mqueue.c revision a609164f7c4eb82b97444871cb47f2f261696dd9 
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@motorola.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 * This file is released under the GPL.
12 */
13
14#include <linux/capability.h>
15#include <linux/init.h>
16#include <linux/pagemap.h>
17#include <linux/file.h>
18#include <linux/mount.h>
19#include <linux/namei.h>
20#include <linux/sysctl.h>
21#include <linux/poll.h>
22#include <linux/mqueue.h>
23#include <linux/msg.h>
24#include <linux/skbuff.h>
25#include <linux/netlink.h>
26#include <linux/syscalls.h>
27#include <linux/signal.h>
28#include <net/sock.h>
29#include "util.h"
30
31#define MQUEUE_MAGIC	0x19800202
32#define DIRENT_SIZE	20
33#define FILENT_SIZE	80
34
35#define SEND		0
36#define RECV		1
37
38#define STATE_NONE	0
39#define STATE_PENDING	1
40#define STATE_READY	2
41
42/* used by sysctl */
43#define FS_MQUEUE 	1
44#define CTL_QUEUESMAX 	2
45#define CTL_MSGMAX 	3
46#define CTL_MSGSIZEMAX 	4
47
48/* default values */
49#define DFLT_QUEUESMAX	256	/* max number of message queues */
50#define DFLT_MSGMAX 	10	/* max number of messages in each queue */
51#define HARD_MSGMAX 	(131072/sizeof(void*))
52#define DFLT_MSGSIZEMAX 8192	/* max message size */
53
54
55struct ext_wait_queue {		/* queue of sleeping tasks */
56	struct task_struct *task;
57	struct list_head list;
58	struct msg_msg *msg;	/* ptr of loaded message */
59	int state;		/* one of STATE_* values */
60};
61
62struct mqueue_inode_info {
63	spinlock_t lock;
64	struct inode vfs_inode;
65	wait_queue_head_t wait_q;
66
67	struct msg_msg **messages;
68	struct mq_attr attr;
69
70	struct sigevent notify;
71	pid_t notify_owner;
72	struct user_struct *user;	/* user who created, for accounting */
73	struct sock *notify_sock;
74	struct sk_buff *notify_cookie;
75
76	/* for tasks waiting for free space and messages, respectively */
77	struct ext_wait_queue e_wait_q[2];
78
79	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
80};
81
82static struct inode_operations mqueue_dir_inode_operations;
83static struct file_operations mqueue_file_operations;
84static struct super_operations mqueue_super_ops;
85static void remove_notification(struct mqueue_inode_info *info);
86
87static spinlock_t mq_lock;
88static kmem_cache_t *mqueue_inode_cachep;
89static struct vfsmount *mqueue_mnt;
90
91static unsigned int queues_count;
92static unsigned int queues_max 	= DFLT_QUEUESMAX;
93static unsigned int msg_max 	= DFLT_MSGMAX;
94static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
95
96static struct ctl_table_header * mq_sysctl_table;
97
98static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
99{
100	return container_of(inode, struct mqueue_inode_info, vfs_inode);
101}
102
103static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
104							struct mq_attr *attr)
105{
106	struct inode *inode;
107
108	inode = new_inode(sb);
109	if (inode) {
110		inode->i_mode = mode;
111		inode->i_uid = current->fsuid;
112		inode->i_gid = current->fsgid;
113		inode->i_blksize = PAGE_CACHE_SIZE;
114		inode->i_blocks = 0;
115		inode->i_mtime = inode->i_ctime = inode->i_atime =
116				CURRENT_TIME;
117
118		if (S_ISREG(mode)) {
119			struct mqueue_inode_info *info;
120			struct task_struct *p = current;
121			struct user_struct *u = p->user;
122			unsigned long mq_bytes, mq_msg_tblsz;
123
124			inode->i_fop = &mqueue_file_operations;
125			inode->i_size = FILENT_SIZE;
126			/* mqueue specific info */
127			info = MQUEUE_I(inode);
128			spin_lock_init(&info->lock);
129			init_waitqueue_head(&info->wait_q);
130			INIT_LIST_HEAD(&info->e_wait_q[0].list);
131			INIT_LIST_HEAD(&info->e_wait_q[1].list);
132			info->messages = NULL;
133			info->notify_owner = 0;
134			info->qsize = 0;
135			info->user = NULL;	/* set when all is ok */
136			memset(&info->attr, 0, sizeof(info->attr));
137			info->attr.mq_maxmsg = DFLT_MSGMAX;
138			info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
139			if (attr) {
140				info->attr.mq_maxmsg = attr->mq_maxmsg;
141				info->attr.mq_msgsize = attr->mq_msgsize;
142			}
143			mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
144			mq_bytes = (mq_msg_tblsz +
145				(info->attr.mq_maxmsg * info->attr.mq_msgsize));
146
147			spin_lock(&mq_lock);
148			if (u->mq_bytes + mq_bytes < u->mq_bytes ||
149		 	    u->mq_bytes + mq_bytes >
150			    p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
151				spin_unlock(&mq_lock);
152				goto out_inode;
153			}
154			u->mq_bytes += mq_bytes;
155			spin_unlock(&mq_lock);
156
157			info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
158			if (!info->messages) {
159				spin_lock(&mq_lock);
160				u->mq_bytes -= mq_bytes;
161				spin_unlock(&mq_lock);
162				goto out_inode;
163			}
164			/* all is ok */
165			info->user = get_uid(u);
166		} else if (S_ISDIR(mode)) {
167			inode->i_nlink++;
168			/* Some things misbehave if size == 0 on a directory */
169			inode->i_size = 2 * DIRENT_SIZE;
170			inode->i_op = &mqueue_dir_inode_operations;
171			inode->i_fop = &simple_dir_operations;
172		}
173	}
174	return inode;
175out_inode:
176	make_bad_inode(inode);
177	iput(inode);
178	return NULL;
179}
180
181static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
182{
183	struct inode *inode;
184
185	sb->s_blocksize = PAGE_CACHE_SIZE;
186	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
187	sb->s_magic = MQUEUE_MAGIC;
188	sb->s_op = &mqueue_super_ops;
189
190	inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
191	if (!inode)
192		return -ENOMEM;
193
194	sb->s_root = d_alloc_root(inode);
195	if (!sb->s_root) {
196		iput(inode);
197		return -ENOMEM;
198	}
199
200	return 0;
201}
202
203static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
204					 int flags, const char *dev_name,
205					 void *data)
206{
207	return get_sb_single(fs_type, flags, data, mqueue_fill_super);
208}
209
210static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
211{
212	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
213
214	if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
215		SLAB_CTOR_CONSTRUCTOR)
216		inode_init_once(&p->vfs_inode);
217}
218
219static struct inode *mqueue_alloc_inode(struct super_block *sb)
220{
221	struct mqueue_inode_info *ei;
222
223	ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
224	if (!ei)
225		return NULL;
226	return &ei->vfs_inode;
227}
228
229static void mqueue_destroy_inode(struct inode *inode)
230{
231	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
232}
233
234static void mqueue_delete_inode(struct inode *inode)
235{
236	struct mqueue_inode_info *info;
237	struct user_struct *user;
238	unsigned long mq_bytes;
239	int i;
240
241	if (S_ISDIR(inode->i_mode)) {
242		clear_inode(inode);
243		return;
244	}
245	info = MQUEUE_I(inode);
246	spin_lock(&info->lock);
247	for (i = 0; i < info->attr.mq_curmsgs; i++)
248		free_msg(info->messages[i]);
249	kfree(info->messages);
250	spin_unlock(&info->lock);
251
252	clear_inode(inode);
253
254	mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
255		   (info->attr.mq_maxmsg * info->attr.mq_msgsize));
256	user = info->user;
257	if (user) {
258		spin_lock(&mq_lock);
259		user->mq_bytes -= mq_bytes;
260		queues_count--;
261		spin_unlock(&mq_lock);
262		free_uid(user);
263	}
264}
265
266static int mqueue_create(struct inode *dir, struct dentry *dentry,
267				int mode, struct nameidata *nd)
268{
269	struct inode *inode;
270	struct mq_attr *attr = dentry->d_fsdata;
271	int error;
272
273	spin_lock(&mq_lock);
274	if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
275		error = -ENOSPC;
276		goto out_lock;
277	}
278	queues_count++;
279	spin_unlock(&mq_lock);
280
281	inode = mqueue_get_inode(dir->i_sb, mode, attr);
282	if (!inode) {
283		error = -ENOMEM;
284		spin_lock(&mq_lock);
285		queues_count--;
286		goto out_lock;
287	}
288
289	dir->i_size += DIRENT_SIZE;
290	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
291
292	d_instantiate(dentry, inode);
293	dget(dentry);
294	return 0;
295out_lock:
296	spin_unlock(&mq_lock);
297	return error;
298}
299
300static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
301{
302  	struct inode *inode = dentry->d_inode;
303
304	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
305	dir->i_size -= DIRENT_SIZE;
306  	inode->i_nlink--;
307  	dput(dentry);
308  	return 0;
309}
310
311/*
312*	This is routine for system read from queue file.
313*	To avoid mess with doing here some sort of mq_receive we allow
314*	to read only queue size & notification info (the only values
315*	that are interesting from user point of view and aren't accessible
316*	through std routines)
317*/
318static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
319				size_t count, loff_t * off)
320{
321	struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
322	char buffer[FILENT_SIZE];
323	size_t slen;
324	loff_t o;
325
326	if (!count)
327		return 0;
328
329	spin_lock(&info->lock);
330	snprintf(buffer, sizeof(buffer),
331			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
332			info->qsize,
333			info->notify_owner ? info->notify.sigev_notify : 0,
334			(info->notify_owner &&
335			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
336				info->notify.sigev_signo : 0,
337			info->notify_owner);
338	spin_unlock(&info->lock);
339	buffer[sizeof(buffer)-1] = '\0';
340	slen = strlen(buffer)+1;
341
342	o = *off;
343	if (o > slen)
344		return 0;
345
346	if (o + count > slen)
347		count = slen - o;
348
349	if (copy_to_user(u_data, buffer + o, count))
350		return -EFAULT;
351
352	*off = o + count;
353	filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
354	return count;
355}
356
357static int mqueue_flush_file(struct file *filp)
358{
359	struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
360
361	spin_lock(&info->lock);
362	if (current->tgid == info->notify_owner)
363		remove_notification(info);
364
365	spin_unlock(&info->lock);
366	return 0;
367}
368
369static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
370{
371	struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
372	int retval = 0;
373
374	poll_wait(filp, &info->wait_q, poll_tab);
375
376	spin_lock(&info->lock);
377	if (info->attr.mq_curmsgs)
378		retval = POLLIN | POLLRDNORM;
379
380	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
381		retval |= POLLOUT | POLLWRNORM;
382	spin_unlock(&info->lock);
383
384	return retval;
385}
386
387/* Adds current to info->e_wait_q[sr] before element with smaller prio */
388static void wq_add(struct mqueue_inode_info *info, int sr,
389			struct ext_wait_queue *ewp)
390{
391	struct ext_wait_queue *walk;
392
393	ewp->task = current;
394
395	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
396		if (walk->task->static_prio <= current->static_prio) {
397			list_add_tail(&ewp->list, &walk->list);
398			return;
399		}
400	}
401	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
402}
403
404/*
405 * Puts current task to sleep. Caller must hold queue lock. After return
406 * lock isn't held.
407 * sr: SEND or RECV
408 */
409static int wq_sleep(struct mqueue_inode_info *info, int sr,
410			long timeout, struct ext_wait_queue *ewp)
411{
412	int retval;
413	signed long time;
414
415	wq_add(info, sr, ewp);
416
417	for (;;) {
418		set_current_state(TASK_INTERRUPTIBLE);
419
420		spin_unlock(&info->lock);
421		time = schedule_timeout(timeout);
422
423		while (ewp->state == STATE_PENDING)
424			cpu_relax();
425
426		if (ewp->state == STATE_READY) {
427			retval = 0;
428			goto out;
429		}
430		spin_lock(&info->lock);
431		if (ewp->state == STATE_READY) {
432			retval = 0;
433			goto out_unlock;
434		}
435		if (signal_pending(current)) {
436			retval = -ERESTARTSYS;
437			break;
438		}
439		if (time == 0) {
440			retval = -ETIMEDOUT;
441			break;
442		}
443	}
444	list_del(&ewp->list);
445out_unlock:
446	spin_unlock(&info->lock);
447out:
448	return retval;
449}
450
451/*
452 * Returns waiting task that should be serviced first or NULL if none exists
453 */
454static struct ext_wait_queue *wq_get_first_waiter(
455		struct mqueue_inode_info *info, int sr)
456{
457	struct list_head *ptr;
458
459	ptr = info->e_wait_q[sr].list.prev;
460	if (ptr == &info->e_wait_q[sr].list)
461		return NULL;
462	return list_entry(ptr, struct ext_wait_queue, list);
463}
464
465/* Auxiliary functions to manipulate messages' list */
466static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
467{
468	int k;
469
470	k = info->attr.mq_curmsgs - 1;
471	while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
472		info->messages[k + 1] = info->messages[k];
473		k--;
474	}
475	info->attr.mq_curmsgs++;
476	info->qsize += ptr->m_ts;
477	info->messages[k + 1] = ptr;
478}
479
480static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
481{
482	info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
483	return info->messages[info->attr.mq_curmsgs];
484}
485
486static inline void set_cookie(struct sk_buff *skb, char code)
487{
488	((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
489}
490
491/*
492 * The next function is only to split too long sys_mq_timedsend
493 */
494static void __do_notify(struct mqueue_inode_info *info)
495{
496	/* notification
497	 * invoked when there is registered process and there isn't process
498	 * waiting synchronously for message AND state of queue changed from
499	 * empty to not empty. Here we are sure that no one is waiting
500	 * synchronously. */
501	if (info->notify_owner &&
502	    info->attr.mq_curmsgs == 1) {
503		struct siginfo sig_i;
504		switch (info->notify.sigev_notify) {
505		case SIGEV_NONE:
506			break;
507		case SIGEV_SIGNAL:
508			/* sends signal */
509
510			sig_i.si_signo = info->notify.sigev_signo;
511			sig_i.si_errno = 0;
512			sig_i.si_code = SI_MESGQ;
513			sig_i.si_value = info->notify.sigev_value;
514			sig_i.si_pid = current->tgid;
515			sig_i.si_uid = current->uid;
516
517			kill_proc_info(info->notify.sigev_signo,
518				       &sig_i, info->notify_owner);
519			break;
520		case SIGEV_THREAD:
521			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
522			netlink_sendskb(info->notify_sock,
523					info->notify_cookie, 0);
524			break;
525		}
526		/* after notification unregisters process */
527		info->notify_owner = 0;
528	}
529	wake_up(&info->wait_q);
530}
531
532static long prepare_timeout(const struct timespec __user *u_arg)
533{
534	struct timespec ts, nowts;
535	long timeout;
536
537	if (u_arg) {
538		if (unlikely(copy_from_user(&ts, u_arg,
539					sizeof(struct timespec))))
540			return -EFAULT;
541
542		if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
543			|| ts.tv_nsec >= NSEC_PER_SEC))
544			return -EINVAL;
545		nowts = CURRENT_TIME;
546		/* first subtract as jiffies can't be too big */
547		ts.tv_sec -= nowts.tv_sec;
548		if (ts.tv_nsec < nowts.tv_nsec) {
549			ts.tv_nsec += NSEC_PER_SEC;
550			ts.tv_sec--;
551		}
552		ts.tv_nsec -= nowts.tv_nsec;
553		if (ts.tv_sec < 0)
554			return 0;
555
556		timeout = timespec_to_jiffies(&ts) + 1;
557	} else
558		return MAX_SCHEDULE_TIMEOUT;
559
560	return timeout;
561}
562
563static void remove_notification(struct mqueue_inode_info *info)
564{
565	if (info->notify_owner != 0 &&
566	    info->notify.sigev_notify == SIGEV_THREAD) {
567		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
568		netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
569	}
570	info->notify_owner = 0;
571}
572
573static int mq_attr_ok(struct mq_attr *attr)
574{
575	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
576		return 0;
577	if (capable(CAP_SYS_RESOURCE)) {
578		if (attr->mq_maxmsg > HARD_MSGMAX)
579			return 0;
580	} else {
581		if (attr->mq_maxmsg > msg_max ||
582				attr->mq_msgsize > msgsize_max)
583			return 0;
584	}
585	/* check for overflow */
586	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
587		return 0;
588	if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
589	    (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
590	    (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
591		return 0;
592	return 1;
593}
594
595/*
596 * Invoked when creating a new queue via sys_mq_open
597 */
598static struct file *do_create(struct dentry *dir, struct dentry *dentry,
599			int oflag, mode_t mode, struct mq_attr __user *u_attr)
600{
601	struct mq_attr attr;
602	int ret;
603
604	if (u_attr) {
605		ret = -EFAULT;
606		if (copy_from_user(&attr, u_attr, sizeof(attr)))
607			goto out;
608		ret = -EINVAL;
609		if (!mq_attr_ok(&attr))
610			goto out;
611		/* store for use during create */
612		dentry->d_fsdata = &attr;
613	}
614
615	mode &= ~current->fs->umask;
616	ret = vfs_create(dir->d_inode, dentry, mode, NULL);
617	dentry->d_fsdata = NULL;
618	if (ret)
619		goto out;
620
621	return dentry_open(dentry, mqueue_mnt, oflag);
622
623out:
624	dput(dentry);
625	mntput(mqueue_mnt);
626	return ERR_PTR(ret);
627}
628
629/* Opens existing queue */
630static struct file *do_open(struct dentry *dentry, int oflag)
631{
632static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
633					MAY_READ | MAY_WRITE };
634
635	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
636		dput(dentry);
637		mntput(mqueue_mnt);
638		return ERR_PTR(-EINVAL);
639	}
640
641	if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
642		dput(dentry);
643		mntput(mqueue_mnt);
644		return ERR_PTR(-EACCES);
645	}
646
647	return dentry_open(dentry, mqueue_mnt, oflag);
648}
649
650asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
651				struct mq_attr __user *u_attr)
652{
653	struct dentry *dentry;
654	struct file *filp;
655	char *name;
656	int fd, error;
657
658	if (IS_ERR(name = getname(u_name)))
659		return PTR_ERR(name);
660
661	fd = get_unused_fd();
662	if (fd < 0)
663		goto out_putname;
664
665	mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
666	dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
667	if (IS_ERR(dentry)) {
668		error = PTR_ERR(dentry);
669		goto out_err;
670	}
671	mntget(mqueue_mnt);
672
673	if (oflag & O_CREAT) {
674		if (dentry->d_inode) {	/* entry already exists */
675			error = -EEXIST;
676			if (oflag & O_EXCL)
677				goto out;
678			filp = do_open(dentry, oflag);
679		} else {
680			filp = do_create(mqueue_mnt->mnt_root, dentry,
681						oflag, mode, u_attr);
682		}
683	} else {
684		error = -ENOENT;
685		if (!dentry->d_inode)
686			goto out;
687		filp = do_open(dentry, oflag);
688	}
689
690	if (IS_ERR(filp)) {
691		error = PTR_ERR(filp);
692		goto out_putfd;
693	}
694
695	set_close_on_exec(fd, 1);
696	fd_install(fd, filp);
697	goto out_upsem;
698
699out:
700	dput(dentry);
701	mntput(mqueue_mnt);
702out_putfd:
703	put_unused_fd(fd);
704out_err:
705	fd = error;
706out_upsem:
707	mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
708out_putname:
709	putname(name);
710	return fd;
711}
712
713asmlinkage long sys_mq_unlink(const char __user *u_name)
714{
715	int err;
716	char *name;
717	struct dentry *dentry;
718	struct inode *inode = NULL;
719
720	name = getname(u_name);
721	if (IS_ERR(name))
722		return PTR_ERR(name);
723
724	mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
725	dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
726	if (IS_ERR(dentry)) {
727		err = PTR_ERR(dentry);
728		goto out_unlock;
729	}
730
731	if (!dentry->d_inode) {
732		err = -ENOENT;
733		goto out_err;
734	}
735
736	inode = dentry->d_inode;
737	if (inode)
738		atomic_inc(&inode->i_count);
739
740	err = vfs_unlink(dentry->d_parent->d_inode, dentry);
741out_err:
742	dput(dentry);
743
744out_unlock:
745	mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
746	putname(name);
747	if (inode)
748		iput(inode);
749
750	return err;
751}
752
753/* Pipelined send and receive functions.
754 *
755 * If a receiver finds no waiting message, then it registers itself in the
756 * list of waiting receivers. A sender checks that list before adding the new
757 * message into the message array. If there is a waiting receiver, then it
758 * bypasses the message array and directly hands the message over to the
759 * receiver.
760 * The receiver accepts the message and returns without grabbing the queue
761 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
762 * are necessary. The same algorithm is used for sysv semaphores, see
763 * ipc/sem.c fore more details.
764 *
765 * The same algorithm is used for senders.
766 */
767
768/* pipelined_send() - send a message directly to the task waiting in
769 * sys_mq_timedreceive() (without inserting message into a queue).
770 */
771static inline void pipelined_send(struct mqueue_inode_info *info,
772				  struct msg_msg *message,
773				  struct ext_wait_queue *receiver)
774{
775	receiver->msg = message;
776	list_del(&receiver->list);
777	receiver->state = STATE_PENDING;
778	wake_up_process(receiver->task);
779	smp_wmb();
780	receiver->state = STATE_READY;
781}
782
783/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
784 * gets its message and put to the queue (we have one free place for sure). */
785static inline void pipelined_receive(struct mqueue_inode_info *info)
786{
787	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
788
789	if (!sender) {
790		/* for poll */
791		wake_up_interruptible(&info->wait_q);
792		return;
793	}
794	msg_insert(sender->msg, info);
795	list_del(&sender->list);
796	sender->state = STATE_PENDING;
797	wake_up_process(sender->task);
798	smp_wmb();
799	sender->state = STATE_READY;
800}
801
802asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
803	size_t msg_len, unsigned int msg_prio,
804	const struct timespec __user *u_abs_timeout)
805{
806	struct file *filp;
807	struct inode *inode;
808	struct ext_wait_queue wait;
809	struct ext_wait_queue *receiver;
810	struct msg_msg *msg_ptr;
811	struct mqueue_inode_info *info;
812	long timeout;
813	int ret;
814
815	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
816		return -EINVAL;
817
818	timeout = prepare_timeout(u_abs_timeout);
819
820	ret = -EBADF;
821	filp = fget(mqdes);
822	if (unlikely(!filp))
823		goto out;
824
825	inode = filp->f_dentry->d_inode;
826	if (unlikely(filp->f_op != &mqueue_file_operations))
827		goto out_fput;
828	info = MQUEUE_I(inode);
829
830	if (unlikely(!(filp->f_mode & FMODE_WRITE)))
831		goto out_fput;
832
833	if (unlikely(msg_len > info->attr.mq_msgsize)) {
834		ret = -EMSGSIZE;
835		goto out_fput;
836	}
837
838	/* First try to allocate memory, before doing anything with
839	 * existing queues. */
840	msg_ptr = load_msg(u_msg_ptr, msg_len);
841	if (IS_ERR(msg_ptr)) {
842		ret = PTR_ERR(msg_ptr);
843		goto out_fput;
844	}
845	msg_ptr->m_ts = msg_len;
846	msg_ptr->m_type = msg_prio;
847
848	spin_lock(&info->lock);
849
850	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
851		if (filp->f_flags & O_NONBLOCK) {
852			spin_unlock(&info->lock);
853			ret = -EAGAIN;
854		} else if (unlikely(timeout < 0)) {
855			spin_unlock(&info->lock);
856			ret = timeout;
857		} else {
858			wait.task = current;
859			wait.msg = (void *) msg_ptr;
860			wait.state = STATE_NONE;
861			ret = wq_sleep(info, SEND, timeout, &wait);
862		}
863		if (ret < 0)
864			free_msg(msg_ptr);
865	} else {
866		receiver = wq_get_first_waiter(info, RECV);
867		if (receiver) {
868			pipelined_send(info, msg_ptr, receiver);
869		} else {
870			/* adds message to the queue */
871			msg_insert(msg_ptr, info);
872			__do_notify(info);
873		}
874		inode->i_atime = inode->i_mtime = inode->i_ctime =
875				CURRENT_TIME;
876		spin_unlock(&info->lock);
877		ret = 0;
878	}
879out_fput:
880	fput(filp);
881out:
882	return ret;
883}
884
885asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
886	size_t msg_len, unsigned int __user *u_msg_prio,
887	const struct timespec __user *u_abs_timeout)
888{
889	long timeout;
890	ssize_t ret;
891	struct msg_msg *msg_ptr;
892	struct file *filp;
893	struct inode *inode;
894	struct mqueue_inode_info *info;
895	struct ext_wait_queue wait;
896
897	timeout = prepare_timeout(u_abs_timeout);
898
899	ret = -EBADF;
900	filp = fget(mqdes);
901	if (unlikely(!filp))
902		goto out;
903
904	inode = filp->f_dentry->d_inode;
905	if (unlikely(filp->f_op != &mqueue_file_operations))
906		goto out_fput;
907	info = MQUEUE_I(inode);
908
909	if (unlikely(!(filp->f_mode & FMODE_READ)))
910		goto out_fput;
911
912	/* checks if buffer is big enough */
913	if (unlikely(msg_len < info->attr.mq_msgsize)) {
914		ret = -EMSGSIZE;
915		goto out_fput;
916	}
917
918	spin_lock(&info->lock);
919	if (info->attr.mq_curmsgs == 0) {
920		if (filp->f_flags & O_NONBLOCK) {
921			spin_unlock(&info->lock);
922			ret = -EAGAIN;
923			msg_ptr = NULL;
924		} else if (unlikely(timeout < 0)) {
925			spin_unlock(&info->lock);
926			ret = timeout;
927			msg_ptr = NULL;
928		} else {
929			wait.task = current;
930			wait.state = STATE_NONE;
931			ret = wq_sleep(info, RECV, timeout, &wait);
932			msg_ptr = wait.msg;
933		}
934	} else {
935		msg_ptr = msg_get(info);
936
937		inode->i_atime = inode->i_mtime = inode->i_ctime =
938				CURRENT_TIME;
939
940		/* There is now free space in queue. */
941		pipelined_receive(info);
942		spin_unlock(&info->lock);
943		ret = 0;
944	}
945	if (ret == 0) {
946		ret = msg_ptr->m_ts;
947
948		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
949			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
950			ret = -EFAULT;
951		}
952		free_msg(msg_ptr);
953	}
954out_fput:
955	fput(filp);
956out:
957	return ret;
958}
959
960/*
961 * Notes: the case when user wants us to deregister (with NULL as pointer)
962 * and he isn't currently owner of notification, will be silently discarded.
963 * It isn't explicitly defined in the POSIX.
964 */
965asmlinkage long sys_mq_notify(mqd_t mqdes,
966				const struct sigevent __user *u_notification)
967{
968	int ret;
969	struct file *filp;
970	struct sock *sock;
971	struct inode *inode;
972	struct sigevent notification;
973	struct mqueue_inode_info *info;
974	struct sk_buff *nc;
975
976	nc = NULL;
977	sock = NULL;
978	if (u_notification != NULL) {
979		if (copy_from_user(&notification, u_notification,
980					sizeof(struct sigevent)))
981			return -EFAULT;
982
983		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
984			     notification.sigev_notify != SIGEV_SIGNAL &&
985			     notification.sigev_notify != SIGEV_THREAD))
986			return -EINVAL;
987		if (notification.sigev_notify == SIGEV_SIGNAL &&
988			!valid_signal(notification.sigev_signo)) {
989			return -EINVAL;
990		}
991		if (notification.sigev_notify == SIGEV_THREAD) {
992			/* create the notify skb */
993			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
994			ret = -ENOMEM;
995			if (!nc)
996				goto out;
997			ret = -EFAULT;
998			if (copy_from_user(nc->data,
999					notification.sigev_value.sival_ptr,
1000					NOTIFY_COOKIE_LEN)) {
1001				goto out;
1002			}
1003
1004			/* TODO: add a header? */
1005			skb_put(nc, NOTIFY_COOKIE_LEN);
1006			/* and attach it to the socket */
1007retry:
1008			filp = fget(notification.sigev_signo);
1009			ret = -EBADF;
1010			if (!filp)
1011				goto out;
1012			sock = netlink_getsockbyfilp(filp);
1013			fput(filp);
1014			if (IS_ERR(sock)) {
1015				ret = PTR_ERR(sock);
1016				sock = NULL;
1017				goto out;
1018			}
1019
1020			ret = netlink_attachskb(sock, nc, 0,
1021					MAX_SCHEDULE_TIMEOUT, NULL);
1022			if (ret == 1)
1023		       		goto retry;
1024			if (ret) {
1025				sock = NULL;
1026				nc = NULL;
1027				goto out;
1028			}
1029		}
1030	}
1031
1032	ret = -EBADF;
1033	filp = fget(mqdes);
1034	if (!filp)
1035		goto out;
1036
1037	inode = filp->f_dentry->d_inode;
1038	if (unlikely(filp->f_op != &mqueue_file_operations))
1039		goto out_fput;
1040	info = MQUEUE_I(inode);
1041
1042	ret = 0;
1043	spin_lock(&info->lock);
1044	if (u_notification == NULL) {
1045		if (info->notify_owner == current->tgid) {
1046			remove_notification(info);
1047			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1048		}
1049	} else if (info->notify_owner != 0) {
1050		ret = -EBUSY;
1051	} else {
1052		switch (notification.sigev_notify) {
1053		case SIGEV_NONE:
1054			info->notify.sigev_notify = SIGEV_NONE;
1055			break;
1056		case SIGEV_THREAD:
1057			info->notify_sock = sock;
1058			info->notify_cookie = nc;
1059			sock = NULL;
1060			nc = NULL;
1061			info->notify.sigev_notify = SIGEV_THREAD;
1062			break;
1063		case SIGEV_SIGNAL:
1064			info->notify.sigev_signo = notification.sigev_signo;
1065			info->notify.sigev_value = notification.sigev_value;
1066			info->notify.sigev_notify = SIGEV_SIGNAL;
1067			break;
1068		}
1069		info->notify_owner = current->tgid;
1070		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1071	}
1072	spin_unlock(&info->lock);
1073out_fput:
1074	fput(filp);
1075out:
1076	if (sock) {
1077		netlink_detachskb(sock, nc);
1078	} else if (nc) {
1079		dev_kfree_skb(nc);
1080	}
1081	return ret;
1082}
1083
1084asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1085			const struct mq_attr __user *u_mqstat,
1086			struct mq_attr __user *u_omqstat)
1087{
1088	int ret;
1089	struct mq_attr mqstat, omqstat;
1090	struct file *filp;
1091	struct inode *inode;
1092	struct mqueue_inode_info *info;
1093
1094	if (u_mqstat != NULL) {
1095		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1096			return -EFAULT;
1097		if (mqstat.mq_flags & (~O_NONBLOCK))
1098			return -EINVAL;
1099	}
1100
1101	ret = -EBADF;
1102	filp = fget(mqdes);
1103	if (!filp)
1104		goto out;
1105
1106	inode = filp->f_dentry->d_inode;
1107	if (unlikely(filp->f_op != &mqueue_file_operations))
1108		goto out_fput;
1109	info = MQUEUE_I(inode);
1110
1111	spin_lock(&info->lock);
1112
1113	omqstat = info->attr;
1114	omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1115	if (u_mqstat) {
1116		if (mqstat.mq_flags & O_NONBLOCK)
1117			filp->f_flags |= O_NONBLOCK;
1118		else
1119			filp->f_flags &= ~O_NONBLOCK;
1120
1121		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1122	}
1123
1124	spin_unlock(&info->lock);
1125
1126	ret = 0;
1127	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1128						sizeof(struct mq_attr)))
1129		ret = -EFAULT;
1130
1131out_fput:
1132	fput(filp);
1133out:
1134	return ret;
1135}
1136
1137static struct inode_operations mqueue_dir_inode_operations = {
1138	.lookup = simple_lookup,
1139	.create = mqueue_create,
1140	.unlink = mqueue_unlink,
1141};
1142
1143static struct file_operations mqueue_file_operations = {
1144	.flush = mqueue_flush_file,
1145	.poll = mqueue_poll_file,
1146	.read = mqueue_read_file,
1147};
1148
1149static struct super_operations mqueue_super_ops = {
1150	.alloc_inode = mqueue_alloc_inode,
1151	.destroy_inode = mqueue_destroy_inode,
1152	.statfs = simple_statfs,
1153	.delete_inode = mqueue_delete_inode,
1154	.drop_inode = generic_delete_inode,
1155};
1156
1157static struct file_system_type mqueue_fs_type = {
1158	.name = "mqueue",
1159	.get_sb = mqueue_get_sb,
1160	.kill_sb = kill_litter_super,
1161};
1162
1163static int msg_max_limit_min = DFLT_MSGMAX;
1164static int msg_max_limit_max = HARD_MSGMAX;
1165
1166static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1167static int msg_maxsize_limit_max = INT_MAX;
1168
1169static ctl_table mq_sysctls[] = {
1170	{
1171		.ctl_name	= CTL_QUEUESMAX,
1172		.procname	= "queues_max",
1173		.data		= &queues_max,
1174		.maxlen		= sizeof(int),
1175		.mode		= 0644,
1176		.proc_handler	= &proc_dointvec,
1177	},
1178	{
1179		.ctl_name	= CTL_MSGMAX,
1180		.procname	= "msg_max",
1181		.data		= &msg_max,
1182		.maxlen		= sizeof(int),
1183		.mode		= 0644,
1184		.proc_handler	= &proc_dointvec_minmax,
1185		.extra1		= &msg_max_limit_min,
1186		.extra2		= &msg_max_limit_max,
1187	},
1188	{
1189		.ctl_name	= CTL_MSGSIZEMAX,
1190		.procname	= "msgsize_max",
1191		.data		= &msgsize_max,
1192		.maxlen		= sizeof(int),
1193		.mode		= 0644,
1194		.proc_handler	= &proc_dointvec_minmax,
1195		.extra1		= &msg_maxsize_limit_min,
1196		.extra2		= &msg_maxsize_limit_max,
1197	},
1198	{ .ctl_name = 0 }
1199};
1200
1201static ctl_table mq_sysctl_dir[] = {
1202	{
1203		.ctl_name	= FS_MQUEUE,
1204		.procname	= "mqueue",
1205		.mode		= 0555,
1206		.child		= mq_sysctls,
1207	},
1208	{ .ctl_name = 0 }
1209};
1210
1211static ctl_table mq_sysctl_root[] = {
1212	{
1213		.ctl_name	= CTL_FS,
1214		.procname	= "fs",
1215		.mode		= 0555,
1216		.child		= mq_sysctl_dir,
1217	},
1218	{ .ctl_name = 0 }
1219};
1220
1221static int __init init_mqueue_fs(void)
1222{
1223	int error;
1224
1225	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1226				sizeof(struct mqueue_inode_info), 0,
1227				SLAB_HWCACHE_ALIGN, init_once, NULL);
1228	if (mqueue_inode_cachep == NULL)
1229		return -ENOMEM;
1230
1231	/* ignore failues - they are not fatal */
1232	mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1233
1234	error = register_filesystem(&mqueue_fs_type);
1235	if (error)
1236		goto out_sysctl;
1237
1238	if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1239		error = PTR_ERR(mqueue_mnt);
1240		goto out_filesystem;
1241	}
1242
1243	/* internal initialization - not common for vfs */
1244	queues_count = 0;
1245	spin_lock_init(&mq_lock);
1246
1247	return 0;
1248
1249out_filesystem:
1250	unregister_filesystem(&mqueue_fs_type);
1251out_sysctl:
1252	if (mq_sysctl_table)
1253		unregister_sysctl_table(mq_sysctl_table);
1254	if (kmem_cache_destroy(mqueue_inode_cachep)) {
1255		printk(KERN_INFO
1256			"mqueue_inode_cache: not all structures were freed\n");
1257	}
1258	return error;
1259}
1260
1261__initcall(init_mqueue_fs);
1262