERR(dentry)) {
803		error = PTR_ERR(dentry);
804		goto out_putfd;
805	}
806	mntget(ipc_ns->mq_mnt);
807
808	if (oflag & O_CREAT) {
809		if (dentry->d_inode) {	/* entry already exists */
810			audit_inode(name, dentry);
811			if (oflag & O_EXCL) {
812				error = -EEXIST;
813				goto out;
814			}
815			filp = do_open(ipc_ns, dentry, oflag);
816		} else {
817			filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
818						dentry, oflag, mode,
819						u_attr ? &attr : NULL);
820		}
821	} else {
822		if (!dentry->d_inode) {
823			error = -ENOENT;
824			goto out;
825		}
826		audit_inode(name, dentry);
827		filp = do_open(ipc_ns, dentry, oflag);
828	}
829
830	if (IS_ERR(filp)) {
831		error = PTR_ERR(filp);
832		goto out_putfd;
833	}
834
835	fd_install(fd, filp);
836	goto out_upsem;
837
838out:
839	dput(dentry);
840	mntput(ipc_ns->mq_mnt);
841out_putfd:
842	put_unused_fd(fd);
843	fd = error;
844out_upsem:
845	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
846out_putname:
847	putname(name);
848	return fd;
849}
850
851SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
852{
853	int err;
854	char *name;
855	struct dentry *dentry;
856	struct inode *inode = NULL;
857	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
858
859	name = getname(u_name);
860	if (IS_ERR(name))
861		return PTR_ERR(name);
862
863	mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
864			I_MUTEX_PARENT);
865	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
866	if (IS_ERR(dentry)) {
867		err = PTR_ERR(dentry);
868		goto out_unlock;
869	}
870
871	if (!dentry->d_inode) {
872		err = -ENOENT;
873		goto out_err;
874	}
875
876	inode = dentry->d_inode;
877	if (inode)
878		ihold(inode);
879	err = mnt_want_write(ipc_ns->mq_mnt);
880	if (err)
881		goto out_err;
882	err = vfs_unlink(dentry->d_parent->d_inode, dentry);
883	mnt_drop_write(ipc_ns->mq_mnt);
884out_err:
885	dput(dentry);
886
887out_unlock:
888	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
889	putname(name);
890	if (inode)
891		iput(inode);
892
893	return err;
894}
895
896/* Pipelined send and receive functions.
897 *
898 * If a receiver finds no waiting message, then it registers itself in the
899 * list of waiting receivers. A sender checks that list before adding the new
900 * message into the message array. If there is a waiting receiver, then it
901 * bypasses the message array and directly hands the message over to the
902 * receiver.
903 * The receiver accepts the message and returns without grabbing the queue
904 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
905 * are necessary. The same algorithm is used for sysv semaphores, see
906 * ipc/sem.c for more details.
907 *
908 * The same algorithm is used for senders.
909 */
910
911/* pipelined_send() - send a message directly to the task waiting in
912 * sys_mq_timedreceive() (without inserting message into a queue).
913 */
914static inline void pipelined_send(struct mqueue_inode_info *info,
915				  struct msg_msg *message,
916				  struct ext_wait_queue *receiver)
917{
918	receiver->msg = message;
919	list_del(&receiver->list);
920	receiver->state = STATE_PENDING;
921	wake_up_process(receiver->task);
922	smp_wmb();
923	receiver->state = STATE_READY;
924}
925
926/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
927 * gets its message and put to the queue (we have one free place for sure). */
928static inline void pipelined_receive(struct mqueue_inode_info *info)
929{
930	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
931
932	if (!sender) {
933		/* for poll */
934		wake_up_interruptible(&info->wait_q);
935		return;
936	}
937	if (msg_insert(sender->msg, info))
938		return;
939	list_del(&sender->list);
940	sender->state = STATE_PENDING;
941	wake_up_process(sender->task);
942	smp_wmb();
943	sender->state = STATE_READY;
944}
945
946SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
947		size_t, msg_len, unsigned int, msg_prio,
948		const struct timespec __user *, u_abs_timeout)
949{
950	struct file *filp;
951	struct inode *inode;
952	struct ext_wait_queue wait;
953	struct ext_wait_queue *receiver;
954	struct msg_msg *msg_ptr;
955	struct mqueue_inode_info *info;
956	ktime_t expires, *timeout = NULL;
957	struct timespec ts;
958	int ret;
959
960	if (u_abs_timeout) {
961		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
962		if (res)
963			return res;
964		timeout = &expires;
965	}
966
967	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
968		return -EINVAL;
969
970	audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
971
972	filp = fget(mqdes);
973	if (unlikely(!filp)) {
974		ret = -EBADF;
975		goto out;
976	}
977
978	inode = filp->f_path.dentry->d_inode;
979	if (unlikely(filp->f_op != &mqueue_file_operations)) {
980		ret = -EBADF;
981		goto out_fput;
982	}
983	info = MQUEUE_I(inode);
984	audit_inode(NULL, filp->f_path.dentry);
985
986	if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
987		ret = -EBADF;
988		goto out_fput;
989	}
990
991	if (unlikely(msg_len > info->attr.mq_msgsize)) {
992		ret = -EMSGSIZE;
993		goto out_fput;
994	}
995
996	/* First try to allocate memory, before doing anything with
997	 * existing queues. */
998	msg_ptr = load_msg(u_msg_ptr, msg_len);
999	if (IS_ERR(msg_ptr)) {
1000		ret = PTR_ERR(msg_ptr);
1001		goto out_fput;
1002	}
1003	msg_ptr->m_ts = msg_len;
1004	msg_ptr->m_type = msg_prio;
1005
1006	spin_lock(&info->lock);
1007
1008	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1009		if (filp->f_flags & O_NONBLOCK) {
1010			spin_unlock(&info->lock);
1011			ret = -EAGAIN;
1012		} else {
1013			wait.task = current;
1014			wait.msg = (void *) msg_ptr;
1015			wait.state = STATE_NONE;
1016			ret = wq_sleep(info, SEND, timeout, &wait);
1017		}
1018		if (ret < 0)
1019			free_msg(msg_ptr);
1020	} else {
1021		receiver = wq_get_first_waiter(info, RECV);
1022		if (receiver) {
1023			pipelined_send(info, msg_ptr, receiver);
1024		} else {
1025			/* adds message to the queue */
1026			if (msg_insert(msg_ptr, info)) {
1027				free_msg(msg_ptr);
1028				ret = -ENOMEM;
1029				spin_unlock(&info->lock);
1030				goto out_fput;
1031			}
1032			__do_notify(info);
1033		}
1034		inode->i_atime = inode->i_mtime = inode->i_ctime =
1035				CURRENT_TIME;
1036		spin_unlock(&info->lock);
1037		ret = 0;
1038	}
1039out_fput:
1040	fput(filp);
1041out:
1042	return ret;
1043}
1044
1045SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1046		size_t, msg_len, unsigned int __user *, u_msg_prio,
1047		const struct timespec __user *, u_abs_timeout)
1048{
1049	ssize_t ret;
1050	struct msg_msg *msg_ptr;
1051	struct file *filp;
1052	struct inode *inode;
1053	struct mqueue_inode_info *info;
1054	struct ext_wait_queue wait;
1055	ktime_t expires, *timeout = NULL;
1056	struct timespec ts;
1057
1058	if (u_abs_timeout) {
1059		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1060		if (res)
1061			return res;
1062		timeout = &expires;
1063	}
1064
1065	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1066
1067	filp = fget(mqdes);
1068	if (unlikely(!filp)) {
1069		ret = -EBADF;
1070		goto out;
1071	}
1072
1073	inode = filp->f_path.dentry->d_inode;
1074	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1075		ret = -EBADF;
1076		goto out_fput;
1077	}
1078	info = MQUEUE_I(inode);
1079	audit_inode(NULL, filp->f_path.dentry);
1080
1081	if (unlikely(!(filp->f_mode & FMODE_READ))) {
1082		ret = -EBADF;
1083		goto out_fput;
1084	}
1085
1086	/* checks if buffer is big enough */
1087	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1088		ret = -EMSGSIZE;
1089		goto out_fput;
1090	}
1091
1092	spin_lock(&info->lock);
1093	if (info->attr.mq_curmsgs == 0) {
1094		if (filp->f_flags & O_NONBLOCK) {
1095			spin_unlock(&info->lock);
1096			ret = -EAGAIN;
1097		} else {
1098			wait.task = current;
1099			wait.state = STATE_NONE;
1100			ret = wq_sleep(info, RECV, timeout, &wait);
1101			msg_ptr = wait.msg;
1102		}
1103	} else {
1104		msg_ptr = msg_get(info);
1105
1106		inode->i_atime = inode->i_mtime = inode->i_ctime =
1107				CURRENT_TIME;
1108
1109		/* There is now free space in queue. */
1110		pipelined_receive(info);
1111		spin_unlock(&info->lock);
1112		ret = 0;
1113	}
1114	if (ret == 0) {
1115		ret = msg_ptr->m_ts;
1116
1117		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1118			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1119			ret = -EFAULT;
1120		}
1121		free_msg(msg_ptr);
1122	}
1123out_fput:
1124	fput(filp);
1125out:
1126	return ret;
1127}
1128
1129/*
1130 * Notes: the case when user wants us to deregister (with NULL as pointer)
1131 * and he isn't currently owner of notification, will be silently discarded.
1132 * It isn't explicitly defined in the POSIX.
1133 */
1134SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1135		const struct sigevent __user *, u_notification)
1136{
1137	int ret;
1138	struct file *filp;
1139	struct sock *sock;
1140	struct inode *inode;
1141	struct sigevent notification;
1142	struct mqueue_inode_info *info;
1143	struct sk_buff *nc;
1144
1145	if (u_notification) {
1146		if (copy_from_user(&notification, u_notification,
1147					sizeof(struct sigevent)))
1148			return -EFAULT;
1149	}
1150
1151	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1152
1153	nc = NULL;
1154	sock = NULL;
1155	if (u_notification != NULL) {
1156		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1157			     notification.sigev_notify != SIGEV_SIGNAL &&
1158			     notification.sigev_notify != SIGEV_THREAD))
1159			return -EINVAL;
1160		if (notification.sigev_notify == SIGEV_SIGNAL &&
1161			!valid_signal(notification.sigev_signo)) {
1162			return -EINVAL;
1163		}
1164		if (notification.sigev_notify == SIGEV_THREAD) {
1165			long timeo;
1166
1167			/* create the notify skb */
1168			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1169			if (!nc) {
1170				ret = -ENOMEM;
1171				goto out;
1172			}
1173			if (copy_from_user(nc->data,
1174					notification.sigev_value.sival_ptr,
1175					NOTIFY_COOKIE_LEN)) {
1176				ret = -EFAULT;
1177				goto out;
1178			}
1179
1180			/* TODO: add a header? */
1181			skb_put(nc, NOTIFY_COOKIE_LEN);
1182			/* and attach it to the socket */
1183retry:
1184			filp = fget(notification.sigev_signo);
1185			if (!filp) {
1186				ret = -EBADF;
1187				goto out;
1188			}
1189			sock = netlink_getsockbyfilp(filp);
1190			fput(filp);
1191			if (IS_ERR(sock)) {
1192				ret = PTR_ERR(sock);
1193				sock = NULL;
1194				goto out;
1195			}
1196
1197			timeo = MAX_SCHEDULE_TIMEOUT;
1198			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1199			if (ret == 1)
1200				goto retry;
1201			if (ret) {
1202				sock = NULL;
1203				nc = NULL;
1204				goto out;
1205			}
1206		}
1207	}
1208
1209	filp = fget(mqdes);
1210	if (!filp) {
1211		ret = -EBADF;
1212		goto out;
1213	}
1214
1215	inode = filp->f_path.dentry->d_inode;
1216	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1217		ret = -EBADF;
1218		goto out_fput;
1219	}
1220	info = MQUEUE_I(inode);
1221
1222	ret = 0;
1223	spin_lock(&info->lock);
1224	if (u_notification == NULL) {
1225		if (info->notify_owner == task_tgid(current)) {
1226			remove_notification(info);
1227			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1228		}
1229	} else if (info->notify_owner != NULL) {
1230		ret = -EBUSY;
1231	} else {
1232		switch (notification.sigev_notify) {
1233		case SIGEV_NONE:
1234			info->notify.sigev_notify = SIGEV_NONE;
1235			break;
1236		case SIGEV_THREAD:
1237			info->notify_sock = sock;
1238			info->notify_cookie = nc;
1239			sock = NULL;
1240			nc = NULL;
1241			info->notify.sigev_notify = SIGEV_THREAD;
1242			break;
1243		case SIGEV_SIGNAL:
1244			info->notify.sigev_signo = notification.sigev_signo;
1245			info->notify.sigev_value = notification.sigev_value;
1246			info->notify.sigev_notify = SIGEV_SIGNAL;
1247			break;
1248		}
1249
1250		info->notify_owner = get_pid(task_tgid(current));
1251		info->notify_user_ns = get_user_ns(current_user_ns());
1252		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1253	}
1254	spin_unlock(&info->lock);
1255out_fput:
1256	fput(filp);
1257out:
1258	if (sock) {
1259		netlink_detachskb(sock, nc);
1260	} else if (nc) {
1261		dev_kfree_skb(nc);
1262	}
1263	return ret;
1264}
1265
1266SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1267		const struct mq_attr __user *, u_mqstat,
1268		struct mq_attr __user *, u_omqstat)
1269{
1270	int ret;
1271	struct mq_attr mqstat, omqstat;
1272	struct file *filp;
1273	struct inode *inode;
1274	struct mqueue_inode_info *info;
1275
1276	if (u_mqstat != NULL) {
1277		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1278			return -EFAULT;
1279		if (mqstat.mq_flags & (~O_NONBLOCK))
1280			return -EINVAL;
1281	}
1282
1283	filp = fget(mqdes);
1284	if (!filp) {
1285		ret = -EBADF;
1286		goto out;
1287	}
1288
1289	inode = filp->f_path.dentry->d_inode;
1290	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1291		ret = -EBADF;
1292		goto out_fput;
1293	}
1294	info = MQUEUE_I(inode);
1295
1296	spin_lock(&info->lock);
1297
1298	omqstat = info->attr;
1299	omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1300	if (u_mqstat) {
1301		audit_mq_getsetattr(mqdes, &mqstat);
1302		spin_lock(&filp->f_lock);
1303		if (mqstat.mq_flags & O_NONBLOCK)
1304			filp->f_flags |= O_NONBLOCK;
1305		else
1306			filp->f_flags &= ~O_NONBLOCK;
1307		spin_unlock(&filp->f_lock);
1308
1309		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1310	}
1311
1312	spin_unlock(&info->lock);
1313
1314	ret = 0;
1315	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1316						sizeof(struct mq_attr)))
1317		ret = -EFAULT;
1318
1319out_fput:
1320	fput(filp);
1321out:
1322	return ret;
1323}
1324
1325static const struct inode_operations mqueue_dir_inode_operations = {
1326	.lookup = simple_lookup,
1327	.create = mqueue_create,
1328	.unlink = mqueue_unlink,
1329};
1330
1331static const struct file_operations mqueue_file_operations = {
1332	.flush = mqueue_flush_file,
1333	.poll = mqueue_poll_file,
1334	.read = mqueue_read_file,
1335	.llseek = default_llseek,
1336};
1337
1338static const struct super_operations mqueue_super_ops = {
1339	.alloc_inode = mqueue_alloc_inode,
1340	.destroy_inode = mqueue_destroy_inode,
1341	.evict_inode = mqueue_evict_inode,
1342	.statfs = simple_statfs,
1343};
1344
1345static struct file_system_type mqueue_fs_type = {
1346	.name = "mqueue",
1347	.mount = mqueue_mount,
1348	.kill_sb = kill_litter_super,
1349};
1350
1351int mq_init_ns(struct ipc_namespace *ns)
1352{
1353	ns->mq_queues_count  = 0;
1354	ns->mq_queues_max    = DFLT_QUEUESMAX;
1355	ns->mq_msg_max       = DFLT_MSGMAX;
1356	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1357	ns->mq_msg_default   = DFLT_MSG;
1358	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1359
1360	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1361	if (IS_ERR(ns->mq_mnt)) {
1362		int err = PTR_ERR(ns->mq_mnt);
1363		ns->mq_mnt = NULL;
1364		return err;
1365	}
1366	return 0;
1367}
1368
1369void mq_clear_sbinfo(struct ipc_namespace *ns)
1370{
1371	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1372}
1373
1374void mq_put_mnt(struct ipc_namespace *ns)
1375{
1376	kern_unmount(ns->mq_mnt);
1377}
1378
1379static int __init init_mqueue_fs(void)
1380{
1381	int error;
1382
1383	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1384				sizeof(struct mqueue_inode_info), 0,
1385				SLAB_HWCACHE_ALIGN, init_once);
1386	if (mqueue_inode_cachep == NULL)
1387		return -ENOMEM;
1388
1389	/* ignore failures - they are not fatal */
1390	mq_sysctl_table = mq_register_sysctl_table();
1391
1392	error = register_filesystem(&mqueue_fs_type);
1393	if (error)
1394		goto out_sysctl;
1395
1396	spin_lock_init(&mq_lock);
1397
1398	error = mq_init_ns(&init_ipc_ns);
1399	if (error)
1400		goto out_filesystem;
1401
1402	return 0;
1403
1404out_filesystem:
1405	unregister_filesystem(&mqueue_fs_type);
1406out_sysctl:
1407	if (mq_sysctl_table)
1408		unregister_sysctl_table(mq_sysctl_table);
1409	kmem_cache_destroy(mqueue_inode_cachep);
1410	return error;
1411}
1412
1413__initcall(init_mqueue_fs);
1414