mqueue.c revision c8308b1c
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
36#include <net/sock.h>
37#include "util.h"
38
39#define MQUEUE_MAGIC	0x19800202
40#define DIRENT_SIZE	20
41#define FILENT_SIZE	80
42
43#define SEND		0
44#define RECV		1
45
46#define STATE_NONE	0
47#define STATE_PENDING	1
48#define STATE_READY	2
49
50struct ext_wait_queue {		/* queue of sleeping tasks */
51	struct task_struct *task;
52	struct list_head list;
53	struct msg_msg *msg;	/* ptr of loaded message */
54	int state;		/* one of STATE_* values */
55};
56
57struct mqueue_inode_info {
58	spinlock_t lock;
59	struct inode vfs_inode;
60	wait_queue_head_t wait_q;
61
62	struct msg_msg **messages;
63	struct mq_attr attr;
64
65	struct sigevent notify;
66	struct pid* notify_owner;
67	struct user_struct *user;	/* user who created, for accounting */
68	struct sock *notify_sock;
69	struct sk_buff *notify_cookie;
70
71	/* for tasks waiting for free space and messages, respectively */
72	struct ext_wait_queue e_wait_q[2];
73
74	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
75};
76
77static const struct inode_operations mqueue_dir_inode_operations;
78static const struct file_operations mqueue_file_operations;
79static const struct super_operations mqueue_super_ops;
80static void remove_notification(struct mqueue_inode_info *info);
81
82static struct kmem_cache *mqueue_inode_cachep;
83
84static struct ctl_table_header * mq_sysctl_table;
85
86static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
87{
88	return container_of(inode, struct mqueue_inode_info, vfs_inode);
89}
90
91/*
92 * This routine should be called with the mq_lock held.
93 */
94static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
95{
96	return get_ipc_ns(inode->i_sb->s_fs_info);
97}
98
99static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
100{
101	struct ipc_namespace *ns;
102
103	spin_lock(&mq_lock);
104	ns = __get_ns_from_inode(inode);
105	spin_unlock(&mq_lock);
106	return ns;
107}
108
109static struct inode *mqueue_get_inode(struct super_block *sb,
110		struct ipc_namespace *ipc_ns, int mode,
111		struct mq_attr *attr)
112{
113	struct user_struct *u = current_user();
114	struct inode *inode;
115
116	inode = new_inode(sb);
117	if (inode) {
118		inode->i_mode = mode;
119		inode->i_uid = current_fsuid();
120		inode->i_gid = current_fsgid();
121		inode->i_mtime = inode->i_ctime = inode->i_atime =
122				CURRENT_TIME;
123
124		if (S_ISREG(mode)) {
125			struct mqueue_inode_info *info;
126			struct task_struct *p = current;
127			unsigned long mq_bytes, mq_msg_tblsz;
128
129			inode->i_fop = &mqueue_file_operations;
130			inode->i_size = FILENT_SIZE;
131			/* mqueue specific info */
132			info = MQUEUE_I(inode);
133			spin_lock_init(&info->lock);
134			init_waitqueue_head(&info->wait_q);
135			INIT_LIST_HEAD(&info->e_wait_q[0].list);
136			INIT_LIST_HEAD(&info->e_wait_q[1].list);
137			info->notify_owner = NULL;
138			info->qsize = 0;
139			info->user = NULL;	/* set when all is ok */
140			memset(&info->attr, 0, sizeof(info->attr));
141			info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
142			info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
143			if (attr) {
144				info->attr.mq_maxmsg = attr->mq_maxmsg;
145				info->attr.mq_msgsize = attr->mq_msgsize;
146			}
147			mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
148			info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
149			if (!info->messages)
150				goto out_inode;
151
152			mq_bytes = (mq_msg_tblsz +
153				(info->attr.mq_maxmsg * info->attr.mq_msgsize));
154
155			spin_lock(&mq_lock);
156			if (u->mq_bytes + mq_bytes < u->mq_bytes ||
157		 	    u->mq_bytes + mq_bytes >
158			    p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
159				spin_unlock(&mq_lock);
160				kfree(info->messages);
161				goto out_inode;
162			}
163			u->mq_bytes += mq_bytes;
164			spin_unlock(&mq_lock);
165
166			/* all is ok */
167			info->user = get_uid(u);
168		} else if (S_ISDIR(mode)) {
169			inc_nlink(inode);
170			/* Some things misbehave if size == 0 on a directory */
171			inode->i_size = 2 * DIRENT_SIZE;
172			inode->i_op = &mqueue_dir_inode_operations;
173			inode->i_fop = &simple_dir_operations;
174		}
175	}
176	return inode;
177out_inode:
178	make_bad_inode(inode);
179	iput(inode);
180	return NULL;
181}
182
183static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
184{
185	struct inode *inode;
186	struct ipc_namespace *ns = data;
187	int error = 0;
188
189	sb->s_blocksize = PAGE_CACHE_SIZE;
190	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
191	sb->s_magic = MQUEUE_MAGIC;
192	sb->s_op = &mqueue_super_ops;
193
194	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
195				NULL);
196	if (!inode) {
197		error = -ENOMEM;
198		goto out;
199	}
200
201	sb->s_root = d_alloc_root(inode);
202	if (!sb->s_root) {
203		iput(inode);
204		error = -ENOMEM;
205	}
206
207out:
208	return error;
209}
210
211static int mqueue_get_sb(struct file_system_type *fs_type,
212			 int flags, const char *dev_name,
213			 void *data, struct vfsmount *mnt)
214{
215	if (!(flags & MS_KERNMOUNT))
216		data = current->nsproxy->ipc_ns;
217	return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
218}
219
220static void init_once(void *foo)
221{
222	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
223
224	inode_init_once(&p->vfs_inode);
225}
226
227static struct inode *mqueue_alloc_inode(struct super_block *sb)
228{
229	struct mqueue_inode_info *ei;
230
231	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
232	if (!ei)
233		return NULL;
234	return &ei->vfs_inode;
235}
236
237static void mqueue_destroy_inode(struct inode *inode)
238{
239	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
240}
241
242static void mqueue_delete_inode(struct inode *inode)
243{
244	struct mqueue_inode_info *info;
245	struct user_struct *user;
246	unsigned long mq_bytes;
247	int i;
248	struct ipc_namespace *ipc_ns;
249
250	if (S_ISDIR(inode->i_mode)) {
251		clear_inode(inode);
252		return;
253	}
254	ipc_ns = get_ns_from_inode(inode);
255	info = MQUEUE_I(inode);
256	spin_lock(&info->lock);
257	for (i = 0; i < info->attr.mq_curmsgs; i++)
258		free_msg(info->messages[i]);
259	kfree(info->messages);
260	spin_unlock(&info->lock);
261
262	clear_inode(inode);
263
264	mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
265		   (info->attr.mq_maxmsg * info->attr.mq_msgsize));
266	user = info->user;
267	if (user) {
268		spin_lock(&mq_lock);
269		user->mq_bytes -= mq_bytes;
270		/*
271		 * get_ns_from_inode() ensures that the
272		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
273		 * to which we now hold a reference, or it is NULL.
274		 * We can't put it here under mq_lock, though.
275		 */
276		if (ipc_ns)
277			ipc_ns->mq_queues_count--;
278		spin_unlock(&mq_lock);
279		free_uid(user);
280	}
281	if (ipc_ns)
282		put_ipc_ns(ipc_ns);
283}
284
285static int mqueue_create(struct inode *dir, struct dentry *dentry,
286				int mode, struct nameidata *nd)
287{
288	struct inode *inode;
289	struct mq_attr *attr = dentry->d_fsdata;
290	int error;
291	struct ipc_namespace *ipc_ns;
292
293	spin_lock(&mq_lock);
294	ipc_ns = __get_ns_from_inode(dir);
295	if (!ipc_ns) {
296		error = -EACCES;
297		goto out_unlock;
298	}
299	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
300			!capable(CAP_SYS_RESOURCE)) {
301		error = -ENOSPC;
302		goto out_unlock;
303	}
304	ipc_ns->mq_queues_count++;
305	spin_unlock(&mq_lock);
306
307	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
308	if (!inode) {
309		error = -ENOMEM;
310		spin_lock(&mq_lock);
311		ipc_ns->mq_queues_count--;
312		goto out_unlock;
313	}
314
315	put_ipc_ns(ipc_ns);
316	dir->i_size += DIRENT_SIZE;
317	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
318
319	d_instantiate(dentry, inode);
320	dget(dentry);
321	return 0;
322out_unlock:
323	spin_unlock(&mq_lock);
324	if (ipc_ns)
325		put_ipc_ns(ipc_ns);
326	return error;
327}
328
329static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
330{
331  	struct inode *inode = dentry->d_inode;
332
333	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
334	dir->i_size -= DIRENT_SIZE;
335  	drop_nlink(inode);
336  	dput(dentry);
337  	return 0;
338}
339
340/*
341*	This is routine for system read from queue file.
342*	To avoid mess with doing here some sort of mq_receive we allow
343*	to read only queue size & notification info (the only values
344*	that are interesting from user point of view and aren't accessible
345*	through std routines)
346*/
347static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
348				size_t count, loff_t *off)
349{
350	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
351	char buffer[FILENT_SIZE];
352	ssize_t ret;
353
354	spin_lock(&info->lock);
355	snprintf(buffer, sizeof(buffer),
356			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
357			info->qsize,
358			info->notify_owner ? info->notify.sigev_notify : 0,
359			(info->notify_owner &&
360			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
361				info->notify.sigev_signo : 0,
362			pid_vnr(info->notify_owner));
363	spin_unlock(&info->lock);
364	buffer[sizeof(buffer)-1] = '\0';
365
366	ret = simple_read_from_buffer(u_data, count, off, buffer,
367				strlen(buffer));
368	if (ret <= 0)
369		return ret;
370
371	filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
372	return ret;
373}
374
375static int mqueue_flush_file(struct file *filp, fl_owner_t id)
376{
377	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
378
379	spin_lock(&info->lock);
380	if (task_tgid(current) == info->notify_owner)
381		remove_notification(info);
382
383	spin_unlock(&info->lock);
384	return 0;
385}
386
387static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
388{
389	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
390	int retval = 0;
391
392	poll_wait(filp, &info->wait_q, poll_tab);
393
394	spin_lock(&info->lock);
395	if (info->attr.mq_curmsgs)
396		retval = POLLIN | POLLRDNORM;
397
398	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
399		retval |= POLLOUT | POLLWRNORM;
400	spin_unlock(&info->lock);
401
402	return retval;
403}
404
405/* Adds current to info->e_wait_q[sr] before element with smaller prio */
406static void wq_add(struct mqueue_inode_info *info, int sr,
407			struct ext_wait_queue *ewp)
408{
409	struct ext_wait_queue *walk;
410
411	ewp->task = current;
412
413	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
414		if (walk->task->static_prio <= current->static_prio) {
415			list_add_tail(&ewp->list, &walk->list);
416			return;
417		}
418	}
419	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
420}
421
422/*
423 * Puts current task to sleep. Caller must hold queue lock. After return
424 * lock isn't held.
425 * sr: SEND or RECV
426 */
427static int wq_sleep(struct mqueue_inode_info *info, int sr,
428			long timeout, struct ext_wait_queue *ewp)
429{
430	int retval;
431	signed long time;
432
433	wq_add(info, sr, ewp);
434
435	for (;;) {
436		set_current_state(TASK_INTERRUPTIBLE);
437
438		spin_unlock(&info->lock);
439		time = schedule_timeout(timeout);
440
441		while (ewp->state == STATE_PENDING)
442			cpu_relax();
443
444		if (ewp->state == STATE_READY) {
445			retval = 0;
446			goto out;
447		}
448		spin_lock(&info->lock);
449		if (ewp->state == STATE_READY) {
450			retval = 0;
451			goto out_unlock;
452		}
453		if (signal_pending(current)) {
454			retval = -ERESTARTSYS;
455			break;
456		}
457		if (time == 0) {
458			retval = -ETIMEDOUT;
459			break;
460		}
461	}
462	list_del(&ewp->list);
463out_unlock:
464	spin_unlock(&info->lock);
465out:
466	return retval;
467}
468
469/*
470 * Returns waiting task that should be serviced first or NULL if none exists
471 */
472static struct ext_wait_queue *wq_get_first_waiter(
473		struct mqueue_inode_info *info, int sr)
474{
475	struct list_head *ptr;
476
477	ptr = info->e_wait_q[sr].list.prev;
478	if (ptr == &info->e_wait_q[sr].list)
479		return NULL;
480	return list_entry(ptr, struct ext_wait_queue, list);
481}
482
483/* Auxiliary functions to manipulate messages' list */
484static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
485{
486	int k;
487
488	k = info->attr.mq_curmsgs - 1;
489	while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
490		info->messages[k + 1] = info->messages[k];
491		k--;
492	}
493	info->attr.mq_curmsgs++;
494	info->qsize += ptr->m_ts;
495	info->messages[k + 1] = ptr;
496}
497
498static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
499{
500	info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
501	return info->messages[info->attr.mq_curmsgs];
502}
503
504static inline void set_cookie(struct sk_buff *skb, char code)
505{
506	((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
507}
508
509/*
510 * The next function is only to split too long sys_mq_timedsend
511 */
512static void __do_notify(struct mqueue_inode_info *info)
513{
514	/* notification
515	 * invoked when there is registered process and there isn't process
516	 * waiting synchronously for message AND state of queue changed from
517	 * empty to not empty. Here we are sure that no one is waiting
518	 * synchronously. */
519	if (info->notify_owner &&
520	    info->attr.mq_curmsgs == 1) {
521		struct siginfo sig_i;
522		switch (info->notify.sigev_notify) {
523		case SIGEV_NONE:
524			break;
525		case SIGEV_SIGNAL:
526			/* sends signal */
527
528			sig_i.si_signo = info->notify.sigev_signo;
529			sig_i.si_errno = 0;
530			sig_i.si_code = SI_MESGQ;
531			sig_i.si_value = info->notify.sigev_value;
532			sig_i.si_pid = task_tgid_nr_ns(current,
533						ns_of_pid(info->notify_owner));
534			sig_i.si_uid = current_uid();
535
536			kill_pid_info(info->notify.sigev_signo,
537				      &sig_i, info->notify_owner);
538			break;
539		case SIGEV_THREAD:
540			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
541			netlink_sendskb(info->notify_sock, info->notify_cookie);
542			break;
543		}
544		/* after notification unregisters process */
545		put_pid(info->notify_owner);
546		info->notify_owner = NULL;
547	}
548	wake_up(&info->wait_q);
549}
550
551static long prepare_timeout(struct timespec *p)
552{
553	struct timespec nowts;
554	long timeout;
555
556	if (p) {
557		if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
558			|| p->tv_nsec >= NSEC_PER_SEC))
559			return -EINVAL;
560		nowts = CURRENT_TIME;
561		/* first subtract as jiffies can't be too big */
562		p->tv_sec -= nowts.tv_sec;
563		if (p->tv_nsec < nowts.tv_nsec) {
564			p->tv_nsec += NSEC_PER_SEC;
565			p->tv_sec--;
566		}
567		p->tv_nsec -= nowts.tv_nsec;
568		if (p->tv_sec < 0)
569			return 0;
570
571		timeout = timespec_to_jiffies(p) + 1;
572	} else
573		return MAX_SCHEDULE_TIMEOUT;
574
575	return timeout;
576}
577
578static void remove_notification(struct mqueue_inode_info *info)
579{
580	if (info->notify_owner != NULL &&
581	    info->notify.sigev_notify == SIGEV_THREAD) {
582		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
583		netlink_sendskb(info->notify_sock, info->notify_cookie);
584	}
585	put_pid(info->notify_owner);
586	info->notify_owner = NULL;
587}
588
589static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
590{
591	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
592		return 0;
593	if (capable(CAP_SYS_RESOURCE)) {
594		if (attr->mq_maxmsg > HARD_MSGMAX)
595			return 0;
596	} else {
597		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
598				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
599			return 0;
600	}
601	/* check for overflow */
602	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
603		return 0;
604	if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
605	    (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
606	    (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
607		return 0;
608	return 1;
609}
610
611/*
612 * Invoked when creating a new queue via sys_mq_open
613 */
614static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
615			struct dentry *dentry, int oflag, mode_t mode,
616			struct mq_attr *attr)
617{
618	const struct cred *cred = current_cred();
619	struct file *result;
620	int ret;
621
622	if (attr) {
623		ret = -EINVAL;
624		if (!mq_attr_ok(ipc_ns, attr))
625			goto out;
626		/* store for use during create */
627		dentry->d_fsdata = attr;
628	}
629
630	mode &= ~current_umask();
631	ret = mnt_want_write(ipc_ns->mq_mnt);
632	if (ret)
633		goto out;
634	ret = vfs_create(dir->d_inode, dentry, mode, NULL);
635	dentry->d_fsdata = NULL;
636	if (ret)
637		goto out_drop_write;
638
639	result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
640	/*
641	 * dentry_open() took a persistent mnt_want_write(),
642	 * so we can now drop this one.
643	 */
644	mnt_drop_write(ipc_ns->mq_mnt);
645	return result;
646
647out_drop_write:
648	mnt_drop_write(ipc_ns->mq_mnt);
649out:
650	dput(dentry);
651	mntput(ipc_ns->mq_mnt);
652	return ERR_PTR(ret);
653}
654
655/* Opens existing queue */
656static struct file *do_open(struct ipc_namespace *ipc_ns,
657				struct dentry *dentry, int oflag)
658{
659	const struct cred *cred = current_cred();
660
661	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
662						  MAY_READ | MAY_WRITE };
663
664	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
665		dput(dentry);
666		mntput(ipc_ns->mq_mnt);
667		return ERR_PTR(-EINVAL);
668	}
669
670	if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
671		dput(dentry);
672		mntput(ipc_ns->mq_mnt);
673		return ERR_PTR(-EACCES);
674	}
675
676	return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
677}
678
679SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
680		struct mq_attr __user *, u_attr)
681{
682	struct dentry *dentry;
683	struct file *filp;
684	char *name;
685	struct mq_attr attr;
686	int fd, error;
687	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
688
689	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
690		return -EFAULT;
691
692	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
693
694	if (IS_ERR(name = getname(u_name)))
695		return PTR_ERR(name);
696
697	fd = get_unused_fd_flags(O_CLOEXEC);
698	if (fd < 0)
699		goto out_putname;
700
701	mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
702	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
703	if (IS_ERR(dentry)) {
704		error = PTR_ERR(dentry);
705		goto out_putfd;
706	}
707	mntget(ipc_ns->mq_mnt);
708
709	if (oflag & O_CREAT) {
710		if (dentry->d_inode) {	/* entry already exists */
711			audit_inode(name, dentry);
712			error = -EEXIST;
713			if (oflag & O_EXCL)
714				goto out;
715			filp = do_open(ipc_ns, dentry, oflag);
716		} else {
717			filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
718						dentry, oflag, mode,
719						u_attr ? &attr : NULL);
720		}
721	} else {
722		error = -ENOENT;
723		if (!dentry->d_inode)
724			goto out;
725		audit_inode(name, dentry);
726		filp = do_open(ipc_ns, dentry, oflag);
727	}
728
729	if (IS_ERR(filp)) {
730		error = PTR_ERR(filp);
731		goto out_putfd;
732	}
733
734	fd_install(fd, filp);
735	goto out_upsem;
736
737out:
738	dput(dentry);
739	mntput(ipc_ns->mq_mnt);
740out_putfd:
741	put_unused_fd(fd);
742	fd = error;
743out_upsem:
744	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
745out_putname:
746	putname(name);
747	return fd;
748}
749
750SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
751{
752	int err;
753	char *name;
754	struct dentry *dentry;
755	struct inode *inode = NULL;
756	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
757
758	name = getname(u_name);
759	if (IS_ERR(name))
760		return PTR_ERR(name);
761
762	mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
763			I_MUTEX_PARENT);
764	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
765	if (IS_ERR(dentry)) {
766		err = PTR_ERR(dentry);
767		goto out_unlock;
768	}
769
770	if (!dentry->d_inode) {
771		err = -ENOENT;
772		goto out_err;
773	}
774
775	inode = dentry->d_inode;
776	if (inode)
777		atomic_inc(&inode->i_count);
778	err = mnt_want_write(ipc_ns->mq_mnt);
779	if (err)
780		goto out_err;
781	err = vfs_unlink(dentry->d_parent->d_inode, dentry);
782	mnt_drop_write(ipc_ns->mq_mnt);
783out_err:
784	dput(dentry);
785
786out_unlock:
787	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
788	putname(name);
789	if (inode)
790		iput(inode);
791
792	return err;
793}
794
795/* Pipelined send and receive functions.
796 *
797 * If a receiver finds no waiting message, then it registers itself in the
798 * list of waiting receivers. A sender checks that list before adding the new
799 * message into the message array. If there is a waiting receiver, then it
800 * bypasses the message array and directly hands the message over to the
801 * receiver.
802 * The receiver accepts the message and returns without grabbing the queue
803 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
804 * are necessary. The same algorithm is used for sysv semaphores, see
805 * ipc/sem.c for more details.
806 *
807 * The same algorithm is used for senders.
808 */
809
810/* pipelined_send() - send a message directly to the task waiting in
811 * sys_mq_timedreceive() (without inserting message into a queue).
812 */
813static inline void pipelined_send(struct mqueue_inode_info *info,
814				  struct msg_msg *message,
815				  struct ext_wait_queue *receiver)
816{
817	receiver->msg = message;
818	list_del(&receiver->list);
819	receiver->state = STATE_PENDING;
820	wake_up_process(receiver->task);
821	smp_wmb();
822	receiver->state = STATE_READY;
823}
824
825/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
826 * gets its message and put to the queue (we have one free place for sure). */
827static inline void pipelined_receive(struct mqueue_inode_info *info)
828{
829	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
830
831	if (!sender) {
832		/* for poll */
833		wake_up_interruptible(&info->wait_q);
834		return;
835	}
836	msg_insert(sender->msg, info);
837	list_del(&sender->list);
838	sender->state = STATE_PENDING;
839	wake_up_process(sender->task);
840	smp_wmb();
841	sender->state = STATE_READY;
842}
843
844SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
845		size_t, msg_len, unsigned int, msg_prio,
846		const struct timespec __user *, u_abs_timeout)
847{
848	struct file *filp;
849	struct inode *inode;
850	struct ext_wait_queue wait;
851	struct ext_wait_queue *receiver;
852	struct msg_msg *msg_ptr;
853	struct mqueue_inode_info *info;
854	struct timespec ts, *p = NULL;
855	long timeout;
856	int ret;
857
858	if (u_abs_timeout) {
859		if (copy_from_user(&ts, u_abs_timeout,
860					sizeof(struct timespec)))
861			return -EFAULT;
862		p = &ts;
863	}
864
865	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
866		return -EINVAL;
867
868	audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
869	timeout = prepare_timeout(p);
870
871	ret = -EBADF;
872	filp = fget(mqdes);
873	if (unlikely(!filp))
874		goto out;
875
876	inode = filp->f_path.dentry->d_inode;
877	if (unlikely(filp->f_op != &mqueue_file_operations))
878		goto out_fput;
879	info = MQUEUE_I(inode);
880	audit_inode(NULL, filp->f_path.dentry);
881
882	if (unlikely(!(filp->f_mode & FMODE_WRITE)))
883		goto out_fput;
884
885	if (unlikely(msg_len > info->attr.mq_msgsize)) {
886		ret = -EMSGSIZE;
887		goto out_fput;
888	}
889
890	/* First try to allocate memory, before doing anything with
891	 * existing queues. */
892	msg_ptr = load_msg(u_msg_ptr, msg_len);
893	if (IS_ERR(msg_ptr)) {
894		ret = PTR_ERR(msg_ptr);
895		goto out_fput;
896	}
897	msg_ptr->m_ts = msg_len;
898	msg_ptr->m_type = msg_prio;
899
900	spin_lock(&info->lock);
901
902	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
903		if (filp->f_flags & O_NONBLOCK) {
904			spin_unlock(&info->lock);
905			ret = -EAGAIN;
906		} else if (unlikely(timeout < 0)) {
907			spin_unlock(&info->lock);
908			ret = timeout;
909		} else {
910			wait.task = current;
911			wait.msg = (void *) msg_ptr;
912			wait.state = STATE_NONE;
913			ret = wq_sleep(info, SEND, timeout, &wait);
914		}
915		if (ret < 0)
916			free_msg(msg_ptr);
917	} else {
918		receiver = wq_get_first_waiter(info, RECV);
919		if (receiver) {
920			pipelined_send(info, msg_ptr, receiver);
921		} else {
922			/* adds message to the queue */
923			msg_insert(msg_ptr, info);
924			__do_notify(info);
925		}
926		inode->i_atime = inode->i_mtime = inode->i_ctime =
927				CURRENT_TIME;
928		spin_unlock(&info->lock);
929		ret = 0;
930	}
931out_fput:
932	fput(filp);
933out:
934	return ret;
935}
936
937SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
938		size_t, msg_len, unsigned int __user *, u_msg_prio,
939		const struct timespec __user *, u_abs_timeout)
940{
941	long timeout;
942	ssize_t ret;
943	struct msg_msg *msg_ptr;
944	struct file *filp;
945	struct inode *inode;
946	struct mqueue_inode_info *info;
947	struct ext_wait_queue wait;
948	struct timespec ts, *p = NULL;
949
950	if (u_abs_timeout) {
951		if (copy_from_user(&ts, u_abs_timeout,
952					sizeof(struct timespec)))
953			return -EFAULT;
954		p = &ts;
955	}
956
957	audit_mq_sendrecv(mqdes, msg_len, 0, p);
958	timeout = prepare_timeout(p);
959
960	ret = -EBADF;
961	filp = fget(mqdes);
962	if (unlikely(!filp))
963		goto out;
964
965	inode = filp->f_path.dentry->d_inode;
966	if (unlikely(filp->f_op != &mqueue_file_operations))
967		goto out_fput;
968	info = MQUEUE_I(inode);
969	audit_inode(NULL, filp->f_path.dentry);
970
971	if (unlikely(!(filp->f_mode & FMODE_READ)))
972		goto out_fput;
973
974	/* checks if buffer is big enough */
975	if (unlikely(msg_len < info->attr.mq_msgsize)) {
976		ret = -EMSGSIZE;
977		goto out_fput;
978	}
979
980	spin_lock(&info->lock);
981	if (info->attr.mq_curmsgs == 0) {
982		if (filp->f_flags & O_NONBLOCK) {
983			spin_unlock(&info->lock);
984			ret = -EAGAIN;
985			msg_ptr = NULL;
986		} else if (unlikely(timeout < 0)) {
987			spin_unlock(&info->lock);
988			ret = timeout;
989			msg_ptr = NULL;
990		} else {
991			wait.task = current;
992			wait.state = STATE_NONE;
993			ret = wq_sleep(info, RECV, timeout, &wait);
994			msg_ptr = wait.msg;
995		}
996	} else {
997		msg_ptr = msg_get(info);
998
999		inode->i_atime = inode->i_mtime = inode->i_ctime =
1000				CURRENT_TIME;
1001
1002		/* There is now free space in queue. */
1003		pipelined_receive(info);
1004		spin_unlock(&info->lock);
1005		ret = 0;
1006	}
1007	if (ret == 0) {
1008		ret = msg_ptr->m_ts;
1009
1010		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1011			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1012			ret = -EFAULT;
1013		}
1014		free_msg(msg_ptr);
1015	}
1016out_fput:
1017	fput(filp);
1018out:
1019	return ret;
1020}
1021
1022/*
1023 * Notes: the case when user wants us to deregister (with NULL as pointer)
1024 * and he isn't currently owner of notification, will be silently discarded.
1025 * It isn't explicitly defined in the POSIX.
1026 */
1027SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1028		const struct sigevent __user *, u_notification)
1029{
1030	int ret;
1031	struct file *filp;
1032	struct sock *sock;
1033	struct inode *inode;
1034	struct sigevent notification;
1035	struct mqueue_inode_info *info;
1036	struct sk_buff *nc;
1037
1038	if (u_notification) {
1039		if (copy_from_user(&notification, u_notification,
1040					sizeof(struct sigevent)))
1041			return -EFAULT;
1042	}
1043
1044	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1045
1046	nc = NULL;
1047	sock = NULL;
1048	if (u_notification != NULL) {
1049		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1050			     notification.sigev_notify != SIGEV_SIGNAL &&
1051			     notification.sigev_notify != SIGEV_THREAD))
1052			return -EINVAL;
1053		if (notification.sigev_notify == SIGEV_SIGNAL &&
1054			!valid_signal(notification.sigev_signo)) {
1055			return -EINVAL;
1056		}
1057		if (notification.sigev_notify == SIGEV_THREAD) {
1058			long timeo;
1059
1060			/* create the notify skb */
1061			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1062			ret = -ENOMEM;
1063			if (!nc)
1064				goto out;
1065			ret = -EFAULT;
1066			if (copy_from_user(nc->data,
1067					notification.sigev_value.sival_ptr,
1068					NOTIFY_COOKIE_LEN)) {
1069				goto out;
1070			}
1071
1072			/* TODO: add a header? */
1073			skb_put(nc, NOTIFY_COOKIE_LEN);
1074			/* and attach it to the socket */
1075retry:
1076			filp = fget(notification.sigev_signo);
1077			ret = -EBADF;
1078			if (!filp)
1079				goto out;
1080			sock = netlink_getsockbyfilp(filp);
1081			fput(filp);
1082			if (IS_ERR(sock)) {
1083				ret = PTR_ERR(sock);
1084				sock = NULL;
1085				goto out;
1086			}
1087
1088			timeo = MAX_SCHEDULE_TIMEOUT;
1089			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1090			if (ret == 1)
1091		       		goto retry;
1092			if (ret) {
1093				sock = NULL;
1094				nc = NULL;
1095				goto out;
1096			}
1097		}
1098	}
1099
1100	ret = -EBADF;
1101	filp = fget(mqdes);
1102	if (!filp)
1103		goto out;
1104
1105	inode = filp->f_path.dentry->d_inode;
1106	if (unlikely(filp->f_op != &mqueue_file_operations))
1107		goto out_fput;
1108	info = MQUEUE_I(inode);
1109
1110	ret = 0;
1111	spin_lock(&info->lock);
1112	if (u_notification == NULL) {
1113		if (info->notify_owner == task_tgid(current)) {
1114			remove_notification(info);
1115			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1116		}
1117	} else if (info->notify_owner != NULL) {
1118		ret = -EBUSY;
1119	} else {
1120		switch (notification.sigev_notify) {
1121		case SIGEV_NONE:
1122			info->notify.sigev_notify = SIGEV_NONE;
1123			break;
1124		case SIGEV_THREAD:
1125			info->notify_sock = sock;
1126			info->notify_cookie = nc;
1127			sock = NULL;
1128			nc = NULL;
1129			info->notify.sigev_notify = SIGEV_THREAD;
1130			break;
1131		case SIGEV_SIGNAL:
1132			info->notify.sigev_signo = notification.sigev_signo;
1133			info->notify.sigev_value = notification.sigev_value;
1134			info->notify.sigev_notify = SIGEV_SIGNAL;
1135			break;
1136		}
1137
1138		info->notify_owner = get_pid(task_tgid(current));
1139		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1140	}
1141	spin_unlock(&info->lock);
1142out_fput:
1143	fput(filp);
1144out:
1145	if (sock) {
1146		netlink_detachskb(sock, nc);
1147	} else if (nc) {
1148		dev_kfree_skb(nc);
1149	}
1150	return ret;
1151}
1152
1153SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1154		const struct mq_attr __user *, u_mqstat,
1155		struct mq_attr __user *, u_omqstat)
1156{
1157	int ret;
1158	struct mq_attr mqstat, omqstat;
1159	struct file *filp;
1160	struct inode *inode;
1161	struct mqueue_inode_info *info;
1162
1163	if (u_mqstat != NULL) {
1164		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1165			return -EFAULT;
1166		if (mqstat.mq_flags & (~O_NONBLOCK))
1167			return -EINVAL;
1168	}
1169
1170	ret = -EBADF;
1171	filp = fget(mqdes);
1172	if (!filp)
1173		goto out;
1174
1175	inode = filp->f_path.dentry->d_inode;
1176	if (unlikely(filp->f_op != &mqueue_file_operations))
1177		goto out_fput;
1178	info = MQUEUE_I(inode);
1179
1180	spin_lock(&info->lock);
1181
1182	omqstat = info->attr;
1183	omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1184	if (u_mqstat) {
1185		audit_mq_getsetattr(mqdes, &mqstat);
1186		spin_lock(&filp->f_lock);
1187		if (mqstat.mq_flags & O_NONBLOCK)
1188			filp->f_flags |= O_NONBLOCK;
1189		else
1190			filp->f_flags &= ~O_NONBLOCK;
1191		spin_unlock(&filp->f_lock);
1192
1193		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1194	}
1195
1196	spin_unlock(&info->lock);
1197
1198	ret = 0;
1199	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1200						sizeof(struct mq_attr)))
1201		ret = -EFAULT;
1202
1203out_fput:
1204	fput(filp);
1205out:
1206	return ret;
1207}
1208
1209static const struct inode_operations mqueue_dir_inode_operations = {
1210	.lookup = simple_lookup,
1211	.create = mqueue_create,
1212	.unlink = mqueue_unlink,
1213};
1214
1215static const struct file_operations mqueue_file_operations = {
1216	.flush = mqueue_flush_file,
1217	.poll = mqueue_poll_file,
1218	.read = mqueue_read_file,
1219};
1220
1221static const struct super_operations mqueue_super_ops = {
1222	.alloc_inode = mqueue_alloc_inode,
1223	.destroy_inode = mqueue_destroy_inode,
1224	.statfs = simple_statfs,
1225	.delete_inode = mqueue_delete_inode,
1226	.drop_inode = generic_delete_inode,
1227};
1228
1229static struct file_system_type mqueue_fs_type = {
1230	.name = "mqueue",
1231	.get_sb = mqueue_get_sb,
1232	.kill_sb = kill_litter_super,
1233};
1234
1235int mq_init_ns(struct ipc_namespace *ns)
1236{
1237	ns->mq_queues_count  = 0;
1238	ns->mq_queues_max    = DFLT_QUEUESMAX;
1239	ns->mq_msg_max       = DFLT_MSGMAX;
1240	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1241
1242	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1243	if (IS_ERR(ns->mq_mnt)) {
1244		int err = PTR_ERR(ns->mq_mnt);
1245		ns->mq_mnt = NULL;
1246		return err;
1247	}
1248	return 0;
1249}
1250
1251void mq_clear_sbinfo(struct ipc_namespace *ns)
1252{
1253	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1254}
1255
1256void mq_put_mnt(struct ipc_namespace *ns)
1257{
1258	mntput(ns->mq_mnt);
1259}
1260
1261static int __init init_mqueue_fs(void)
1262{
1263	int error;
1264
1265	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1266				sizeof(struct mqueue_inode_info), 0,
1267				SLAB_HWCACHE_ALIGN, init_once);
1268	if (mqueue_inode_cachep == NULL)
1269		return -ENOMEM;
1270
1271	/* ignore failues - they are not fatal */
1272	mq_sysctl_table = mq_register_sysctl_table();
1273
1274	error = register_filesystem(&mqueue_fs_type);
1275	if (error)
1276		goto out_sysctl;
1277
1278	spin_lock_init(&mq_lock);
1279
1280	init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1281	if (IS_ERR(init_ipc_ns.mq_mnt)) {
1282		error = PTR_ERR(init_ipc_ns.mq_mnt);
1283		goto out_filesystem;
1284	}
1285
1286	return 0;
1287
1288out_filesystem:
1289	unregister_filesystem(&mqueue_fs_type);
1290out_sysctl:
1291	if (mq_sysctl_table)
1292		unregister_sysctl_table(mq_sysctl_table);
1293	kmem_cache_destroy(mqueue_inode_cachep);
1294	return error;
1295}
1296
1297__initcall(init_mqueue_fs);
1298