mqueue.c revision a3ed2a15
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/slab.h>
36
37#include <net/sock.h>
38#include "util.h"
39
40#define MQUEUE_MAGIC	0x19800202
41#define DIRENT_SIZE	20
42#define FILENT_SIZE	80
43
44#define SEND		0
45#define RECV		1
46
47#define STATE_NONE	0
48#define STATE_PENDING	1
49#define STATE_READY	2
50
51struct ext_wait_queue {		/* queue of sleeping tasks */
52	struct task_struct *task;
53	struct list_head list;
54	struct msg_msg *msg;	/* ptr of loaded message */
55	int state;		/* one of STATE_* values */
56};
57
58struct mqueue_inode_info {
59	spinlock_t lock;
60	struct inode vfs_inode;
61	wait_queue_head_t wait_q;
62
63	struct msg_msg **messages;
64	struct mq_attr attr;
65
66	struct sigevent notify;
67	struct pid* notify_owner;
68	struct user_struct *user;	/* user who created, for accounting */
69	struct sock *notify_sock;
70	struct sk_buff *notify_cookie;
71
72	/* for tasks waiting for free space and messages, respectively */
73	struct ext_wait_queue e_wait_q[2];
74
75	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
76};
77
78static const struct inode_operations mqueue_dir_inode_operations;
79static const struct file_operations mqueue_file_operations;
80static const struct super_operations mqueue_super_ops;
81static void remove_notification(struct mqueue_inode_info *info);
82
83static struct kmem_cache *mqueue_inode_cachep;
84
85static struct ctl_table_header * mq_sysctl_table;
86
87static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
88{
89	return container_of(inode, struct mqueue_inode_info, vfs_inode);
90}
91
92/*
93 * This routine should be called with the mq_lock held.
94 */
95static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
96{
97	return get_ipc_ns(inode->i_sb->s_fs_info);
98}
99
100static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
101{
102	struct ipc_namespace *ns;
103
104	spin_lock(&mq_lock);
105	ns = __get_ns_from_inode(inode);
106	spin_unlock(&mq_lock);
107	return ns;
108}
109
110static struct inode *mqueue_get_inode(struct super_block *sb,
111		struct ipc_namespace *ipc_ns, int mode,
112		struct mq_attr *attr)
113{
114	struct user_struct *u = current_user();
115	struct inode *inode;
116
117	inode = new_inode(sb);
118	if (inode) {
119		inode->i_mode = mode;
120		inode->i_uid = current_fsuid();
121		inode->i_gid = current_fsgid();
122		inode->i_mtime = inode->i_ctime = inode->i_atime =
123				CURRENT_TIME;
124
125		if (S_ISREG(mode)) {
126			struct mqueue_inode_info *info;
127			struct task_struct *p = current;
128			unsigned long mq_bytes, mq_msg_tblsz;
129
130			inode->i_fop = &mqueue_file_operations;
131			inode->i_size = FILENT_SIZE;
132			/* mqueue specific info */
133			info = MQUEUE_I(inode);
134			spin_lock_init(&info->lock);
135			init_waitqueue_head(&info->wait_q);
136			INIT_LIST_HEAD(&info->e_wait_q[0].list);
137			INIT_LIST_HEAD(&info->e_wait_q[1].list);
138			info->notify_owner = NULL;
139			info->qsize = 0;
140			info->user = NULL;	/* set when all is ok */
141			memset(&info->attr, 0, sizeof(info->attr));
142			info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
143			info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
144			if (attr) {
145				info->attr.mq_maxmsg = attr->mq_maxmsg;
146				info->attr.mq_msgsize = attr->mq_msgsize;
147			}
148			mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
149			info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
150			if (!info->messages)
151				goto out_inode;
152
153			mq_bytes = (mq_msg_tblsz +
154				(info->attr.mq_maxmsg * info->attr.mq_msgsize));
155
156			spin_lock(&mq_lock);
157			if (u->mq_bytes + mq_bytes < u->mq_bytes ||
158		 	    u->mq_bytes + mq_bytes >
159			    task_rlimit(p, RLIMIT_MSGQUEUE)) {
160				spin_unlock(&mq_lock);
161				/* mqueue_delete_inode() releases info->messages */
162				goto out_inode;
163			}
164			u->mq_bytes += mq_bytes;
165			spin_unlock(&mq_lock);
166
167			/* all is ok */
168			info->user = get_uid(u);
169		} else if (S_ISDIR(mode)) {
170			inc_nlink(inode);
171			/* Some things misbehave if size == 0 on a directory */
172			inode->i_size = 2 * DIRENT_SIZE;
173			inode->i_op = &mqueue_dir_inode_operations;
174			inode->i_fop = &simple_dir_operations;
175		}
176	}
177	return inode;
178out_inode:
179	make_bad_inode(inode);
180	iput(inode);
181	return NULL;
182}
183
184static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
185{
186	struct inode *inode;
187	struct ipc_namespace *ns = data;
188	int error;
189
190	sb->s_blocksize = PAGE_CACHE_SIZE;
191	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
192	sb->s_magic = MQUEUE_MAGIC;
193	sb->s_op = &mqueue_super_ops;
194
195	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
196				NULL);
197	if (!inode) {
198		error = -ENOMEM;
199		goto out;
200	}
201
202	sb->s_root = d_alloc_root(inode);
203	if (!sb->s_root) {
204		iput(inode);
205		error = -ENOMEM;
206		goto out;
207	}
208	error = 0;
209
210out:
211	return error;
212}
213
214static int mqueue_get_sb(struct file_system_type *fs_type,
215			 int flags, const char *dev_name,
216			 void *data, struct vfsmount *mnt)
217{
218	if (!(flags & MS_KERNMOUNT))
219		data = current->nsproxy->ipc_ns;
220	return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
221}
222
223static void init_once(void *foo)
224{
225	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
226
227	inode_init_once(&p->vfs_inode);
228}
229
230static struct inode *mqueue_alloc_inode(struct super_block *sb)
231{
232	struct mqueue_inode_info *ei;
233
234	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
235	if (!ei)
236		return NULL;
237	return &ei->vfs_inode;
238}
239
240static void mqueue_destroy_inode(struct inode *inode)
241{
242	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
243}
244
245static void mqueue_delete_inode(struct inode *inode)
246{
247	struct mqueue_inode_info *info;
248	struct user_struct *user;
249	unsigned long mq_bytes;
250	int i;
251	struct ipc_namespace *ipc_ns;
252
253	if (S_ISDIR(inode->i_mode)) {
254		clear_inode(inode);
255		return;
256	}
257	ipc_ns = get_ns_from_inode(inode);
258	info = MQUEUE_I(inode);
259	spin_lock(&info->lock);
260	for (i = 0; i < info->attr.mq_curmsgs; i++)
261		free_msg(info->messages[i]);
262	kfree(info->messages);
263	spin_unlock(&info->lock);
264
265	clear_inode(inode);
266
267	/* Total amount of bytes accounted for the mqueue */
268	mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
269	    + info->attr.mq_msgsize);
270	user = info->user;
271	if (user) {
272		spin_lock(&mq_lock);
273		user->mq_bytes -= mq_bytes;
274		/*
275		 * get_ns_from_inode() ensures that the
276		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
277		 * to which we now hold a reference, or it is NULL.
278		 * We can't put it here under mq_lock, though.
279		 */
280		if (ipc_ns)
281			ipc_ns->mq_queues_count--;
282		spin_unlock(&mq_lock);
283		free_uid(user);
284	}
285	if (ipc_ns)
286		put_ipc_ns(ipc_ns);
287}
288
289static int mqueue_create(struct inode *dir, struct dentry *dentry,
290				int mode, struct nameidata *nd)
291{
292	struct inode *inode;
293	struct mq_attr *attr = dentry->d_fsdata;
294	int error;
295	struct ipc_namespace *ipc_ns;
296
297	spin_lock(&mq_lock);
298	ipc_ns = __get_ns_from_inode(dir);
299	if (!ipc_ns) {
300		error = -EACCES;
301		goto out_unlock;
302	}
303	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
304			!capable(CAP_SYS_RESOURCE)) {
305		error = -ENOSPC;
306		goto out_unlock;
307	}
308	ipc_ns->mq_queues_count++;
309	spin_unlock(&mq_lock);
310
311	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
312	if (!inode) {
313		error = -ENOMEM;
314		spin_lock(&mq_lock);
315		ipc_ns->mq_queues_count--;
316		goto out_unlock;
317	}
318
319	put_ipc_ns(ipc_ns);
320	dir->i_size += DIRENT_SIZE;
321	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
322
323	d_instantiate(dentry, inode);
324	dget(dentry);
325	return 0;
326out_unlock:
327	spin_unlock(&mq_lock);
328	if (ipc_ns)
329		put_ipc_ns(ipc_ns);
330	return error;
331}
332
333static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
334{
335  	struct inode *inode = dentry->d_inode;
336
337	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
338	dir->i_size -= DIRENT_SIZE;
339  	drop_nlink(inode);
340  	dput(dentry);
341  	return 0;
342}
343
344/*
345*	This is routine for system read from queue file.
346*	To avoid mess with doing here some sort of mq_receive we allow
347*	to read only queue size & notification info (the only values
348*	that are interesting from user point of view and aren't accessible
349*	through std routines)
350*/
351static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
352				size_t count, loff_t *off)
353{
354	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
355	char buffer[FILENT_SIZE];
356	ssize_t ret;
357
358	spin_lock(&info->lock);
359	snprintf(buffer, sizeof(buffer),
360			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
361			info->qsize,
362			info->notify_owner ? info->notify.sigev_notify : 0,
363			(info->notify_owner &&
364			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
365				info->notify.sigev_signo : 0,
366			pid_vnr(info->notify_owner));
367	spin_unlock(&info->lock);
368	buffer[sizeof(buffer)-1] = '\0';
369
370	ret = simple_read_from_buffer(u_data, count, off, buffer,
371				strlen(buffer));
372	if (ret <= 0)
373		return ret;
374
375	filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
376	return ret;
377}
378
379static int mqueue_flush_file(struct file *filp, fl_owner_t id)
380{
381	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
382
383	spin_lock(&info->lock);
384	if (task_tgid(current) == info->notify_owner)
385		remove_notification(info);
386
387	spin_unlock(&info->lock);
388	return 0;
389}
390
391static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
392{
393	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
394	int retval = 0;
395
396	poll_wait(filp, &info->wait_q, poll_tab);
397
398	spin_lock(&info->lock);
399	if (info->attr.mq_curmsgs)
400		retval = POLLIN | POLLRDNORM;
401
402	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
403		retval |= POLLOUT | POLLWRNORM;
404	spin_unlock(&info->lock);
405
406	return retval;
407}
408
409/* Adds current to info->e_wait_q[sr] before element with smaller prio */
410static void wq_add(struct mqueue_inode_info *info, int sr,
411			struct ext_wait_queue *ewp)
412{
413	struct ext_wait_queue *walk;
414
415	ewp->task = current;
416
417	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
418		if (walk->task->static_prio <= current->static_prio) {
419			list_add_tail(&ewp->list, &walk->list);
420			return;
421		}
422	}
423	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
424}
425
426/*
427 * Puts current task to sleep. Caller must hold queue lock. After return
428 * lock isn't held.
429 * sr: SEND or RECV
430 */
431static int wq_sleep(struct mqueue_inode_info *info, int sr,
432			long timeout, struct ext_wait_queue *ewp)
433{
434	int retval;
435	signed long time;
436
437	wq_add(info, sr, ewp);
438
439	for (;;) {
440		set_current_state(TASK_INTERRUPTIBLE);
441
442		spin_unlock(&info->lock);
443		time = schedule_timeout(timeout);
444
445		while (ewp->state == STATE_PENDING)
446			cpu_relax();
447
448		if (ewp->state == STATE_READY) {
449			retval = 0;
450			goto out;
451		}
452		spin_lock(&info->lock);
453		if (ewp->state == STATE_READY) {
454			retval = 0;
455			goto out_unlock;
456		}
457		if (signal_pending(current)) {
458			retval = -ERESTARTSYS;
459			break;
460		}
461		if (time == 0) {
462			retval = -ETIMEDOUT;
463			break;
464		}
465	}
466	list_del(&ewp->list);
467out_unlock:
468	spin_unlock(&info->lock);
469out:
470	return retval;
471}
472
473/*
474 * Returns waiting task that should be serviced first or NULL if none exists
475 */
476static struct ext_wait_queue *wq_get_first_waiter(
477		struct mqueue_inode_info *info, int sr)
478{
479	struct list_head *ptr;
480
481	ptr = info->e_wait_q[sr].list.prev;
482	if (ptr == &info->e_wait_q[sr].list)
483		return NULL;
484	return list_entry(ptr, struct ext_wait_queue, list);
485}
486
487/* Auxiliary functions to manipulate messages' list */
488static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
489{
490	int k;
491
492	k = info->attr.mq_curmsgs - 1;
493	while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
494		info->messages[k + 1] = info->messages[k];
495		k--;
496	}
497	info->attr.mq_curmsgs++;
498	info->qsize += ptr->m_ts;
499	info->messages[k + 1] = ptr;
500}
501
502static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
503{
504	info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
505	return info->messages[info->attr.mq_curmsgs];
506}
507
508static inline void set_cookie(struct sk_buff *skb, char code)
509{
510	((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
511}
512
513/*
514 * The next function is only to split too long sys_mq_timedsend
515 */
516static void __do_notify(struct mqueue_inode_info *info)
517{
518	/* notification
519	 * invoked when there is registered process and there isn't process
520	 * waiting synchronously for message AND state of queue changed from
521	 * empty to not empty. Here we are sure that no one is waiting
522	 * synchronously. */
523	if (info->notify_owner &&
524	    info->attr.mq_curmsgs == 1) {
525		struct siginfo sig_i;
526		switch (info->notify.sigev_notify) {
527		case SIGEV_NONE:
528			break;
529		case SIGEV_SIGNAL:
530			/* sends signal */
531
532			sig_i.si_signo = info->notify.sigev_signo;
533			sig_i.si_errno = 0;
534			sig_i.si_code = SI_MESGQ;
535			sig_i.si_value = info->notify.sigev_value;
536			sig_i.si_pid = task_tgid_nr_ns(current,
537						ns_of_pid(info->notify_owner));
538			sig_i.si_uid = current_uid();
539
540			kill_pid_info(info->notify.sigev_signo,
541				      &sig_i, info->notify_owner);
542			break;
543		case SIGEV_THREAD:
544			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
545			netlink_sendskb(info->notify_sock, info->notify_cookie);
546			break;
547		}
548		/* after notification unregisters process */
549		put_pid(info->notify_owner);
550		info->notify_owner = NULL;
551	}
552	wake_up(&info->wait_q);
553}
554
555static long prepare_timeout(struct timespec *p)
556{
557	struct timespec nowts;
558	long timeout;
559
560	if (p) {
561		if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
562			|| p->tv_nsec >= NSEC_PER_SEC))
563			return -EINVAL;
564		nowts = CURRENT_TIME;
565		/* first subtract as jiffies can't be too big */
566		p->tv_sec -= nowts.tv_sec;
567		if (p->tv_nsec < nowts.tv_nsec) {
568			p->tv_nsec += NSEC_PER_SEC;
569			p->tv_sec--;
570		}
571		p->tv_nsec -= nowts.tv_nsec;
572		if (p->tv_sec < 0)
573			return 0;
574
575		timeout = timespec_to_jiffies(p) + 1;
576	} else
577		return MAX_SCHEDULE_TIMEOUT;
578
579	return timeout;
580}
581
582static void remove_notification(struct mqueue_inode_info *info)
583{
584	if (info->notify_owner != NULL &&
585	    info->notify.sigev_notify == SIGEV_THREAD) {
586		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
587		netlink_sendskb(info->notify_sock, info->notify_cookie);
588	}
589	put_pid(info->notify_owner);
590	info->notify_owner = NULL;
591}
592
593static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
594{
595	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
596		return 0;
597	if (capable(CAP_SYS_RESOURCE)) {
598		if (attr->mq_maxmsg > HARD_MSGMAX)
599			return 0;
600	} else {
601		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
602				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
603			return 0;
604	}
605	/* check for overflow */
606	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
607		return 0;
608	if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
609	    + sizeof (struct msg_msg *))) <
610	    (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
611		return 0;
612	return 1;
613}
614
615/*
616 * Invoked when creating a new queue via sys_mq_open
617 */
618static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
619			struct dentry *dentry, int oflag, mode_t mode,
620			struct mq_attr *attr)
621{
622	const struct cred *cred = current_cred();
623	struct file *result;
624	int ret;
625
626	if (attr) {
627		if (!mq_attr_ok(ipc_ns, attr)) {
628			ret = -EINVAL;
629			goto out;
630		}
631		/* store for use during create */
632		dentry->d_fsdata = attr;
633	}
634
635	mode &= ~current_umask();
636	ret = mnt_want_write(ipc_ns->mq_mnt);
637	if (ret)
638		goto out;
639	ret = vfs_create(dir->d_inode, dentry, mode, NULL);
640	dentry->d_fsdata = NULL;
641	if (ret)
642		goto out_drop_write;
643
644	result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
645	/*
646	 * dentry_open() took a persistent mnt_want_write(),
647	 * so we can now drop this one.
648	 */
649	mnt_drop_write(ipc_ns->mq_mnt);
650	return result;
651
652out_drop_write:
653	mnt_drop_write(ipc_ns->mq_mnt);
654out:
655	dput(dentry);
656	mntput(ipc_ns->mq_mnt);
657	return ERR_PTR(ret);
658}
659
660/* Opens existing queue */
661static struct file *do_open(struct ipc_namespace *ipc_ns,
662				struct dentry *dentry, int oflag)
663{
664	int ret;
665	const struct cred *cred = current_cred();
666
667	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
668						  MAY_READ | MAY_WRITE };
669
670	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
671		ret = -EINVAL;
672		goto err;
673	}
674
675	if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
676		ret = -EACCES;
677		goto err;
678	}
679
680	return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
681
682err:
683	dput(dentry);
684	mntput(ipc_ns->mq_mnt);
685	return ERR_PTR(ret);
686}
687
688SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
689		struct mq_attr __user *, u_attr)
690{
691	struct dentry *dentry;
692	struct file *filp;
693	char *name;
694	struct mq_attr attr;
695	int fd, error;
696	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
697
698	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
699		return -EFAULT;
700
701	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
702
703	if (IS_ERR(name = getname(u_name)))
704		return PTR_ERR(name);
705
706	fd = get_unused_fd_flags(O_CLOEXEC);
707	if (fd < 0)
708		goto out_putname;
709
710	mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
711	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
712	if (IS_ERR(dentry)) {
713		error = PTR_ERR(dentry);
714		goto out_putfd;
715	}
716	mntget(ipc_ns->mq_mnt);
717
718	if (oflag & O_CREAT) {
719		if (dentry->d_inode) {	/* entry already exists */
720			audit_inode(name, dentry);
721			if (oflag & O_EXCL) {
722				error = -EEXIST;
723				goto out;
724			}
725			filp = do_open(ipc_ns, dentry, oflag);
726		} else {
727			filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
728						dentry, oflag, mode,
729						u_attr ? &attr : NULL);
730		}
731	} else {
732		if (!dentry->d_inode) {
733			error = -ENOENT;
734			goto out;
735		}
736		audit_inode(name, dentry);
737		filp = do_open(ipc_ns, dentry, oflag);
738	}
739
740	if (IS_ERR(filp)) {
741		error = PTR_ERR(filp);
742		goto out_putfd;
743	}
744
745	fd_install(fd, filp);
746	goto out_upsem;
747
748out:
749	dput(dentry);
750	mntput(ipc_ns->mq_mnt);
751out_putfd:
752	put_unused_fd(fd);
753	fd = error;
754out_upsem:
755	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
756out_putname:
757	putname(name);
758	return fd;
759}
760
761SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
762{
763	int err;
764	char *name;
765	struct dentry *dentry;
766	struct inode *inode = NULL;
767	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
768
769	name = getname(u_name);
770	if (IS_ERR(name))
771		return PTR_ERR(name);
772
773	mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
774			I_MUTEX_PARENT);
775	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
776	if (IS_ERR(dentry)) {
777		err = PTR_ERR(dentry);
778		goto out_unlock;
779	}
780
781	if (!dentry->d_inode) {
782		err = -ENOENT;
783		goto out_err;
784	}
785
786	inode = dentry->d_inode;
787	if (inode)
788		atomic_inc(&inode->i_count);
789	err = mnt_want_write(ipc_ns->mq_mnt);
790	if (err)
791		goto out_err;
792	err = vfs_unlink(dentry->d_parent->d_inode, dentry);
793	mnt_drop_write(ipc_ns->mq_mnt);
794out_err:
795	dput(dentry);
796
797out_unlock:
798	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
799	putname(name);
800	if (inode)
801		iput(inode);
802
803	return err;
804}
805
806/* Pipelined send and receive functions.
807 *
808 * If a receiver finds no waiting message, then it registers itself in the
809 * list of waiting receivers. A sender checks that list before adding the new
810 * message into the message array. If there is a waiting receiver, then it
811 * bypasses the message array and directly hands the message over to the
812 * receiver.
813 * The receiver accepts the message and returns without grabbing the queue
814 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
815 * are necessary. The same algorithm is used for sysv semaphores, see
816 * ipc/sem.c for more details.
817 *
818 * The same algorithm is used for senders.
819 */
820
821/* pipelined_send() - send a message directly to the task waiting in
822 * sys_mq_timedreceive() (without inserting message into a queue).
823 */
824static inline void pipelined_send(struct mqueue_inode_info *info,
825				  struct msg_msg *message,
826				  struct ext_wait_queue *receiver)
827{
828	receiver->msg = message;
829	list_del(&receiver->list);
830	receiver->state = STATE_PENDING;
831	wake_up_process(receiver->task);
832	smp_wmb();
833	receiver->state = STATE_READY;
834}
835
836/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
837 * gets its message and put to the queue (we have one free place for sure). */
838static inline void pipelined_receive(struct mqueue_inode_info *info)
839{
840	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
841
842	if (!sender) {
843		/* for poll */
844		wake_up_interruptible(&info->wait_q);
845		return;
846	}
847	msg_insert(sender->msg, info);
848	list_del(&sender->list);
849	sender->state = STATE_PENDING;
850	wake_up_process(sender->task);
851	smp_wmb();
852	sender->state = STATE_READY;
853}
854
855SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
856		size_t, msg_len, unsigned int, msg_prio,
857		const struct timespec __user *, u_abs_timeout)
858{
859	struct file *filp;
860	struct inode *inode;
861	struct ext_wait_queue wait;
862	struct ext_wait_queue *receiver;
863	struct msg_msg *msg_ptr;
864	struct mqueue_inode_info *info;
865	struct timespec ts, *p = NULL;
866	long timeout;
867	int ret;
868
869	if (u_abs_timeout) {
870		if (copy_from_user(&ts, u_abs_timeout,
871					sizeof(struct timespec)))
872			return -EFAULT;
873		p = &ts;
874	}
875
876	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
877		return -EINVAL;
878
879	audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
880	timeout = prepare_timeout(p);
881
882	filp = fget(mqdes);
883	if (unlikely(!filp)) {
884		ret = -EBADF;
885		goto out;
886	}
887
888	inode = filp->f_path.dentry->d_inode;
889	if (unlikely(filp->f_op != &mqueue_file_operations)) {
890		ret = -EBADF;
891		goto out_fput;
892	}
893	info = MQUEUE_I(inode);
894	audit_inode(NULL, filp->f_path.dentry);
895
896	if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
897		ret = -EBADF;
898		goto out_fput;
899	}
900
901	if (unlikely(msg_len > info->attr.mq_msgsize)) {
902		ret = -EMSGSIZE;
903		goto out_fput;
904	}
905
906	/* First try to allocate memory, before doing anything with
907	 * existing queues. */
908	msg_ptr = load_msg(u_msg_ptr, msg_len);
909	if (IS_ERR(msg_ptr)) {
910		ret = PTR_ERR(msg_ptr);
911		goto out_fput;
912	}
913	msg_ptr->m_ts = msg_len;
914	msg_ptr->m_type = msg_prio;
915
916	spin_lock(&info->lock);
917
918	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
919		if (filp->f_flags & O_NONBLOCK) {
920			spin_unlock(&info->lock);
921			ret = -EAGAIN;
922		} else if (unlikely(timeout < 0)) {
923			spin_unlock(&info->lock);
924			ret = timeout;
925		} else {
926			wait.task = current;
927			wait.msg = (void *) msg_ptr;
928			wait.state = STATE_NONE;
929			ret = wq_sleep(info, SEND, timeout, &wait);
930		}
931		if (ret < 0)
932			free_msg(msg_ptr);
933	} else {
934		receiver = wq_get_first_waiter(info, RECV);
935		if (receiver) {
936			pipelined_send(info, msg_ptr, receiver);
937		} else {
938			/* adds message to the queue */
939			msg_insert(msg_ptr, info);
940			__do_notify(info);
941		}
942		inode->i_atime = inode->i_mtime = inode->i_ctime =
943				CURRENT_TIME;
944		spin_unlock(&info->lock);
945		ret = 0;
946	}
947out_fput:
948	fput(filp);
949out:
950	return ret;
951}
952
953SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
954		size_t, msg_len, unsigned int __user *, u_msg_prio,
955		const struct timespec __user *, u_abs_timeout)
956{
957	long timeout;
958	ssize_t ret;
959	struct msg_msg *msg_ptr;
960	struct file *filp;
961	struct inode *inode;
962	struct mqueue_inode_info *info;
963	struct ext_wait_queue wait;
964	struct timespec ts, *p = NULL;
965
966	if (u_abs_timeout) {
967		if (copy_from_user(&ts, u_abs_timeout,
968					sizeof(struct timespec)))
969			return -EFAULT;
970		p = &ts;
971	}
972
973	audit_mq_sendrecv(mqdes, msg_len, 0, p);
974	timeout = prepare_timeout(p);
975
976	filp = fget(mqdes);
977	if (unlikely(!filp)) {
978		ret = -EBADF;
979		goto out;
980	}
981
982	inode = filp->f_path.dentry->d_inode;
983	if (unlikely(filp->f_op != &mqueue_file_operations)) {
984		ret = -EBADF;
985		goto out_fput;
986	}
987	info = MQUEUE_I(inode);
988	audit_inode(NULL, filp->f_path.dentry);
989
990	if (unlikely(!(filp->f_mode & FMODE_READ))) {
991		ret = -EBADF;
992		goto out_fput;
993	}
994
995	/* checks if buffer is big enough */
996	if (unlikely(msg_len < info->attr.mq_msgsize)) {
997		ret = -EMSGSIZE;
998		goto out_fput;
999	}
1000
1001	spin_lock(&info->lock);
1002	if (info->attr.mq_curmsgs == 0) {
1003		if (filp->f_flags & O_NONBLOCK) {
1004			spin_unlock(&info->lock);
1005			ret = -EAGAIN;
1006			msg_ptr = NULL;
1007		} else if (unlikely(timeout < 0)) {
1008			spin_unlock(&info->lock);
1009			ret = timeout;
1010			msg_ptr = NULL;
1011		} else {
1012			wait.task = current;
1013			wait.state = STATE_NONE;
1014			ret = wq_sleep(info, RECV, timeout, &wait);
1015			msg_ptr = wait.msg;
1016		}
1017	} else {
1018		msg_ptr = msg_get(info);
1019
1020		inode->i_atime = inode->i_mtime = inode->i_ctime =
1021				CURRENT_TIME;
1022
1023		/* There is now free space in queue. */
1024		pipelined_receive(info);
1025		spin_unlock(&info->lock);
1026		ret = 0;
1027	}
1028	if (ret == 0) {
1029		ret = msg_ptr->m_ts;
1030
1031		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1032			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1033			ret = -EFAULT;
1034		}
1035		free_msg(msg_ptr);
1036	}
1037out_fput:
1038	fput(filp);
1039out:
1040	return ret;
1041}
1042
1043/*
1044 * Notes: the case when user wants us to deregister (with NULL as pointer)
1045 * and he isn't currently owner of notification, will be silently discarded.
1046 * It isn't explicitly defined in the POSIX.
1047 */
1048SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1049		const struct sigevent __user *, u_notification)
1050{
1051	int ret;
1052	struct file *filp;
1053	struct sock *sock;
1054	struct inode *inode;
1055	struct sigevent notification;
1056	struct mqueue_inode_info *info;
1057	struct sk_buff *nc;
1058
1059	if (u_notification) {
1060		if (copy_from_user(&notification, u_notification,
1061					sizeof(struct sigevent)))
1062			return -EFAULT;
1063	}
1064
1065	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1066
1067	nc = NULL;
1068	sock = NULL;
1069	if (u_notification != NULL) {
1070		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1071			     notification.sigev_notify != SIGEV_SIGNAL &&
1072			     notification.sigev_notify != SIGEV_THREAD))
1073			return -EINVAL;
1074		if (notification.sigev_notify == SIGEV_SIGNAL &&
1075			!valid_signal(notification.sigev_signo)) {
1076			return -EINVAL;
1077		}
1078		if (notification.sigev_notify == SIGEV_THREAD) {
1079			long timeo;
1080
1081			/* create the notify skb */
1082			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1083			if (!nc) {
1084				ret = -ENOMEM;
1085				goto out;
1086			}
1087			if (copy_from_user(nc->data,
1088					notification.sigev_value.sival_ptr,
1089					NOTIFY_COOKIE_LEN)) {
1090				ret = -EFAULT;
1091				goto out;
1092			}
1093
1094			/* TODO: add a header? */
1095			skb_put(nc, NOTIFY_COOKIE_LEN);
1096			/* and attach it to the socket */
1097retry:
1098			filp = fget(notification.sigev_signo);
1099			if (!filp) {
1100				ret = -EBADF;
1101				goto out;
1102			}
1103			sock = netlink_getsockbyfilp(filp);
1104			fput(filp);
1105			if (IS_ERR(sock)) {
1106				ret = PTR_ERR(sock);
1107				sock = NULL;
1108				goto out;
1109			}
1110
1111			timeo = MAX_SCHEDULE_TIMEOUT;
1112			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1113			if (ret == 1)
1114				goto retry;
1115			if (ret) {
1116				sock = NULL;
1117				nc = NULL;
1118				goto out;
1119			}
1120		}
1121	}
1122
1123	filp = fget(mqdes);
1124	if (!filp) {
1125		ret = -EBADF;
1126		goto out;
1127	}
1128
1129	inode = filp->f_path.dentry->d_inode;
1130	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1131		ret = -EBADF;
1132		goto out_fput;
1133	}
1134	info = MQUEUE_I(inode);
1135
1136	ret = 0;
1137	spin_lock(&info->lock);
1138	if (u_notification == NULL) {
1139		if (info->notify_owner == task_tgid(current)) {
1140			remove_notification(info);
1141			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1142		}
1143	} else if (info->notify_owner != NULL) {
1144		ret = -EBUSY;
1145	} else {
1146		switch (notification.sigev_notify) {
1147		case SIGEV_NONE:
1148			info->notify.sigev_notify = SIGEV_NONE;
1149			break;
1150		case SIGEV_THREAD:
1151			info->notify_sock = sock;
1152			info->notify_cookie = nc;
1153			sock = NULL;
1154			nc = NULL;
1155			info->notify.sigev_notify = SIGEV_THREAD;
1156			break;
1157		case SIGEV_SIGNAL:
1158			info->notify.sigev_signo = notification.sigev_signo;
1159			info->notify.sigev_value = notification.sigev_value;
1160			info->notify.sigev_notify = SIGEV_SIGNAL;
1161			break;
1162		}
1163
1164		info->notify_owner = get_pid(task_tgid(current));
1165		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1166	}
1167	spin_unlock(&info->lock);
1168out_fput:
1169	fput(filp);
1170out:
1171	if (sock) {
1172		netlink_detachskb(sock, nc);
1173	} else if (nc) {
1174		dev_kfree_skb(nc);
1175	}
1176	return ret;
1177}
1178
1179SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1180		const struct mq_attr __user *, u_mqstat,
1181		struct mq_attr __user *, u_omqstat)
1182{
1183	int ret;
1184	struct mq_attr mqstat, omqstat;
1185	struct file *filp;
1186	struct inode *inode;
1187	struct mqueue_inode_info *info;
1188
1189	if (u_mqstat != NULL) {
1190		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1191			return -EFAULT;
1192		if (mqstat.mq_flags & (~O_NONBLOCK))
1193			return -EINVAL;
1194	}
1195
1196	filp = fget(mqdes);
1197	if (!filp) {
1198		ret = -EBADF;
1199		goto out;
1200	}
1201
1202	inode = filp->f_path.dentry->d_inode;
1203	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1204		ret = -EBADF;
1205		goto out_fput;
1206	}
1207	info = MQUEUE_I(inode);
1208
1209	spin_lock(&info->lock);
1210
1211	omqstat = info->attr;
1212	omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1213	if (u_mqstat) {
1214		audit_mq_getsetattr(mqdes, &mqstat);
1215		spin_lock(&filp->f_lock);
1216		if (mqstat.mq_flags & O_NONBLOCK)
1217			filp->f_flags |= O_NONBLOCK;
1218		else
1219			filp->f_flags &= ~O_NONBLOCK;
1220		spin_unlock(&filp->f_lock);
1221
1222		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1223	}
1224
1225	spin_unlock(&info->lock);
1226
1227	ret = 0;
1228	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1229						sizeof(struct mq_attr)))
1230		ret = -EFAULT;
1231
1232out_fput:
1233	fput(filp);
1234out:
1235	return ret;
1236}
1237
1238static const struct inode_operations mqueue_dir_inode_operations = {
1239	.lookup = simple_lookup,
1240	.create = mqueue_create,
1241	.unlink = mqueue_unlink,
1242};
1243
1244static const struct file_operations mqueue_file_operations = {
1245	.flush = mqueue_flush_file,
1246	.poll = mqueue_poll_file,
1247	.read = mqueue_read_file,
1248};
1249
1250static const struct super_operations mqueue_super_ops = {
1251	.alloc_inode = mqueue_alloc_inode,
1252	.destroy_inode = mqueue_destroy_inode,
1253	.statfs = simple_statfs,
1254	.delete_inode = mqueue_delete_inode,
1255	.drop_inode = generic_delete_inode,
1256};
1257
1258static struct file_system_type mqueue_fs_type = {
1259	.name = "mqueue",
1260	.get_sb = mqueue_get_sb,
1261	.kill_sb = kill_litter_super,
1262};
1263
1264int mq_init_ns(struct ipc_namespace *ns)
1265{
1266	ns->mq_queues_count  = 0;
1267	ns->mq_queues_max    = DFLT_QUEUESMAX;
1268	ns->mq_msg_max       = DFLT_MSGMAX;
1269	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1270
1271	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1272	if (IS_ERR(ns->mq_mnt)) {
1273		int err = PTR_ERR(ns->mq_mnt);
1274		ns->mq_mnt = NULL;
1275		return err;
1276	}
1277	return 0;
1278}
1279
1280void mq_clear_sbinfo(struct ipc_namespace *ns)
1281{
1282	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1283}
1284
1285void mq_put_mnt(struct ipc_namespace *ns)
1286{
1287	mntput(ns->mq_mnt);
1288}
1289
1290static int __init init_mqueue_fs(void)
1291{
1292	int error;
1293
1294	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1295				sizeof(struct mqueue_inode_info), 0,
1296				SLAB_HWCACHE_ALIGN, init_once);
1297	if (mqueue_inode_cachep == NULL)
1298		return -ENOMEM;
1299
1300	/* ignore failures - they are not fatal */
1301	mq_sysctl_table = mq_register_sysctl_table();
1302
1303	error = register_filesystem(&mqueue_fs_type);
1304	if (error)
1305		goto out_sysctl;
1306
1307	spin_lock_init(&mq_lock);
1308
1309	init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1310	if (IS_ERR(init_ipc_ns.mq_mnt)) {
1311		error = PTR_ERR(init_ipc_ns.mq_mnt);
1312		goto out_filesystem;
1313	}
1314
1315	return 0;
1316
1317out_filesystem:
1318	unregister_filesystem(&mqueue_fs_type);
1319out_sysctl:
1320	if (mq_sysctl_table)
1321		unregister_sysctl_table(mq_sysctl_table);
1322	kmem_cache_destroy(mqueue_inode_cachep);
1323	return error;
1324}
1325
1326__initcall(init_mqueue_fs);
1327