mqueue.c revision f1eb1332
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			    task_rlimit(p, RLIMIT_MSGQUEUE)) {
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;
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		goto out;
206	}
207	error = 0;
208
209out:
210	return error;
211}
212
213static int mqueue_get_sb(struct file_system_type *fs_type,
214			 int flags, const char *dev_name,
215			 void *data, struct vfsmount *mnt)
216{
217	if (!(flags & MS_KERNMOUNT))
218		data = current->nsproxy->ipc_ns;
219	return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
220}
221
222static void init_once(void *foo)
223{
224	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
225
226	inode_init_once(&p->vfs_inode);
227}
228
229static struct inode *mqueue_alloc_inode(struct super_block *sb)
230{
231	struct mqueue_inode_info *ei;
232
233	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
234	if (!ei)
235		return NULL;
236	return &ei->vfs_inode;
237}
238
239static void mqueue_destroy_inode(struct inode *inode)
240{
241	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
242}
243
244static void mqueue_delete_inode(struct inode *inode)
245{
246	struct mqueue_inode_info *info;
247	struct user_struct *user;
248	unsigned long mq_bytes;
249	int i;
250	struct ipc_namespace *ipc_ns;
251
252	if (S_ISDIR(inode->i_mode)) {
253		clear_inode(inode);
254		return;
255	}
256	ipc_ns = get_ns_from_inode(inode);
257	info = MQUEUE_I(inode);
258	spin_lock(&info->lock);
259	for (i = 0; i < info->attr.mq_curmsgs; i++)
260		free_msg(info->messages[i]);
261	kfree(info->messages);
262	spin_unlock(&info->lock);
263
264	clear_inode(inode);
265
266	/* Total amount of bytes accounted for the mqueue */
267	mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
268	    + info->attr.mq_msgsize);
269	user = info->user;
270	if (user) {
271		spin_lock(&mq_lock);
272		user->mq_bytes -= mq_bytes;
273		/*
274		 * get_ns_from_inode() ensures that the
275		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
276		 * to which we now hold a reference, or it is NULL.
277		 * We can't put it here under mq_lock, though.
278		 */
279		if (ipc_ns)
280			ipc_ns->mq_queues_count--;
281		spin_unlock(&mq_lock);
282		free_uid(user);
283	}
284	if (ipc_ns)
285		put_ipc_ns(ipc_ns);
286}
287
288static int mqueue_create(struct inode *dir, struct dentry *dentry,
289				int mode, struct nameidata *nd)
290{
291	struct inode *inode;
292	struct mq_attr *attr = dentry->d_fsdata;
293	int error;
294	struct ipc_namespace *ipc_ns;
295
296	spin_lock(&mq_lock);
297	ipc_ns = __get_ns_from_inode(dir);
298	if (!ipc_ns) {
299		error = -EACCES;
300		goto out_unlock;
301	}
302	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
303			!capable(CAP_SYS_RESOURCE)) {
304		error = -ENOSPC;
305		goto out_unlock;
306	}
307	ipc_ns->mq_queues_count++;
308	spin_unlock(&mq_lock);
309
310	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
311	if (!inode) {
312		error = -ENOMEM;
313		spin_lock(&mq_lock);
314		ipc_ns->mq_queues_count--;
315		goto out_unlock;
316	}
317
318	put_ipc_ns(ipc_ns);
319	dir->i_size += DIRENT_SIZE;
320	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
321
322	d_instantiate(dentry, inode);
323	dget(dentry);
324	return 0;
325out_unlock:
326	spin_unlock(&mq_lock);
327	if (ipc_ns)
328		put_ipc_ns(ipc_ns);
329	return error;
330}
331
332static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
333{
334  	struct inode *inode = dentry->d_inode;
335
336	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
337	dir->i_size -= DIRENT_SIZE;
338  	drop_nlink(inode);
339  	dput(dentry);
340  	return 0;
341}
342
343/*
344*	This is routine for system read from queue file.
345*	To avoid mess with doing here some sort of mq_receive we allow
346*	to read only queue size & notification info (the only values
347*	that are interesting from user point of view and aren't accessible
348*	through std routines)
349*/
350static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
351				size_t count, loff_t *off)
352{
353	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
354	char buffer[FILENT_SIZE];
355	ssize_t ret;
356
357	spin_lock(&info->lock);
358	snprintf(buffer, sizeof(buffer),
359			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
360			info->qsize,
361			info->notify_owner ? info->notify.sigev_notify : 0,
362			(info->notify_owner &&
363			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
364				info->notify.sigev_signo : 0,
365			pid_vnr(info->notify_owner));
366	spin_unlock(&info->lock);
367	buffer[sizeof(buffer)-1] = '\0';
368
369	ret = simple_read_from_buffer(u_data, count, off, buffer,
370				strlen(buffer));
371	if (ret <= 0)
372		return ret;
373
374	filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
375	return ret;
376}
377
378static int mqueue_flush_file(struct file *filp, fl_owner_t id)
379{
380	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
381
382	spin_lock(&info->lock);
383	if (task_tgid(current) == info->notify_owner)
384		remove_notification(info);
385
386	spin_unlock(&info->lock);
387	return 0;
388}
389
390static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
391{
392	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
393	int retval = 0;
394
395	poll_wait(filp, &info->wait_q, poll_tab);
396
397	spin_lock(&info->lock);
398	if (info->attr.mq_curmsgs)
399		retval = POLLIN | POLLRDNORM;
400
401	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
402		retval |= POLLOUT | POLLWRNORM;
403	spin_unlock(&info->lock);
404
405	return retval;
406}
407
408/* Adds current to info->e_wait_q[sr] before element with smaller prio */
409static void wq_add(struct mqueue_inode_info *info, int sr,
410			struct ext_wait_queue *ewp)
411{
412	struct ext_wait_queue *walk;
413
414	ewp->task = current;
415
416	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
417		if (walk->task->static_prio <= current->static_prio) {
418			list_add_tail(&ewp->list, &walk->list);
419			return;
420		}
421	}
422	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
423}
424
425/*
426 * Puts current task to sleep. Caller must hold queue lock. After return
427 * lock isn't held.
428 * sr: SEND or RECV
429 */
430static int wq_sleep(struct mqueue_inode_info *info, int sr,
431			long timeout, struct ext_wait_queue *ewp)
432{
433	int retval;
434	signed long time;
435
436	wq_add(info, sr, ewp);
437
438	for (;;) {
439		set_current_state(TASK_INTERRUPTIBLE);
440
441		spin_unlock(&info->lock);
442		time = schedule_timeout(timeout);
443
444		while (ewp->state == STATE_PENDING)
445			cpu_relax();
446
447		if (ewp->state == STATE_READY) {
448			retval = 0;
449			goto out;
450		}
451		spin_lock(&info->lock);
452		if (ewp->state == STATE_READY) {
453			retval = 0;
454			goto out_unlock;
455		}
456		if (signal_pending(current)) {
457			retval = -ERESTARTSYS;
458			break;
459		}
460		if (time == 0) {
461			retval = -ETIMEDOUT;
462			break;
463		}
464	}
465	list_del(&ewp->list);
466out_unlock:
467	spin_unlock(&info->lock);
468out:
469	return retval;
470}
471
472/*
473 * Returns waiting task that should be serviced first or NULL if none exists
474 */
475static struct ext_wait_queue *wq_get_first_waiter(
476		struct mqueue_inode_info *info, int sr)
477{
478	struct list_head *ptr;
479
480	ptr = info->e_wait_q[sr].list.prev;
481	if (ptr == &info->e_wait_q[sr].list)
482		return NULL;
483	return list_entry(ptr, struct ext_wait_queue, list);
484}
485
486/* Auxiliary functions to manipulate messages' list */
487static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
488{
489	int k;
490
491	k = info->attr.mq_curmsgs - 1;
492	while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
493		info->messages[k + 1] = info->messages[k];
494		k--;
495	}
496	info->attr.mq_curmsgs++;
497	info->qsize += ptr->m_ts;
498	info->messages[k + 1] = ptr;
499}
500
501static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
502{
503	info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
504	return info->messages[info->attr.mq_curmsgs];
505}
506
507static inline void set_cookie(struct sk_buff *skb, char code)
508{
509	((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
510}
511
512/*
513 * The next function is only to split too long sys_mq_timedsend
514 */
515static void __do_notify(struct mqueue_inode_info *info)
516{
517	/* notification
518	 * invoked when there is registered process and there isn't process
519	 * waiting synchronously for message AND state of queue changed from
520	 * empty to not empty. Here we are sure that no one is waiting
521	 * synchronously. */
522	if (info->notify_owner &&
523	    info->attr.mq_curmsgs == 1) {
524		struct siginfo sig_i;
525		switch (info->notify.sigev_notify) {
526		case SIGEV_NONE:
527			break;
528		case SIGEV_SIGNAL:
529			/* sends signal */
530
531			sig_i.si_signo = info->notify.sigev_signo;
532			sig_i.si_errno = 0;
533			sig_i.si_code = SI_MESGQ;
534			sig_i.si_value = info->notify.sigev_value;
535			sig_i.si_pid = task_tgid_nr_ns(current,
536						ns_of_pid(info->notify_owner));
537			sig_i.si_uid = current_uid();
538
539			kill_pid_info(info->notify.sigev_signo,
540				      &sig_i, info->notify_owner);
541			break;
542		case SIGEV_THREAD:
543			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
544			netlink_sendskb(info->notify_sock, info->notify_cookie);
545			break;
546		}
547		/* after notification unregisters process */
548		put_pid(info->notify_owner);
549		info->notify_owner = NULL;
550	}
551	wake_up(&info->wait_q);
552}
553
554static long prepare_timeout(struct timespec *p)
555{
556	struct timespec nowts;
557	long timeout;
558
559	if (p) {
560		if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
561			|| p->tv_nsec >= NSEC_PER_SEC))
562			return -EINVAL;
563		nowts = CURRENT_TIME;
564		/* first subtract as jiffies can't be too big */
565		p->tv_sec -= nowts.tv_sec;
566		if (p->tv_nsec < nowts.tv_nsec) {
567			p->tv_nsec += NSEC_PER_SEC;
568			p->tv_sec--;
569		}
570		p->tv_nsec -= nowts.tv_nsec;
571		if (p->tv_sec < 0)
572			return 0;
573
574		timeout = timespec_to_jiffies(p) + 1;
575	} else
576		return MAX_SCHEDULE_TIMEOUT;
577
578	return timeout;
579}
580
581static void remove_notification(struct mqueue_inode_info *info)
582{
583	if (info->notify_owner != NULL &&
584	    info->notify.sigev_notify == SIGEV_THREAD) {
585		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
586		netlink_sendskb(info->notify_sock, info->notify_cookie);
587	}
588	put_pid(info->notify_owner);
589	info->notify_owner = NULL;
590}
591
592static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
593{
594	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
595		return 0;
596	if (capable(CAP_SYS_RESOURCE)) {
597		if (attr->mq_maxmsg > HARD_MSGMAX)
598			return 0;
599	} else {
600		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
601				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
602			return 0;
603	}
604	/* check for overflow */
605	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
606		return 0;
607	if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
608	    + sizeof (struct msg_msg *))) <
609	    (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
610		return 0;
611	return 1;
612}
613
614/*
615 * Invoked when creating a new queue via sys_mq_open
616 */
617static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
618			struct dentry *dentry, int oflag, mode_t mode,
619			struct mq_attr *attr)
620{
621	const struct cred *cred = current_cred();
622	struct file *result;
623	int ret;
624
625	if (attr) {
626		if (!mq_attr_ok(ipc_ns, attr)) {
627			ret = -EINVAL;
628			goto out;
629		}
630		/* store for use during create */
631		dentry->d_fsdata = attr;
632	}
633
634	mode &= ~current_umask();
635	ret = mnt_want_write(ipc_ns->mq_mnt);
636	if (ret)
637		goto out;
638	ret = vfs_create(dir->d_inode, dentry, mode, NULL);
639	dentry->d_fsdata = NULL;
640	if (ret)
641		goto out_drop_write;
642
643	result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
644	/*
645	 * dentry_open() took a persistent mnt_want_write(),
646	 * so we can now drop this one.
647	 */
648	mnt_drop_write(ipc_ns->mq_mnt);
649	return result;
650
651out_drop_write:
652	mnt_drop_write(ipc_ns->mq_mnt);
653out:
654	dput(dentry);
655	mntput(ipc_ns->mq_mnt);
656	return ERR_PTR(ret);
657}
658
659/* Opens existing queue */
660static struct file *do_open(struct ipc_namespace *ipc_ns,
661				struct dentry *dentry, int oflag)
662{
663	int ret;
664	const struct cred *cred = current_cred();
665
666	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
667						  MAY_READ | MAY_WRITE };
668
669	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
670		ret = -EINVAL;
671		goto err;
672	}
673
674	if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
675		ret = -EACCES;
676		goto err;
677	}
678
679	return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
680
681err:
682	dput(dentry);
683	mntput(ipc_ns->mq_mnt);
684	return ERR_PTR(ret);
685}
686
687SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
688		struct mq_attr __user *, u_attr)
689{
690	struct dentry *dentry;
691	struct file *filp;
692	char *name;
693	struct mq_attr attr;
694	int fd, error;
695	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
696
697	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
698		return -EFAULT;
699
700	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
701
702	if (IS_ERR(name = getname(u_name)))
703		return PTR_ERR(name);
704
705	fd = get_unused_fd_flags(O_CLOEXEC);
706	if (fd < 0)
707		goto out_putname;
708
709	mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
710	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
711	if (IS_ERR(dentry)) {
712		error = PTR_ERR(dentry);
713		goto out_putfd;
714	}
715	mntget(ipc_ns->mq_mnt);
716
717	if (oflag & O_CREAT) {
718		if (dentry->d_inode) {	/* entry already exists */
719			audit_inode(name, dentry);
720			if (oflag & O_EXCL) {
721				error = -EEXIST;
722				goto out;
723			}
724			filp = do_open(ipc_ns, dentry, oflag);
725		} else {
726			filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
727						dentry, oflag, mode,
728						u_attr ? &attr : NULL);
729		}
730	} else {
731		if (!dentry->d_inode) {
732			error = -ENOENT;
733			goto out;
734		}
735		audit_inode(name, dentry);
736		filp = do_open(ipc_ns, dentry, oflag);
737	}
738
739	if (IS_ERR(filp)) {
740		error = PTR_ERR(filp);
741		goto out_putfd;
742	}
743
744	fd_install(fd, filp);
745	goto out_upsem;
746
747out:
748	dput(dentry);
749	mntput(ipc_ns->mq_mnt);
750out_putfd:
751	put_unused_fd(fd);
752	fd = error;
753out_upsem:
754	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
755out_putname:
756	putname(name);
757	return fd;
758}
759
760SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
761{
762	int err;
763	char *name;
764	struct dentry *dentry;
765	struct inode *inode = NULL;
766	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
767
768	name = getname(u_name);
769	if (IS_ERR(name))
770		return PTR_ERR(name);
771
772	mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
773			I_MUTEX_PARENT);
774	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
775	if (IS_ERR(dentry)) {
776		err = PTR_ERR(dentry);
777		goto out_unlock;
778	}
779
780	if (!dentry->d_inode) {
781		err = -ENOENT;
782		goto out_err;
783	}
784
785	inode = dentry->d_inode;
786	if (inode)
787		atomic_inc(&inode->i_count);
788	err = mnt_want_write(ipc_ns->mq_mnt);
789	if (err)
790		goto out_err;
791	err = vfs_unlink(dentry->d_parent->d_inode, dentry);
792	mnt_drop_write(ipc_ns->mq_mnt);
793out_err:
794	dput(dentry);
795
796out_unlock:
797	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
798	putname(name);
799	if (inode)
800		iput(inode);
801
802	return err;
803}
804
805/* Pipelined send and receive functions.
806 *
807 * If a receiver finds no waiting message, then it registers itself in the
808 * list of waiting receivers. A sender checks that list before adding the new
809 * message into the message array. If there is a waiting receiver, then it
810 * bypasses the message array and directly hands the message over to the
811 * receiver.
812 * The receiver accepts the message and returns without grabbing the queue
813 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
814 * are necessary. The same algorithm is used for sysv semaphores, see
815 * ipc/sem.c for more details.
816 *
817 * The same algorithm is used for senders.
818 */
819
820/* pipelined_send() - send a message directly to the task waiting in
821 * sys_mq_timedreceive() (without inserting message into a queue).
822 */
823static inline void pipelined_send(struct mqueue_inode_info *info,
824				  struct msg_msg *message,
825				  struct ext_wait_queue *receiver)
826{
827	receiver->msg = message;
828	list_del(&receiver->list);
829	receiver->state = STATE_PENDING;
830	wake_up_process(receiver->task);
831	smp_wmb();
832	receiver->state = STATE_READY;
833}
834
835/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
836 * gets its message and put to the queue (we have one free place for sure). */
837static inline void pipelined_receive(struct mqueue_inode_info *info)
838{
839	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
840
841	if (!sender) {
842		/* for poll */
843		wake_up_interruptible(&info->wait_q);
844		return;
845	}
846	msg_insert(sender->msg, info);
847	list_del(&sender->list);
848	sender->state = STATE_PENDING;
849	wake_up_process(sender->task);
850	smp_wmb();
851	sender->state = STATE_READY;
852}
853
854SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
855		size_t, msg_len, unsigned int, msg_prio,
856		const struct timespec __user *, u_abs_timeout)
857{
858	struct file *filp;
859	struct inode *inode;
860	struct ext_wait_queue wait;
861	struct ext_wait_queue *receiver;
862	struct msg_msg *msg_ptr;
863	struct mqueue_inode_info *info;
864	struct timespec ts, *p = NULL;
865	long timeout;
866	int ret;
867
868	if (u_abs_timeout) {
869		if (copy_from_user(&ts, u_abs_timeout,
870					sizeof(struct timespec)))
871			return -EFAULT;
872		p = &ts;
873	}
874
875	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
876		return -EINVAL;
877
878	audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
879	timeout = prepare_timeout(p);
880
881	filp = fget(mqdes);
882	if (unlikely(!filp)) {
883		ret = -EBADF;
884		goto out;
885	}
886
887	inode = filp->f_path.dentry->d_inode;
888	if (unlikely(filp->f_op != &mqueue_file_operations)) {
889		ret = -EBADF;
890		goto out_fput;
891	}
892	info = MQUEUE_I(inode);
893	audit_inode(NULL, filp->f_path.dentry);
894
895	if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
896		ret = -EBADF;
897		goto out_fput;
898	}
899
900	if (unlikely(msg_len > info->attr.mq_msgsize)) {
901		ret = -EMSGSIZE;
902		goto out_fput;
903	}
904
905	/* First try to allocate memory, before doing anything with
906	 * existing queues. */
907	msg_ptr = load_msg(u_msg_ptr, msg_len);
908	if (IS_ERR(msg_ptr)) {
909		ret = PTR_ERR(msg_ptr);
910		goto out_fput;
911	}
912	msg_ptr->m_ts = msg_len;
913	msg_ptr->m_type = msg_prio;
914
915	spin_lock(&info->lock);
916
917	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
918		if (filp->f_flags & O_NONBLOCK) {
919			spin_unlock(&info->lock);
920			ret = -EAGAIN;
921		} else if (unlikely(timeout < 0)) {
922			spin_unlock(&info->lock);
923			ret = timeout;
924		} else {
925			wait.task = current;
926			wait.msg = (void *) msg_ptr;
927			wait.state = STATE_NONE;
928			ret = wq_sleep(info, SEND, timeout, &wait);
929		}
930		if (ret < 0)
931			free_msg(msg_ptr);
932	} else {
933		receiver = wq_get_first_waiter(info, RECV);
934		if (receiver) {
935			pipelined_send(info, msg_ptr, receiver);
936		} else {
937			/* adds message to the queue */
938			msg_insert(msg_ptr, info);
939			__do_notify(info);
940		}
941		inode->i_atime = inode->i_mtime = inode->i_ctime =
942				CURRENT_TIME;
943		spin_unlock(&info->lock);
944		ret = 0;
945	}
946out_fput:
947	fput(filp);
948out:
949	return ret;
950}
951
952SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
953		size_t, msg_len, unsigned int __user *, u_msg_prio,
954		const struct timespec __user *, u_abs_timeout)
955{
956	long timeout;
957	ssize_t ret;
958	struct msg_msg *msg_ptr;
959	struct file *filp;
960	struct inode *inode;
961	struct mqueue_inode_info *info;
962	struct ext_wait_queue wait;
963	struct timespec ts, *p = NULL;
964
965	if (u_abs_timeout) {
966		if (copy_from_user(&ts, u_abs_timeout,
967					sizeof(struct timespec)))
968			return -EFAULT;
969		p = &ts;
970	}
971
972	audit_mq_sendrecv(mqdes, msg_len, 0, p);
973	timeout = prepare_timeout(p);
974
975	filp = fget(mqdes);
976	if (unlikely(!filp)) {
977		ret = -EBADF;
978		goto out;
979	}
980
981	inode = filp->f_path.dentry->d_inode;
982	if (unlikely(filp->f_op != &mqueue_file_operations)) {
983		ret = -EBADF;
984		goto out_fput;
985	}
986	info = MQUEUE_I(inode);
987	audit_inode(NULL, filp->f_path.dentry);
988
989	if (unlikely(!(filp->f_mode & FMODE_READ))) {
990		ret = -EBADF;
991		goto out_fput;
992	}
993
994	/* checks if buffer is big enough */
995	if (unlikely(msg_len < info->attr.mq_msgsize)) {
996		ret = -EMSGSIZE;
997		goto out_fput;
998	}
999
1000	spin_lock(&info->lock);
1001	if (info->attr.mq_curmsgs == 0) {
1002		if (filp->f_flags & O_NONBLOCK) {
1003			spin_unlock(&info->lock);
1004			ret = -EAGAIN;
1005			msg_ptr = NULL;
1006		} else if (unlikely(timeout < 0)) {
1007			spin_unlock(&info->lock);
1008			ret = timeout;
1009			msg_ptr = NULL;
1010		} else {
1011			wait.task = current;
1012			wait.state = STATE_NONE;
1013			ret = wq_sleep(info, RECV, timeout, &wait);
1014			msg_ptr = wait.msg;
1015		}
1016	} else {
1017		msg_ptr = msg_get(info);
1018
1019		inode->i_atime = inode->i_mtime = inode->i_ctime =
1020				CURRENT_TIME;
1021
1022		/* There is now free space in queue. */
1023		pipelined_receive(info);
1024		spin_unlock(&info->lock);
1025		ret = 0;
1026	}
1027	if (ret == 0) {
1028		ret = msg_ptr->m_ts;
1029
1030		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1031			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1032			ret = -EFAULT;
1033		}
1034		free_msg(msg_ptr);
1035	}
1036out_fput:
1037	fput(filp);
1038out:
1039	return ret;
1040}
1041
1042/*
1043 * Notes: the case when user wants us to deregister (with NULL as pointer)
1044 * and he isn't currently owner of notification, will be silently discarded.
1045 * It isn't explicitly defined in the POSIX.
1046 */
1047SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1048		const struct sigevent __user *, u_notification)
1049{
1050	int ret;
1051	struct file *filp;
1052	struct sock *sock;
1053	struct inode *inode;
1054	struct sigevent notification;
1055	struct mqueue_inode_info *info;
1056	struct sk_buff *nc;
1057
1058	if (u_notification) {
1059		if (copy_from_user(&notification, u_notification,
1060					sizeof(struct sigevent)))
1061			return -EFAULT;
1062	}
1063
1064	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1065
1066	nc = NULL;
1067	sock = NULL;
1068	if (u_notification != NULL) {
1069		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1070			     notification.sigev_notify != SIGEV_SIGNAL &&
1071			     notification.sigev_notify != SIGEV_THREAD))
1072			return -EINVAL;
1073		if (notification.sigev_notify == SIGEV_SIGNAL &&
1074			!valid_signal(notification.sigev_signo)) {
1075			return -EINVAL;
1076		}
1077		if (notification.sigev_notify == SIGEV_THREAD) {
1078			long timeo;
1079
1080			/* create the notify skb */
1081			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1082			if (!nc) {
1083				ret = -ENOMEM;
1084				goto out;
1085			}
1086			if (copy_from_user(nc->data,
1087					notification.sigev_value.sival_ptr,
1088					NOTIFY_COOKIE_LEN)) {
1089				ret = -EFAULT;
1090				goto out;
1091			}
1092
1093			/* TODO: add a header? */
1094			skb_put(nc, NOTIFY_COOKIE_LEN);
1095			/* and attach it to the socket */
1096retry:
1097			filp = fget(notification.sigev_signo);
1098			if (!filp) {
1099				ret = -EBADF;
1100				goto out;
1101			}
1102			sock = netlink_getsockbyfilp(filp);
1103			fput(filp);
1104			if (IS_ERR(sock)) {
1105				ret = PTR_ERR(sock);
1106				sock = NULL;
1107				goto out;
1108			}
1109
1110			timeo = MAX_SCHEDULE_TIMEOUT;
1111			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1112			if (ret == 1)
1113				goto retry;
1114			if (ret) {
1115				sock = NULL;
1116				nc = NULL;
1117				goto out;
1118			}
1119		}
1120	}
1121
1122	filp = fget(mqdes);
1123	if (!filp) {
1124		ret = -EBADF;
1125		goto out;
1126	}
1127
1128	inode = filp->f_path.dentry->d_inode;
1129	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1130		ret = -EBADF;
1131		goto out_fput;
1132	}
1133	info = MQUEUE_I(inode);
1134
1135	ret = 0;
1136	spin_lock(&info->lock);
1137	if (u_notification == NULL) {
1138		if (info->notify_owner == task_tgid(current)) {
1139			remove_notification(info);
1140			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1141		}
1142	} else if (info->notify_owner != NULL) {
1143		ret = -EBUSY;
1144	} else {
1145		switch (notification.sigev_notify) {
1146		case SIGEV_NONE:
1147			info->notify.sigev_notify = SIGEV_NONE;
1148			break;
1149		case SIGEV_THREAD:
1150			info->notify_sock = sock;
1151			info->notify_cookie = nc;
1152			sock = NULL;
1153			nc = NULL;
1154			info->notify.sigev_notify = SIGEV_THREAD;
1155			break;
1156		case SIGEV_SIGNAL:
1157			info->notify.sigev_signo = notification.sigev_signo;
1158			info->notify.sigev_value = notification.sigev_value;
1159			info->notify.sigev_notify = SIGEV_SIGNAL;
1160			break;
1161		}
1162
1163		info->notify_owner = get_pid(task_tgid(current));
1164		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1165	}
1166	spin_unlock(&info->lock);
1167out_fput:
1168	fput(filp);
1169out:
1170	if (sock) {
1171		netlink_detachskb(sock, nc);
1172	} else if (nc) {
1173		dev_kfree_skb(nc);
1174	}
1175	return ret;
1176}
1177
1178SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1179		const struct mq_attr __user *, u_mqstat,
1180		struct mq_attr __user *, u_omqstat)
1181{
1182	int ret;
1183	struct mq_attr mqstat, omqstat;
1184	struct file *filp;
1185	struct inode *inode;
1186	struct mqueue_inode_info *info;
1187
1188	if (u_mqstat != NULL) {
1189		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1190			return -EFAULT;
1191		if (mqstat.mq_flags & (~O_NONBLOCK))
1192			return -EINVAL;
1193	}
1194
1195	filp = fget(mqdes);
1196	if (!filp) {
1197		ret = -EBADF;
1198		goto out;
1199	}
1200
1201	inode = filp->f_path.dentry->d_inode;
1202	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1203		ret = -EBADF;
1204		goto out_fput;
1205	}
1206	info = MQUEUE_I(inode);
1207
1208	spin_lock(&info->lock);
1209
1210	omqstat = info->attr;
1211	omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1212	if (u_mqstat) {
1213		audit_mq_getsetattr(mqdes, &mqstat);
1214		spin_lock(&filp->f_lock);
1215		if (mqstat.mq_flags & O_NONBLOCK)
1216			filp->f_flags |= O_NONBLOCK;
1217		else
1218			filp->f_flags &= ~O_NONBLOCK;
1219		spin_unlock(&filp->f_lock);
1220
1221		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1222	}
1223
1224	spin_unlock(&info->lock);
1225
1226	ret = 0;
1227	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1228						sizeof(struct mq_attr)))
1229		ret = -EFAULT;
1230
1231out_fput:
1232	fput(filp);
1233out:
1234	return ret;
1235}
1236
1237static const struct inode_operations mqueue_dir_inode_operations = {
1238	.lookup = simple_lookup,
1239	.create = mqueue_create,
1240	.unlink = mqueue_unlink,
1241};
1242
1243static const struct file_operations mqueue_file_operations = {
1244	.flush = mqueue_flush_file,
1245	.poll = mqueue_poll_file,
1246	.read = mqueue_read_file,
1247};
1248
1249static const struct super_operations mqueue_super_ops = {
1250	.alloc_inode = mqueue_alloc_inode,
1251	.destroy_inode = mqueue_destroy_inode,
1252	.statfs = simple_statfs,
1253	.delete_inode = mqueue_delete_inode,
1254	.drop_inode = generic_delete_inode,
1255};
1256
1257static struct file_system_type mqueue_fs_type = {
1258	.name = "mqueue",
1259	.get_sb = mqueue_get_sb,
1260	.kill_sb = kill_litter_super,
1261};
1262
1263int mq_init_ns(struct ipc_namespace *ns)
1264{
1265	ns->mq_queues_count  = 0;
1266	ns->mq_queues_max    = DFLT_QUEUESMAX;
1267	ns->mq_msg_max       = DFLT_MSGMAX;
1268	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1269
1270	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1271	if (IS_ERR(ns->mq_mnt)) {
1272		int err = PTR_ERR(ns->mq_mnt);
1273		ns->mq_mnt = NULL;
1274		return err;
1275	}
1276	return 0;
1277}
1278
1279void mq_clear_sbinfo(struct ipc_namespace *ns)
1280{
1281	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1282}
1283
1284void mq_put_mnt(struct ipc_namespace *ns)
1285{
1286	mntput(ns->mq_mnt);
1287}
1288
1289static int __init init_mqueue_fs(void)
1290{
1291	int error;
1292
1293	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1294				sizeof(struct mqueue_inode_info), 0,
1295				SLAB_HWCACHE_ALIGN, init_once);
1296	if (mqueue_inode_cachep == NULL)
1297		return -ENOMEM;
1298
1299	/* ignore failures - they are not fatal */
1300	mq_sysctl_table = mq_register_sysctl_table();
1301
1302	error = register_filesystem(&mqueue_fs_type);
1303	if (error)
1304		goto out_sysctl;
1305
1306	spin_lock_init(&mq_lock);
1307
1308	init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1309	if (IS_ERR(init_ipc_ns.mq_mnt)) {
1310		error = PTR_ERR(init_ipc_ns.mq_mnt);
1311		goto out_filesystem;
1312	}
1313
1314	return 0;
1315
1316out_filesystem:
1317	unregister_filesystem(&mqueue_fs_type);
1318out_sysctl:
1319	if (mq_sysctl_table)
1320		unregister_sysctl_table(mq_sysctl_table);
1321	kmem_cache_destroy(mqueue_inode_cachep);
1322	return error;
1323}
1324
1325__initcall(init_mqueue_fs);
1326